adbi series on asian economic integration and · pdf filemonetary and currency policy ......

ADBI SERIES ON ASIAN ECONOMIC INTEGRATION AND COOPERATION

Previous titles published in association with the ADBI include:

Infrastructure’s Role in Lowering Asia’s Trade CostsBuilding for TradeEdited by Douglas H. Brooks and David Hummels

Trade Facilitation and Regional Cooperation in AsiaEdited by Douglas H. Brooks and Susan F. Stone

Managing Capital FlowsThe Search for a FrameworkEdited by Masahiro Kawai and Mario B. Lamberte

The Asian TsunamiAid and Reconstruction after a DisasterEdited by Sisira Jayasuriya and Peter McCawley

Asia’s Free Trade AgreementsHow is Business Responding?Edited by Masahiro Kawai and Ganeshan Wignaraja

Monetary and Currency Policy Management in AsiaEdited by Masahiro Kawai, Peter J. Morgan and Shinji Takagi

Implications of the Global Financial Crisis for Financial Reform and Regulation in AsiaEdited by Masahiro Kawai, David G. Mayes and Peter J. Morgan

Infrastructure for Asian ConnectivityEdited by Biswa Nath Bhattacharyay, Masahiro Kawai and Rajat M. Nag

A World Trade Organization for the 21st CenturyThe Asian PerspectiveEdited by Richard Baldwin, Masahiro Kawai and Ganeshan Wignaraja

New Global Economic ArchitectureThe Asian PerspectiveEdited by Masahiro Kawai, Peter J. Morgan and Pradumna B. Rana

Connecting AsiaInfrastructure for Integrating South and Southeast AsiaEdited by Ganeshan Wignaraja, Michael G. Plummer and Peter J. Morgan

The Asian ‘Poverty Miracle’Impressive Accomplishments or Incomplete Achievements?Edited by Jacques Silber and Guanghua Wan

The Asian Development Bank Institute (ADBI), located in Tokyo, is the think tank of the Asian Development Bank (ADB). ADBI’s mission is to identify effective development strate-gies and improve development management in ADB’s developing member countries. ADBI has an extensive network of partners in the Asia and Pacific region and globally. ADBI’s activities are aligned with ADB’s strategic focus, which includes poverty reduction and inclu-sive economic growth, the environment, regional cooperation and integration, infrastructure development, middle-income countries, and private sector development and operations.

M4017 SILBER TEXT.indd iiM4017 SILBER TEXT.indd ii 16/08/2016 16:5016/08/2016 16:50

The Asian ‘Poverty Miracle’Impressive Accomplishments or Incomplete Achievements?

Edited by

Jacques SilberDepartment of Economics, Bar-Ilan University, Israel

Guanghua WanAsian Development Bank Institute, Tokyo, Japan

A JOINT PUBLICATION OF THE ASIAN DEVELOPMENT BANK AND THE ASIAN DEVELOPMENT BANK INSTITUTE

Cheltenham, UK • Northampton, MA, USA

M4017 SILBER TEXT.indd iiiM4017 SILBER TEXT.indd iii 16/08/2016 16:5016/08/2016 16:50

© Asian Development Bank and the Asian Development Bank Institute 2016

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical or photocopying, recording, or otherwise without the prior permission of the publisher.

Publ ished byEdward Elgar Publishing Limited Edward Elgar Publishing, Inc.The Lypiatts William Pratt House15 Lansdown Road 9 Dewey CourtCheltenham NorthamptonGlos GL50 2JA Massachusetts 01060UK USA

The views expressed in this book are those of the authors and do not necessarily refl ect the views and policies of the Asian Development Bank (ADB), its Board of Governors or the governments they represent.

ADB does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequences of their use.

By making any designation of or reference to a particular territory or geographic area, or by using the term “country” in this document, ADB does not intend to make any judgments as to the legal or other status of any territory or area.

ADB encourages printing or copying information exclusively for personal and noncommercial use with proper acknowledgment of ADB. Users are restricted from reselling, redistributing, or creating derivative works for commercial purposes without the express, written consent of ADB.

Asian Development Bank6 ADB Avenue, Mandaluyong City1550 Metro Manila, PhilippinesTel +63 2 632 4444Fax +63 2 636 2444www.adb.org

A catalogue record for this bookis available from the British Library

Library of Congress Control Number: 2016932486

This book is available electronically in the Economics subject collectionDOI 10.4337/9781785369155

ISBN 978 1 78536 914 8 (cased)ISBN 978 1 78536 915 5 (eBook)

Typeset by Servis Filmsetting Ltd, Stockport, Cheshire

M4017 SILBER TEXT.indd ivM4017 SILBER TEXT.indd iv 16/08/2016 16:5016/08/2016 16:50

v

Contents

List of contributors vii

Introduction 1Jacques Silber and Guanghua Wan

PART I IS THERE A CASE FOR A POVERTY LINE SPECIFIC TO ASIA?

1 An Asian poverty line? Issues and options 13 Stephan Klasen

2 A poverty line contingent on reference groups: implications for the extent of poverty in some Asian countries 30

Satya R. Chakravarty, Nachiketa Chattopadhyay and Jacques Silber

PART II POVERTY AND VULNERABILITY IN ASIA

3 Concepts and measurement of vulnerability to poverty and other issues: a review of literature 53

Tomoki Fujii

4 Measuring the impact of vulnerability on the number of poor: a new methodology with empirical illustrations 84

Satya R. Chakravarty, Nachiketa Chattopadhyay, Jacques Silber and Guanghua Wan

5 Climate change and vulnerability to poverty: an empirical investigation in rural Indonesia 118

Tomoki Fujii

M4017 SILBER TEXT.indd vM4017 SILBER TEXT.indd v 16/08/2016 16:5016/08/2016 16:50

vi The Asian ‘poverty miracle’

PART III THE MULTIDIMENSIONALITY OF POVERTY IN ASIA

6 Measuring multidimensional poverty in three Southeast Asian countries using ordinal variables 149

Valérie Bérenger

7 Poverty and nutrition: a case study of rural households in Thailand and Viet Nam 215

Hermann Waibel and Lena Hohfeld

PART IV POVERTY AND INEQUITY

8 Poverty and ethnicity in Asian countries 253 Carlos Gradín

Index 321

M4017 SILBER TEXT.indd viM4017 SILBER TEXT.indd vi 16/08/2016 16:5016/08/2016 16:50

vii

Contributors

Valérie Bérenger is a Professor at the University of Toulon, France. Her research interests include social policies, measurement of poverty and inequalities, assessment of indicators related to human development and gender issues. She has been involved in several scientific projects address-ing the methodological issues and policy- oriented applications of multidi-mensional poverty and pro- poor growth measures.

Satya R. Chakravarty is a Professor of Economics at the Indian Statistical Institute, Kolkata, India. He has published articles in many internationally known journals and edited volumes on welfare issues, cooperative game theory, industrial organization and mathematical finance, and his books have been published by Cambridge University Press, Springer, Anthem Press and Avebury. He is an Associate Editor of Social Choice and Welfare and a member of the Editorial Board of Journal of Economic Inequality. He worked as a consultant to the Asian Development Bank, an external adviser of the World Bank, and also as an adviser of the National Council of Social Policy Evaluation, Mexico. He was awarded the Mahalanobis memorial prize by the Indian Econometric Society in 1994 and is a fellow of the Human Development and Capability Association.

Nachiketa Chattopadhyay is an Associate Professor at the Sampling and Official Statistics Unit of the Indian Statistical Institute, Kolkata, India. He has published articles in many internationally known journals and edited volumes.

Tomoki Fujii is an Associate Professor of Economics at Singapore Management University, Singapore. His main areas of research are devel-opment economics and environmental economics, and his papers have appeared in leading journals in these fields. He has also consulted for leading international development agencies, including the World Bank, World Food Programme and Asian Development Bank, and holds a number of editorial appointments with international journals such as Singapore Economic Review.

Carlos Gradín (PhD in Economics, Universitat Autònoma de Barcelona, 1999) is a Professor of Applied Economics at the University of Vigo in

M4017 SILBER TEXT.indd viiM4017 SILBER TEXT.indd vii 16/08/2016 16:5016/08/2016 16:50

viii The Asian ‘poverty miracle’

Galicia, Spain and a member of the EQUALITAS network of researchers on income distribution. His main research interests are distributive issues such as poverty, inequality, mobility or polarization, as well as gender and ethnic economics. His research has been published in journals such as the Journal of Economic Inequality, Review of Income and Wealth, Review of Household Economics, Journal of Development Studies, Regional Studies, Journal of African Economics, and Industrial Relations, among others.

Lena Hohfeld is an evaluator at DEval, the German Institute for Development Evaluation, Bonn, Germany. A development economist interested in issues of vulnerability to poverty, food security, gender and migration in Southeast Asia, she has work experience as researcher and lecturer at the Institute of Development and Agricultural Economics, Leibniz University of Hanover and as consultant for the World Bank in Myanmar and Thailand. She holds a Diploma and a PhD in economics from the University of Hanover, Germany.

Stephan Klasen is a Professor of Development Economics at the University of Göttingen in Germany. He holds a PhD from Harvard University, USA and has since held positions at the World Bank, King’s College, Cambridge, UK and the University of Munich. His research focuses on issues of poverty, inequality and gender in developing countries. He is a member of the United Nations Committee on Development Policy and has been a member of the Intergovernmental Panel on Climate Change for the 5th Assessment Report.

Jacques Silber (PhD, University of Chicago, 1975), Professor Emeritus of Economics at Bar- Ilan University, Israel, a specialist in the measurement of income inequality, poverty and segregation, is the author of more than 100 scientific papers and of several books, among which is the Handbook on Income Inequality Measurement. He was the founding Editor of the Journal of Economic Inequality, is the Editor of the book series Economic Studies in Inequality, Social Exclusion and Well- Being (Springer) and edited several special issues of academic journals. He was President of ECINEQ, the (International) Society for the Study of Economic Inequality, between 2011 and 2013.

Hermann Waibel is a Professor of Agricultural Economics and Director of the Institute of Development and Agricultural Economics, School of Economics and Management, Leibniz University of Hanover, Germany. He has 30 years of experience in research and development in Southeast Asia. Formerly Associate Professor with the Asian Institute of Technology (AIT) in Bangkok he holds a PhD in agricultural economics from the

M4017 SILBER TEXT.indd viiiM4017 SILBER TEXT.indd viii 16/08/2016 16:5016/08/2016 16:50

Contributors ix

University of Hohenheim, Germany. His current research focus is on vul-nerability to poverty and rural development in Asia.

Guanghua Wan is the Director of Research at the Asian Development Bank Institute (ADBI), Tokyo, Japan. Previously, he was Principal Economist and Head of the Poverty/Inequality Group, Asian Development Bank (ADB). Prior to ADB/ADBI, he was a senior economist at the United Nations and taught in a number of universities in Australia and the People’s Republic of China. Trained in development economics and econo-metrics, he is a multi- award- winning scholar on the Chinese economy and an expert on Asia, with an outstanding publication record of more than 100 professional articles and a dozen books including two by Oxford University Press. An honorary professor of over ten top institutions in the People’s Republic of China, including Fudan and Zhejiang universities, he is among the top 8 percent of economists globally and top 4 percent in Asia according to the latest ranking of REPEC.

M4017 SILBER TEXT.indd ixM4017 SILBER TEXT.indd ix 16/08/2016 16:5016/08/2016 16:50

M4017 SILBER TEXT.indd xM4017 SILBER TEXT.indd x 16/08/2016 16:5016/08/2016 16:50

1

IntroductionJacques Silber and Guanghua Wan

As a consequence of the rapid economic growth in recent decades, Asia and the Pacific have experienced an impressive reduction in extreme poverty, when measured at the conventional $1.25/day/person poverty line. Whereas in 1981, 1.59 billion Asians were poor (corresponding to a poverty rate of 69.8 percent), in 1990 the number of poor in Asia had fallen to 1.48 billion (a 54.7 percent poverty rate). In fact, by 2005, Asia had suc-ceeded in halving its extreme poverty because its 26.9 percent poverty rate was already less than half the 1990 level and, by 2010, the extreme poverty rate reduced further to 20.7 percent.1

The early attainment, in the world as a whole, of the Millennium Development Goals (MDGs) (halving extreme poverty globally) would certainly not have been possible without Asia. Although, for the develop-ing world as a whole, the poverty rate fell from 43.1 percent in 1990 to 25.1 percent in 2005 and to 20.6 percent in 2010, thus meeting the MDG global target, when excluding Asia the extreme poverty rate would have been 24.9 percent in 1990 and 20.5 percent in 2010. Clearly, the rest of the developing world would not have been able to halve its 1990 poverty rate before, say, 2030.

In absolute terms, between 1990 and 2010, the number of extremely poor declined by 745.42 million in Asia, but only by 693.47 million glo-bally. Thus, the number of extremely poor actually increased in the rest of the developing world by 51.95 million, partly because of population growth. Similar conclusions may be drawn when adopting a more ‘moder-ate’ poverty line of $2 per day per person. The number of moderately poor in Asia declined, between 1990 and 2010, by 566.31 million but, during the same period, it increased by 97.73 million in other regions of the world.

The drop in extreme poverty was, however, not uniform across Asia. It has been especially impressive in East Asia because, over a 20- year period, extreme poverty in East Asia fell from about 60.2 percent in 1990 to 11.6 percent in 2010 – with the People’s Republic of China (the PRC) reducing the number of extremely poor by 527.64 million. The poverty reduction was also quite important in Central and West Asia (39.4 percent)

M4017 SILBER TEXT.indd 1M4017 SILBER TEXT.indd 1 16/08/2016 16:5016/08/2016 16:50

2 The Asian ‘poverty miracle’

and Southeast Asia (31.0 percent). In South Asia, the reduction was 19.8 percent and, in the Pacific, it was 10.9 percent. Note also that India, the second most populous country, reduced the number of ‘extremely poor’ by 48.26 million during this period.

This remarkable decrease in poverty in Asia, when poverty is measured on the basis of $1.25, or even $2, per day poverty line, should, however, not imply that poverty is no longer an important issue in Asia.2 This book aims at showing that, as impressive as Asia’s accomplishments in poverty reduc-tion are, the achievements are still incomplete. A more objective picture of the extent of poverty in Asia would show that several issues remain critical. At least four elements have to be taken into consideration.

First, it is likely that the $1.25 poverty line is not adequate for Asia- Pacific, mainly because it underestimates the costs for the poor to maintain a minimal standard of living. Remember here that the $1.25 poverty line was derived from the world’s 15 poorest countries, only two of them being Asian countries. In addition, this $1.25 poverty line was based on 1988–2005 consumption data. Because consumption patterns are not uniform across regions and change over time, there is a case for refining and updating the poverty line accordingly. In fact, several Asian countries (including the PRC and India) recently raised their national poverty lines to make them more relevant for policy making.

There is an additional reason for adjusting the traditional $1.25 poverty line. In the past two decades, numerous studies have shown that indi-vidual well- being depends on relative as well as absolute income. There is, thus, a case for taking relative income into account when determining a poverty line because individual life- satisfaction is not independent of the standard of living of some relevant ‘reference group’. Such an idea had already been suggested by Duesenberry (1949) who argued that the utility of an individual was negatively affected by the income of anyone with a higher income. It is difficult to identify the relevant reference group. Some researchers, for example, have considered colleagues at work (Senik 2009). Ferrer- i- Carbonell (2005) assumed that the reference group consisted of people with the same characteristics such as age, level of education and region of residence. Others have used space- based reference incomes such as the average income of individuals of the same race in the cluster and district where the individuals surveyed live (Kingdon and Knight 2007).

The need to modify the $1.25 poverty line to adapt it to the new stand-ards of living in Asia, and the idea of taking relative income into account when determining a poverty line, is the object of the two chapters that form Part I of this book.

A second issue, which is important for poverty targeting, is that of vul-nerability, whether it is vulnerability to natural disasters, to the increasing


Introduction 3

impact of climate change, to economic crises or other types of shocks. Following unexpected shocks, such as earthquakes, extreme weather events, job losses, or illnesses, it is not uncommon to observe that many individu-als, whose standards of living before the shock classified them as non- poor, end up with standards of living below the poverty line. In recent years, vulnerability to natural calamities has been increasing in both frequency and severity, especially in East, South, and Southeast Asia. In fact, seven of the world’s ten most vulnerable countries are in Asia. Many low- income households live just above extreme poverty and, as a consequence, can easily fall back into poverty following an unexpected shock. Conventional poverty assessments tend, however, to ignore these vulnerabilities. In Asia, where formal insurance is not common, poverty reduction policies should thus take into account such vulnerabilities. Part II of this book includes three chapters that deal with various features of vulnerability in Asia, with special emphasis being put on approaches that could allow incorporating shocks or risks into setting poverty lines and estimating ‘vulnerability’ to poverty in Asia.

In Part III of this book, an attempt is made to estimate the extent of poverty, not on the basis of a traditional money- metric poverty line, but by taking into account the fact that well- being is multifaceted so that poverty is intrinsically multidimensional. Many studies have indeed shown that money- metric measures do not provide a satisfactory picture of well- being for either individuals or households and that other dimensions should be taken into account (see Baulch and Masset 2003; McKay and Lawson 2003; Carter and Barrett 2006; Hulme and McKay 2007). Money does not mean much when there is market failure or markets do not exist.

Several studies have shown that the correlation between monetary income and other dimensions of human well- being is often quite low (Baulch and Masset 2003; McKay and Lawson 2003; Günther and Klasen 2009). Non- monetary poverty is generally more persistent than monetary poverty. For example, once a child is stunted, the impact of such malnu-trition is often irreversible, even when the income status of the individual improves. Similar conclusions may be drawn with respect to education because most school dropouts remain poor in terms of human capital, even if some become richer later in life (Baulch and Masset 2003; Stifel et al. 1999).

Taking a multidimensional approach to poverty is, however, not a simple task (see Kakwani and Silber 2007, 2008a, 2008b) because, among other reasons, it implies selecting poverty indicators and their weights. The Multidimensional Poverty Index (MPI) published by the United Nations Development Programme (UNDP) is an interesting attempt to create this type of comparable poverty measure. It uses a so- called ‘dual cutoff’



method (Alkire and Foster 2011), where the first cutoff defines whether a household is deprived in a particular dimension and the second determines whether a household has passed the threshold of deprivation that defines it as multidimensionally poor. Many details remain to be worked out (Dotter and Klasen 2014) but computing an Asia- specific version of an MPI could be considered, although this will not be a simple task given Asia’s great heterogeneity. The two chapters of Part III of this book attempt to present illustrations of the relevance of multidimensional poverty for Asia.

Given the unprecedented reduction in poverty levels in Asia in recent decades, it is also important to investigate whether the benefits of this higher well- being have reached the entire population or whether differ-ences in economic opportunities are related to characteristics such as gender, race or ethnicity. This question is particularly relevant in Asia, given its large ethnic diversity. We may, thus, expect disadvantaged groups to be overrepresented among population subgroups that live in remote areas, were historically denied access to proper education and basic infra-structure, or suffered from segregation and wage discrimination in the labor market. Attempting to estimate the extent of the ethnic differential in poverty is, hence, a crucial issue that should be of utmost interest to policy makers who wish to decrease such gaps. This topic is covered in Part IV of this book. This part consists of only one chapter that deals with poverty and ethnicity in Asian countries.

OUTLINE OF INDIVIDUAL CHAPTERS

In the first chapter, which is titled ‘An Asian poverty line? Issues and options’, Stephan Klasen wonders, in view of Asia’s particular economic situation, whether there is merit in developing an Asia- specific poverty line that would address some of the shortcomings of the $1.25 per day poverty line. The author reviews various ways of determining an Asia- specific poverty line, including an Asia- specific international income poverty line (using purchasing power parity, PPP, adjusted dollars) that would be derived from Asian national poverty lines. He suggests grounding such an Asian- specific poverty line in a consistent method of generating national poverty lines using national currencies rather than a PPP- adjusted poverty line in international dollars that would be specific to Asia. Klasen thinks that it is also important for such a poverty line to take into account relative poverty to reflect the rising aspirations of Asian societies. Finally, Klasen discusses the possibility of developing an Asia- specific multidimensional poverty index that would take into account the specific living conditions of Asian societies.


Introduction 5

The second chapter is titled ‘A poverty line contingent on reference groups: implications for the extent of poverty in some Asian coun-tries’. Its authors, Satya R. Chakravarty, Nachiketa Chattopadhyay and Jacques Silber, start by defining what they call an ‘amalgam poverty line’, that is, a poverty line that is a weighted average of an absolute poverty line (such as $1.25 per day) and a reference income (such as the mean or the median income). They then compute the number of poor in various coun-tries and subnational regions in Asia, on the basis of such an ‘amalgam poverty line’. They examine various scenarios, depending on the relative weight of the absolute poverty line and of the reference income. They also analyze the impact on poverty of selecting an absolute poverty line of $1.45 (rather than $1.25) per day, a threshold that seems more adapted to the Asian case. The chapter provides, for the countries and subnational regions involved and under the various scenarios examined, estimates of the headcount ratio, together with the income poverty gap and the average income of the poor.

The next three chapters, in Part II of the book, examine different aspects of vulnerability in Asia. In the chapter titled ‘Concepts and measurement of vulnerability to poverty and other issues: a review of literature’ Tomoki Fujii reviews the growing body of literature on vulnerability. He first sum-marizes various concepts of vulnerability that have appeared in the litera-ture, namely, the welfarist, expected poverty and axiomatic approaches. He then reviews a number of empirical studies on vulnerability to poverty, mainly in Asia and, in particular, in the PRC. Finally, the author briefly reviews other areas of vulnerability analysis, such as vulnerability to climate change, to nonmonetary outcomes (for example, nutritional out-comes). The chapter ends with a discussion of some policy implications of vulnerability analysis.

In a chapter titled ‘Measuring the impact of vulnerability on the number of poor: a new methodology with empirical illustrations’ Satya R. Chakravarty, Nachiketa Chattopadhyay, Jacques Silber and Guanghua Wan examine the possibility of defining a poverty line that would also depend on the extent of vulnerability. The basic idea is to adjust the poverty line, in the presence of vulnerability, in such a way that the utility of a person at the current poverty line and that at the adjusted poverty line are identical. Using an additive model of vulnerability, it is shown that if the utility function is assumed to have the property of constant absolute risk aversion, then the adjusted poverty line becomes a simple absolute augmentation of the current poverty line. On the other hand, under a multiplicative model of vulnerability, assuming constant relative risk aversion, the revised poverty line is a simple relative augmen-tation of the current poverty line. Given that uncertainty plays a central



role, a section of the chapter is devoted to the estimation of the variance of the noise which characterizes uncertain income. The chapter ends with a detailed empirical illustration covering many Asian countries.

In the final chapter in Part II, titled ‘Climate change and vulnerability to poverty: an empirical investigation in rural Indonesia’, Tomoki Fujii takes a closer look at the impact of climate change on vulnerability. More pre-cisely, given that that anthropogenic climate change may lead to increased surface temperature, sea- level rise, more frequent and significant extreme weather and climate events, among others, the author investigates how climate change can potentially affect vulnerability to poverty. His analysis is based on panel data from Indonesia and the focus of his study is on the effects of droughts and floods, two of the commonly observed disasters there. Fujii’s simulation results indicate that vulnerability to poverty may increase substantially as a result of climate change in Indonesia.

The following two chapters, in Part III of the book, deal with the need to take a multidimensional approach to poverty in Asia. In the chapter titled ‘Measuring multidimensional poverty in three Southeast Asian countries using ordinal variables’ Valérie Bérenger looks at the evolution of poverty in Cambodia, Indonesia and the Philippines. Her approach is based on recent methodological refinements of poverty measurement that are based on counting approaches using ordinal variables. Her chapter compares mul-tidimensional poverty measures such as the MPI proposed by Alkire and Foster (2011) with other measures, such as those proposed by Chakravarty and D’Ambrosio (2006), Rippin (2010) and those based on the exten-sion of the Aaberge and Peluso (2012) approach, suggested by Silber and Yalonetzky (2013) – approaches that have the advantage of being sensitive to the distribution of deprivation counts across individuals. The empirical illustrations of Bérenger are based on the Demographic and Health Surveys for the years 2000, 2005 and 2010 for Cambodia, 1997, 2003 and 2007 for Indonesia, and 1997, 2003 and 2008 for the Philippines. The emphasis of her analysis is on deprivations in education, health and standard of living.

The second chapter covering the topic of multidimensional poverty, written by Hermann Waibel and Lena Hohfeld, is titled ‘Poverty and nutrition: a case study of rural households in Thailand and Viet Nam’. In this chapter, the authors analyze the link between nutrition and poverty in Thailand and Viet Nam, where monetary- based poverty reduction was especially successful. They are two emerging market economies where poverty rates are now below 10 percent and are declining further. Despite this impressive decline in income poverty, the authors wonder whether this success was translated into similar improvements in the nutritional status of the people and especially that of children. They conclude that malnutrition continues to be a problem in Viet Nam and Thailand where


Introduction 7

child underweight rates are 27 percent and 19 percent, respectively, and, are much higher than the traditional headcount ratios. The authors inves-tigated factors that influence nutrition outcomes and found that, although poverty and income have an impact on nutrition outcomes, other factors such as mother’s height, education, migration and sanitation also have important influences. The authors conclude that, even under the assump-tion of high growth, income growth alone will not be able to reduce mal-nutrition in the near future.

The final chapter of this book, in Part IV, deals with the issue of poverty and what is often called ‘horizontal inequities’. The chapter, written by Carlos Gradín, is titled ‘Poverty and ethnicity in Asian countries’. Based on data from Health and Demographic Surveys, it compares the extent and nature of the higher prevalence of poverty among disadvantaged ethnic groups in several Asian countries. The author first estimates a composite wealth index which serves as a proxy for economic status. He then analyzes the magnitude of the ethnic gap in absolute and relative poverty levels across countries and ethnicities. Then, using regression- based counterfac-tual analysis, he compares the actual differential in poverty with the gap that remains after disadvantaged ethnic groups are given the distribution of characteristics of the advantaged ethnic groups. The results of this empiri-cal investigation show that there is substantial cross- country variability in the extent, evolution and nature of the ethnic poverty gap, which can be as high as 50 percentage points or more in Nepal, Pakistan and India. Ethnic disadvantaged groups are poorer because of the strongly persisting high inequalities in education (for example, in India, Nepal and Pakistan), differences in regional development (for example, in the Philippines) or a persistent large urban- rural gap (for example, in Pakistan).

In an interview given for the magazine, Prospect, on 18 August 2013, Nobel Prize winner Amartya Sen stressed the following points:

[I]n order to judge how a country is doing you can’t just talk about per capita income. India used to be 50 per cent richer than Bangladesh in per capita income terms but is now 100 per cent richer. Yet, in the same period . . . when, in the early 1990s, India was three years ahead of Bangladesh in life expectancy, it is now three or four years behind. In India it is 65 or 66, in Bangladesh it’s 69. Similarly, immunization: India is 72 per cent, Bangladesh is more like 95 per cent. Similarly, for the ratio of girls to boys in school. So in all these respects, we are looking at capability. We’re looking at the capability to lead a healthy life, an educated life, to lead a secure life (with immunization making people immune to some preventable illnesses), having the capability to read and write, for girls as well as boys.

Expanding and safeguarding human capability is central to thinking about policy making. . . . human capability is not only important in itself,



but that human capability expansion is also a kind of classic Asian way of having sustained economic growth. It started in Japan, just after the Meiji restoration, where the Japanese said: ‘We Japanese are no different from the Europeans or the Americans; the only reason we’re behind is that they are educated and we are not.’ They then had this dramatic expansion in universal education and then, later, widespread enhancement of health-care. They found that a healthy, educated population served the purpose of economic growth very well.3

This long citation seems to justify, in a certain way, the title we chose for this book, because clearly, as far as poverty reduction is concerned, the accomplishments of most Asian countries are impressive but they certainly remain incomplete.

NOTES

1. See ADB (2014) for additional data on poverty reduction in Asia in recent decades.2. See ADB (2014) for more details.3. The full transcript of Jonathan Derbyshire’s interview with renowned Nobel Prize-

winning economist, Amartya Sen, Prospect Magazine, August issue, 2013. http://www.prospectmagazine.co.uk/magazine/prospect- interviews- amartya- sen- the- full- transcript- jonathan- derbyshire.

REFERENCES

Aaberge, R., and E. Peluso (2012), ‘A counting approach for measuring multi-dimensional deprivation’, Discussion Paper No. 700, Research Department, Statistics Norway.

Alkire, S., and J. Foster (2011), ‘Counting and Multidimensional Poverty Measurement’, Journal of Public Economics, 95 (7–8), 476–87.

Asian Development Bank (ADB) (2014), Key Indicators for Asia and the Pacific 2014, Manila: Asian Development Bank.

Baulch, B. and E. Masset (2003), ‘Do monetary and nonmonetary indicators tell the same story about chronic poverty? A study of Vietnam in the 1990s’, World Development, 31 (3), 441–53.

Carter, M. and C. Barrett (2006), ‘The economics of poverty traps and persist-ent poverty: an asset- based approach’, Journal of Development Studies, 42 (2): 178–99.

Chakravarty, S. and C. D’Ambrosio (2006), ‘The measurement of social exclusion’, Review of Income and Wealth, 52 (3), 377–98.

Derbyshire, J. (2013), ‘Prospect interviews Amartya Sen’, Prospect Magazine, 18 July.

Dotter, C. and S. Klasen (2014), ‘The Multidimensional Poverty Index: achieve-ments, conceptual and empirical issues’, UNDP Human Development Report Office, Occasional Paper, December.


Introduction 9

Duesenberry, J. (1949), Income, Savings, and the Theory of Consumer Behavior, Cambridge, MA: Harvard University Press.

Ferrer- i- Carbonell, A. (2005), ‘Income and well- being: an empirical analysis of the comparison income effect’, Journal of Public Economics, 86 (1), 43–60.

Günther, I. and S. Klasen (2009), ‘Measuring chronic non- income poverty’, in T. Addison, D. Hulme, and R. Kanbur (eds), Poverty Dynamics: Interdisciplinary Perspectives, Oxford: Oxford University Press.

Hulme, D. and A. McKay (2007), ‘Identifying and measuring chronic poverty: beyond monetary measures?’, in N. Kakwani and J. Silber (eds), The Many Dimensions of Poverty, New York: Palgrave Macmillan.

Kakwani, N. and J. Silber (eds) (2007), The Many Dimensions of Poverty, New York: Palgrave Macmillan.

Kakwani, N. and J. Silber (eds) (2008a), Quantitative Approaches to Multidimensional Poverty Measurement, New York: Palgrave Macmillan.

Kakwani, N. and J. Silber (eds) (2008b), ‘Introduction to the special issue on mul-tidimensional poverty analysis: conceptual issues, empirical illustrations and policy implications’, World Development, 36 (6), 987–91.

Kingdon, G. and J. Knight (2007), ‘Community, comparisons and subjective well- being in a divided society’, Journal of Economic Behavior and Organization, 64 (1), 69–90.

McKay, A. and D. Lawson (2003), ‘Assessing the extent and nature of chronic poverty in low income countries: issues and evidence’, World Development, 31 (3), 425–39.

Rippin, N. (2010), ‘Poverty severity in a multidimensional framework: the issue of inequality between dimensions’, Discussion Paper No. 47, Courant Research Centre, Georg- August- University Göttingen.

Senik, C. (2009), Income Distribution and Subjective Happiness: A Survey, Paris: OECD.

Silber, J. and G. Yalonetzky (2013), ‘Measuring Multidimensional Deprivation with Dichotomized and Ordinal Variables’, in G. Betti, and A. Lemmi (eds), Poverty and Social Exclusion: New Methods of Analysis. Routledge Frontiers of Political Economy, London and New York: Routledge.

Stiffel, D., D. Sahn and S. Younger (1999), ‘Inter- temporal changes in welfare: preliminary results for nine African countries’, CFNPP Working Paper No. 94, Cornell University, Ithaca, NY.


PART I

Is There a Case for a Poverty Line Specific to Asia?


13

1. An Asian poverty line? Issues and optionsStephan Klasen1

1 INTRODUCTION

Using the widely used international income poverty line ($1.25 per person per day), poverty in Asia has fallen dramatically in recent decades. In fact, the very rapid progress on absolute income poverty reduction in Asia is largely responsible that the first Millennium Development Goal (MDG) aiming to halve the incidence of absolute poverty between 1990 and 2015 has been reached four years ahead of schedule. This was achieved by par-ticularly rapid progress in many populous Asian economies (particularly, the People’s Republic of China, Indonesia, and Viet Nam), overcom-pensating for much slower progress in poverty reduction in sub- Saharan Africa (Chen and Ravallion 2013).

Despite this progress, it is too early to declare victory on the poverty front in Asia for various reasons. First, progress in poverty reduction remains fragile in many Asian countries and the vulnerability to poverty remains high (Klasen and Waibel 2013, 2014). Second, there is the recogni-tion that poverty captures more than a lack of incomes, an issue covered by the literature on multidimensional poverty (for example, Rippin 2013; Alkire and Santos 2014). Progress in reducing multidimensional poverty in Asia has generally been more uneven (although there are substantial uncer-tainties about the data, particularly comparable data over time). Finally, in many Asian countries national poverty lines are substantially higher than the international $1.25- a- day poverty line; in some, including the People’s Republic of China (PRC) and India, they have been revised upwards to also reflect the rising aspirations of the populations in these societies. At these higher (and increasing) poverty lines, poverty is far from defeated.

Partly as a result of these factors, the Asian Development Bank (ADB) is considering whether there is merit in developing an Asia- specific poverty line. In addition, it is considering ways to derive such an Asian poverty line, closely related to the methods developed and applied by the World Bank (Ravallion et al. 2009) in deriving the international $1.25- a- day



poverty line but specific for the Asian situation. This chapter first dis-cusses whether there is merit in developing an Asia- specific poverty line. We then discuss various options of developing such a poverty line, con-sidering income and multidimensional versions of such a poverty line. We argue that there can be some merit in developing an Asian poverty line and that, in the case of income poverty, it would be best to ground such an Asian- specific poverty line in a consistent method of generating national poverty lines using national currencies rather than generating a purchasing power parity (PPP) adjusted poverty line in international dollars (see also Klasen 2013a; Klasen et al. 2015). It is important that such a poverty line also considers relative poverty in its assessment to reflect the rising aspirations of Asian societies (see Ravallion and Chen 2011; Chen and Ravallion 2013). In terms of multidimensional poverty lines, there is merit in developing an Asia- specific multidimensional poverty index (MPI) that takes into account the specific living conditions of Asian societies.

2 ADVANTAGES AND DISADVANTAGES OF AN ‘ASIAN’ POVERTY LINE

Before discussing options to derive a poverty line for Asia, it is important to first discuss whether it is useful to develop such a line to begin with. We consider four possible arguments for an Asia- specific poverty line. First, it could be argued that conditions in Asia are so different from other parts of the world that it justifies a different poverty line, in the sense that it would reflect these particular circumstances. For example, households tend to be smaller than in Africa, family ties are quite strong, and the provision of public services by the state is substantial. To the extent that this is the case, it might justify a lower poverty line, measured in terms of private per capita incomes because fewer private incomes are required to achieve a certain level of well- being. However, it is not obvious that these apparent differences justify a peculiar Asian poverty line because the heterogeneity within Asia in these economic and social arrangements is very large. Also, it would first be necessary to investigate the empirical importance of these claims and their relevance to particular Asian countries before drawing any firm conclusions on this. It should also logically lead to different poverty lines within Asia, depending on the particular circumstances. It would thus be particularly difficult to use this argument as a motivation for a uniform income poverty line appropriate for all of Asia.

A second argument relates to differences in levels and trends of eco-nomic performance that ought to be reflected in the setting of a poverty


An Asian poverty line? 15

line. Average incomes in Asia are higher than in Africa (but lower than in Latin America and the Middle East) and most economies in Asia have experienced rapid growth in the past three decades. This might justify the use of an Asian poverty line that reflects the average income level and, more importantly, reflects its rapid economic performance. We argue below that such a poverty line should contain a relative element, that is, should increase with rising prosperity in Asia. The high heterogeneity in Asia’s income levels and economic growth experience might be seen as a counterargument to a single and uniquely Asian income poverty line, but to the extent that neighboring countries benchmark their performance against each other, an argument for a unique line reflecting these special features can be made.

A third argument is that an Asian poverty line would be more closely aligned with national poverty lines in Asia and, thus, the disconnect between national and international poverty measurement would be cor-respondingly smaller (see Klasen 2013a; Dotter 2014). This is essentially an empirical question. Clearly, the current $1.25 international poverty line is only very loosely linked to Asian realities. The only Asian country included in the sample of the 15 poorest countries that were used to derive the $1.25 the poverty line is Tajikistan (Ravallion et al. 2009). In fact, it can be argued that the $1.25 poverty line is much more a reflection of national poverty realities in Africa than in Asia (which includes only South, East, and Southeast Asia; Central Asia is included with Europe here). This can also be seen in Figure 1.1. This figure shows the differ-ence in the poverty headcount using the national poverty line minus the headcount using the international poverty line. A negative number means that the international poverty line is higher than the national poverty line. This is the case in a number of Asian countries, including large ones (the size of the bubble indicates the number of poor people there). For these countries, the international poverty line appears high. In Europe and Central Asia, the international poverty line is far too low. An extreme case is Tajikistan where the poverty headcount using the national poverty line is 40 percentage points higher than when using the international poverty line (even though this country is included in the sample used to calculate the international poverty line). Thus, the mismatch between Asian national poverty lines and the international poverty line appears substantial, which would argue for an Asia- specific poverty line. If an ‘Asian’ poverty line was directly grounded in country- specific poverty lines (in contrast to the international poverty line that is an average of poverty lines across the world), the linkage between national and pan- Asian poverty measurement could be even closer. This is an issue we discuss below.



A fourth argument relates to a multidimensional poverty measure. The most prominent internationally comparable multidimensional poverty measure, UNDP’s MPI (see UNDP 2010, ch. 5; Alkire and Santos 2014), uses the same indicators and cutoffs across the entire developing world. Owing to differences in climate, economic and social arrangements, social preferences, and the nature and state of public services, it might be argued that an Asian MPI should reflect this in terms of indicators and cutoffs. For example, it could be argued that the role of education as key to per-sonal advancement is seen as particularly important in Asian societies, and an MPI should reflect this by giving education more weight and pos-sibly argue for a higher cutoff. As before, the heterogeneity within Asia is a problem for this line of reasoning. When considering multidimensional poverty measures below, we revisit this issue again.

0

–50 0 50

200

400

600 Asia

0

–50 0 50

200

400

600 MENA

0

–50 0 50

200

400

600 SSA

0

–50 0 50

200

400

600 Europe

0

–50 0 50

200

400

600 LAC

Difference in poverty headcount (NPL – IPL)

Mea

n co

nsum

ptio

n fr

om N

A

Notes: NPL refers to national poverty lines; IPL to the international poverty line ($1.25 a day); the y- axis measures mean consumption per capita per month, taken from the national accounts (NA). Europe includes Eastern European and Central Asian countries; LAC refers to Latin American and Caribbean countries; and MENA refers to countries in the Middle East and North Africa; where these regions refer to World Bank regional classifications.

Source: Dotter (2014).

Figure 1.1 Mismatch in poverty headcount by region using national and international poverty lines



There are also disadvantages to generating a continent- specific poverty line. The most important is that comparisons across continents are diffi-cult and lack transparency both in terms of levels and trends.

To conclude this section, it is not obvious that a specific Asian poverty line is desirable. The most compelling arguments are that it could reflect income levels and faster economic progress better than could a global measure; that it can be linked more closely to national poverty lines in Asia; and that it might reflect uniquely Asian conditions and settings in a multidimensional measure. However, there are costs to it and this suggests that a global measure should not be dropped for an Asian measure, but only treat an Asian poverty line as complementary to a global assessment.

3 OPTIONS TO CONSTRUCT AN ASIAN POVERTY LINE

There are different options to generate an Asia- specific poverty line. First, it is necessary to distinguish between an income and a multidimensional poverty line. When constructing an income poverty line, we consider three options. The first is to mimic the estimation method of the World Bank of generating the $1.25- a- day poverty line, but only using Asian countries in the estimation; the second option uses the same set of countries to produce a ‘weakly relative’ poverty line (Ravallion and Chen 2011; Chen and Ravallion 2013); and the third option grounds an Asian poverty line in national poverty measurement (see Klasen 2013a, 2013b). Thus, together with an Asia- specific multidimensional poverty line, altogether four options are considered.

3.1 An Absolute Income Poverty Line Using the World Bank’s Methods

The World Bank has been generating an international poverty line since 1990 (World Bank 1990; Ravallion et al. 1991). In 1990, it stood at $1.02 in 1985 PPP- adjusted dollars; in 2000, it was adjusted to $1.08 in 1993 PPP- adjusted dollars (World Bank 2000; Chen and Ravallion 2001); and, in 2008, it was adjusted to $1.25 in 2005 PPP- adjusted dollars (Ravallion et al. 2009). Now, the World Bank is working on adjusting its estimated poverty line using 2011 PPP- adjusted dollars.

The methods of deriving the international poverty line have essentially been the same (although differing in some details of data used) and we focus on the latest completed revision done in 2008. Ravallion et al. (2009) explain how the World Bank derives the international income poverty line using the following steps. First, available national poverty lines for



74 developing countries are translated into poverty lines expressed in PPP- adjusted international dollars in 2005 prices. Then, these national poverty lines (expressed in 2005 international dollars) are lined up against the log of consumption per capita (in 2005 international dollars) of the 74 countries (see Figure 1.2(a)). They then observe that, below a certain

4.0 4.5 5.0 5.5 6.0

0

50

100

150

Log PCE

Nat

iona

l pov

erty

line

Armenia

Azerbaijan

Bangladesh

Cambodia

PRC

Georgia

IndiaIndonesia

Jordan

Kazakhstan

Kyrgyz Republic Mongolia

Nepal

PakistanPhilippines Sri Lanka

Tajikistan Thailand

Turkey

Viet Nam

3.5 4.0 4.5 5.0 5.5 6.0 6.5

50

100

150

200

250

log (C)

Paraguay Paraguay

Mauritius Mauritius

Pov

erty

line

($

per

mon

th)

Yemen

100 200 300 400 500 600

50

100

150

200

250

C

Pov

erty

line

($

per

mon

th)

(a) Threshold model for log(consumption) (b) Threshold model for consumption

(c) National poverty lines for 21 Asian countries plotted against mean consumption in international 2005 PPP

Note: PCE refers to per capita expenditures; all figures refer to dollars per capita per month. PRC 5 the People’s Republic of China.

Sources: (a) and (b): analysis based on Klasen et al. (2015), data from Ravallion et al. (2009); (c) elaboration of ADB (2014: 8), based on data from Ravallion et al. (2009).

Figure 1.2 National poverty lines for 74 developing countries plotted against mean consumption in international 2005 PPP



threshold, the relationship of per capita consumption, national poverty lines are very similar, indicated by the flat portion of the curve below, whereas above a threshold they rise, less than proportionately, with mean incomes. The $1.25- a- day poverty line is then simply the average poverty line of the flat portion.

For an Asian poverty line, simply limit the sample to 21 national poverty lines from Asian countries, as done by ADB (2014) in an illustrative exer-cise reproduced in Figure 1.2(c). As can be seen, also here the relationship between log per capita consumption and the national poverty lines is also nonlinear, with a flat portion below a certain threshold level of per capita incomes and an increasing portion above that. The average poverty line of the flat portion in Asia extends substantially further to the right. Whereas, in the global sample, only two Asian economies are in included in the flat portion which constitutes the reference line for the poverty line, in the Asian sample, nine countries are included in the reference group. This also leads to a substantially higher ‘Asian’ poverty line of around $1.51. However, this difference between the international and the Asian poverty line is actu-ally due to differences in the estimation method between Ravallion et al. (2009) and the estimate produced by ADB (2014). Whereas the former estimate the relationship between the level of the national poverty line and per capita consumption from the national accounts (that is, a linear model, in effect modeling the relationship in Figure 1.2(b)), ADB (2014) estimates the relationship between national poverty lines and the log of per capita consumption (that is, a log- linear model). As shown by Greb et al. (2011), using a log- linear model also leads to a larger reference group in the global model and a global poverty line of $1.45 per day. Thus, using a sample of Asian countries does not lead to a different poverty line from using the global sample if the same estimation method is used. Nevertheless, the question arises which estimation method is to be preferred.

However, both the estimations by Ravallion et al. (2009) and by ADB (2014) are problematic from a statistical point of view. In particular, in the linear model used by Ravallion et al. (2009), there is no statistical evidence of a kink in the curve so that the kink is imposed on the data rather than observed (which is visible from Figure 1.2(b), see Greb et al. 2011; Klasen et al. 2015). In the log- linear model, the residuals are not normally distrib-uted so that the inference, especially regarding location and significance of the kink which separates the flat from the rising portion, is problematic. As shown by Klasen et al. (2015), the preferred statistical specification is actually a log- log model where there is statistical evidence for a kink as well as normally distributed residuals. Using the latter model, a global poverty line of about $1.21 is obtained, with a slightly larger reference group (of 19 countries). Applying this to the estimation of the Asian poverty line, we



could then take the average of the Asian countries included in the refer-ence group for the global line. That would result in a poverty line of about $1.28 but only three Asian countries are included in the reference group (Tajikistan, Nepal, and Bangladesh) which makes this approach unreliable. If we instead estimated the linear or the log- log model using the 21 Asian observations, this would yield an ‘Asian’ poverty line of $1.41–$1.43;2 this is driven largely by the relative high national poverty lines in Tajikistan, Yemen, and Mongolia. Also here, reliability is an issue because a nonlinear threshold model is estimated on just 21 observations where outliers and small data problems could have a large impact; it is also likely that the switch to the 2011 PPP round would lead to a substantial change in this estimate. Overall, we suggest that this method would not generate a very reliable and robust estimate for an Asia- specific income poverty line.

Besides these estimation issues, there are more serious concerns and criticisms of this entire approach, which have been discussed extensively in the literature (for example, Reddy and Pogge 2009; Deaton 2010; Klasen 2013a, 2013b; Dotter 2014; Klasen et al. 2015). We highlight four of the most important issues that have been discussed in the literature. First, this method is unstable and highly dependent on the sample of countries included in the estimation and the PPP exchange rates used. When, in 2008, the World Bank switched from using the 1993 PPPs and the sample of countries used for estimating the poverty line, it led to the switch of the international poverty line from $1.08 in 1993 dollars to $1.25 in 2005 dollars. Currently, similar issues are arising with the new 2011 PPPs which could lead to serious reassessments of poverty levels in the world and in different regions (Klasen et al. 2015). More seriously, the 2008 revision led to a massive upward shift in global poverty for all years, for example from about 29 percent in 1990 to about 41 percent in the same year; thus, the base year of the first MDG was changed substantially with a large impact on what halving global poverty would mean. The pace of poverty reduction was, however, less affected (Chen and Ravallion 2010). As shown by Deaton (2010) and Greb et al. (2011), the main reason for the massive increase in levels of observed global poverty was not the switch of the PPPs, but the switch in the sample of countries used to estimate global poverty. Deaton (2010) additionally noted that the change in the sample led to some perverse effects. In particular, he noted the case of India. Whereas India was part of the reference group of countries that made up the global poverty line using 1993 dollars, high subsequent growth ensured that India was no longer in the reference group in the assessment using 2005 dollars. Because India’s poverty line is rather low, the exclusion of India from the reference group led to an increase in the global poverty line, which, in turn, led to an increase in measured poverty in India using



that new line. In a sense, rapidly rising incomes in India have led to higher observed poverty in India using the international poverty line, clearly a problematic effect. In short, there appear to be substantial problems and uncertainties associated with switches in PPPs and national poverty lines used to estimate the global poverty line. The 2011 PPPs suggest that prior assessments of PPP- adjusted incomes underestimated per capita incomes in the PRC and India, and some other Asian economies. If these are used to generate a new international poverty line, this could have substantial implications for poverty in those countries, compared with other regions, as well as on global poverty.

A second line of criticism relates to the use of PPPs more generally for this type of assessment (Reddy and Pogge 2009; Deaton 2010; Klasen 2013b). One criticism is that PPPs are generated to compare overall price levels, not price levels for the poor; worse, they can be sensitive to changes in the price level for goods unrelated to the poor (Reddy and Pogge 2009). Another criticism is that PPPs are only valid for a particular benchmark year, but not over time. Thus, the question arises whether one should use only one PPP benchmark year (as currently being done in the World Bank’s approach to poverty measurement), or several benchmark years (as done for the Penn World Tables that also use PPP- adjusted income data).

A third line of criticism is that the international income poverty line has limited relevance for country- level poverty assessments because the difference between country- level income poverty lines and the interna-tional income poverty line is substantial (Dotter 2014). This point, already alluded to above, is nicely visible in the estimation of the Asian poverty line in Figure 1.2(c). As can be seen, the difference between country- level poverty lines and the estimated Asian poverty line is substantial. In Tajikistan and Yemen, poverty using the Asian poverty line is much lower than when using national poverty estimates, whereas in Nepal and India it is much higher. In fact, there is a clear regional pattern to the difference between national poverty lines and an Asian poverty line. All South Asian countries are below the estimated line, that is, poverty is lower using national poverty lines than the international poverty line. The converse is the case for all Western and Central Asian countries that were part of the former Soviet Union. There, national poverty lines are all above the line; thus, poverty is much lower using the common Asian poverty line. This clear regional pattern appears problematic and suggests substantial prob-lems with one common Asian poverty line.

A fourth criticism of such an approach, closely related to the one just discussed, is the increasing irrelevance of the $1.25 poverty line for an increasing number of Asian countries. In many Asian economies, this poverty line is simply too low to be relevant for policy makers there. In



fact, several Asian countries, including the PRC and India, have recently increased their national poverty line to make it more relevant for national policy making. In this context, the question arises whether one should adjust the poverty line because of increasing prosperity. This is precisely the suggestion by Ravallion and Chen (2011) of a ‘weakly relative’ interna-tional poverty line to which we turn below.

To summarize, the case for an Asian poverty line using the World Bank’s method of deriving the $1.25 poverty line is weak. It would not lead to a substantially different poverty line, it is poorly linked to national poverty lines, it is unstable due to the link to the PPPs and the estimation method, and it would be increasingly irrelevant for fast- growing Asian economies.

3.2 A ‘Weakly Relative’ Poverty Line Using the World Bank’s Approach

Ravallion and Chen (2011) proposed a ‘weakly relative’ international poverty line. It can be derived in various ways, but the easiest is to con-sider Figure 1.2(a) again. The suggestion is that for countries on the flat portion of the curve, the $1.25- a- day should be the relevant line. For those on the ascending portion of the curve, the poverty line should rise with the increase in mean income. It turns out that the best empirical fit is that, above the threshold, the poverty line should increase by $0.33 for every $1 increase in per capita consumption of a country above the threshold (Ravallion and Chen 2011; Chen and Ravallion 2013). The elasticity of the weakly relative poverty line is substantially below 1 (but increases with increasing incomes), which distinguishes it from a purely relative line.

Such a weakly relative poverty line has several features that make it advantageous to be used for an Asian poverty line (see Klasen 2013a; Klasen et al. 2015). First, it adjusts the poverty line ‘automatically’ with increasing prosperity in Asia, thereby addressing the problem of the increasing irrelevance of the very low $1.25- a- day poverty line. As shown by Chen and Ravallion (2013), the weakly relative poverty line in East Asia, for example, is about $2.34 in 2008, and, in South Asia, it is $1.94 (see Appendix Tables A1.1 and A1.2). Given that this poverty line increases under- proportionately with mean income, distribution- neutral growth will still lower ‘weakly relative’ poverty, but will do so at a slower pace than when using a purely absolute line. As can be seen in Appendix Tables A1.1 and A1.2, there was still substantial poverty reduction in Asia using this approach, but poverty remains a very serious issue.

Thus, the ‘weakly relative’ poverty line has some advantages. At the same time, all the other disadvantages of the World Bank’s method remain,



so it is not clear whether this is the best way forward. However, it clearly seems to be superior to simply deriving an Asian absolute poverty line using the World Bank’s method.

3.3 Using National Poverty Lines to Measure Poverty in Asia

A third approach of setting an income poverty line is to coordinate a process in Asia of setting national poverty lines using a standardized methodology. These national poverty lines would be expressed in national currency but could still aggregate the poor across countries in a consist-ent fashion if the poverty lines were consistently derived. This proposal was made by Reddy et al. (2008) and later by Klasen (2013a, 2013b) and Klasen et al. (2015). One advantage is that the problems associated with the PPP exchange rates would be avoided. A second advantage is that such a poverty line would be more closely linked to national poverty measure-ment and, thus, would have a higher relevance.

At the same time, a range of questions would need to be addressed before such a proposal could be implemented (see Klasen 2013b for an extensive discussion). First, how should such a poverty line be grounded? The most promising approach would be to use the method most com-monly used to set national poverty lines in developing countries, that is, the cost of basic needs method (Ravallion 1994). This method involves first identifying a reference group of households (which should be close to the poverty line) whose spending pattern would be used to derive expendi-ture shares on a basket of goods and services used to assess poverty. In a second step, the food expenditures in that basket are turned into calories and then the basket is scaled up (or down) to reach the required caloric norm for households. This basket (including nonfood items) then defines the quantities of food and nonfood items to be consumed at the poverty line. The cost of that basket then yields the poverty line. This poverty line is then updated for price changes of goods included in the basket over the years. However, over longer time periods the basket is adjusted to reflect changing expenditure patterns. In a rapidly growing economy, this usually means that the basket changes by reducing the food share and increasing more higher- quality goods. In this way, relative poverty considerations can be brought in when the poverty basket is adjusted.

Although the methods are straightforward and have been applied in many countries (including in Asia), setting these poverty lines in a consist-ent fashion across countries is challenging. The first- best option would be for participating countries to agree on a consistent system of poverty measurement using this approach. It would ideally also include coordinat-ing household surveys so that the questionnaires are similar enough that



they can be used consistently. The model would be the System of National Accounts where a similarly coordinated process of standard methods is accepted across the world.

At the same time, it is unlikely that such a coordinated way to set national poverty lines would be agreed upon quickly. In the meantime, a second- best option would be to use existing household surveys from these Asian countries and apply consistent poverty lines in these surveys, even if these lines are not the current approaches used by the governments. In this way, the feasibility of this approach could be demonstrated, thereby moving the debate forward.

Therefore, this approach is promising but requires a longer- term process to implement it fully. However, as suggested, a short- cut is possible and it is useful to illustrate the feasibility of this approach.

3.4 An Asia- specific MPI?

Finally, there is the Asia- specific MPI option to consider. It is widely rec-ognized that poverty is a multidimensional phenomenon. The challenge has always been to come up with a set of indicators and weights that would allow for a consistent analysis of poverty over time and across space. With the publication of the MPI in 2010 (UNDP 2010, ch. 5), a first attempt to create such a comparable poverty measure was made. It uses a so- called dual cutoff method proposed by Alkire and Foster (2011) where the first cutoff defines whether a household is deprived in a particular dimension, and a second cutoff defines whether a household has passed the threshold of deprivations to be called multidimensionally poor.

Although there are many questions of details that still need to be addressed (Dotter and Klasen 2014), it now appears feasible to generate an Asia- specific version of such an MPI. As discussed above, it would first be necessary to think through why and how an Asian MPI would have different indicators, cutoffs, or weights. This is not a straightforward ques-tion and has to deal with the great heterogeneity among Asian countries. Although it may be argued that, because of differences in climate, social structures, or values in particular subregions (for example, South Asia, Southeast Asia and Central Asia), appropriate indicators, cutoffs, and weights could be chosen to generate MPIs for these different subregions, it would be hard to develop an MPI for all of Asia. The only way out of this dilemma would be to initiate a process, possibly at the level of the ADB, to develop a common understanding for indicators, weights, and cutoffs, although it is expected that such a consensus would not be reached easily.

A second way by which one could construct an Asian MPI that differed less fundamentally would be to adjust cutoffs to better reflect the average



performance of Asian economies in these MPI indicators. For example, a cutoff of five years of education of a single household member to render the entire household non- poor, as currently used in the MPI, might be too low for many Asian countries. Thus, the idea would be to move away from an absolute indicator of acute multidimensional poverty to a (weakly?) relative one that considers the performance of Asian economies in these indicators.

A third approach would be to change the weights used for an Asian- based multidimensional poverty line. Here the results from the illustrative exercise of Pasha (2014) are interesting. She uses principal components analysis to derive statistical weights for the indicators included in the MPI. Pasha (2014) finds substantial differences across countries in these weights. In India, the weights for child mortality and nutrition are higher than in all the 22 countries included in her sample. Conversely, the weights for educa-tion are low. The weights for standard of living are very high (altogether 80 percent), with quality of floor and access to cooking fuel particularly important. Using such country- specific weights would lead to quite differ-ent multidimensional poverty measures and might provide interesting new insights. It would also lower the ability to compare levels and trends across countries. Clearly this is an issue well worth exploring further.

4 SOME TENTATIVE CONCLUSIONS

This discussion suggests that the case for developing an Asian poverty line is not straightforward. In particular, we have argued that there are no good reasons to adjust the World Bank’s $1- a- day approach to an Asian setting. Many of the problems of the World Bank’s international poverty line would carry over to its Asian version; in addition, the database to estimate such a poverty line would be even smaller, leading to questions of reliabil-ity and robustness. Also, the large heterogeneity in existing poverty lines in Asia would militate against this proposal. A more promising option is to consider a ‘weakly relative’ Asian income or non- income poverty line that takes into account the rapid growth in living conditions and aspira-tions in many of Asia’s economies. However, many of the drawbacks of the current international poverty line would carry over to the ‘weakly relative’ case. Even more promising could be a coordinated process for setting national income poverty lines where national poverty measurement is based on a common conception of poverty. This is a long- term agenda that would need a great deal of coordination between Asian economies but is well worth pursuing further. Another option would be the creation of an Asia- specific MPI, maybe one that adjusts itself automatically to improving living conditions by adjusting the cutoffs. All of these proposals



would have to be rigorously tested to see whether they can be implemented and yield new insights that are not visible in current approaches to poverty measurement in Asia.

The discussions about the changes in the international poverty line to reflect the results of the 2011 PPP show the difficulty of maintaining a reli-able, consistent, and robust international poverty line (Klasen et al. 2015). Thus, it is all the more important to consider alternatives.

NOTES

1. I would like to thank Xuehin Han, Guanghua Wan, Jacques Silber, Tatyana Krivobokova, and George Battese for helpful inputs, comments, and discussion on earlier versions of the chapter.

2. I thank Tatyana Krivobokova for providing these estimates, based on the methods described in Klasen et al. (2015).

REFERENCES

Alkire, S. and J. Foster (2011a), ‘Counting and multidimensional poverty measure-ment’, Journal of Public Economies, 95 (7), 476–87.

Alkire, S. and M.E. Santos (2014), ‘Measuring acute poverty in the developing world: robustness and scope of the Multidimensional Poverty Index’, World Development, 59, 251–74.

Asian Development Bank (ADB) (2014), Poverty in Asia: A Deeper Look, Manila: Asia Development Bank.

Chen, S. and M. Ravallion (2001), ‘How did the world’s poor fare in the 1990s?’, Review of Income and Wealth, 47 (3), 283–300.

Chen, S. and M. Ravallion (2010), ‘The developing world is poorer than we thought, but no less successful in the fight against poverty’, Quarterly Journal of Economics, 125 (4), 1577–625.

Chen, S. and M. Ravallion (2013), ‘More relatively poor in a less absolutely- poor world’, Review of Income and Wealth, 59 (1), 1–28.

Deaton, A. (2010), ‘Price indexes, inequality, and the measurement of world poverty, American Economic Review, 100 (1), 5–34.

Dotter, C. (2014), ‘The (ir- )relevance of the international poverty line for national poverty assessment’, mimeograph, University of Göttingen.

Dotter, C. and S. Klasen (2014), ‘The Multidimensional Poverty Index: achieve-ments, conceptual and empirical issues’, UNDP HDRO Occasional Paper, UNDP, New York.

Greb, F., S. Klasen, S. Pasaribu and M. Wiesenfarth (2011), ‘Dollar a day re- revisited’, Courant Research Center Discussion Paper No. 91, University of Göttingen.

Klasen, S. (2013a), ‘Is it time for a new international poverty measure?’, in E. Solheim (ed.), Development Cooperation Report 2013: Ending Poverty, Paris: OECD.



Klasen, S. (2013b), ‘Measuring levels and trends in absolute poverty in the world: open questions and possible alternatives’, in G. Betti and A. Lemmi (eds), Poverty and Social Exclusion: New Methods of Analysis, London: Taylor and Francis.

Klasen, S., T. Krivobokova, F. Greb, R. Lahot, S. Pasaribu and M, Wiesenfarth (2015), ‘International poverty measurement: which way now?’, Courant Research Center: Poverty, Equity, and Growth Discussion Paper No. 184, University of Göttingen.

Klasen, S. and H. Waibel (2013), Vulnerability to Poverty, London: Palgrave.Klasen, S. and H. Waibel (2014), Vulnerability to poverty in South- East Asia:

drivers, measurement, responses, and policy issues, World Development, DOI: http://dx.doi.org/10.1016/j.worlddev.2014.01.007.

Pasha, A. (2014), ‘Regional perspectives to the Multidimensional Poverty Index’, mimeograph, University of Göttingen.

Ravallion, M. (1994), Poverty Comparisons, Fundamentals of Pure and Applied Economics, vol. 56, Chur: Harwood Academic.

Ravallion, M. and S. Chen (2011), ‘Weakly relative poverty’, Review of Economics and Statistics, 93 (4), 1251–61.

Ravallion, M., S. Chen and P. Sangraula (2009), ‘Dollar a day revisited’, World Bank Economic Review, 23 (2), 163–84.

Ravallion, M., G. Datt and D. van de Walle (1991), ‘Quantifying absolute poverty in the developing world’, Review of Income and Wealth, 37 (4), 345–61.

Reddy, S. and T. Pogge (2009), ‘How not to count the poor’, in S. Anand, P. Segal and J. Stiglitz (eds), Debates on the Measurement of Global Poverty, Oxford: Oxford University Press.

Reddy, S., S. Visaria and M. Attali (2008), ‘Inter- country comparisons of income poverty based on a capability approach’, in K. Basu and R. Kanbur (eds), Arguments for a Better World, vol. 2, Oxford: Oxford University Press.

Rippin, N. (2013), ‘Considerations of efficiency and distributive justice in multi-dimensional poverty measurement’, PhD dissertation, University of Göttingen.

United Nations Development Programme (UNDP) (2010), Human Development Report, New York: UNDP.

World Bank (1990), World Development Report 1990: Poverty, New York: Oxford University Press.

World Bank (2000), World Development Report 2000/01: Attacking Poverty, Washington DC: World Bank.


28

APP

EN

DIX

1A

.1

Tabl

e 1A

.1

Aver

age

rela

tive

pove

rty

line

by re

gion

and

yea

r

Reg

ion

Mea

n po

vert

y lin

e $

per p

erso

n pe

r day

at 2

005

PPP

1981

1984

1987

1990

1993

1996

1999

2002

2005

2008

Met

hod

1E

ast A

sia a

nd P

acifi

c1.

331.

341.

351.

391.

431.

571.

661.

822.

032.

34PR

C1.

251.

251.

251.

251.

251.

351.

481.

641.

852.

20E

aste

rn E

urop

e an

d C

entr

al A

sia4.

054.

214.

354.

213.

783.

793.

984.

545.

616.

99L

atin

Am

eric

a an

d th

e C

arib

bean

4.32

4.25

4.07

4.00

4.28

4.41

4.68

4.76

5.09

5.79

Mid

dle

Eas

t and

Nor

th A

fric

a2.

422.

562.

402.

412.

472.

502.

592.

803.

053.

37So

uth

Asia

1.27

1.27

1.30

1.35

1.38

1.47

1.54

1.58

1.74

1.94

Sub-

Saha

ran

Afr

ica

1.55

1.55

1.53

1.51

1.49

1.51

1.51

1.53

1.55

1.60

Tota

l2.

002.

012.

002.

001.

992.

082.

172.

302.

542.

90To

tal e

xcl.

PRC

2.29

2.30

2.28

2.28

2.26

2.34

2.41

2.52

2.77

3.13

Met

hod

2E

ast A

sia a

nd P

acifi

c1.

361.

381.

411.

451.

541.

731.

802.

062.

392.

72PR

C1.

261.

291.

331.

341.

421.

611.

712.

012.

402.

81E

aste

rn E

urop

e an

d C

entr

al A

sia4.

194.

364.

494.

274.

764.

373.

964.

395.

226.

71L

atin

Am

eric

a an

d th

e C

arib

bean

4.51

4.17

4.36

4.42

4.47

4.68

4.70

4.66

5.17

5.93

Mid

dle

Eas

t and

Nor

th A

fric

a2.

843.

042.

842.

852.

982.

983.

063.

353.

023.

24So

uth

Asia

1.30

1.32

1.35

1.37

1.38

1.41

1.46

1.48

1.54

1.58

Sub-

Saha

ran

Afr

ica

1.71

1.66

1.65

1.60

1.54

1.56

1.58

1.61

1.63

1.78

Tota

l2.

092.

092.

122.

112.

182.

242.

232.

372.

602.

94To

tal e

xcl.

PRC

2.41

2.39

2.41

2.38

2.46

2.46

2.41

2.50

2.66

2.98

Sour

ce:

Rep

rodu

ced

from

Rav

allio

n an

d C

hen

(201

3, ta

ble

4).


29

Tabl

e 1A

.2

Wee

kly

pove

rty

mea

sure

for t

he d

evel

opin

g wo

rld, 1

981–

2008

Reg

ion

1981

1984

1987

1990

1993

1996

1999

2002

2005

2008

% o

f po

pula

tion

belo

w re

lativ

e po

vert

y lin

eE

ast A

sia a

nd P

acifi

c80

.570

.060

.463

.660

.151

.952

.148

.843

.442

.4PR

C85

.272

.659

.065

.261

.151

.251

.348

.441

.941

.0E

aste

rn E

urop

e an

d C

entr

al A

sia22

.021

.421

.525

.432

.234

.032

.230

.429

.328

.2L

atin

Am

eric

a an

d th

e C

arib

bean

49.6

50.3

46.9

46.8

50.0

49.9

51.1

51.2

47.9

45.9

Mid

dle

Eas

t and

Nor

th A

fric

a42

.041

.240

.739

.338

.638

.438

.837

.936

.635

.0So

uth

Asia

64.0

61.6

60.9

60.3

58.9

58.0

56.9

56.8

55.1

53.5

Sub-

Saha

ran

Afr

ica

62.3

64.3

64.2

65.1

66.9

66.6

66.5

65.3

63.6

61.1

Tota

l62

.858

.554

.456

.055

.752

.752

.651

.248

.246

.9To

tal e

xcl.

PRC

54.6

53.4

52.8

52.9

53.8

53.2

53.0

52.0

50.2

48.6

Num

ber o

f re

lativ

e po

or (i

n m

illio

ns)

Eas

t Asia

and

Pac

ific

1143

.410

44.1

946.

210

47.0

1031

.892

5.2

959.

292

4.4

841.

784

0.4

PRC

847.

075

3.1

639.

873

9.9

720.

062

3.1

642.

762

0.0

546.

554

2.6

Eas

tern

Eur

ope

and

Cen

tral

Asia

94.7

94.7

97.5

118.

015

0.8

159.

715

1.3

143.

113

8.0

133.

6L

atin

Am

eric

a an

d th

e C

arib

bean

180.

719

5.4

193.

620

4.3

230.

024

1.1

259.

127

0.9

263.

426

1.5

Mid

dle

Eas

t and

Nor

th A

fric

a72

.477

.883

.988

.593

.398

.710

5.6

109.

011

1.1

111.

9So

uth

Asia

594.

761

5.7

653.

469

1.6

720.

375

2.0

781.

682

1.4

836.

684

9.4

Sub-

Saha

ran

Afr

ica

248.

127

8.6

302.

833

3.6

371.

440

0.6

431.

945

7.9

479.

949

6.4

Tota

l23

33.9

2306

.122

77.3

2483

.025

97.6

2577

.126

88.7

2726

.626

71.0

2692

.9To

tal e

xcl.

PRC

1486

.915

53.0

1637

.517

43.1

1877

.619

53.9

2046

.121

06.6

2124

.521

50.3

Not

e:

Rel

ativ

e po

vert

y lin

es b

ased

on

met

hod

2. R

egio

ns w

ith su

rvey

cov

erag

e le

ss th

an 5

0 pe

rcen

t are

em

bold

ened

.

Sour

ce:

Rep

rodu

ced

from

Rav

allio

n an

d C

hen

(201

3, ta

ble

5).


30

2. A poverty line contingent on reference groups: implications for the extent of poverty in some Asian countries1

Satya R. Chakravarty, Nachiketa Chattopadhyay and Jacques Silber

1 INTRODUCTION

In his Wealth of Nations, Adam Smith stated that:

by necessaries I understand, not only the commodities which are indispensably necessary for the support of life, but whatever the custom of the country renders it indecent for creditable people, even of the lowest order, to be without. A linen shirt, for example, is, strictly speaking, not a necessary of life. The Greeks and Romans lived, I suppose, very comfortably, though they had no linen. But in the present times, through the greater part of Europe, a creditable day- laborer would be ashamed to appear in public without a linen shirt, the want of which would be supposed to denote that disgraceful degree of poverty, which, it is presumed, nobody can well fall into without extreme bad conduct. . . . Under necessaries therefore, I comprehend, not only those things which nature, but those things which established rules of decency have rendered necessary to the lowest rank of people. (Smith 1937: 821–2)

In fact, absolute poverty lines are generally used in poor countries (for example, $1.25 per day, which is an updated figure of the earlier pro-posal of $1 per day). On the other hand, in rich countries, such as in Western Europe, the poverty line corresponds to some proportion (60 percent) of the median income. Ravallion and Chen (2011) have argued that both approaches are justified because, in poor countries, it makes sense that those who are able to feed and clothe themselves should not be considered poor, whereas, in rich countries, the idea of social exclusion should be of prime importance (see Sen 2000 for more details).

Ravallion and Chen (2012: 3) have actually argued that:


A poverty line contingent on reference groups 31

if one thinks that it is really only social norms that differ, with welfare depend-ing solely on own consumption, then one would probably prefer an absolute measure, imposing a common norm (though one would presumably also be drawn to consider more than one possible line). However, if one is convinced that that there are social effects on welfare then one would be more inclined to use a relative line in the consumption or income space, anchored to a common welfare standard. The problem for global poverty comparisons is that we do not know which of these two interpretations – differing social norms or social effects on welfare – is right. And we may never resolve the matter from conven-tional empirical evidence. This uncertainty makes it compelling to consider both approaches when measuring global poverty.

This is why Ravallion and Chen (2011), generalizing somehow the meas-ures proposed by Atkinson and Bourguignon (2001), suggested that there should be a positive lower bound to the costs of social inclusion so that the poverty line would rise with the mean income only above some critical value and it then would do so with an elasticity less than 1. A different but still combined approach to the selection of a poverty line was proposed more recently by Chakravarty et al. (2015) who developed axiomatically what they called ‘an amalgam poverty line’. The Atkinson–Bourguignon (2001) and European Union (see Lelkes and Gasior 2011) standard sugges-tions for basing the poverty lines on some location parameter are particu-lar cases of this formulation.

Essential to the methodology adopted in the chapter is the notion of ‘reference group’. By a reference group we mean a subgroup of population within which an individual confines his or her aspiration. Since we are con-cerned with poverty, we identify the reference group by a reference income level, say, the median or mean income.

The novelty of the present chapter is that it offers an empirical illustra-tion of the proposal of Chakravarty et al. (2015). Using data on the shares in total expenditures of the deciles of the distribution of expenditures in different Asian countries around 2010, it indicates what the headcount ratio, the number of poor and the poverty gap ratio would be under various scenarios. These scenarios are a function of the absolute poverty line (taken as $1.25 per day or $1.45 per day), the ‘reference income’ chosen as either the mean or the median of the distribution of expenditures, and the weight given to the absolute poverty line.

The chapter is organized as follows. Section 2 briefly summarizes the main elements of the paper by Chakravarty et al. (2015). Section 3 reviews the role played by reference groups in the growing economic literature on happiness. Section 4 presents the results of the empirical investigation, while concluding comments are given in section 5.



2 MAKING THE POVERTY LINE DEPENDENT ON REFERENCE GROUPS: AN AXIOMATIC APPROACH

Chakravarty et al. (2015) developed an axiomatic approach to the deter-mination of what they call ‘an amalgam poverty line’. Given a reference income, say, the mean or the median, this ‘amalgam poverty line’ is derived as a weighted average of the existing absolute poverty line and the refer-ence income, the choice of the weight being guided by the policy maker’s preferences for aggregating the two components. The individual utility is assumed to be an increasing concave function of the absolute poverty line but a decreasing convex function of the reference standard. Following Clark and Oswald (1996), Chakravarty et al. (2015) considered both an additive and a multiplicative form of the utility function, using two differ-ent sets of intuitively reasonable axioms.

The general idea of their approach is as follows. Imagine some reference income and a person with an income equal to some arbitrarily set poverty line. They first determine the level of the corresponding utility. They then consider an alternative situation where this person has an income identi-cal to some given poverty line. Moreover they suppose that for this indi-vidual his or her own income is actually his or her reference income. If it is assumed that the person is equally satisfied in both cases, it is possible to equate the utilities in both states of affairs and to then determine uniquely the arbitrary poverty line. This presumption of equal satisfaction in both situations is plausible because in each case the individual is at the existing poverty line income.

Chakravarty et al. (2015) then proceed as follows. Following Clark and Oswald (1998) they examine two options. First they assume that an indi-vidual’s utility function depends in part on his or her absolute income. Assuming that the utility function has the properties of linear translatabil-ity and linear homogeneity (see Chakravarty et al. 2015, for more details), they proved that the utility function has the form

U(y, (y − r)) 5 (k − a)y + ar (2.1)

where y > 0 is the individual income, r > 0 is the reference income, k > 0 and a < 0 are constants. Positivity of k reflects the view that as income increases, satisfaction increases, whereas negativity of a ensures that satis-faction decreases as the reference income increases.

Chakravarty et al. (2015) show that the ‘amalgam poverty line’ is a weighted average of the traditional absolute poverty line and the specified reference income. But they also explore the case where utility depends on



the relative income, that is, income relative to some reference standard. In other words, an individual’s utility depends also on his or her relative position (or ‘status’) in the society in terms of some measure of well- being. They assumed linear homogeneity, normalization, and continuity (see Chakravarty et al. 2015 for more details) and then proved that one form of the utility function was of the form

U 5 yab 2ryb 5 by 2 r (2.2)

where b > 1 is the ratio between the upper bound u and lower bound l > 0 of incomes.

It follows that the ‘amalgam poverty line’ z1 is a weighted average of the traditional absolute poverty line z0 and the reference income r. More precisely they derive that

z1 5 wz0 + (1 − w)r (2.3)

with

w 5b 2 1

b. (2.4)

Since b > 1, it follows that, 0 < w < 1.As mentioned previously, Clark and Oswald (1996) had suggested

taking into account this relativity by making either difference or ratio comparisons. In the empirical section of this chapter, we, however, base our analysis on the ‘ratio comparisons model’. In such a model, the extent of an individual’s feeling of deprivation arising out comparison of its own income with respect to a reference income is expressed in terms of the ratio between the two incomes (see the next section).

3 HAPPINESS, OWN STANDARD OF LIVING AND REFERENCE GROUPS

The analysis of subjective well- being has been a growing field of inquiry during the last two decades, especially in recent years. Of particular interest in this literature is the analysis of the impact of so- called reference groups on life satisfaction or satisfaction with income. This question is related to a much older hypothesis, which assumes that utility depends not only on our own income but also on that of others. The importance of relative income



had been stressed in the work of Duesenberry (1949) who assumed that the utility of an individual is negatively affected by the income of anyone richer than him, as well as the work of Runciman (1966) whose focus was on the concept of relative deprivation.

An empirical application of these ideas may be summarized by the fol-lowing equation (see Clark et al. 2008: 100):

U(t)5b1 lny(t) 1 b2 lna y(t)y*(t)

b 1 Z r (t)g (2.5)

where U(t) is the individual’s utility, y(t) is his income, y*(t) is some refer-ence income, Z9 is a vector of additional determinants and g is a vector of the coefficients of these determinants, all the variables being measured at time t.

There are various ways of obtaining measures for y*(t). We can estimate wage equations, controlling for individual characteristic, such as age, gender, education, area of residence, and then obtain a predicted value of y*(t) for each individual. Another possibility is to compute cell aver-ages to obtain an estimate of the average wage by, say, gender, education and region, on the basis of either the dataset itself or some external data source. Finally, more recently information about the reference income can be obtained directly from the survey itself. This information may be of a qualitative nature and the respondents are then asked how much higher or smaller (on some ordinal scale) their incomes are with respect to their reference incomes (see, for example, Knight et al. 2009). There may even be some quantitative information on the income of the reference group such as that available in a Japanese survey where those who participated in this survey were asked to estimate the income of people who had the same age, sex and educational level as theirs (see Clark et al. 2013). This direct source of information is still very rare, although van Praag pleaded recently ‘for an extension of the happiness paradigm by setting up a new additional agenda for empirical research in order to get quantified knowledge about the referencing process’ (van Praag 2011: 111).

Reference groups have also been introduced in studies of the determi-nants of the ‘subjective economic ladder’ where people are asked to define their position on some scale of standard of living. For instance, using an Indonesian survey, Powdthavee (2009), rather than selecting relative income as a determinant of this subjective economic ladder, introduced a variable measuring the rank of the individual in the distribution of income/wealth at the local level (see Powdthavee 2009).

The next question concerns the determination of the reference group. A first possibility is to consider that the reference group is made of colleagues



in which case the emphasis is on ‘the relation between income gaps in the professional sphere and various notions of satisfaction ranging from job to life satisfaction’ (Senik 2009: 8). Clark and Oswald (1996), for example, analyzed job satisfaction on the basis of the British Household Panel Survey (BHPS) and defined the reference group of a worker as the income of employees who had the same age and level of qualification as the worker and were doing the same kind of job. Other studies have assumed that the reference group was composed of people with the same characteristics as the individual, with, for example, the same age, level of education and region of residence (see Ferrer- i- Carbonell 2005). Some authors have also used space- based reference incomes such as the average income of indi-viduals of the same race in the cluster and district where the individuals surveyed live (see Kingdon and Knight 2007).

The objective rank of an individual in the area where he lives has also been shown to affect the satisfaction he gets from his consumption level, as stressed by Fafchamps and Shilpi (2008) in their work on subjective welfare in Nepal.

The impact of the reference income on subjective well- being is an impor-tant issue to be examined. The literature makes a distinction between two possible impacts, one reflecting a signaling effect, the other the role of status. The idea that other people’s income may have a positive effect on satisfaction was originally introduced by Hirschman and Rothschild (1973: 545–6):

Suppose that the individual has very little information about his future income, but at some point a few of his relatives, neighbours, or acquaintances improve their economic or social position. Now, he has something to go on: expect-ing that his turn will come in due course, he will draw gratification from the advances of others – for a while. It will be helpful to refer to this initial gratifica-tion as the ‘tunnel effect’.

Evidence confirming the existence of such signaling effects was provided by Senik (2004, 2008). The more common impact of reference income seems nevertheless to be a status effect: ceteris paribus a higher reference income negatively affects satisfaction from life or income (see, for example, the studies of Senik 2009 and Clark and Senik 2009).

As far as empirical results are concerned, there are hitherto very few papers in the literature on subjective welfare that estimate the impact on happiness, ceteris paribus, of an increase in one’s own income, on the one hand, and of a rise in the reference group’s income, on the other hand. Moreover the effect of a change in the reference income, when estimated, was generally derived indirectly. Knight et al. (2009), for example, who looked at subjective well- being in the People’s Republic



of China, introduced in their regression a dummy variable indicating whether the household income was much above, above, below or much below the village average. Clark and Senik (2009), using the third wave of the European Social Survey, defined two types of variables to take into account other people’s income: a dummy variable indicating how impor-tant it was for the respondent to compare his or her own income with that of others and another dummy variable showing with what popula-tion category the comparison was made (friends, work colleagues, family members, others). However, in a more recent paper Clark et al. (2013) were able to introduce a variable referring directly to the income of some refer-ence group. They analyzed an Internet survey that was conducted in Japan and in which the respondent was asked to indicate what she thought was the average personal income before taxes of people of the same age, gender and educational level as hers. The authors were also able to estimate this individual reference income by looking at the mean values observed in cells corresponding to individuals with the same, age, education, gender, and labor force status. Finally, Clark et al. (2013) also used external sources to compute the actual income of individuals by labor force status (civil serv-ants, self- employed, and so on). In table 4 of their paper Clark et al. report the results of a regression where the dependent variable refers to satisfac-tion with income. It then appears that the coefficient of own income is about three times as high as that of self- reported reference income, and of opposite sign, even when a variable measuring the ‘comparison intensity’ of the individual (how important it is for the respondent to compare her income with that of others) is introduced.

We can now attempt to use this result (a ratio of about three between the coefficient of own income and that of the reference income) and introduce it in equations (2.2) and (2.3) above. More precisely, this empirical result would imply that the coefficient b in (2.2) would be equal to 3.

Using (2.2) we derive that

dU 50U0y

dy 10U0r

dr 5 bdy 2 dr (2.6)

so that for a given utility level,

drdy

5 b (2.7)

Using (2.4), we then conclude that the weight w would be equal to 2/3, one of the values which is used for w in the empirical section of the present chapter.



The data we use do not provide any information on the reference income of individuals. We know only the shares in total income of the various deciles as well as the mean and median of the income distribution (or rather the distribution of expenditures) in the various countries for which data are available. We therefore decided that the reference income would be either the mean or the median. If the mean is selected, we implicitly assume that the extent of poverty should also be a function of the income of those who are not poor, or more generally of the standards of living of all the individuals in the population. If the reference chosen is the median income, then, since the latter does not depend on the incomes of those who are not poor, we really assume that the extent of poverty depends on the standards of living of those individuals who belong to the middle class, and are in the middle of the income distribution.

We prefer the choice of the median as the reference income. As Aristotle (Politics, book 4, part XI: 96) argued: ‘the best political community is formed by citizens of the middle class, and that those states are likely to be well- administered, in which the middle class is large’. A large and rich middle class contributes significantly to the welfare of a society in many ways, for instance, with respect to high economic growth, higher contribu-tion to the country’s tax revenue, a better infrastructure and higher level of education. Therefore, a person with a low income may view the median as a reference income and be hopeful about achieving this income (see Chakravarty 2015).

4 THE EXTENT OF POVERTY WITH AN ‘AMALGAM POVERTY LINE’: THE CASE OF ASIAN COUNTRIES

In this section, we present several measures of the extent of poverty in various Asian countries, when an ‘amalgam poverty line’, is used. For an absolute poverty line, we first use a monthly income of $38 (at 2005 PPP) which corresponds to $1.25 per day, as originally suggested by Ravallion et al. (2009). However, following some of the objections raised by Deaton (2001, 2010) in his criticism of a unique poverty line of $1 per day or $1.25 per day, we have also introduced, on the basis of the estimations of Han (2014), an absolute poverty line of $44, which is based only on Asian data and corresponds to $1.45 a day. We also assumed various pos-sible weights. More precisely, we supposed that the weight w given to the absolute poverty line (the weight of the median or of the mean then being (1 − w)), could be 1, 0.9, 0.66 and 0.5.

The database consisted of information on the income shares of ten



deciles in the various countries and years for which these figures were available. The computation method used is based on the algorithm pro-posed by Shorrocks and Wan (2009), which makes it possible to ‘ungroup’ income distributions, that is, to derive, for example, the share of each centile when the only data available originally are the income shares of deciles. The first step of this algorithm consists of building an initial sample with unit mean which is generated from a parametric form fitted to the grouped data. In the second stage the algorithm adjusts the observa-tions generated in the initial sample to the true values available from the grouped data.

In Table 2.1, we present the value of the headcount ratio in the differ-ent Asian countries for which data were available, under several possible scenarios. As expected, for a given weight, the headcount ratio is higher when the weight (1 − w) refers to the mean rather than the median. The headcount ratio increases with the weight w and is higher with an absolute poverty line of $44 than with one of $38.

We then combined the data on the headcounts given in Table 2.1 with data on the total population around 2010 of the countries examined to derive an estimate of the total number of poor in each country. All these results are given in Table 2.2. It is then easy to compare the number of poor under various scenarios with those obtained on the basis of a $38 absolute poverty line and a value of w equal to 1 (so that the ‘amalgam poverty line’ is also equal to $38).

Finally, Table 2.3 gives the income gap ratios in the different countries under the various scenarios. This index is an indicator of poverty depths of different individuals. When multiplied by the poverty line and the total number of poor, this summary measure has a direct policy interpretation in the sense that the multiplied formula determines the total amount of money required to put all the poor persons at the poverty line. Now, for any country, with a given poverty line and the reference income, we deter-mine the amalgam poverty line using a specific weighting scheme. Given an amalgam poverty line for a country, we can directly estimate the amount of money necessary to place the poor persons of the country at its poverty line using the country’s income gap ratio from Table 2.3 and number of poor from Table 2.2.


39

Tabl

e 2.

1 H

eadc

ount

ratio

s und

er v

ario

us sc

enar

ios

Wei

ghtin

g sc

hem

e (w

eigh

t giv

en to

the

abso

lute

pov

erty

lin

e)

Arm

enia

(2

010)

Aze

rbai

jan

(200

8)B

angl

ades

h (2

010)

Bhu

tan

(201

2)C

ambo

dia

(200

9)F

iji (2

008)

Geo

rgia

(2

010)

Indo

nesia

, R

ural

(2

011)

Indo

nesia

, U

rban

(2

011)

Abs

olut

e po

vert

y

line:

$38

(Wei

ghte

d

with

the

med

ian)

100%

0.02

0.00

0.43

0.02

0.19

0.04

0.18

0.15

0.18

90%

0.04

0.01

0.44

0.04

0.22

0.09

0.21

0.19

0.21

66%

0.16

0.09

0.46

0.17

0.30

0.22

0.29

0.28

0.30

50%

0.24

0.19

0.47

0.26

0.35

0.29

0.35

0.34

0.36

Abs

olut

e po

vert

y

line:

$38

(Wei

ghte

d

with

the

mea

n)90

%0.

060.

010.

460.

060.

250.

130.

230.

210.

2566

%0.

220.

160.

520.

260.

390.

330.

360.

360.

4150

%0.

340.

290.

560.

380.

480.

440.

440.

450.

49A

bsol

ute

pove

rty

lin

e: $

44(W

eigh

ted

w

ith th

e m

edia

n)10

0%0.

050.

000.

550.

030.

280.

090.

230.

230.

2490

%0.

090.

010.

540.

060.

300.

150.

250.

260.

2766

%0.

200.

110.

530.

200.

360.

250.

320.

330.

3450

%0.

270.

210.

520.

270.

390.

310.

370.

380.

38


40

Tabl

e 2.

1 (c

ontin

ued)

Wei

ghtin

g sc

hem

e (w

eigh

t giv

en to

the

abso

lute

pov

erty

lin

e)

Arm

enia

(2

010)

Aze

rbai

jan

(200

8)B

angl

ades

h (2

010)

Bhu

tan

(201

2)C

ambo

dia

(200

9)F

iji (2

008)

Geo

rgia

(2

010)

Indo

nesia

, R

ural

(2

011)

Indo

nesia

, U

rban

(2

011)

Abs

olut

e po

vert

y

line:

$44

(Wei

ghte

d

with

the

mea

n)90

%0.

110.

020.

560.

090.

330.

180.

280.

290.

3166

%0.

260.

180.

590.

280.

440.

360.

390.

410.

4450

%0.

370.

300.

610.

390.

510.

460.

470.

480.

52

Wei

ghtin

g sc

hem

e (w

eigh

t giv

en to

the

abso

lute

pov

erty

lin

e)

Kaz

akhs

tan

(200

9)K

yrgy

z R

epub

lic

(201

1)

Lao

PD

R

(200

8)M

alay

sia

(200

9)M

aldi

ves

(200

4)Fe

dera

ted

Stat

es o

f M

icro

nesia

(2

000)

Nep

al

(201

0)Pa

kist

an

(200

8)Pa

pua

New

G

uine

a (1

996)

Abs

olut

e po

vert

y

line:

$38

(Wei

ghte

d

with

the

med

ian)

100%

0.00

0.04

0.34

0.00

0.02

0.31

0.25

0.21

0.36

90%

0.00

0.08

0.36

0.03

0.04

0.33

0.27

0.24

0.37

66%

0.05

0.20

0.40

0.17

0.18

0.38

0.34

0.31

0.41

50%

0.16

0.27

0.42

0.26

0.26

0.41

0.38

0.36

0.43


41

Abs

olut

e po

vert

y lin

e: $

38(W

eigh

ted

with

the

m

ean)

90%

0.00

0.10

0.38

0.05

0.06

0.38

0.29

0.27

0.41

66%

0.10

0.25

0.48

0.28

0.24

0.51

0.40

0.39

0.52

50%

0.24

0.35

0.50

0.40

0.36

0.58

0.47

0.47

0.58

Abs

olut

e po

vert

y

line:

$44

(Wei

ghte

d

with

the

med

ian)

100%

0.00

0.08

0.44

0.01

0.03

0.35

0.34

0.33

0.42

90%

0.01

0.13

0.45

0.03

0.07

0.37

0.36

0.35

0.43

66%

0.07

0.22

0.46

0.18

0.20

0.41

0.40

0.39

0.45

50%

0.18

0.29

0.47

0.27

0.27

0.43

0.42

0.42

0.46

Abs

olut

e po

vert

y

line:

$44

(Wei

ghte

d

with

the

mea

n)90

%0.

010.

140.

470.

060.

090.

410.

380.

370.

4766

%0.

120.

280.

540.

280.

260.

520.

460.

460.

5650

%0.

250.

380.

580.

410.

370.

590.

510.

520.

61


42

Tabl

e 2.

1 (c

ontin

ued)

Wei

ghtin

g sc

hem

e (w

eigh

t giv

en to

the

abso

lute

pov

erty

lin

e)

Phili

ppin

es

(200

9)Sr

i Lan

ka

(200

9)Ta

jikist

an

(200

9)T

haila

nd

(201

0)T

imor

- Les

te

(200

7)Tu

rkm

enist

an

(199

8)V

iet N

am

(200

8)

Abs

olut

e po

vert

y

line:

$38

(Wei

ghte

d

with

the

med

ian)

100%

0.19

0.03

0.06

0.00

0.37

0.25

0.17

90%

0.22

0.06

0.10

0.02

0.39

0.28

0.20

66%

0.31

0.20

0.21

0.11

0.42

0.34

0.29

50%

0.36

0.27

0.28

0.21

0.44

0.38

0.34

Abs

olut

e po

vert

y

line:

$38

(Wei

ghte

d

with

the

mea

n)90

%0.

260.

090.

120.

030.

410.

300.

2266

%0.

410.

280.

260.

220.

490.

420.

3650

%0.

500.

400.

350.

370.

540.

500.

44


43

Abs

olut

e po

vert

y

line:

$44

(Wei

ghte

d

with

the

med

ian)

100%

0.25

0.07

0.12

0.01

0.49

0.32

0.24

90%

0.28

0.12

0.16

0.03

0.49

0.34

0.27

66%

0.35

0.23

0.24

0.13

0.49

0.39

0.33

50%

0.39

0.30

0.31

0.23

0.50

0.41

0.38

Abs

olut

e po

vert

y

line:

$44

(Wei

ghte

d

with

the

mea

n)90

%0.

310.

160.

170.

040.

510.

360.

2966

%0.

440.

320.

290.

240.

560.

460.

4050

%0.

520.

430.

380.

380.

590.

520.

47

Not

e:

The

com

plet

e in

com

e di

strib

utio

ns w

ere

deriv

ed o

n th

e ba

sis o

f th

e Sh

orro

cks a

nd W

an (2

009)

pro

posa

l for

‘ung

roup

ing

inco

me

dist

ribut

ions

’. T

he fi

rst c

olum

n gi

ves t

he w

eigh

t (in

per

cent

age)

giv

en to

the

abso

lute

pov

erty

line

(eith

er $

38 o

r $44

), th

e co

mpl

emen

t (in

pe

rcen

tage

) giv

ing

the

wei

ght g

iven

to th

e m

edia

n or

the

mea

n of

the

inco

me

dist

ribut

ions

.


44

Tabl

e 2.

2 N

umbe

r of

poor

(in

mill

ions

) in

eac

h co

untr

y, d

epen

ding

on

the

weig

htin

g sc

hem

e*

Cou

ntry

$38,

med

ian,

10

0%$3

8, m

edia

n,

90%

$38,

med

ian,

66

%$3

8, m

edia

n,

50%

$38,

mea

n,

90%

$38,

mea

n, 6

6%$3

8, m

ean,

50

%

Arm

enia

(201

0)0.

070.

130.

480.

710.

170.

661.

00A

zerb

aija

n (2

008)

0.01

0.07

0.77

1.70

0.11

1.40

2.55

Ban

glad

esh

(201

0)65

.43

66.4

968

.99

70.6

369

.65

79.1

084

.95

Bhu

tan

(201

2)0.

010.

030.

130.

190.

040.

190.

28C

ambo

dia

(200

9)2.

723.

134.

264.

993.

555.

536.

75F

iji (2

008)

0.04

0.08

0.18

0.25

0.11

0.28

0.38

Geo

rgia

(201

0)0.

810.

941.

311.

541.

041.

621.

98In

done

sia, R

ural

(201

1)18

.21

23.0

833

.59

40.4

925

.43

43.2

953

.99

Indo

nesia

, Urb

an (2

011)

21.8

126

.00

37.0

743

.89

30.6

250

.06

60.9

7K

azak

hsta

n (2

009)

0.00

0.04

0.80

2.61

0.06

1.61

3.80

Kyr

gyz

Rep

ublic

(201

1)0.

240.

451.

081.

470.

541.

371.

94L

ao P

DR

(200

8)2.

102.

202.

442.

602.

362.

943.

29M

alay

sia (2

009)

0.12

0.75

4.75

7.26

1.32

7.64

11.1

3M

aldi

ves (

2004

)0.

000.

010.

050.

080.

020.

070.

10M

icro

nesia

, Fed

erat

ed S

tate

s

of, u

rban

(200

0)0.

010.

010.

010.

010.

010.

010.

01

Nep

al (2

010)

6.63

7.35

9.07

10.1

97.

9110

.87

12.7

1Pa

kist

an (2

008)

35.5

940

.61

52.5

660

.52

44.4

965

.51

78.6

1Pa

pua

New

Gui

nea

(199

6)1.

731.

811.

992.

102.

002.

542.

83Ph

ilipp

ines

(200

9)17

.20

20.2

728

.20

33.1

523

.68

37.7

145

.54

Sri L

anka

(200

9)0.

621.

294.

095.

611.

965.

848.

25Ta

jikist

an (2

009)

0.45

0.76

1.54

2.07

0.87

1.90

2.62

Tha

iland

(201

0)0.

321.

207.

4414

.04

1.90

14.3

824

.61

Tim

or- L

este

(200

7)0.

390.

400.

440.

460.

430.

510.

56Tu

rkm

enist

an (1

998)

1.10

1.22

1.50

1.68

1.33

1.86

2.18

Vie

t Nam

(200

8)14

.79

17.3

0 24

.36

28.9

9 19

.12

30.4

8 37

.70


45

Cou

ntry

$44,

med

ian,

10

0%$4

4, m

edia

n,

90%

$44,

med

ian,

66

%$4

4, m

edia

n,

50%

$44,

mea

n,

90%

$44,

mea

n,

66%

$44,

mea

n,

50%

Arm

enia

(201

0)0.

140.

260.

590.

800.

320.

771.

09A

zerb

aija

n (2

008)

0.03

0.13

0.95

1.80

0.18

1.58

2.66

Ban

glad

esh

(201

0)82

.77

82.0

780

.37

79.2

284

.78

89.3

992

.30

Bhu

tan

(201

2)0.

020.

050.

150.

200.

070.

210.

29C

ambo

dia

(200

9)3.

904.

245.

045.

574.

616.

267.

23F

iji (2

008)

0.08

0.12

0.21

0.27

0.15

0.30

0.39

Geo

rgia

(201

0)1.

001.

131.

441.

641.

231.

752.

07In

done

sia, R

ural

(201

1)27

.95

31.6

339

.79

45.1

434

.62

49.1

658

.02

Indo

nesia

, Urb

an (2

011)

30.1

633

.90

42.3

347

.55

37.9

554

.52

63.9

4K

azak

hsta

n (2

009)

0.01

0.09

1.09

2.87

0.12

1.98

4.06

Kyr

gyz

Rep

ublic

(201

1)0.

460.

701.

231.

610.

801.

552.

08L

ao P

DR

(200

8)2.

702.

742.

832.

892.

883.

293.

54M

alay

sia (2

009)

0.23

0.96

5.04

7.46

1.70

7.90

11.2

9M

aldi

ves (

2004

)0.

010.

020.

060.

080.

030.

080.

11M

icro

nesia

, Fed

erat

ed S

tate

s

of, u

rban

(200

0)0.

010.

010.

010.

010.

010.

010.

01

Nep

al (2

010)

9.16

9.62

10.6

911

.37

10.1

512

.39

13.7

8Pa

kist

an (2

008)

54.7

357

.81

65.0

269

.67

61.6

177

.24

86.8

0Pa

pua

New

Gui

nea

(199

6)2.

062.

092.

192.

242.

262.

692.

93Ph

ilipp

ines

(200

9)23

.12

25.8

331

.93

35.6

928

.77

40.8

147

.55

Sri L

anka

(200

9)1.

472.

574.

756.

203.

226.

608.

79Ta

jikist

an (2

009)

0.86

1.17

1.82

2.29

1.28

2.19

2.84

Tha

iland

(201

0)0.

671.

828.

6615

.11

2.67

15.8

025

.55

Tim

or- L

este

(200

7)0.

510.

510.

520.

520.

530.

580.

61Tu

rkm

enist

an (1

998)

1.41

1.49

1.69

1.82

1.60

2.04

2.30

Vie

t Nam

(200

8)20

.39

22.7

428

.29

31.9

624

.57

34.3

340

.40

Not

e:

* T

he h

eadi

ng o

f ea

ch c

olum

n in

dica

tes w

hich

pov

erty

line

is u

sed

($38

or $

44),

whi

ch o

ther

indi

cato

r is w

eigh

ted

(med

ian

or m

ean)

and

w

hich

wei

ght i

s giv

en to

the

abso

lute

pov

erty

line

. The

com

puta

tions

wer

e ba

sed

on th

e Sh

orro

cks a

nd W

an (2

009)

app

roac

h.


46

Tabl

e 2.

3 Po

vert

y ga

p ra

tio in

eac

h co

untr

y, d

epen

ding

on

the

weig

htin

g sc

hem

e*

Cou

ntry

$38,

med

ian,

10

0%$3

8, m

edia

n,

90%

$38,

med

ian,

66

%$3

8, m

edia

n,

50%

$38,

mea

n,

90%

$38,

mea

n,

66%

$38,

mea

n,

50%

Arm

enia

(201

0)0.

000.

010.

030.

050.

010.

050.

08A

zerb

aija

n (2

008)

0.00

0.00

0.01

0.04

0.00

0.03

0.07

Ban

glad

esh

(201

0)0.

110.

110.

120.

130.

120.

150.

17B

huta

n (2

012)

0.00

0.01

0.04

0.07

0.01

0.07

0.12

Cam

bodi

a (2

009)

0.03

0.04

0.07

0.09

0.05

0.10

0.14

Fiji

(200

8)0.

010.

020.

050.

080.

020.

090.

15G

eorg

ia (2

010)

0.06

0.07

0.11

0.13

0.08

0.14

0.18

Indo

nesia

Rur

al (2

011)

0.02

0.03

0.06

0.08

0.04

0.09

0.13

Indo

nesia

Urb

an (2

011)

0.03

0.04

0.08

0.10

0.06

0.13

0.17

Kaz

akhs

tan

(200

9)0.

000.

000.

010.

030.

000.

010.

05K

yrgy

z R

epub

lic (2

011)

0.01

0.02

0.04

0.07

0.02

0.06

0.10

Lao

PD

R (2

008)

0.09

0.10

0.11

0.12

0.11

0.15

0.18

Mal

aysia

(200

9)0.

000.

010.

050.

090.

010.

090.

16M

aldi

ves (

2004

)0.

000.

010.

040.

070.

010.

070.

12M

icro

nesia

, Fed

erat

ed S

tate

s

of, u

rban

(200

0)0.

160.

180.

210.

230.

200.

290.

33

Nep

al (2

010)

0.05

0.06

0.09

0.10

0.07

0.11

0.14

Paki

stan

(200

8)0.

030.

040.

060.

080.

050.

090.

12Pa

pua

New

Gui

nea

(199

6)0.

120.

130.

150.

160.

150.

220.

26Ph

ilipp

ines

(200

9)0.

040.

050.

090.

110.

060.

140.

18Sr

i Lan

ka (2

009)

0.00

0.01

0.04

0.06

0.01

0.07

0.11

Tajik

istan

(200

9)0.

010.

020.

050.

070.

020.

060.

10T

haila

nd (2

010)

0.00

0.00

0.02

0.05

0.01

0.05

0.11

Tim

or- L

este

(200

7)0.

090.

090.

110.

110.

100.

130.

16Tu

rkm

enist

an (1

998)

0.07

0.08

0.11

0.13

0.09

0.15

0.18

Vie

t Nam

(200

8)0.

030.

050.

070.

100.

050.

100.

14


47

Cou

ntry

$44,

med

ian,

10

0%$4

4, m

edia

n,

90%

$44,

med

ian,

66

%$4

4, m

edia

n,

50%

$44,

mea

n,

90%

$44,

mea

n,

66%

$44,

mea

n,

50%

Arm

enia

(201

0)0.

140.

110.

090.

090.

110.

100.

12A

zerb

aija

n (2

008)

0.14

0.10

0.06

0.07

0.09

0.07

0.09

Ban

glad

esh

(201

0)0.

230.

220.

200.

190.

220.

220.

22B

huta

n (2

012)

0.14

0.11

0.09

0.10

0.10

0.11

0.14

Cam

bodi

a (2

009)

0.16

0.15

0.14

0.14

0.15

0.16

0.18

Fiji

(200

8)0.

140.

120.

110.

120.

120.

140.

18G

eorg

ia (2

010)

0.18

0.17

0.17

0.17

0.18

0.19

0.21

Indo

nesia

Rur

al (2

011)

0.15

0.14

0.13

0.13

0.14

0.15

0.17

Indo

nesia

Urb

an (2

011)

0.16

0.15

0.15

0.15

0.16

0.18

0.21

Kaz

akhs

tan

(200

9)0.

140.

100.

050.

050.

090.

050.

07K

yrgy

z R

epub

lic (2

011)

0.14

0.12

0.11

0.11

0.12

0.12

0.14

Lao

PD

R (2

008)

0.21

0.20

0.19

0.18

0.21

0.21

0.22

Mal

aysia

(200

9)0.

140.

090.

080.

110.

080.

120.

17M

aldi

ves (

2004

)0.

140.

110.

090.

100.

100.

110.

14M

icro

nesia

, Fed

erat

ed S

tate

s

of, u

rban

(200

0)0.

270.

270.

260.

260.

280.

320.

35

Nep

al (2

010)

0.18

0.17

0.16

0.15

0.17

0.18

0.19

Paki

stan

(200

8)0.

160.

150.

140.

130.

150.

160.

16Pa

pua

New

Gui

nea

(199

6)0.

240.

230.

220.

220.

240.

270.

30Ph

ilipp

ines

(200

9)0.

160.

160.

150.

160.

160.

190.

22Sr

i Lan

ka (2

009)

0.14

0.12

0.10

0.10

0.11

0.12

0.14

Tajik

istan

(200

9)0.

140.

120.

110.

110.

120.

120.

13T

haila

nd (2

010)

0.14

0.10

0.07

0.08

0.09

0.09

0.13

Tim

or- L

este

(200

7)0.

200.

200.

180.

170.

200.

200.

21Tu

rkm

enist

an (1

998)

0.19

0.18

0.18

0.17

0.19

0.20

0.22

Vie

t Nam

(200

8)0.

160.

150.

140.

140.

16

Not

e:

* T

he h

eadi

ng o

f ea

ch c

olum

n in

dica

tes w

hich

pov

erty

line

is u

sed

($38

or $

44),

whi

ch o

ther

indi

cato

r is w

eigh

ted

(med

ian

or m

ean)

and

w

hich

wei

ght i

s giv

en to

the

abso

lute

pov

erty

line

. The

com

puta

tions

wer

e ba

sed

on th

e Sh

orro

cks a

nd W

an (2

009)

app

roac

h.



5 CONCLUSIONS

This chapter builds on the previous work of Chakravarty et al. (2015) and estimates measures of poverty such as the headcount ratio and the poverty gap index when an absolute poverty line is adjusted to take account of the existence of reference groups. Given the scarcity of available data on ref-erence groups it was assumed that either the median or the mean income would be the reference income but several scenarios were considered with different weights for the absolute poverty line ($1.25 or $1.45 per day) and the reference income (either the mean or the median). This empirical analysis covered many Asian countries, generally around the year 2010. Given the well- known asymmetry of an income distribution, the adjust-ment of the poverty line was evidently higher when the reference income was the mean rather than the median, and the adjusted headcount ratios were clearly higher when the absolute poverty line was $44 rather than $38 a month. This chapter presented the results of only four weighting schemes (giving weights of 100 percent, 90 percent, 66 percent and 50 percent to the absolute poverty line); other weights can easily be introduced. The choice of these weights should clearly be guided by the empirical evidence about the importance individuals give to the incomes of others, and by budgetary and political constraints that are faced by policy makers because increasing the number of poor has financial as well as political consequences.

NOTE

1. The authors are grateful to Iva Sebastian- Samanieo of the Asian Development Bank for helping them with the computations based on the Shorrocks–Wan algorithm.

REFERENCES11*

Aristotle, Politics, Kitchener, ON: Batoche Books.Atkinson, A.B. and F. Bourguignon (2001), ‘Poverty and inclusion from a world

perspective’, in J. Stiglitz and P.- A. Muet (eds), Governance, Equity and Global Markets, Oxford: Oxford University Press, pp. 179–92.

Chakravarty, S.R. (2015), Inequality, Polarization and Conflict: An Analytical Study, New York: Springer.

Chakravarty, S.R., N. Chattopadhyay, Z. Nissanov and J. Silber (2015), ‘Reference groups and the poverty line: an axiomatic approach with an empirical illustra-tion’, UNU- WIDER Working Paper 2015/002, United Nation University’s World Institute for Development Economics Research, Helsinki.

* The Asian Development Bank recognizes China as the People’s Republic of China.



Clark, A., C. Senik and K. Yamada (2013), ‘The Joneses in Japan: income compari-sons and financial satisfaction’, Discussion Paper No. 866, Institute of Social and Economic Research, Osaka University, Japan.

Clark, A.E. and A.J. Oswald (1996), ‘Satisfaction and comparison income’, Journal of Public Economics, 61 (3), 359–81.

Clark, A.E. and C. Senik (2009), ‘Who compares to whom? The anatomy of income comparisons in Europe’, Economic Journal, 120 (544), 573–94.

Clark, A.E., P. Frijters and M.A. Shields (2008), ‘Relative income, happiness and utility: an explanation for the Easterlin paradox and other puzzles’, Journal of Economic Literature, 46 (1), 95–144.

Deaton, A. (2001), ‘Counting the world’s poor: problems and possible solutions’, World Bank Research Observer, 16 (2), 125–48.

Deaton, A. (2010), ‘Price indexes, inequality, and the measurement of world poverty’, American Economic Review, 100 (1), 3–34.

Duesenberry, J.S. (1949), Income, Saving and the Theory of Consumer Behavior, Cambridge, MA: Harvard University Press.

Fafchamps, M. and F. Shilpi (2008), ‘Subjective welfare, isolation and relative con-sumption’, Journal of Development Economics, 86 (1), 43–60.

Ferrer- i- Carbonell, A. (2005), ‘Income and well- being: an empirical analysis of the comparison income effect’, Journal of Public Economics, 89 (5–6), 997–1019.

Han, X. (2014), ‘Estimating an Asian poverty line’, mimeo, Asian Development Bank, Manila.

Hirschman, A.O. and M. Rothschild (1973), ‘The changing tolerance for income inequality in the course of economic development’, Quarterly Journal of Economics, 87 (4), 544–66.

Kingdon, G. and J. Knight (2007), ‘Community comparisons and subjective well- being in a divided society’, Journal of Economic Behavior and Organization, 64 (1), 69–90.

Knight, J., L. Song and R. Gunatilaka (2009), ‘Subjective well- being and its deter-minants in rural China’, China Economic Review, 20 (4), 635–49.

Lelkes, O. and K. Gasior (2011), ‘Income poverty in the EU situation in 2007 and trends (based on EU- SILC 2005–2008)’, Policy Brief, European Centre for Social Welfare Policy and Research, Vienna.

Powdthavee, N. (2009), ‘How important is rank to individual perception of eco-nomic standing? A within- community analysis’, Journal of Economic Inequality, 7 (3), 225–48.

Ravallion, M. and S. Chen (2011), ‘Weakly relative poverty’, Review of Economics and Statistics, 93 (4), 1251–61.

Ravallion, M. and S. Chen (2012), ‘More relatively- poor people in a less absolutely- poor world’, World Bank, Policy Research Working Paper 6114.

Ravallion, M., S. Chen and P. Sangraula (2009), ‘Dollar a day revisited’, World Bank Economic Review, 23 (2), 163–84.

Runciman, W.G. (1966), Relative Deprivation and Social Justice, Berkeley, CA: University of California Press.

Sen, A. (2000), ‘Social exclusion: concept, application, and scrutiny’, Social Development Papers No. 1, Office of Environment and Social Development, Asian Development Bank, Manila.

Senik, C. (2004), ‘When information dominates comparison. Learning from Russian subjective panel data’, Journal of Public Economics, 88 (9–10), 2009–133.



Senik, C. (2008), ‘Ambition and jealousy. income interactions in the “Old Europe” versus the “New Europe” and the United States’, Economica, 75 (299), 495–513.

Senik, C. (2009), ‘Income distribution and subjective happiness: a survey’, OECD Social, Employment and Migration Working Papers, No. 96, OECD, Paris.

Shorrocks, A. and G. Wan (2009), ‘Ungrouping income distributions. synthesizing samples for inequality and poverty analysis’, in K. Basu and R. Kanbur (eds), Arguments for a Better World. Essays in Honor of Amartya Sen, vol. 1, Oxford: Oxford University Press, pp. 414–34.

Smith, A. (1937), An Inquiry into the Nature and Causes of the Wealth of Nations, New York: Modern Library, pp. 821–2.

Van Praag, B. (2011), ‘Well- being inequality and reference groups: an agenda for new research’, Journal of Economic Inequality, 9 (1), 111–27.


PART II

Poverty and Vulnerability in Asia


53

3. Concepts and measurement of vulnerability to poverty and other issues: a review of literatureTomoki Fujii1

1 I NTRODUCTION

There has been a surge in interest in vulnerability analysis among develop-ment economists in recent years. For example, the number of academic journal articles indexed in EconLit and containing the word ‘vulnerability’ in the title was only 76 in the last half of the twentieth century. The correspond-ing number between 2001 and 2013 was 444. A sizable fraction of these is also related to poverty.2 The purpose of this chapter is to review this growing body of literature on vulnerability. We primarily focus on vulnerability to poverty, but we also discuss its relationship with other vulnerability studies.

The trend of increasing interest in vulnerability is not surprising. While progress has been uneven, the developing world has witnessed a massive reduction in extreme poverty since the end of the Second World War. The fight against poverty has been particularly successful in East and Southeast Asia. However, the threat of poverty has not yet become a thing of the past. This remains true even in relatively successful regions such as East and Southeast Asia. A noticeable fraction of people remain below the poverty line and even those who are above the poverty line can be pulled back into poverty when they are hit by a large negative shock such as a natural disaster or an economic crisis.

Vulnerability is a topic of interest on its own but it also has important implications for economic efficiency and long- run welfare of households. Those who are under the constant threat of poverty are often observed to choose to make safer, but less lucrative, investments than those who are free from the fear of poverty. As pointed out by Eswaran and Kotwal (1990), when the poor have less access to credit than the rich, the former may engage in less risky and less profitable behavior than the latter, even if everyone has the same preference. Therefore, in the presence of credit constraints, bad shocks can lead to a poverty trap (Morduch 1994).



Empirical evidence also supports this possibility. For example, in Tanzania, households with low asset holdings allocate more of their land to low- risk crops (Dercon 1996) and richer households make a substantial investment in cattle, which is profitable but lumpy, whereas poorer house-holds specialize in low- risk low- return activities (Dercon 1998). In India, Rosenzweig and Binswanger (1993) find that uninsured weather risk is a significant cause of lower efficiency and lower average income.

Vulnerability to poverty also affects the accumulation of assets. On the one hand, the lack of credit access can be mitigated by accumulating assets over time because the poor can sell assets at bad times and buy assets at good times to smooth consumption over time (Carter and Zimmerman 2000). On the other hand, when the poor face a survival constraint, they may respond to negative shocks by adjusting consumption to defend or smooth their asset value to ensure their survival (Zimmerman and Carter 2003). Therefore, it may be useful to look at asset holdings to assess the vulnerability of households.

We start our review from the discussion on vulnerability to poverty in the next section. In section 3, we provide a brief overview of other areas of vulnerability studies. One important area is vulnerability to climate change. Although this body of literature has grown largely independent of the studies on vulnerability to poverty, it is interlinked with and arguably becom-ing increasingly more important to the analysis of vulnerability to poverty. Therefore, we briefly review the vulnerability issue related to climate change and its significance in the analysis of vulnerability to poverty. We also review several other aspects of vulnerability, including assets and nutrition. Section 4 offers some discussions, including policy relevance of vulnerability studies.

2 V ULNERABILITY TO POVERTY

We begin this section with discussion on the concepts and measurements of vulnerability to poverty in section 2.1. These are important topics for two reasons. First, there has not yet been a universally accepted definition of vulnerability. Therefore, it is useful to review different formalizations of vulnerability to highlight the similarities and differences of vulnerability concepts proposed by various authors.

Second, measurement is important for the understanding of the situa-tion and sources of vulnerability, which, in turn, is essential for formulating the policies to remove or reduce the risks and impacts of negative shocks.

In section 2.2, we provide a survey on empirical applications of the concepts and measurements of vulnerability to poverty discussed in section 2.1 and other related studies.


Concepts and measurement of vulnerability to poverty 55

2.1 C oncepts and Measurements

There have been a number of studies that aim to conceptualize and measure vulnerability to poverty. Partly using the terminology of Calvo and Dercon (2005), we classify the approaches to define vulnerability measures into the following three categories: the welfarist approach, the expected poverty approach, and the axiomatic approach. As discussed subsequently, these categories are not mutually exclusive.

Welfarist approachSome of the earlier studies such as Ligon and Schechter (2003) and Elbers and Gunning (2003) develop a measure of vulnerability based on explicit welfare foundations. In Ligon and Schechter (2003), vulnerability vi for individual i is defined as3

ui ; ui(z) − E[ui(ci(w))] (3.1)

where ui is the instantaneous utility function, E[∙] is the expectation opera-tor, z(≥ 0) is the threshold- level certainty- equivalent consumption below which the individual is deemed vulnerable, and ci(w)(≥ 0) is the consump-tion expenditure per capita for individual i that depends on the state of the world w([ Ω) for state space Ω.

Notice that z corresponds to the poverty line in the analysis of poverty in the static framework. Therefore, we also interpret z as the poverty line below when it is appropriate to do so. Also, while we interpret ci as the consumption expenditure per capita, most of our presentation remains unchanged even if it is interpreted as income or other cardinal and observ-able measure of individual welfare.

Ligon and Schechter (2003) decompose vulnerability into poverty, aggregate risk, idiosyncratic risk, and unexplained risk based on a model of linear consumption equation, where poverty in their study refers to the difference between ui(z) and ui(E[ci(w)]). Applying this decomposition method to a panel data set in Bulgaria, they find that poverty is the largest single component of vulnerability, accounting for more than half of the observed vulnerability. They also find that aggregate risk is more impor-tant than idiosyncratic risk, though unexplained risk is much larger than these two.

The analytical framework of Ligon and Schechter (2003) is static. In contrast, Elbers and Gunning (2003) define vulnerability in the frame-work of a Ramsey model with income and asset shocks. Their measure of vulnerability has a form similar to equation (3.1), but ui is taken as the welfare of the individual, which is the sum of the present- discounted



instantaneous utility over an infinite time horizon. Unlike most other studies discussed in this chapter, they explicitly incorporate the future streams of consumption and deal with the effects of risk on both the mean consumption and the volatility of consumption around the mean. The latter point is particularly important because the exposure to risks and the risk- coping strategies available to the individual affect not only the current volatility of consumption but also the investment decision and thus the future streams of consumption.

Their analytical framework allows for the explicit distinction between ex ante and ex post effects of risk, where the former arises from the anticipa-tion of the risk that the individual is going to face and the latter arises from the shock that has been realized. They apply their model to a panel data for smallholders in Zimbabwe and show that the failure to account for the distinction between ex ante and ex post effects may lead to large errors in the estimates of chronic and transient poverty.

One obvious drawback of these welfarist measures is that they require explicit specification of the utility or welfare function. While both Ligon and Schechter (2003) and Elbers and Gunning (2003) use the constant relative risk aversion utility function, this is clearly not the only choice. Further, the estimation of the coefficient of relative risk aversion often poses a challenge.

Expected poverty approachAnother approach to vulnerability to poverty is to regard vulnerability as expected poverty. More precisely, given the current condition, vulnerability measures or relates to how likely it is for the individual to fall into poverty in a given time horizon. Thus, the time horizon is inherently relevant in the expected poverty approach. This point contrasts with Elbers and Gunning (2003) mentioned above, who consider the infinite time horizon. For the benefit of simplicity, we choose to discuss the expected poverty approach in a static framework by fixing the time horizon, even though the choice of time horizon is important. It should be noted here that the consumption measure used in the definition of vulnerability always refers to the ex ante consump-tion, whereas it is the ex post (realized) consumption in the case of poverty.

The seminal idea of using expected poverty measures to analyze vulner-ability can be seen in Ravallion (1988), who analyzes the marginal impact of a random variable influencing the individual welfare on the poverty in society. He proposes a decomposition of the marginal impact into tran-sient and chronic (persistent) poverty, which respectively refer to the mar-ginal impact on extensive and intensive margins.

Chaudhuri et al. (2002) and Suryahadi and Sumarto (2003) formulate vulnerability as the probability of consumption per capita falling below the



poverty line z given the current set of observable individual characteristics Xi for individual i. They essentially define vulnerability vi for individual i in the following manner

ui ; Pr(ci < z | Xi), (3.2)

where ci is the (ex ante) consumption per capita. It should be noted that this is different from poverty, because the poor households are those house-holds whose (ex post) consumption per capita falls below the poverty line.

To operationalize equation (3.2), both Chaudhuri et al. (2002) and Suryahadi and Sumarto (2003) assume that the logarithmic consumption is conditionally linear such that

lnci = Xib + ei, (3.3)

where b is a vector of coefficients and ei is an idiosyncratic error term.Further, denoting the standard deviation of ei by si, which may be

heteroskedastic across individuals, and assuming that ei is normally dis-tributed, eq. (2) reduces to

vi 5 Fa lnz 2 Xib

sib, (3.4)

where �(∙) is the cumulative distribution function for the standard normal distribution.

Replacing the parameters (b, si) with their estimates in equation (3.4), we obtain a measure of vulnerability. Individuals can be then classified into high vulnerability (ui ≥ /) and low vulnerability (ui < /) groups, where the threshold value of vulnerability is denoted by /.

An obvious question that arises here is how to choose /. Suryahadi and Sumarto (2003) choose / = 0.5. While this choice is somewhat arbitrary, some justifications can be made. As Pritchett et al. (2000) argue, 50- 50 odds has a nice focal point and it makes intuitive sense to say an individual is vulnerable if he or she faces even odds or worse. Second, if an individual is just at the poverty line and faces a symmetric shock with a zero mean, this individual has a vulnerability of 0.5.4 It should be noted that the definition of vulnerability in Pritchett et al. (2000) is slightly different from that of Suryahadi and Sumarto (2003), because the former defines vulnerability as a risk of falling into poverty at least in one period in the next n periods being greater than the threshold probability level. However, the justifica-tions for choosing / = 0.5 explained above are nevertheless applicable to Suryahadi and Sumarto (2003).



Suryahadi and Sumarto (2003) propose to further categorize individuals according to Table 3.1. This table helps us to understand the difference between poverty and vulnerability. They classify the vulnerable groups into high mean consumption (E [c] ≥ z) and low mean consumption groups. For example, B corresponds to a group of individuals who are poor and have high mean consumption and high vulnerability in their categorization. Note that even when the mean consumption is high, an individual may still fall below the poverty line for a given period because of a negative idiosyncratic shock. Such a possibility is higher for individuals with high vulnerability.

Using this framework, Suryahadi and Sumarto (2003) divide the poor (A + B + C) into chronic and transient poor, which are respectively A (poor with expected consumption below poverty line) and B + C (poor with expected consumption above poverty line) in Table 3.1.5 They also divide the high vulnerability group (A + B + D + E) into two groups, one characterized by low expected consumption (A + D) and the other charac-terized by high variability of the consumption group (B + E). They define the total vulnerability group (A + B + C + D + E) as those individuals who are either poor and/or in the high vulnerability group.

Suryahadi and Sumarto (2003) apply this framework to Indonesia. They first describe the profile of the poor and vulnerable individuals in Indonesia and then compare the change in poverty and vulnerability between 1996 and 1999 across geographic locations, sector of individual head’s occupation, education level, and gender. They find that the vulner-ability to poverty among Indonesian individuals after the Asian financial crisis has unambiguously increased and the proportion of the total vulner-able group almost doubled.

Kamanou and Morduch (2004) also use expected poverty to measure vulnerability, though they take vulnerability as the difference between the expected poverty in the future and the current poverty. They use a Monte Carlo method to simulate the possible future outcomes for indi-viduals based on their observed characteristics and observed consumption fluctuations of similar individuals. Their measure, however, can be difficult to interpret because it could take a negative value.

Table 3.1 Poverty and vulnerability categories by Suryahadi and Sumarto (2003)

Poor Non- poor

E [c] < z, u ≥ / A DE [c] ≥ z, u ≥ / B EE [c] ≥ z, u < / C F



Christiaensen and Subbarao (2005) define vulnerability as the expected value of the Foster–Greer–Thorbecke (FGT) poverty measure due to Foster et al. (1984), which is given as follows:

vi,g 5 E c az 2 ci

z bg# Ind(ci , z) d 5 3

z

0c z 2 ci

z d gf (ci)dci, (3.5)

where lnd(∙) is the indicator function, which is equal to one if the argu-ment is true and zero otherwise, f(∙) is the probability density function for consumption, and g is a parameter for the FGT measure. Because f(∙) is not known in general, we need to make additional assumptions to calcu-late vulnerability based on equation (3.5). As with Chaudhuri et al. (2002) and Suryahadi and Sumarto (2003), they estimate the parameters for the conditional mean and the variance of ci. Hence, the vulnerability measure considered by Christiaensen and Subbarao (2005) can be thought of as an extension of equation (3.4), because equation (3.5) reduces to equa-tion (3.4) under a log- linearity condition in equation (3.3) and normality assumption when g = 0.

Unlike Chaudhuri et al. (2002) and Suryahadi and Sumarto (2003), Christiaensen and Subbarao (2005) utilize a repeated cross- sectional data that is augmented with historical information on the shocks. They find that individuals in arid areas, who experience large rainfall volatility, appear more vulnerable than those in non- arid areas in Kenya.

The vulnerability studies mentioned above typically use either some form of consumption or income regressions to estimate parameters such as b in equation (3.3). It is not immediately clear, though, how vulnerability estimated via a regression approach actually matches the expected poverty. Using a multi- period panel data for rural areas in the People’s Republic of China (PRC), Zhang and Wan (2009) attempt to answer how accurately vulnerability can be computed.6

To this end, they define vulnerability as the probability of being in poverty in the future and calculate vulnerability assuming that income is log- normally distributed. Exploiting the panel structure, they evaluate the precision of the estimated vulnerability by comparing the vulnerability computed from earlier rounds of data against the actual observed poverty based on later rounds. They find that the precision of estimated vulner-ability depends on / and the poverty line. They obtain a more precise esti-mate under the US$2 per day poverty line than the US$1 per day poverty line. They also argue that the choice of / = 0.5 is appropriate because the vulnerability under this threshold appears to be more precise than other choices they tried.

It is worth pointing out here that the expected poverty measure can be



considered a welfarist measure by treating the individual- level poverty measure as the individual utility. However, these two types of measures differ in the following two aspects (Christiaensen and Subbarao 2005). First, the welfarist approach explicitly considers the risk preference, while the expected poverty measure does not. Second, the former considers the entire distribution of c including the states in which c exceeds z, whereas the latter only focuses on what is below z.

It is also worth noting that the mathematical expression of the expected poverty measure in the form of equation (3.5) is similar to the total poverty, or the sum of the transient poverty and chronic poverty, as pro-posed by Jalan and Ravallion (1998, 2000). They define total poverty to be simply the poverty averaged over all periods, whereas chronic poverty is at the level of consumption averaged over all periods. Therefore, transient poverty, which is the difference between total poverty and chronic poverty, comes from the nonlinearity of poverty with respect to consumption in their definition.7

To further elucidate the relationship between vulnerability and chronic/transient poverty, suppose that the poverty measure of interest is the FGT measure with parameter g and the vulnerability measure is equation (3.5). Consider a situation where vulnerability coincides with total poverty.8 Then, the chronic poverty CPi,g and transient poverty TPi,g can be written as follows:

CPi,g 5 az 2 E [ci ]z bg

,

TPi,g = vi,g − CPi,g.

This result also points to the fact that high vulnerability to poverty may be due to low mean consumption (or high chronic poverty), high consump-tion variability (or high transient poverty), or a combination of both. Therefore, this result qualitatively relates to Table 3.1.

Axiomatic approachInstead of basing the definition of vulnerability on utility or poverty at the individual level, it is also possible to derive a vulnerability measure from a set of axioms, which lists the properties that an ideal vulnerability measure would satisfy. Calvo and Dercon (2005, 2007, 2013) make seminal contri-butions to the derivation of vulnerability measures from a set of axioms. Our discussion on axiomatic approach is mainly based on these studies. We then discuss the relationship between these studies and others.



Because we hereafter focus on a particular individual, we drop the subscript i for the time being to simplify the notation. We also intro-duce additional notations to present the axioms formally. There are K possible states in Ω such that Ω = {s1, . . ., sK}. Further, we denote the consumption in state sk by ck ; c(sk) and the probability that the state sk arises by pk(; Pr(w = sk)) for k [ {1, . . ., K}. We denote the K- vectors of c and p by c ; (c1, . . ., cK) and p ; (p1, . . ., pK), respectively. We define the consumption right- censored at the poverty line by c|k ; min(ck,z) and its vector analogue by c| ; (c|1,. . .,c|K) . We denote the k- th unit vector in a K- dimensional space by ek, whose elements are all zero except for the k- th element, which is one. For example, e1 = (1, 0, . . ., 0).

Calvo and Dercon (2005, 2013) consider a class of vulnerability meas-ures that can be written as a function of z, c, and p such that vulnerability measures in this class can be written as v(z, c, p). One assumption that is implicit here is that the poverty line is common across states. We also main-tain this assumption here to avoid unnecessary complications. Calvo and Dercon (2005, 2013) require the following properties as basic properties of individual vulnerability measures.

Axiom 1 (Focus): For every (z, c, p), u satisfies v(z,c,p) 5 v(z,c|,p) .

Axiom 2 (Symmetry): For every (z,c|, p) and K × K permutation matrix B, u satisfies

v(z, c|, p)5v(z, Bc|, Bp) .

Axiom 3 (State- dependent effect of outcomes): Suppose that we have 1 ≤ k ≤ K, c|k

a 5 c|kb . 2 d, pk

apkb 2 0. Then, pk

a 5 pkb if and only if

v(z,c|a,pa) 2v(z,c|a 1 dek,pa) 5v(z, c|b,pb) 2v(z,c|b 1dek,pb) . (3.6)

Axiom 4 (Probability transfer): Suppose that we have 1 ≤ k, l ≤ K, a ≠ b, pk ≥ d > 0, and 1 − d ≥ pl ≥ 0.9 For every (z,c|,p) , u satisfies

v(z,c|,p)v(z,c|, p 2 dek 1 del) if and only if c|lc|k.

Axiom 5 (Risk sensitivity): For every (z, c|,p) , u satisfies

v(z, c|,p) $ v(z, c 1K, p) , for c ; pTc|, (3.7)



where 1K is a K- vector of ones and the equation is held with equality if and only if c|k 5 c for all 1 ≤ k ≤ K.10

Axiom 6 (Scale invariance): For every (z,c|,p) and l > 0, u satisfies

v(z, c|, p) 5 v(lz,lc|,p) .

Axiom 7 (Differentiability): v(z, c|,p) is twice differentiable in c|.

Axiom 1 states that the change in consumption measure in a particular state makes no difference so long as it is above the poverty line. In other words, the outcome of interest is not the consumption itself but the censored consumption. From a technical perspective, this axiom is not essential because the results presented below including equations (3.9) and (3.10) hold by appropriately replacing c| with c.

It is worth pointing out that welfarist measures do not satisfy this axiom in general. This means that the possibility of severe destitution can be compensated by another state that is sufficiently good under the welfarist measures. Therefore, individuals are not necessarily deemed vulnerable, even in the presence of the possibility of severe destitution. This feature appears unattractive when we are concerned with vulnerability to poverty. Hence, we regard Axiom 1 as an essential requirement for our purpose.

Axiom 2 states that the states of the world can swap their indices without any impact on vulnerability. That is, only the censored consumption and probability in each state matter. Therefore, given c|k and pk, all states are treated equally.

To interpret axiom 3, imagine d > 0 such that equation (3.6) is positive. The ‘if ’ part of the axiom states that the probability of k- th state is the same if the reduction in vulnerability is the same for the same change in consumption in the k- th state (that is, from c|k

a (5 c|kb) to c|k

a 1 d(5 c|kb 1 d)).

The ‘only if ’ part requires that the change in vulnerability is the same if the probability of the k- th state is the same and the consumption in the k- th state changes in the same way.

Axiom 4 says that if the probability is hypothetically transferred from a good (bad) state, in which the censored consumption is high (low), to a bad (good) state, then the vulnerability would increase (decrease). Axiom 4 also implies that increases in vulnerability are monotonically related to decreases in consumption as long as outcomes are below the poverty line. Note that the expected poverty rate given in equation (3.2) fails to satisfy this axiom.

Axiom 5 requires that vulnerability is lower if the (stochastic) censored consumption is replaced with its expected value c. In this axiom, the risk



is implicitly taken as a probability transfer from the middle to the tails. That is, the right- hand side of equation (3.7) assumes that the probabilistic weight falls entirely on c, whereas the left- hand side spreads that weight away from the expected outcome towards the tails. The risk and vulner-ability are higher as a consequence.

Alternatively, Axiom 5 can be interpreted in the following manner. Define the certainty- equivalent consumption c* 5 c*(z,c|,p) by:

v(z,c|,p) 5 v(z,c*1K,p) . (3.8)

Thus, the certainty- equivalent consumption c* 5 c*(z,c|, p) is a fixed amount of consumption that gives rise to the same vulnerability. By axioms 4 and 5, we have c* , c. Therefore, if perfect insurance becomes available so that the individual gets the expected consumption for sure, the individual would be willing to pay up to c 2 c* as a premium to reduce its vulnerability.

Axiom 6 implies that the individual becomes neither more nor less vul-nerable when both the poverty line and consumption change by the same proportion. This makes intuitive sense, because this axiom requires that the vulnerability measure is not affected by the currency unit used for the poverty line and consumption.

Axiom 7 implies that small changes in consumption cause no abrupt reactions in u and the marginal impact of consumption on vulnerability is also smooth. Calvo and Dercon (2005) show that vulnerability measures satisfying axioms 1–7 can be written in the following form:

v(z,c,p) 5 E [�(q) ] 5 aK

k51pk�(c|k/z) , (3.9)

where q ; c|/z is the (random) censored consumption normalized by the poverty line, which necessarily lies on the unit interval, and �(∙) is a monotonically decreasing and convex function. We can interpret �(∙) as a state- dependent deprivation index because it tends to increase as ck falls when ck < z.

The expected FGT measure given in equation (3.5) fails to satisfy axiom 5 if g ≤ 1 because it means that the poor individuals are risk- neutral or risk- loving below the poverty line. If g > 1, the expected FGT measure satisfies all of axioms 1–6 (Calvo and Dercon 2005). However, the expected FGT measure with g > 1 is not without problems. As pointed out by Ligon and Schechter (2003), this implies that poor individuals are implicitly assumed to have increasing absolute risk aversion, which is at odds with empirical evidence.

To address this point and pin down the desirable vulnerability index,



Calvo and Dercon (2013) propose to require the following two additional axioms:

Axiom 8 (Normalization): If c = z1K, v(z, c, p) = 0 for all (z, p).

Axiom 9 (Constant relative risk sensitivity): For every l > 0 and (z, c, p), u satisfies

v(z, lc, p) = v(z, lc*1K, p)

Axiom 8 states that the vulnerability measure should be equal to zero if the individual’s consumption is equal to poverty line for sure. This axiom makes intuitive sense because the individual barely escapes from the threat of poverty in this case. Note that the welfarist measures generally do not satisfy this axiom.

Axiom 9 essentially states that if the consumption increases by the pro-portion l in all possible states of the world, then the certainty- equivalent consumption must also increase by the same proportion. Further, because the expected consumption also increases by the proportion l in this case, the ratio of the certainty- equivalent consumption to the expected con-sumption is independent of l. This requirement also addresses the short-comings of the expected FGT measure with g > 1 discussed above.

Calvo and Dercon (2013) show that the vulnerability measure u that sat-isfies Axioms 1–9 can be written as a multiple of the following expression:

v(z,c, p) 5e (1 2 E [qq ]) /q for q , 1 and q 2 0.2E [lnq ] for q 5 0

(3.10)

Note that the first and second cases above are the expected Chakravarty measure of poverty (Chakravarty 1983) and the expected Watts measure of poverty (Watts 1968), ignoring the factor q−1 in the first case. Therefore, the individual- level vulnerability measure axiomatically derived by Calvo and Dercon (2005, 2013) can be also regarded as an expected poverty measure.

As with Calvo and Dercon (2005, 2013), Dutta et al. (2011) also derive a vulnerability measure at the individual level from a set of axioms, which are: (i) decomposability; (ii) transferability; (iii) monotonicity of (future) consumption; (iv) monotonicity of current consumption; and (v) independence.11 It is worth noting that, unlike Calvo and Dercon (2005), Dutta et al. (2011) let the deprivation explicitly depend on both current and future consumption. Therefore, the critical difference between these two studies lie in axiom (iv) of the monotonicity of current consumption.

In Calvo and Dercon (2005), the current (ex post) consumption plays



no role in the measurement of vulnerability. However, Dutta et al. (2011) require that vulnerability can only either monotonically increase or decrease compared with the status quo when there is an increase in the current living standard. The monotonic increase is possible when, for example, individuals who enjoy higher current consumption find it hard to cope with negative shocks compared with current poor people because of the lack of previous experience of coping with poverty. On the other hand, the monotonic decrease is also possible when, for example, lower current consumption means the lack of assets and networks that individuals can count on at the time of distress. Also, axiom (iv) on the monotonicity of current consumption in Dutta et al. (2011) implies that their vulnerability measure is, in general, not an expected poverty measure unlike Calvo and Dercon (2005).

However, if vulnerability is assumed to be independent of the current standards of living, the axioms presented in Calvo and Dercon (2005, 2013) and Dutta et al. (2011) are strikingly similar. For example, axiom 4 implies axiom (iii) of monotonicity in consumption, which states that an increase in ck for a particular state k does not affect vulnerability ordering of two consumption- probability profiles, (c|a, pa) and (c|b, pb) . Similarly, axiom 5 is closely related to axiom (ii) of transferability, which states that the transfer of consumption from a bad state to a equally- likely good state increases vulnerability. Axiom 3 relates to axiom (i) of decomposability, which restricts vulnerability to be a expected deprivation function, and axiom (v) of independence, which requires that the vulnerability ordering of two consumption profiles for given probability profile is the same after consumption increases in a particular state.12

A study related to the above- mentioned studies is Chakravarty et al. (2015), who explore a partial ordering of vulnerability to poverty based on expected poverty measures. They find, among other things, that the condi-tion that situation a (ca, pa) is no more vulnerable than situation b (cb, pb) is equivalent to the condition that the deprivation function in each meager state k [ {k | 1 ≤ k ≤ Ka, ck < z} in situation a is obtained by a smoothing of the meager states in situation b.

Hardeweg et al. (2013) also propose a method that leads to a partial ordering of vulnerability. In their approach, two groups are compared by the first- , second- , and third- order stochastic dominance of consumption (or income) distribution up to a certain threshold such as the poverty line. When a higher- order stochastic dominance is used, it is more likely to be able to rank the two different groups but the set of vulnerability measures that is consistent with the ranking shrinks. This approach has an attraction that the comparison of vulnerability across groups does not depend on the (arbitrary) choice of the vulnerability measure.



The discussion thus far has been concerned with the individual- level vul-nerability to unidimensional poverty. However, the vulnerability measure given in equation (3.10) has been extended in at least two directions. The first direction is due to Calvo (2008), who extends to multidimensional poverty in a spirit similar to Calvo and Dercon (2005), even though its development is not fully based on a set of axioms. This extension is impor-tant because consumption poverty cannot possibly capture every relevant dimension of poverty.

Formally, the outcome (‘consumption’) in the j- th dimension for indi-vidual i is denoted by cij and the threshold- level outcome (‘poverty line’), below which the outcome is deemed ‘deprived’ by zj for 1 ≤ j ≤ J, such that we can define the multidimensional counterpart of qi by qij ; min(cij, zj) / zj. Dimension j has a weight gj where the sum of weights is equal to one. Calvo (2008) considers constant- elasticity- of- substitution aggregation across different outcomes such that the index of vulnerability to multidi-mensional poverty vMP

i is given by the following:

vMPi 512E c aaJ

j51gjqr

ijb ar d

with a [(0,1) and r[ [0,1]. (3.11)

Because qij does not exceed one, it is not possible to (fully) compensate a bad outcome in one dimension by a good outcome in another. Applying this index to a panel dataset in Peru for the dimensions of consumption and leisure, Calvo (2008) finds that the gap in the multidimensional vul-nerability between rural and urban areas tends to become larger as the substitutability between leisure and consumption decreases (that is, when r is lower). This is because the idiosyncratic shocks in rural areas exhibit stronger negative correlation than urban areas, which in turn means that rural areas (relative to urban areas) depend more heavily on rare positive shocks in both dimensions to escape from poverty as r gets lower.

The second direction of extension is due to Calvo and Dercon (2007, 2013), who consider vulnerability to poverty at an aggregate level. The reason why we may need a measure for the society is that a simple aggrega-tion of individual- level vulnerability may not be an appropriate measure for the society.13

To further elaborate on this point, we introduce some notations. Suppose that there are I individuals in the society. We denote the state- contingent consumption profile by a (K × I)- matrix C, whose i- th column vector is a K- vector ci of state- contingent consumption for individual i.

Now, consider a simple example with I = K = 3 and p = 1K / 3. We assume c = 0 means poor and c = 1 means non- poor. Now, consider the following two state- contingent consumption profiles:



Ca 5 £ 1 0 00 1 00 0 1

§ , Cb 5 £ 1 1 10 0 00 0 0

§ .

From each individual’s perspective, vulnerability is the same for these two profiles because each individual falls into poverty with probability 1/3. However, from the social perspective, they are not the same. In profile a, exactly one of the three individuals is poor in each of the three possible states. On the other hand, everyone is poor in state 1 and no one is poor in the two other states in profile b. Arguably, the latter situation is less desir-able because there is a catastrophic state in which everyone is poor.

Based on this idea, Calvo and Dercon (2013) propose a set of axioms for aggregate vulnerability similar to axioms 1–7. However, there are three important differences. First, axioms 3 and 4 must be modified to the case where everyone faces the same state- contingent, censored consumption. In other words, these axioms are focused on covariate risk in a world where the risk is fully shared in the society.

Second, they do not require axiom 6 for aggregate vulnerability. They instead require sensitivity to correlation, which requires avoidance of cata-strophic states. This alternative requirement is sufficient to secure that an increase in covariant risk raises vulnerability.

Finally, they require symmetry over individuals and replication invari-ance. The former states that all individuals are treated equally and the latter requires that population size plays no role. Calvo and Dercon (2013) have shown that these requirements are satisfied if and only if the aggregate measure of vulnerability u can be written as a positive multiple of the fol-lowing expression:

V(z, p,C) 51qa1 2E c aqI

i51q1/I

i bq d b, with q , 0. (3.12)

It is interesting to note that equation (3.12) becomes the expected value of the poverty measure proposed by Clark et al. (1981) when q is set equal to one, though this possibility is excluded by the condition q < 0. As we have seen, a number of measures derived from a set of axioms can also be interpreted as expected poverty measures.

Calvo and Dercon (2013) also compute various poverty and vulnerability statistics, including the FGT poverty measures, the average individual- level vulnerability measure in equation (3.10), and the aggregate vulnerability measure in equation (3.12) using a panel data survey from Ethiopia. Their finding underscores the importance of distinguishing between vulnerabil-ity and poverty, because their profiles can be very different.



2.2 Empirical Studies

In this subsection, we review a number of empirical studies on vulnerability to poverty. We start our discussion with the PRC, because there are multi-ple studies on vulnerability to poverty and several other related studies in the PRC. We then discuss the rest of Asia and the rest of the world.

PRCMcCulloch and Calandrino (2003), Zhang and Wan (2006), and Imai et al. (2010) study vulnerability to poverty at the household level in the PRC. All of these studies adopt an expected poverty approach and use equation (3.4) or a similar form to estimate vulnerability. However, the data source, geographic coverage, and focus of these studies are different.

Using a five- year panel data of rural Sichuan households for 1991–95, McCulloch and Calandrino (2003) investigate the factors that affect vul-nerability. They find that demographic characteristics, education, the value of assets, and location are important for vulnerability. They also find that some factors such as education and location are a significant determinant for transient poverty but not for chronic poverty.

Zhang and Wan (2006) analyze vulnerability in six rural districts of Shanghai between 2000 and 2004. They compare vulnerability across education levels and whether the share of income from agricultural activities exceeds the sample average in a given year. They find that low- education households are substantially more vulnerable than high- education households.

Imai et al. (2010) use a large repeated cross- sectional survey dataset for 1988, 1995, and 2002 collected under the Chinese Household Income Project and study the effect of a regressive tax system on poverty and vulnerability in rural PRC. They find that poverty and vulnerability have been significantly reduced during their study period in the PRC. The after- tax poverty and vulnerability dropped more than their before- tax counterparts, because the tax system has become less regressive, but the geographic disparity of poverty and vulnerability increased during the same period. Imai et al. (2010) also find that head’s education and access to electric power supply are found to be negatively associated with both poverty and vulnerability. On the other hand, a few factors, including farm land size and the share of the farm land irrigated, are associated with vul-nerability but not poverty.

As mentioned previously, vulnerability, defined as expected poverty, is closely related to chronic poverty and transient poverty. Using rural house-hold surveys, Jalan and Ravallion (1998) find that a substantial fraction of poverty in rural PRC is transient. Poverty regression results reported in



Jalan and Ravallion (2000) indicate that some factors such as demograph-ics and wealth are important for both chronic and transient poverty but other factors only matter for one of them.

The above- mentioned studies show that vulnerability to poverty is heterogeneous across households. Education and location appear to be among the factors that consistently emerge as the significant covariates of vulnerability to poverty in the PRC.

Asia outside the PRCCurrently, studies on vulnerability to poverty in Asia outside the PRC are limited. One notable exception, however, is Viet Nam. In addition to Hardeweg et al. (2013), discussed in the previous subsection, Imai et al. (2011a, 2011b) compute various vulnerability measures in Viet Nam. In Imai et al. (2011b), expected poverty measures for various ethnic groups in Viet Nam are calculated using equations (3.2) and (3.3). They find that households in an ethnic minority group are not only poor but also more vulnerable than those in an ethnic majority group such as Chinese and Kinh. Imai et al. (2011a) use the vulnerability measure calculated in this way as a regressor. They run a probit regression of future poverty as well as a multinomial logit regression of the poverty transition between current and future poverty. In both cases, vulnerability to poverty was found to be statistically significant.

Jha et al. (2010) analyze poverty and vulnerability in Tajikistan using a panel data set for 2004–05. They use the expected poverty approach to describe the profile of vulnerable households. Their analysis indicates that rural households tend to be poorer and more vulnerable than urban house-holds. They also adopt the vulnerability measure proposed by Ligon and Schechter (2003) to conduct a decompose analysis. Their analysis indicates that vulnerability comes mostly from poverty.

Gaiha and Imai (2004) study the vulnerability to poverty of rural house-holds in South India during 1975–1984 using a variant of the expected poverty approach. They first use a dynamic- panel income regression model and simulate the effects of negative crop shocks of various sizes and duration. They find that even relatively rich households are highly vulner-able to long spells of poverty when severe crop shocks occur.

Using panel data from two Bangladeshi villages, Amin et al. (2003) analyze vulnerability as the inability of households to insure against idi-osyncratic risks, which is measured by the comovement between house-hold income and household consumption based on a risk- sharing test (Townsend 1994). Using this test, they find that microcredit is successful at reaching the poor. However, it is less successful at reaching the vulnerable and unsuccessful at reaching the vulnerable poor. This may be because the



forces that make some poor households vulnerable may also make them greater risks for microcredit providers. Their study suggests that the neces-sary anti- poverty intervention may be different between the vulnerable and non- vulnerable poor.

Note that the vulnerability measure used in Amin et al. (2003) is a measure of uninsured exposure to risk and not a direct measure of vulner-ability to poverty. Further, as pointed out by Klasen and Povel (2013), the vulnerability measure used in Amin et al. (2003) is at odds with the concept of vulnerability to poverty in the literature, because it is not an ex ante measure and ignores the current consumption level and the likelihood of adverse idiosyncratic and covariate shocks.

Despite these drawbacks, similar methods have been used in a number of other studies. For example, Skoufias and Quisumbing (2005) study vulner-ability as uninsured exposure to risk in five countries including Bangladesh. They find that there is no perfect risk sharing and that food consumption tends to fluctuate less than nonfood consumption by idiosyncratic shocks. Using panel data from Pakistan, Kurosaki (2006) also studies vulnerability based on a risk- sharing test. His study, however, allows for the asymmetry between positive and negative income shocks. His results show that the ability to cope with negative income shocks tends to be lower for those house-holds which are aged, landless, and without regular remittance receipts.

Rest of the worldUsing cross- sectional data in Madagascar and equation (3.4) as a measure of vulnerability, Günther and Harttgen (2009) propose a method to assess relative importance of various sources of vulnerability. They find, among other things, that risk- induced vulnerability is relatively more important than poverty- induced vulnerability in urban areas but the opposite is true for rural areas. They also find that the relative importance of covariate vulnerability to idiosyncratic vulnerability in rural areas is higher than urban areas.

Milcher (2010) also uses the expected poverty in equation (3.4) as a measure of vulnerability. He compares the profile of vulnerability to poverty for Roma and non- Roma households in Southeast Europe. He finds that Roma tends to have higher levels of vulnerability than non- Roma. The characteristics of vulnerable households include large house-holds, households with a poorly- educated head, households whose main source of income is benefits or informal activities.

Using a panel dataset of villages in rural Ethiopia, Dercon and Krishnan (2000) compute (predicted) poverty measures under a combina-tion of various possibilities, such as (1) whether there is a safety net (food aid and consumption from food- for- work), (2) whether the rainfall that households face is ‘normal’ (at the long- term mean) or ‘bad’ (half thereof),



and (3) whether there is seasonal price fluctuations. Comparison of these scenarios indicates that poverty can change substantially within a relatively short period of time.

As with Amin et al. (2003), Skoufias and Quisumbing (2005), and Kurosaki (2006) discussed above, there are also studies that take vulner-ability as uninsured exposure to risk. Using a panel dataset from Peru, Glewwe and Hall (1998) analyze the effect of macroeconomic shock between 1985 and 1990. They find that households headed by relatively well- educated persons, households headed by females, and households with fewer children tend to be less vulnerable.

In the Russian Federation, Gerry and Li (2010) apply quantile regres-sion to a model similar to Glewwe and Hall (1998). They find that a well- functioning labor market is highly valuable, because individuals entering unemployment faced heightened levels of vulnerability among those experiencing the severest consumption shocks, whereas households con-taining individuals entering the labor market are well equipped to smooth consumption.

Gerry and Li (2010) also find that personal networks are important for the most vulnerable. Those in receipt of increased support from relatives were better able to smooth consumption at lower quantiles. They find no evidence that social welfare benefits, such as childcare allowances, unem-ployment benefits and disability benefits, cushion individuals against declining consumption but pension benefits appear to help individuals smooth consumption, particularly for higher quantiles.

In Papua New Guinea, Jha and Dang (2010) estimate poverty and vulnerability, where the latter is computed as expected poverty. Using a subsample of households with an observation in the second round of the survey, they compare the vulnerability derived from the cross- sectional estimation in the first round against the realized poverty in the second round and find that the prediction is reasonably good. Their results are reassuring because vulnerability studies based on cross- sectional data may still be informative.

Empirical studies of vulnerability discussed above are either purely descriptive or try to identify the causes of vulnerability. In contrast, de la Fuente (2010) uses vulnerability as an explanatory variable. He investigates the impact of vulnerability, as measured by the probability of poverty in the future, on remittance flows in Mexico. His findings indicate that money remitted from abroad does not end up with those who are more likely to be needy in the future. While this would be less of a problem if an injection of remittances anywhere within the village would trickle down to those in most need but such social exchanges proved almost inexistent in his study households.



We have reviewed a broad range of empirical studies on vulnerability to poverty in this subsection. The geographic coverage, techniques used, and the covariates considered all differ across studies. However, we make three points that emerge out of this review.

First, poverty and vulnerability to poverty are related but different. Therefore, it is important to understand the underlying causes of poverty and vulnerability. Some policies such as one- off food aid are likely to alle-viate current poverty but do little to reduce vulnerability. Other policies such as improved access to credit would help those entrepreneurial poor facing a credit constraint but will not help reduce vulnerability of farmers who lack the knowledge to diversify crops.

Second, many of the studies discussed above indicate that education is among the important factors that help reduce both poverty and vulner-ability to poverty. One possible explanation is that educated people are able to exploit and adapt to the changes in the economic environment and use assets more efficiently (Schultz 1975).

Finally, location is an important determinant of vulnerability to poverty in many of the studies reviewed above. This is not surprising given that the economic conditions are different across different locations. However, there is currently little knowledge about which location- specific character-istics affect vulnerability. Certain characteristics, such as access to markets, are possible to change by policies. Other characteristics, such as the pattern of rainfall, are more difficult to change, in which case policies should focus on the mitigation of rainfall variations. Hence, understanding the underly-ing cause of vulnerability at each location is a first step to determine the appropriate location- specific policy to cope with vulnerability. We revisit policy issues in section 4.

3 O THER AREAS OF VULNERABILITY

The study of vulnerability is not limited to the vulnerability to poverty. In this section, we briefly review other areas of vulnerability that are related to vulnerability to poverty. In the first subsection, we review studies on vulnerability to climate change. We review this literature because climate change is becoming increasingly important and has implications for poverty. In the second subsection, we review vulnerability studies in which the outcome of interest is not household income or consumption but other measures such as nutrition, assets, and some aggregate- level outcomes.



3.1 V ulnerability to Climate Change

There is now a wide agreement among scientists that the rapid increases in the atmospheric concentration of greenhouse gases, such as carbon dioxide and methane, since the industrial revolution are largely anthro-pogenic. The impacts of the increased concentration of greenhouse gases are already apparent. The global surface temperature is estimated to have risen by more than 0.5 degrees Celsius over the last century and the global average sea level rose at an average rate of 1.8 millimeters per year between 1961 and 2003 (Solomon et al. 2007). Even if stringent climate policies are implemented immediately, global mean surface temperature is expected to rise in years to come.

Climate change affects, among others, agriculture, forestry, water resources, human health, and industry. The impact of climate change is complex because it varies across regions and may be positive or nega-tive. For example, in Asia, crop yields could increase up to 20 percent in East and Southeast Asia, whereas they could decrease up to 30 percent in Central and South Asia by the mid twenty- first century (Parry et al. 2007). Although it is beyond the scope of this chapter to discuss specific impacts of climate change,14 it is evident that climate change affects various aspects of social, economic, and ecological systems, and may have profound impacts on the lives of the poor. Therefore, it is useful to review studies on vulnerability to climate change in relation to poverty.

To understand the relationship between vulnerability to poverty and vulnerability to climate change, Adger (2006) provides a useful overview of these two strands of literature. He argues that the idea of vulnerability to poverty originates from the school of thought that views vulnerability as absence of entitlements (for example, Sen 1981). On the other hand, the roots of studies on vulnerability to climate change are the analysis of vulnerability to hazards (for example, Burton et al. 1993). Adger (2006) suggests that the conceptualization and measurement of vulnerability to poverty discussed above complements the hazard- based approach. While there is a dearth of studies linking climate change and vulnerability to poverty,15 this is potentially a fruitful area of research.16

To bring insights in the study of vulnerability to climate change to the context of vulnerability to poverty, it is useful to consider the following four dimensions to describe a vulnerability situation (Fussel 2007): system, attribute of concern, hazard, and temporal reference. All of these are important for considering the impact of climate change on poverty and its policy implications. They also offer potentially fruitful areas of research.

First, the system of analysis, which may be, for example, a population group, an economic sector, or a geographic region, has important policy



implications. This is because a policy that makes a particular group less vulnerable may make other groups more vulnerable. Therefore, the analysis of vulnerability to poverty discussed in the previous section may become misleading if the system, or the population relevant for the analysis, is not appropriately identified.

Second, the valued attribute of the vulnerable system threatened by its exposure to a hazard is also important. In the previous section, the attribute of concern was taken as consumption, but it may include other dimensions such as nutrition.17 We briefly discuss vulnerability in nutrition outcome in the next subsection.

Third, it is also important to clarify what type of hazard – or potentially damaging physical event, phenomenon, or human activity that may cause the loss of life or injury, property damage, social and economic disrup-tion, or environmental degradation (United Nations Office for Disaster Risk Reduction 2004) – is being considered. Most of the studies on vul-nerability to poverty presented in the previous section abstract from spe-cific hazards and analyze vulnerability from the perspective of stochastic consumption. Because appropriate policies to reduce or remove vulner-ability depend on the specific hazards at issue, more research is needed to identify the link between various hazards that climate change brings about and poverty.

Finally, temporal reference is particularly relevant in the context of climate change. Most of the studies mentioned in the previous section only have a rudimentary treatment of time with only one or two periods in their models. However, intertemporal tradeoffs are fundamentally important for mitigation of climate change. Temporal reference is also important from the perspective of adaptation, because long- term impact of climate change depends on how the economy and society are able to respond. Therefore, a careful examination of the relevant time frame is essential for appropri-ately dealing with vulnerability to poverty due to climate change.

3.2 V ulnerability in Non- monetary Outcomes

The studies discussed in the previous section are based primarily on an individual- level money- metric outcome measure such as consumption per capita. However, vulnerability can be analyzed by other observable outcomes. First, there is a critical relationship between vulnerability and asset ownership. As Moser (1998) argues, analyzing vulnerability involves identifying not only the threat but also the ‘resilience,’ or the responsive-ness in exploiting opportunities and in resisting or recovering from the negative effects of a changing environment. Therefore, the assets and entitlements available to individuals and households are critically related



to vulnerability. This point is consistent with the theoretical argument put forward by Elbers and Gunning (2003).

Chiwaula et al. (2011) propose a variant of the expected poverty approach discussed in the previous section, which includes asset indica-tors in an income regression. They decompose expected poverty into structural- chronic (that is, vulnerable and mean consumption more than one standard deviation below the poverty line), structural- transient (that is, vulnerable and mean consumption less than one standard deviation below the poverty line), and stochastic- transient (that is, not vulnerable and mean consumption above the poverty line). In their empirical applica-tion to Cameroon and Nigeria, they find that the majority of households are vulnerable for structural reasons. That is, their asset base is so low that even if favorable production conditions would occur or risk- reducing measures would be introduced they are unlikely to be able to move out of poverty permanently. Their study underscores the importance of building productive assets to increase income and decrease the variance of income to escape from the threat of poverty.

It is also possible to analyze vulnerability with nutritional outcomes. Using six nutritional outcomes, Stillman and Thomas (2008) examine the effect of dramatic income change on nutritional well- being during the crisis in 1998 in the Russian Federation. They test whether young women and the elderly are particularly vulnerable to worsening economic condi-tions and find that there is no significant difference in nutritional intakes between males and females nor across different demographic groups.

The discussion of vulnerability so far has been mainly concerned with vulnerability at the individual or household level. However, it is also useful to consider vulnerability at a more aggregate level. For example, consider vulnerability from trade openness (Montalbano 2011). While trade open-ness has been generally found to be beneficial to economic growth, it can adversely affect the lives of the poor, for example, when the prices of goods that the poor consume increase or when prices of goods that the poor produce decrease. Further, trade openness can increase the volatility of prices of certain goods.

The impact of trade openness on vulnerability to poverty can be ana-lyzed at the household level, because it would manifest itself in relative prices and their volatilities. However, it is also useful to consider vulner-ability from openness to trade at the country level. Briguglio et al. (2009), for example, define economic vulnerability as the exposure of an economy to exogenous shocks arising out of economic openness. They then present an index of vulnerability and resilience, where the latter is defined as the policy- induced ability of an economy to withstand or recover from the effect of shocks.



Montalbano (2011) argues that a meso approach, which is between household and country levels, is important for holistic welfare analysis of the risks induced by trade liberalization. Montalbano (2011) identi-fies two main strands of the literature: the ‘vulnerability of subnational regions approach’ and the ‘industry- level volatility approach’. The former strand includes Naudé et al. (2009), who construct a local vulnerability index using principal component analysis for 354 magisterial districts in South Africa. The latter includes Koren and Tenreyro (2007), who decom-pose the volatility of gross domestic product growth into various sources and quantify their contributions to volatility. According to their findings, as countries develop, their productive structure moves from more volatile to less volatile sectors.18

4 DI SCUSSION

In this chapter, we have reviewed vulnerability studies primarily in relation to poverty. While there is some agreement on what characterizes vulner-ability across various studies, there is as yet no concept or measurement of vulnerability that is widely accepted. This is true even within the narrowly defined literature on vulnerability to poverty discussed in section 2. As discussed in section 3, there are even larger varieties of concepts of vulner-ability originating from different disciplines and traditions.

Therefore, one obvious area of research that arises from this review is further refinement of the vulnerability concept and its measurement, par-ticularly those based on the axiomatic approach discussed in section 2.1. We argue that the measures proposed by Calvo and Dercon (2013) provide an excellent starting point because they satisfy a set of desirable axioms. However, their analytical framework abstracts from the time dimension. It may be fruitful to explicitly incorporate intertemporal tradeoffs, especially when we consider the vulnerability of households to poverty induced by climate change.

This review also indicates that there is still a dearth of empirical studies on vulnerability. This is true for most countries in Asia and elsewhere. One obvious reason for this observation is the lack of high- quality data. While the availability of socioeconomic surveys with a panel structure is rapidly improving, the availability is still limited and most available panel data that could be used for vulnerability analysis contain only a few time periods at best. To seriously evaluate the risk of falling into poverty, a longer and pos-sibly more frequent data collection is desirable.

From the perspective of data availability, the situation surrounding vulnerability studies is somewhat similar to poverty analysis in the early



1980s when there was a lack of relevant high- quality consumption survey data. Just as subsequent expansion of consumption survey data stimulated poverty research, better availability of long panel data is almost sure to stimulate vulnerability research.

Long panel data may also create new areas of research. For example, long panel data would also allow us to consider a distinction between vulnerability to chronic poverty and vulnerability to transient poverty. This distinction may be important because certain negative shocks may be persistent (for example, disability) while others may be transient (for example, diarrhea). This distinction is also important because vulnerabil-ity to chronic poverty and vulnerability to transient poverty are likely to require different solutions and different targeting policies.

Besides the lack of long panel data, current surveys often do not contain sufficient information about the shocks that households face to estimate the impact of these shocks on vulnerability. From this perspective, the study by Günther and Harttgen (2009) would be useful. They collect infor-mation about important shocks that households face including malaria, tuberculosis, typhoid, cholera, rice pest, swine flu, Newcastle disease, flooding, impassible bridge or road, drought, and cyclones. There may be other shocks such as asset losses, labor market disturbances, harvest failure and civil unrest. Hence, collecting data on some of these and other relevant indicators may prove valuable for the analysis of vulnerability.

The current state of research on vulnerability is also inadequate for designing appropriate policies to deal with vulnerability. As noted earlier, there are some common factors, including education and location, that help to explain vulnerability. However, existing studies provide little guid-ance on the appropriate choice of policies to reduce or remove vulnerabil-ity. Therefore, more research is needed to understand the impact of policy on vulnerability.

There are a number of policies that can potentially reduce individual vulnerability. As Morduch (1999) argues, increasing macroeconomic sta-bility, reining in inflation, securing property rights, improving transport and communications, and creating a stable political environment can go a long way toward reducing the frequency and size of downturns and creat-ing a supportive environment to facilitate private risk- reducing activities. Similarly, risk can be reduced through public health campaigns for immu-nization and sanitation, civil works projects and, in some cases, price sta-bilization. Higher incomes and stable employment opportunities further enhance the ability to cope with risk. However, the primary purpose of these policies is not to reduce individual vulnerability and thus they are best judged by other criteria. Therefore, we focus below on several policies that could directly address vulnerability.



First, one can insure oneself by building assets and using them to smooth consumption. Therefore, the saving technology available to individuals is crucial for mitigating vulnerability. Relevant policies for promoting savings include ensuring long- term security of saving and improving convenience Morduch (1999). Providing households with access to more attractive and more diversified assets could improve the functioning of self- insurance Dercon (2002). Note, however, that large negative shocks cannot be easily insured by self- insurance.

Second, provision of microcredit can help those poor who are entrepre-neurial but credit- constrained to increase the income and also diversify the sources of income. As a result, it may help them increase the mean income and reduce the variance of income. However, the results of recent randomized control trial studies suggest that the provision of microcredit will not benefit all poor individuals equally. Hence, it is also unlikely to be sufficient to eliminate vulnerability to poverty.

Third, employment- guarantee schemes such as rural public works pro-grams can also help to reduce vulnerability (Morduch 1999). In this type of program, employment is offered to (ideally) anyone who is willing to work for a low wage rate. Under such a scheme, the program is self- targeted. That is, workers would participate only when there is no better option elsewhere. Hence, employment guarantee schemes essentially provide a self- targeted fallback option.

Finally, a well- designed social safety net is likely to help reduce vulner-ability. For example, Devarajan and Jack (2007) argue that a simple public insurance scheme that pays a fixed benefit to all households that suffer a negative shock is an effective redistributional instrument of public policy even when there is a well- functioning private insurance market.

The experience in Indonesia during the Asian financial crisis also high-lights the potential importance of a social safety net. Dhanani and Islam (2002) find that vulnerability could have worsened in the absence of gov-ernment intervention, even though some of the social safety- net programs did not appear to work well. Despite its potential usefulness for addressing vulnerability, social protection policies have to be carefully crafted because they may crowd out the existing informal insurance.

To conclude, there are a number of policy options to address vulner-ability. However, little is known about the policy impact of specific poli-cies. Further research is needed to better understand the interplay between informal insurance and public policies as well as its impact on vulnerability.



NOTES

1. The author thanks Satya Chakraverty, Indranil Dutta, Jacques Silber, Hermann Waibel, and Guanghua Wan for their helpful comments. The author has benefitted from useful discussion with Chris Elbers at the initial stage of this research. Xu Sijia is gratefully acknowledged for providing research assistance.

2. Based on the author’s search of Econlit on 6 September 2014. 3. In Ligon and Schechter (2003) and various other studies, vulnerability is defined for

households and not individuals. Despite the fact that household is often the unit of measurement in surveys by which vulnerability is measured, we chose to use individual as the unit for which vulnerability is defined, because vulnerability may vary even within the household, at least in principle.

4. While Pritchett et al. (2000) require only a zero mean and not symmetry, but this is clearly inappropriate. If an individual at the poverty line receives a small negative shock with a high probability and a large positive shock with a low probability, the probability of falling below the poverty line is higher than 0.5 even when the shock has a zero mean.

5. Note that chronic and transient poverty are typically defined as ex post concepts. However, they are treated as ex ante concepts here as with vulnerability.

6. Using wage as a welfare variable, Bourguignon et al. (2004) compare the accuracy of estimation of expected poverty based on a repeated cross- sectional data against true- panel data. See also Jha and Dang (2010) discussed below.

7. Duclos et al. (2010) propose alternative measures of chronic poverty and transient poverty.

8. Then, if each period is an independent trial and the observation period is arbitrarily long, vulnerability and total poverty make no practical difference despite the fact that they are respectively ex ante and ex post concepts. Further, we can also obtain an arbi-trarily accurate estimate of E[ci].

9. We require 1 − d ≥ pl ≥ 0, which was not explicitly required in Calvo and Dercon (2005, 2013), to ensure that the probability for the l- th component after the transfer is still on a unit interval.

10. The possibility of equality was not included in Calvo and Dercon (2005, 2013) presum-ably because it is a trivial case. We include it here to be complete. It should be noted that c| can be a constant even when c is random. This occurs if consumption is always above the poverty line.

11. Dutta et al. (2011) use income instead of consumption. We use consumption to be con-sistent with the rest of the paper.

12. Using our notations, the axiom of independence requires v(z, ca, p) ≤ v(z, cb, p) 1 v(z, ca + d1k, p) ≤ v(z, cb + d1k, p) for d > 0 and 1 ≤ k ≤ K.

13. It is also possible to argue that the additive decomposability is a desirable property for a social measure of vulnerability. Dutta and Mishra (2013) derive a social measure of vulnerability from a set of axioms that includes the axiom of decomposability.

14. See Parry et al. (2007) for a detailed description of the impacts that have already been observed and are likely to occur under various scenarios.

15. To the best of our knowledge, Fujii’s Chapter 5 in this volume is currently the only study that directly links future climate change to vulnerability to poverty based on a household- level data set.

16. Incidentally, the Fifth Assessment Report by the Interregional Panel on Climate Change (IPCC) Working Group II, which focuses on vulnerability and adaptation, has a new chapter on ‘livelihoods and poverty’ (see IPCC 2014).

17. It may be useful to consider a composite index to describe vulnerability to climate change. Brooks et al. (2005) construct a vulnerability index as a combination of various health, education and governance indicators. According to their index, the most vulner-able countries are nearly all situated in sub- Saharan Africa.

18. See also Naudé et al. (2009) for additional discussion on vulnerability in nonmonetary outcomes.



REFERENCES22*

Adger, N. (2006), ‘Vulnerability’, Global Environment Change, 16 (3), 268–81.Amin, S., A.S. Rai and G. Topa (2003), ‘Does microcredit reach the poor and

vulnerable? Evidence from Northern Bangladesh’, Journal of Development Economics, 70 (1), 59–82.

Bourguignon, F., C. Goh and D.I. Kim (2004), ‘estimating individual vulnerability to poverty with pseudo- panel data’, World Bank Policy Research Working Paper 3375, World Bank, Washington, DC.

Briguglio, L., G. Cordina, N. Farrugia and S. Vella (2009), ‘Economic vulnerability and resilience: concepts and measurements’, Oxford Development Papers, 37 (3), 229–47.

Brooks, N., N. Adger and M. Kelly (2005), ‘The determinants of vulnerability and adaptive capacity at the national level and the implications for adaptation’, Global Environmental Changes, 15 (2), 151–63.

Burton, I., R.W. Kates and G.F. White (1993), The Environment as Hazard, 2nd edn, New York: Guilford.

Calvo, C. (2008), ‘Vulnerability to multidimensional poverty: Peru, 1998–2002’, World Development, 36 (6), 1011–20.

Calvo, C. and S. Dercon (2005), ‘Measuring individual vulnerability’, Department of Economics Discussion Paper 229, University of Oxford.

Calvo, C. and S. Dercon (2007), ‘Vulnerability to poverty’, CSAE Working Paper 2007- 03, University of Oxford.

Calvo, C. and S. Dercon (2013), ‘Vulnerability of individual and aggregate poverty’, Social Choice and Welfare, 41 (4), 721–40.

Carter, M. and F. Zimmerman (2000), ‘The dynamic cost and persistence of asset inequality in an agrarian economy’, Journal of Development Economics, 63 (2), 265–302.

Chakravarty, S.R. (1983), ‘A new index of poverty’, Mathematical Social Sciences, 6, 307–13.

Chakravarty, S.R., N. Chattopadhyay and L. Qingbin (2015), ‘Vulnerability orderings for expected poverty indices’, Japanese Economic Review, 66 (3), 300–310.

Chaudhuri, S., J. Jyotsna and A. Suryahadi (2002), ‘Assessing household vulner-ability to poverty from cross- sectional data: a methodology and estimates from Indonesia’, Discussion Paper Series 0102- 52, Department of Economics, Columbia University.

Chiwaula, L.S., R. Witt and W. Hermann (2011), ‘An asset- based approach to vulnerability: the case of small- scale fishing areas in Cameroon and Nigeria’, Journal of Development Studies, 47 (2), 338–53.

Christiaensen, L. and K. Subbarao (2005), ‘Towards an understanding of house-hold vulnerability in rural Kenya’, Journal of African Economies, 14 (4), 520–58.

Clark, S., R. Hemming and D. Ulph (1981), ‘On indices for the measurement of poverty’, Economic Journal, 91 (362), 515–26.

De la Fuente, A. (2010), ‘Remittances and vulnerability to poverty in rural Mexico’, World Development, 38 (6), 828–39.

* The Asian Development Bank recognizes China as the People’s Republic of China and Vietnam as Viet Nam.



Dercon, S. (1996), ‘Risk, crop choice and savings: evidence from Tanzania’, Economic Development and Cultural Change, 44 (3), 485–513.

Dercon, S. (1998), ‘Wealth, risk, and activity choice: cattle in western Tanzania’, Journal of Development Economics, 55 (1), 1–42.

Dercon, S. (2002), ‘Income risk, coping strategies, and safety nets’, World Bank Research Observer 17(2), 141–166.

Dercon, S. and P. Krishnan (2000), ‘Vulnerability, seasonality and poverty in Ethiopia’, Journal of Development Studies, 36 (6), 25–53.

Devarajan, S. and W. Jack (2007), ‘Protecting the vulnerable: the tradeoff between risk reduction and public insurance’, World Bank Economic Review, 21 (1), 73–91.

Dhanani, S. and I. Islam (2002), ‘Poverty, vulnerability and social protection in a period of crisis: the case of Indonesia’, World Development, 30 (7), 1211–31.

Duclos, J.- Y., A. Arrar and J. Giles (2010), ‘Chronic and transient poverty: meas-urement and estimation, with evidence from China’, Journal of Development Economics, 91 (2), 266–77.

Dutta, I. and A. Mishra (2013), ‘Measuring vulnerability using the counting approach’, mimeo, Faculty of Humanities, University of Manchester and Department of Economics, University of Bath.

Dutta, I., J. Foster and A. Mishra (2011), ‘On measuring vulnerability to poverty’, Social Choice and Welfare, 37 (4), 743–61.

Elbers, C. and J.W. Gunning (2003), ‘Vulnerability in a stochastic dynamic model’, Tinbergen Institute Discussion Paper TI 2003- 070/2, Tinbergen Institute, Amsterdam.

Eswaran, M. and A. Kotwal (1990), ‘Implications of credit constraints for risk behaviour in less developed economies’, Oxford Economic Papers, 42 (2), 473–82.

Foster, J., J. Greer and E. Thorbecke (1984), ‘A class of decomposable poverty measures’, Econometrica, 52 (3), 761–6.

Fussel, H.- M. (2007), ‘Vulnerability: a generally applicable conceptual framework for climate change research’, Global Environmental Change, 17 (2), 155–67.

Gaiha, R. and K. Imai (2004), ‘Vulnerability, shocks and persistence of poverty: estimates for semi- arid rural south India’, Oxford Development Studies, 32 (2), 261–81.

Gerry, C. and C. Li (2010), ‘Consumption smoothing and vulnerability in Russia’, Applied Economics, 42 (16–18), 1995–2007.

Glewwe, P. and G. Hall (1998), ‘Are some groups more vulnerable to macroeco-nomic shocks than others? Hypothesis tests based on panel data from Peru’, Journal of Development Economics, 56 (1), 181–206.

Günther, I. and K. Harttgen (2009), ‘Estimating households vulnerability to idi-osyncratic and covariate shocks: a novel method applied in Madagascar’, World Development, 37 (7), 1222–34.

Hardeweg, B., A. Wagener and H. Waibel (2013), ‘A distributional approach to comparing vulnerability, applied to rural provinces in Thailand and Vietnam’, Journal of Asian Economics, 25 (April), 53–65.

Imai, K., R. Gaiha and W. Kang (2011a), ‘Poverty, dynamics and vulnerability in Vietnam’, Applied Economics, 43 (25–27), 3603–18.

Imai, K., R. Gaiha and W. Kang (2011b), ‘Poverty, inequality and ethnic minorities in Vietnam’, International Review of Applied Economics, 25 (3), 249–82.

Imai, K., X. Wang and W. Kang (2010), ‘Poverty and vulnerability in rural China:



effects of taxation’, Journal of Chinese Economic and Business Studies, 8 (4), 399–425.

Interregional Panel on Climate Change (IPCC) (2014), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, eds C.B. Field, V.R. Barros, D.J. Dokken, K.J. Mach, M.D. Mastrandrea, T.E. Bilir et al., Cambridge and New York: Cambridge University Press.

Jalan, J. and M. Ravallion (1998), ‘Transient poverty in postreform rural China’, Journal of Comparative Economics, 26 (2), 338–57.

Jalan, J. and M. Ravallion (2000), ‘Is transient poverty different? Evidence for rural China’, Journal of Development Studies, 36 (6), 82–99.

Jha, R. and T. Dang (2010), ‘Vulnerability to poverty in Papua New Guinea in 1996’, Asian Economic Journal, 24 (3), 235–51.

Jha, R., T. Dang and Y. Tashrifov (2010), ‘Economic vulnerability and poverty in Tajikistan’, Economic Change and Restructuring, 43 (2), 95–112.

Kamanou, G. and J. Morduch (2004), ‘Measuring vulnerability to poverty’, in S. Dercon (ed.), Insurance Against Poverty, Oxford: Oxford University Press, pp. 155–75.

Klasen, S. and F. Povel (2013), ‘Defining and measuring vulnerability: state of the art and new proposals’, in S. Klasen and H. Waibel (eds), Vulnerability to Poverty: Theory, Measurement and Determinants, with Case Studies from Thailand and Vietnam, New York: Palgrave- Macmillan.

Koren, M. and S. Tenreyro (2007), ‘Volatility and development’, Quarterly Journal of Economics, 122 (1), 243–87.

Kurosaki, T. (2006), ‘Consumption vulnerability to risk in rural Pakistan’, Journal of Development Studies, 42 (1), 70–89.

Ligon, E. and L. Schechter (2003), ‘Measuring vulnerability’, Economic Journal, 113 (486), C95–C102.

McCulloch, N. and M. Calandrino (2003), ‘Vulnerability and chronic poverty in rural Sichuan’, World Development, 31 (3), 611–28.

Milcher, S. (2010), ‘Household vulnerability estimates of Roma in southeast Europe’, Cambridge Journal of Economics, 34 (4), 773–92.

Montalbano, P. (2011), ‘Trade openness and developing countries’ vulnerability: concepts, misconceptions, and directions for research’, World Development, 39 (9), 1489–502.

Morduch, J. (1994), ‘Poverty and vulnerability’, American Economic Review, 84 (2), 221–5.

Morduch, J. (1999), ‘Between the state and the market: can informal insurance patch the safety net?’, World Bank Research Observer, 14 (2), 187–207.

Moser, C. (1998), ‘The asset vulnerability framework: reassessing urban poverty reduction strategies’, World Development, 26 (1), 1–19.

Naudé, W., M. McGillivray and S. Rossouw (2009), ‘Measuring the vulnerability of subnational regions in South Africa’, Oxford Development Studies, 37 (3), 249–76.

Naudé, W., A.U. Santos- Paulino and M. McGillivray (2009), ‘Measuring vulner-ability: an overview and introduction’, Oxford Development Studies, 37 (3), 183–91.

Parry, M.L., O.F. Canziani, J.P. Palutikof, P. van der Linden and C.E. Hanson (eds) (2007), Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution



of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge: Cambridge University Press.

Pritchett, L., A. Suryahadi and S. Sumarto (2000), ‘Quantifying vulnerability to poverty: a proposed measure, applied to Indonesia’, World Bank Policy Research Working Paper 2437, World Bank, Washington, DC.

Ravallion, M. (1988), ‘Expected poverty under risk- induced welfare variability’, Economic Journal, 98 (393), 1171–82.

Rosenzweig, M. and H. Binswanger (1993), ‘Wealth, weather risk and agricultural investments’, Economic Journal, 103 (416), 56–78.

Schultz, T.W. (1975), ‘Human capital and disequilibrium’, Journal of Economic Literature, 13 (3), 827–46.

Sen, A.K. (1981), Poverty and Famines: An Essay on Entitlement and Deprivation, Oxford: Clarendon Press.

Skoufias, E. and A.R. Quisumbing (2005), ‘Consumption insurance and vulner-ability to poverty: a synthesis of the evidence from Bangladesh, Ethiopia, Mali, Mexico and Russia’, European Journal of Development Research, 17 (1), 24–58.

Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt et al. (eds) (2007), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge: Cambridge University Press.

Stillman, S. and D. Thomas (2008), ‘Nutritional status during an economic crisis: evidence from Russia’, Economic Journal, 118 (53), 1385–417.

Suryahadi, A. and S. Sumarto (2003), ‘Poverty and vulnerability in Indonesia before and after the economic crisis’, Asian Economic Journal, 17 (1), 45–64.

Townsend, R. (1994), ‘Risk and insurance in village India’, Econometrica, 62 (3), 539–91.

United Nations Office for Disaster Risk Reduction (2004), Living with Risk: A Global Review of Disaster Reduction Initiatives, vol. 1, New York: United Nations.

Watts, H.W. (1968), ‘An economic definition of poverty’, in D.P. Moynihan (ed.), On Understanding Poverty, New York: Basic Books, pp. 316–29.

Zhang, Y. and G. Wan (2006), ‘An empirical analysis of household vulnerability in rural China’, Journal of the Asia Pacific Economy, 11 (2), 196–212.

Zhang, Y. and G. Wan (2009), ‘How precisely can we estimate vulnerability to poverty?’, Oxford Development Studies, 37 (3), 277–87.

Zimmerman, F. and M. Carter (2003), ‘Asset smoothing, consumption smooth-ing and the reproduction of inequality under risk and subsistence constraints’, Journal of Development Economics, 71 (2), 233–60.


84

4. Measuring the impact of vulnerability on the number of poor: a new methodology with empirical illustrations1

Satya R. Chakravarty, Nachiketa Chattopadhyay, Jacques Silber and Guanghua Wan

1 INTRODUCTION

In the dimensions of income and health, vulnerability is the risk that a household or an individual will experience an episode of income or health poverty over time. However, vulnerability also means the probability of being exposed to a number of other risks (violence, crime, natural disas-ters, having to leave school) (World Bank 2000: 19). The focus of vulner-ability should hence be on the risk of negative outcomes in the future (Hoddinott and Quisumbing 2003), where most generally ‘negativity’ refers to a situation in which an individual is below the poverty line (Calvo and Dercon 2013). Vulnerability, thus, imposes a security risk on individu-als in the sense that it affects their well- being negatively. It may cause long- term deprivation for individuals.

‘The challenge of development includes not only the elimination of per-sistent and endemic deprivation, but also the removal of vulnerability to sudden and severe destitution’ (Sen 1999: 1). ‘Protecting vulnerable groups during episodes of macroeconomic contraction is vital to poverty reduc-tions in developing countries’ (World Bank 1997: 1).

In the measurement of vulnerability, we need to be concerned, not only with current conditions, such as current income and consumption, but also with the risks individuals face and their ability to avoid, reduce and overcome these. This shows that an indicator of vulnerability should take several appropriate factors into account. For concreteness, in the remain-der of this chapter we assume that the unit of analysis is an individual and income represents the underlying economic variable.


Measuring the impact of vulnerability on the number of poor 85

As Klasen and Povel (2013) pointed out, vulnerability at the household/individual level can be broadly subdivided into the following categories: (1) vulnerability as uninsured exposure to risk; (2) vulnerability as low expected utility; (3) vulnerability as expected poverty; and (4) vulner-ability to poverty. The first three categorizations of vulnerability were analyzed, among others, by Hoddinott and Quisumbing (2003), Ligon and Schechter (2004) and Gaiha and Imai (2009) (see also Hoogeveen et al. 2004). Vulnerability to poverty was introduced and discussed by Calvo and Dercon (2013). (See Fujii 2013 for a recent discussion.)

Vulnerability, as uninsured exposure to risk, indicates whether income shocks induce changes in consumption (see Townsend 1994; Amin et al. 2003; Skoufias and Quisumbing 2005). This notion of vulnerability is concerned with changes in the current level of consumption and not with the levels of consumption. It does not take into account an individual’s attitudes towards risks.

Vulnerability, as low expected utility, relates vulnerability with vari-ability. There is a long history of the use of the variance as a measure of risk in statistical decision theory (Rothschild and Stiglitz 1970). It was rigorously formulated by Ligon and Schechter (2003). The Ligon and Schechter notion of vulnerability is measured by the difference between the utility derived from a threshold income and the individual’s expected utility derived from incomes in a vulnerable situation. The higher the dif-ference between the two utility values, the more vulnerable the person is. The individual is non- vulnerable in this situation if his income is above the threshold limit (see also Glewwe and Hall 1998; Dercon 2002; Coudouel and Hentschel 2000). The major advantage of this approach is that it incorporates an individual’s attitudes towards risks explicitly by making the formulation directly dependent on the von Neumann–Morgenstern utility function. In view of the non- constancy of the utility function and probabilistic formulation, the approach takes into account the severity and likelihood of shocks on individual welfare. One limitation of this approach is that all individuals are assumed to possess the same attitudes towards risks. It is, however, true that under non- comparability of indi-vidual utility functions, aggregation is not possible under usual Arrowian axioms (Sen 1977; Boadway and Bruce 1984; Blackorby et al. 1984). The Ligon– Schechter framework was also assumed by Elbers and Gunning (2003) by incorporating explicitly the future streams of income over an infinite time horizon.

Vulnerability, as expected poverty, refers to the risk of an individ-ual’s income falling below the poverty line. The idea was initiated by Ravallion (1988) and advanced and analyzed further by Holzmann and Jorgensen (1999). A formal analysis of this approach was developed by



Chaudhuri et al. (2002), which indicates the probability that an individual’s income will be below an exogenously given poverty line. However, it does not take into account the sensitivity towards risks. An individual’s position with respect to vulnerability is simply decided in terms of some expected income. Hoddinott and Quisumbing (2003) address this shortcoming by expressing vulnerability as expected poverty using the Foster–Greer–Thorbecke (Foster et al. 1984) poverty index. Interpreting the negative of poverty as utility, we note that the Arrow–Pratt absolute risk- aversion measure for this utility function increases as the value of the underlying parameter increases. However, such a risk preference is not unambiguously supported by empirical findings (see Binswanger 1981; Hoddinott and Quisumbing 2003). Empirical applications of this approach can be found in Hoddinott and Quisumbing (2003), Suryahadi and Sumarto (2003), Christiaensen and Subbarao (2005), Kamanou and Morduch (2004), and Günther and Harttgen (2009).

The notion of vulnerability to poverty was introduced by Calvo and Dercon (2013). Instead of starting from individual poverty or a utility function, they developed an axiomatic characterization of a vulnerability measure. In this framework, vulnerability is a weighted average of future state- wise deprivations, where the weights are the probabilities of out-comes associated with different states of the world in the future. The two measures that were characterized by Calvo and Dercon (2013) are the expected measures of Chakravarty (1983) and Watts (1968). These meas-ures are, in fact, expected poverty measures.2 These measures explicitly take into account risk aversion. They rely on the poverty line, assigned prob-abilities and relevant states of the world. Previously, Dutta et al. (2011) axiomatically derived a vulnerability measure, which unlike the Calvo–Dercon measure, assumes that deprivation depends explicitly on current and future incomes. Therefore, this measure allows us to look at relative changes under vulnerability.

The objective of this chapter is to study the implications of vulnerability on the poverty line. More precisely, we investigate the issue of adjusting the poverty threshold under vulnerability so that the corrected poverty line also represents the subsistence standard of living in an environment of vulnerability. Essential to the adjustment is the assumption that the utility derived from the existing poverty line is the same as the expected utility generated by the new poverty line affected by a random error (noise) repre-senting vulnerability. Thus, the formulation relies on the implicit assump-tion that vulnerability is treated as low expected utility. Under certain realistic assumptions about the noise, in an additive model the improved poverty line is shown to exceed the existing poverty line by a constant amount if the utility function displays constant Arrow–Pratt absolute risk



aversion (see Arrow 1965; Pratt 1964). Likewise, in a multiplicative model, the adjusted poverty line becomes a scale transformation, where the under-lying scalar is greater than unity, if the utility function exhibits constant Arrow–Pratt relative risk aversion.

In a recent contribution, Dang and Lanjouw (2014) suggested two formal approaches to setting the vulnerability line. In the first approach, they identified a subgroup of a population which is clearly not vulnerable and defined the vulnerability line as the lower- bound income for this popu-lation subgroup. The second approach considers a subgroup which is not poor but faces a real risk of falling into poverty. They set the upper- bound income for this subgroup as the vulnerability line. While our approach relies on the Arrow–Pratt theory of risk aversion, the Dang–Lanjouw approach is based on a probabilistic formulation. Therefore, neither sup-plements the other; the two approaches are clearly different.

This chapter is organized as follows. The next section presents a brief overview of the background material involving the Arrow–Pratt meas-ures of risk aversion. Section 3 formally presents the derivation of the vulnerability- adjusted poverty lines under alternative assumptions about the Arrow–Pratt measures. The focus of section 4 is on the estimation of the variance of the noise which characterizes the uncertain income, whereas section 5 presents an empirical illustration using data from the Asian- Pacific region. Section 6 concludes.

2 BACKGROUND

It is often useful to have an indicator of risk aversion. A risk- aversion indi-cator is a measure of the extent to which an individual becomes averse to risky situations. It is helpful to make a comparison between two individuals in terms of their attitudes towards risk.

Let U: (0, ∞) → R denote the utility function of the individual under consideration, where R denotes the real line. The utility function U is assumed to be continuous, increasing and strictly concave. For our pur-poses, we assume also that it is at least twice differentiable. We denote the first and second derivatives of U by U9 and U0 respectively. Since it is assumed that U is a monotone increasing and strictly concave function, U9 > 0 and U0 < 0 are satisfied.

The Arrow–Pratt measure of absolute risk aversion APA(M), for a person with utility function U and level of income M, is defined as:

APA (M) 5 2Us (M)U r (M) . (4.1)APA (M) 5 2



The indicator APA(M) takes on a positive, zero or negative value depend-ing on whether an individual is risk- averse, risk- neutral or risk- preferring, that is, the utility function is strictly concave, affine or strictly convex. A higher value of APA indicates that an individual’s aversion towards risk is higher.

If incomes are expressed in relative terms, an appropriate measure that indicates attitudes to risk is the Arrow–Pratt relative risk aversion measure defined as:

APR (M) 5 2M 3 Us (M)

U r (M) . (4.2)

The measure APR takes on positive, zero or negative values depending on whether an individual is risk- averse, risk- neutral or risk- preferring.

Highly significant implications of constant and strictly monotonic risk- aversion measures that can motivate a focus on these concepts arise in the context of analysis of the cost of risk and portfolio formation. One way of looking at the cost of risk or the risk premium is to define it as the dif-ference between the expected income on a risky prospect and the certainty equivalent, the certain amount of income that is equally preferred to the prospect. It shows how much the individual would be willing to pay rather than face the risky prospect (see Gravelle and Rees 2004). Formally, it is defined as:

CA (p,x) 5 ak

i51pixi 2 xe, (4.3)

where x 5 (x1,x2,. . .,xk) is the vector of state- contingent returns on the risky prospect; k is the number of states; pi is the probability of state i; and p 5 (p1, p2, . . . , pk) . Now ak

i51 pixi is the expected return and the certainty equivalent xe is implicitly defined by:

ak

i51piU(xe) 5 a

k

i51piU(xi) .

The indicator CA is a cost of risk because, in the absence of uncertainty, it is zero and it is positive for a risk- averse person if the environment is char-acterized by uncertainty. This cost remains invariant under equal absolute changes in outcomes on the prospect if and only if APA of the underlying utility function is a constant (Chakravarty 2013).

Likewise, the relative cost can be defined as the proportionate gap between the expected return on the prospect and the certainty equivalent. Formally, it is given by:

APR (M) 5 2



CR (p,x) 5 1 2xe

ak

i51pixi

. (4.4)

If uncertainty prevails, this cost is positive under strict concavity of the utility function. The cost measure CR remains invariant when the scale of outcomes changes by a positive scalar if and only if APR of the underlying utility function is a constant (Chakravarty 2013).

In a portfolio consisting of one risky asset and one risk- free asset, the amount invested in the risky prospect increases with an increase in his or her wealth if the absolute risk aversion measure is decreasing. That is, as a person becomes less risk averse with an increase in the level of wealth, his or her demand for the risky asset increases. This means that the risky prospect is a normal good (Arrow 1970). Likewise, if the relative risk aver-sion measure is increasing, then the share of wealth invested in the risky prospect decreases with an increase in the level of wealth (see Demange and Laroque 2006).

3 FORMAL FRAMEWORK

In this section, we investigate the impact of vulnerability on the poverty line. A person at the poverty line z0 without vulnerability has a certain utility U(z0). On the other hand, in a vulnerable situation, he or she is subjected to an uncertain income. We deal with the cases of constant and relative risk aversion in the next two subsections respectively.

3.1 Constant Absolute Risk Aversion

In this subsection we assume that the individual’s income is characterized by an additive noise e, a random variable whose mean is 0 and variance is s2. Such an assumption for the error process in consumption was made by Ligon and Schechter (2003). Rothschild and Stiglitz (1970) assumed this type of additive noise in their well- known study on defining increasing risk. In our case, the noise term represents vulnerability. Hence, the indi-vidual’s income is now z1 + e and the corresponding state- dependent utility is U(z1 + e), where z1 is the new poverty line. We refer to this formulation as the additive noise model.

We assume a utility consistency condition, which says that utility derived from the given poverty line z0 and the expected utility from the poverty line z1, accompanied by the noise, should be equal. That is, the poverty line z1 should be such that the person becomes indifferent between the expected



utility from the vulnerable income z1 + e and the certain utility from z0. Thus, U(z0) = E(U(z1 + e)), where E stands for the expectation operator. This idea is similar in spirit to the notion of certainty equivalent and risk- neutral valuation employed in the theory of finance. According to risk- neutral valu-ation, the current period stock price is the discounted present value of the expected value of the future period stock prices, where the discounting is done using a risk- free rate of interest (Demange and Laroque 2006).

Expanding the right- hand side of the expression U(z0) = E(U(z1 + e)) by Taylor’s expansion around z1, we have:

U(z0) 5 EaU(z1) 1 eU r (z1) 1 ae2

2bUs (z1) 1 . . .b.

Ignoring higher- order terms greater than 2, we have:

U(z0) 5 U(z1) 1 as2

2bUs (z1) . (4.5)

Given that U0 < 0, we obtain,

U(z0) 2 U(z1) 5 as2

2bUs (z1) , 0,

which implies that z1 > z0. Intuitively, this is a quite reasonable result. Because z1 is the poverty line in the presence of vulnerability, the value of z1 should be higher that of z0 so that with the additional income the indi-vidual can cope with the disturbance in income generated by vulnerability and becomes equally well off as he was with z0.

We can rewrite equation (4.5) as F(z0, z1) = 0, where F is a real- valued function defined on the positive part of the two- dimensional Euclidean space. By the implicit function theorem, we can solve F(z0, z1) = 0 for z1 as a function of z0 (Apostol 1971).

We are interested in finding non- trivial solutions and we try some special cases for which non- trivial solutions can be found by inspection. As a simple trial, suppose U(.) satisfies constant absolute risk aversion. That is, U(z) = A − Be−az, where a > 0, B > 0 and A are constants. In fact, a is the constant value of the absolute risk aversion measure.

The first derivative is written as U9(z) = −B(−a)e−az = Bae−az.The second derivative is expressed as U0(z) = aB(−a)e−az = −Ba2e−az.Thus, equation (4.5) implies that:

A 2 Be2az0 5 A 2 Be2az1 1 as2A

2b (2Ba2)e2az1



where s2A denotes the variance in this absolute case.

This leads to the result:

z1 5 z0 1 a 1ablna1 1 as2

A

2ba2b. (4.6)

Thus the adjusted poverty line z1 is easily estimated3 on the basis of:

● the original poverty line, z0; ● the variance s2

A of the error e; and ● the coefficient of absolute risk aversion, a.

From equation (4.6), it follows that z1 = z0 + b, where b > 0 is a constant, depending only on a and s. That is, z1 is a positive translation of z0. Thus, with a constant absolute aversion to risk we get that the new poverty line is an absolute positive shift of the existing poverty line. The term, b, may be regarded as a compensation factor for vulnerability. For instance, for a poor country where the poverty line is assumed to cover only basic needs, if one wants to take vulnerability into account, one just moves upward the original poverty line (the one that ignores vulnerability) by a constant. This absolute shift does not depend on the existing poverty line. It is explicitly dependent on the noise representing vulnerability and the nature of risk aversion given by the utility function.

3.2 Constant Relative Risk Aversion

Assume again that a person at the poverty line z0 without vulnerability has a certain utility U(z0). Assume now that the individual is subjected to an uncertain income characterized by the proportional noise e defined above. The individual’s income is now z2(1 + e) and the corresponding state- dependent utility is U(z2(1 + e)). Given that there is indifference in the two situations, U(z0) = E[U(z2(1 + e))]. Expanding the right- hand side by a Taylor’s expansion, we have:

U(z0)5E eU(z2) 1 ez2U r (z2) 1e2

2z2

2 Us (z2) 1 cf .

Taking approximations, we obtain:

U(z0) 5 U(z2) 1s2

R

2z2

2Us (z2) (4.7)



where s2R is now the variance of e.

We thus have

U(z0) 2 U(z2) 5s2

R

2z2

2Us (z2) , 0

because U0 < 0. Hence, z2 > z0. Again, in view of the implicit function theorem, we can always solve z2 in terms of z0.

Let the utility function be defined as

U(z) 5 A1 1 B1z12d

1 2 d, where B1 > 0, A1

and 0 < d ≠ 1 are constants.The first and second derivatives are:

U r (z) 5 B1 1

1 2 d (1 2 d)z2d 5 B1z2d; and U0(z) = B1(−d)z−d − 1.

On the basis of the utility function previously defined, using equation (4.7), we obtain:

A1 1 B1

(z0) 12d

1 2 d5 A1 1 B1

(z2) 12d

1 2 d1

s2R

2(z2) 2B1 (2d) (z2)2d21.

Again, by simple but tedious algebra, we obtain:

z2 5 z0 c1 2 d (1 2 d)s2

R

2d 21/(12d).

(4.8)

Here also it is easy to calculate4 the adjusted poverty line z2 on the basis of:

● the original poverty line z0;● the variance s2

R; and● the coefficient of relative risk aversion, d.

We now consider the particular case where d = 1. We may then write that:

U(z) = A1 + B1ln z.

Thus we obtain:

U r (z) 5 B1a1zb and Us (z) 5 2B1

1z2.



Again, using equation (4.7), we derive that:

A1 1 B1ln z0 5 A1 1 B1ln z2 1s2

R

2(z2) 2 (2B1)

1(z2) 2

which leads to:

ln z2 5 ln z0 1s2

R

2

And

z2 5 z0b 5 z0e(s2R/2) (4.9)

where b is defined as ln b 5s2

R

2 .To compute z2, we need only to know s2

R and the original poverty line z0.From equation (4.9), it is clear that the constant relative risk aversion

utility is not consistent with an additive shift of the poverty line.In short, we note that the proportionally adjusted poverty line can be

justified by assuming a constant relative risk aversion utility under a mul-tiplicative model of vulnerability to poverty. For instance, for a country where the poverty line is also assumed to take into account the ‘cost of social inclusion’ (relative poverty), if one wants to take vulnerability into account one would have to implement a scale transformation of the origi-nal poverty line (the one ignoring vulnerability).

We may now summarize the previous observations in the following two propositions:

Proposition 1: In the additive noise model, under constant absolute risk aversion, the vulnerability- adjusted poverty line is a positive translation of the existing poverty line. In this additive model, the translation shift is not supported by a constant relative risk aversion utility function.

Proposition 2: In the multiplicative noise model, under constant relative risk aversion, the vulnerability- adjusted poverty line is a relatively aug-mented transformation of the existing poverty line. In this multiplicative model, the scale transformation is not supported by a constant absolute risk aversion utility function.



4 ESTIMATION OF THE VARIANCE IN THE MULTIPLICATIVE CASE5

Given some income distribution which is supposed to be subject to vulner-ability, we need to derive the variance V(e) (denoted above by s2

R) in the multiplicative case.

Let X denote income that would be observed if there was no vulnerabil-ity and let z0 be the poverty line in such a case at, say, time 0. Assume that, at some time t, the appropriate income variable is Yt but this is assumed to be subject to vulnerability, in the sense that it is generated by taking into account the presence of a noise term et in addition to the existing distribu-tion at time 0. We assume a multiplicative model:

Yt = X(1 + et) (4.10)

where X and et are assumed to be uncorrelated.Hence:

lnYt = lnX + ln(1 + et) (4.11)

Assume we have information on T distributions, Yt, t = 1, 2,. . ., T.We therefore write:

a 1TbaT

t51lnYt 5 lnX 1 a 1

TbaT

t51ln(1 1 et) .

Under first- order approximation of ln(1 + et) we may write that ln(1 + et) ≈ et so that:

a 1TbaT

t51lnYt 5 lnX 1 a 1

TbaT

t51et. (4.12)

We rewrite equation (4.12) as

a 1TbaT

t51lnYt < lnX 1 e

where e is the average of all the et. Assume that the variance s2Rt of et is the

same for all t, so that we can write:

s2Rt 5 s2

R.



The variance Var(e) of e will then be expressed as

Var(e) 5s2

R

T,

so that Var(e) S 0 if s is small or if T → ∞.It then follows that:

a 1TbaT

t51lnYt < lnX. (4.13)

From equation (4.13), we obtain:

V(lnX) > V c a 1TbaT

t51lnYt d 5 a 1

T2baTt51

V(lnYt) . (4.14)

We assume that we have data, for each period t, and that the observations on Yt have as typical element, an income, Yit, with i varying from 1 to n (for example, n = 100 000). We, therefore, also have, for each period t, obser-vations on lnYt whose typical element is lnYit. As a consequence, we can approximate the variance V(lnYt) of these lnYit and, using equation (4.10), after having estimated this variance for each time period t, we are able to estimate the variance we require, namely, V(lnX).

On the basis of the observations on lnYt, as outlined above, we can also estimate, for each period t, the expectation, E(lnYt).

Using equation (4.13), we then also obtain:

E(lnX) 5 Ea 1Ta

T

t51lnYtb 5

1Ta

T

t51E(lnYt) . (4.15)

We estimate E(lnYt) by [ (1n)a

n

i51lnYit], the sample mean of the log observa-

tions for the tth period. The mean of these sample means is then the esti-mate of E(lnX), so that we not only estimate V(lnX) but also E(lnX).

We may now use the Taylor’s expansion of lnX to obtain

E(lnX) < lnE(X) 21

2[E(X) ]2 V(X) (4.16)

and

V(lnX) <1

[E(X) ]2V(X) . (4.17)



Combining equations (4.16) and (4.17), we obtain:

12

V(lnX) 1 E(lnX) <1

2[E(X) ]2V(X) 1 E(lnX) < lnE(X) . (4.18)

Because we previously estimated both V(lnX) and E(lnX), we have now estimated lnE(X).

We then derive also that:

E(X) 5 elnE(X) (4.19)

From equation (4.19), we derive also [E(X)]2.Using equation (4.20), we conclude that

V(X) < V(lnX)[E(X)]2 (4.20)

which enables us to estimate V(X).From equation (4.10), and using the well- known formula for the vari-

ance of the product of two uncorrelated random variables, we then obtain:

V(Yt) = V(X) + V(Xet) = V(X) + {[V(X)V(et)] + [V(X)[E(et)]2] + [V(et)[E(X)]2]}.

Because E(et)50, the third term on the right- hand side is zero, and so we obtain:

V(et) 5V(Yt) 2 V(X)

V(X) 1 [E(X) ]2. (4.21)

Because we previously estimated the values of V(Yt), V(X) and [E(X)]2, using equation (4.21), we are able to estimate V(et) 5 s2

R.This allows then, using equation (4.9), to estimate the adjusted poverty

line in the case of vulnerability and constant relative risk aversion.

5 EMPIRICAL ILLUSTRATIONS

Implementing either the additive or the multiplicative case requires esti-mating (or approximating) the value of the risk aversion parameter. In the additive case, the parameter value depends on the unit of measurement of the income, consumption or other well- being variable. To the best of our knowledge, no prior estimates of this parameter have been obtained using



consumption or income in 2005 purchasing power parities (PPPs). In this chapter, we implement the framework with multiplicative risks and present estimates for the case of constant relative risk aversion.

There have been numerous attempts to estimate the coefficient of rela-tive risk aversion. Hartley et al. (2013) start their study by reviewing the literature on this topic, and mention, among the many papers they cite, the following results. Szpiro (1986) derived his estimates of the coefficient of relative risk aversion (CRRA) from time- series data on insurance pre-miums and concluded that the CRRA was close to 2. Barsky et al. (1997) worked with the US Health and Retirement Survey and estimated the CRRA to have a mean of about 12. Hersch and McDougall (1997) used data from the Illinois Instant Riches television game show and found evi-dence of a high value for the CRRA, up to 15. Jianakoplos and Bernasek (1998) analyzed US household portfolio data on risky assets and con-cluded that single women are more risk- averse than single men because the former had a CRRA of 9, with 6 for the latter. Beetsma and Schotman (2001) used a Dutch television game show called Lingo and derived a range of 3 to 7 for the CRRA. Attanasio et al. (2002) used a large UK sample survey and obtained an estimate of the CRRA of 1.44. Chetty (2003) derived estimates of the CRRA on the basis of labor supply elasticities and found that the CRRA was close to 1. Fullenkamp et al. (2003) took the Hoosier Millionaire television game show as the data base and found that the CRRA varied between 0.64 and 1.76. Chiappori and Paiella (2011) preferred to use panel data because these data allow one to disentangle the impact of the shape of individual preferences and that of the correlation between preferences and wealth. They found that the median of the CRRA was around 2 but, for one- fourth of the population, the CRRA was larger than 3. Gandelman and Hernández- Murillo (2011) used information on self- reports of subjective personal well- being from three datasets: the Gallup World Poll, the European Social Survey and the World Values Survey. They concluded that the CRRA varied between 0.79 and 1.44. Hartley et al. (2013) themselves analyzed data of the famous game show Who Wants to Be a Millionaire and reached the conclusion that the CRRA was close to 1.

The short survey above on the CRRA shows clearly that there is quite a wide range of possible values for this coefficient, more or less from 0.5 to 15. The empirical illustration below covers mainly poor countries in Asia and it is reasonable to apply medium CRRA values to generate poverty lines. We therefore assume that the CRRA is equal to 3. More precisely, starting from an original poverty line of $1.25 (the official poverty line derived by the World Bank) which serves as a benchmark, we estimate what a vulnerability- adjusted poverty line would be for each country examined.



The computations, presented in Tables 4.1 to 4.3, as well as those given in Table 4A.2 in Appendix 4A.2, are based on a technique originally pro-posed by Shorrocks and Wan (2009), which allows one to considerably increase the number of observations, even when starting from, say, only ten observations for a given country and year (for example, income deciles). This technique, known as ‘ungrouping income distributions’, is described in Table 4A.1 in Appendix 4A.2.

In Appendix Table 4A.2, we present additional results where the CRRA is equal to 1.8, 5 and 10. Assuming that the CRRA is equal to 3, Table 4.1 shows that, for 2005, large values for vulnerability- adjusted poverty lines are observed for the People’s Republic of China (PRC) ($1.88), Thailand ($1.56), Turkmenistan ($1.56), Georgia ($1.51), Malaysia ($1.51), and Viet Nam ($1.50). In 2010, the order did not change much – countries with high poverty lines include the PRC ($2.26), Malaysia ($1.82), Azerbaijan ($1.66), Viet Nam ($1.60), Thailand ($1.59), Tajikistan (1.58), and Turkmenistan ($1.56).

Using these vulnerability- adjusted poverty lines, we computed the poverty rates and the number of poor (see Table 4.2). In the PRC, for example, we observe that once vulnerability is incorporated, its poverty rate becomes equal to 31.8 percent in 2005 and 28.7 percent in 2010. The correspond-ing rates are respectively: 30.6 percent in 2005 and 24.5 percent in 2010 for Pakistan; 56.4 percent in 2005 and 50.9 percent in 2010 for Bangladesh; 48.1 percent in 2005 and 41.6 percent in 2010 for India; 54.4 percent in 2005 and 39.6 percent in 2010 for Nepal; 29.6 percent in 2005 and 27.1 percent in 2010 for Indonesia; 30.0 percent in 2005 and 26.4 percent in 2010 for the Philippines; and 35.1 percent in 2005 and 25.4 percent in 2010 for Viet Nam.

In Table 4.3, we present a summary of the results concerning the head-count ratios and the number of poor for developing Asia as a whole in 2005, 2008 and 2010, under various assumptions regarding the coefficient of relative risk aversion. We also show what the headcount ratio and the number of poor are, when no adjustment is made for vulnerability so that the poverty line is assumed to be equal to $1.25 (our benchmark). We observe the important increases in the headcount ratio when vulnerability is taken into account, even when the CRRA is equal to 1.2. Note also that the headcount ratio (and the number of poor) increases with the CRRA but only up to a value of 5. When the CRRA is equal to 10, the values of the headcount ratio (and the number of poor) are smaller than when the CRRA is equal to 5. Assuming a coefficient of relative of risk aversion equal to 3, we can see that in 2005 there are 350 000 more poor than under the benchmark case (poverty line equal to $1.25). In 2010 the difference is even higher (more than 400 000 additional poor).



If we now take a look at specific countries (see Table 4A.1), for example, for the year 2010, we observe that under the benchmark of a poverty line of $1.25, the headcount ratio is equal to 11.6 percent in the PRC, 32.7 percent in India, 18.2 percent in Indonesia, 13.5 percent in Pakistan and 18.4

Table 4.1 Vulnerability- adjusted poverty lines for countries in Asia and the Pacific (CRRA = 3)

Sub- region/country 2005 2008 2010

Central and West AsiaArmenia 1.39 1.45 1.39Azerbaijan 1.46 1.60 1.66Georgia 1.51 1.53 1.51Kazakhstan 1.38 1.41 1.42Kyrgyz Republic 1.36 1.56 1.49Pakistan 1.40 1.39 1.47Tajikistan 1.46 1.57 1.58Turkmenistan 1.56 1.56 1.56East Asia (PRC) 1.88 2.15 2.26

South AsiaBangladesh 1.35 1.37 1.38Bhutan 1.36 1.44 1.50India 1.37 1.39 1.40Maldives 1.47 1.38 1.46Nepal 1.43 1.50 1.56Sri Lanka 1.42 1.45 1.45

Southeast AsiaCambodia 1.37 1.43 1.46Indonesia 1.44 1.43 1.49Lao PDR 1.38 1.41 1.47Malaysia 1.51 1.81 1.82Philippines 1.48 1.48 1.49Thailand 1.56 1.55 1.59Viet Nam 1.50 1.56 1.60

PacificFiji 1.41 1.46 1.48Federated States of Micronesia, (Urban)

1.38 1.40 1.41

Papua New Guinea 1.38 1.40 1.41Timor- Leste 1.35 1.35 1.34

Note: CRRA = coefficient of constant relative risk aversion.


100

Tabl

e 4.

2 Po

vert

y in

cou

ntri

es o

f A

sia

and

the

Paci

fic u

nder

vul

nera

bilit

y- ad

just

ed p

over

ty li

nes (

CR

RA

= 3

)

Sub-

regi

on/c

ount

ryPo

vert

y ra

te (%

)N

umbe

r of

poor

(mill

ion)

2005

2008

2010

2005

2008

2010

Cen

tral

and

Wes

t Asi

a25

.823

.920

.453

.12

51.5

745

.59

Arm

enia

6.6

2.8

4.1

0.20

0.09

0.13

Aze

rbai

jan

2.8

1.0

0.6

0.23

0.09

0.06

Geo

rgia

21.7

21.1

23.7

0.95

0.93

1.05

Kaz

akhs

tan

1.4

0.1

0.5

0.21

0.02

0.07

Kyr

gyz

Rep

ublic

26.5

12.2

12.1

1.36

0.64

0.65

Paki

stan

30.6

29.0

24.5

48.4

848

.57

42.4

8Ta

jikist

an25

.018

.316

.41.

611.

221.

13Tu

rkm

enist

an1.

60.

40.

20.

080.

020.

01E

ast A

sia

(PR

C)

31.8

30.3

28.7

414.

3940

1.53

384.

05So

uth

Asi

a48

.646

.042

.064

6.82

637.

3859

9.28

Ban

glad

esh

56.4

53.6

50.9

79.2

477

.92

75.6

2B

huta

n22

.815

.29.

10.

150.

110.

07In

dia

48.1

45.6

41.6

549.

2054

3.56

509.

96M

aldi

ves

4.2

0.6

0.9

0.01

0.00

0.00

Nep

al54

.446

.539

.614

.83

13.4

311

.87

Sri L

anka

17.0

11.6

8.5

3.38

2.37

1.77


101

Sout

heas

t Asi

a26

.024

.422

.013

1.93

128.

3611

8.54

Cam

bodi

a39

.430

.823

.35.

274.

253.

29In

done

sia29

.630

.927

.167

.18

72.4

964

.94

Lao

PD

R47

.242

.036

.52.

712.

532.

26M

alay

sia0.

91.

21.

20.

230.

320.

35Ph

ilipp

ines

30.0

27.0

26.4

25.6

824

.30

24.6

3T

haila

nd2.

91.

41.

51.

910.

961.

01V

iet N

am35

.127

.625

.428

.95

23.5

122

.06

Paci

fic47

.643

.040

.53.

793.

653.

59F

iji21

.69.

212

.30.

180.

080.

11Fe

dera

ted

Stat

es o

f M

icro

nesia

, (U

rban

)33

.235

.135

.20.

010.

010.

01Pa

pua

New

Gui

nea

51.1

47.8

44.1

3.12

3.13

3.02

Tim

or- L

este

47.9

40.5

40.3

0.48

0.44

0.45

Dev

elop

ing

Asi

a37

.335

.332

.612

50.0

412

22.5

011

51.0

5

Not

e:

The

dat

a in

col

umns

2 to

4 in

dica

te th

e va

lue

of th

e he

adco

unt r

atio

s whe

n th

e or

igin

al p

over

ty li

ne o

f $1

.25

is ad

just

ed fo

r vul

nera

bilit

y w

ith a

coe

ffici

ent o

f co

nsta

nt re

lativ

e ris

k av

ersio

n (C

RR

A) e

qual

to 3

. The

dat

a in

col

umns

5 to

7 g

ive

the

corr

espo

ndin

g nu

mbe

rs o

f po

or. T

he

term

‘Dev

elop

ing

Asia

’ cov

ers a

ll th

e co

untr

ies t

hat a

ppea

r in

the

tabl

e.


102

Tabl

e 4.

3 V

ulne

rabi

lity-

adju

sted

hea

dcou

nt ra

tios a

nd n

umbe

r of

poor

for d

evel

opin

g A

sia:

sum

mar

y of

resu

lts

Hea

dcou

nt ra

tios

(200

5)H

eadc

ount

ratio

s (2

008)

Hea

dcou

nt ra

tios

(201

0)N

umbe

r of

poor

(2

005)

Num

ber o

f po

or

(200

8)N

umbe

r of

poor

(2

010)

Ben

chm

ark

(p

over

ty li

ne =

$1

.25)

26.9

23.9

20.7

901.

9682

7.57

733.

06

CR

RA

= 1

.232

.130

.227

.510

77.8

210

46.1

797

3.39

CR

RA

= 1

.533

.331

.628

.911

18.5

710

93.6

710

21.4

2C

RR

A =

1.8

34.4

32.7

30.0

1154

.45

1132

.99

1061

.90

CR

RA

= 2

35.0

33.4

30.7

1175

.12

1154

.82

1084

.06

CR

RA

= 3

37.3

35.3

32.6

1250

.04

1222

.50

1151

.05

CR

RA

= 5

38.9

36.5

33.6

1306

.34

1262

.14

1186

.69

CR

RA

= 1

038

.535

.532

.412

93.2

712

29.0

611

46.6

7



percent in the Philippines. When we adjust the poverty line for vulnerability and select a CRRA of 1.8, the corresponding percentages are 23.2 (PRC), 36.7 (India), 21.8 (Indonesia), 18.8 (Pakistan) and 21.8 (the Philippines). With a CRRA equal to 5 the headcount ratios in these countries become 26.7 percent (PRC), 44.9 percent (India), 29.5 percent (Indonesia), 27.9 percent (Pakistan) and 26.4 percent (the Philippines). Finally, when the CRRA is equal to 10 the headcount ratios are 22.0 percent in the PRC, 46.6 percent in India, 29.7 percent in Indonesia, 28.4 percent in Pakistan, and 28.6 percent in the Philippines.

We thus observe more than a doubling of the headcount ratio in Pakistan and the PRC when we compare the situation under a poverty line of $1.25 with that adjusted for vulnerability with a CRRA equal to 5. For India the increase is higher than 35 percent, for the Philippines it is higher than 50 percent and for Indonesia the rise is of 60 percent. Therefore, when vulner-ability is taken into account, the extent of poverty in the most populated countries of Asia, and hence in Asia as a whole is modified significantly.

6 CONCLUSIONS

In this chapter, we have addressed the problem of modifying the poverty line when the income distribution is affected by vulnerability. The formal framework considered in the chapter relies on the Ligon and Schechter (2003) definition of vulnerability as expected utility loss. Under alternative assumptions about the uncertainty (noise) that indicates vulnerability, it is shown that for the constant absolute or relative Arrow–Pratt risk aver-sions, the tailored poverty line becomes either an absolute or relative shift of the current poverty line. The empirical illustration, based on data from various Asian and Pacific countries, assumed constant relative risk aver-sion and showed the important impact of vulnerability on the number of poor in various Asian countries.

We thus observed generally important increases in the headcount ratio when vulnerability is taken into account. For example, assuming a coef-ficient of relative of risk aversion equal to 3, in 2005 there were 350 000 more poor than under the benchmark case (poverty line equal to $1.25). In 2010, the difference was even higher (more than 400 000 additional poor). Looking at specific countries, we observed more than a doubling of the headcount ratio in Pakistan and the PRC when we compared the situation under a poverty line of $1.25 with that adjusted for vulnerability with a CRRA equal to 5. For India, the Philippines and Indonesia, the increases were respectively higher than 35 percent, 50 percent and 60 percent. It is thus clear that when vulnerability is taken into account, the extent of



poverty in the most populated countries of Asia, and, hence, in Asia as a whole, is modified significantly.

NOTES

1. The authors thank Iva Sebastian- Samaniego for her tremendous help in deriving the results of the empirical section of this chapter.

2. Chakravarty et al. (2015) explored a partial ordering of vulnerability to poverty induced by expected poverty indices. See also Hardeweg et al. (2013) for a stochastic dominance- based partial ordering.

3. Using the definition of equation (4.1), it is easily shown that APA(z) = a for this model.4. It may be noted that, using equation (4.2), we obtain APR(q) = d for this proportional

model.5. As stressed below at the beginning of section 5, in the additive case, the parameter value

depends on the unit of measurement of the income, consumption or other well- being variable, so that implementing such an approach becomes very difficult. Hence, we con-centrate our attention on the multiplicative case.

REFERENCES33*

Amin, S., A.S. Rai and G. Topa (2003), ‘Does microcredit reach the poor and vulnerable? Evidence from northern Bangladesh’, Journal of Development Economics, 70 (1), 59–82.

Apostol, T. (1971), Mathematical Analysis, London: Addison Wesley.Arrow, K.J. (1965), Aspects of the Theory of Risk Bearing, Helsinki: Yrjo Jahnssonin

Saatio.Arrow, K.J. (1970), Essays in the Theory of Risk Bearing, Amsterdam:

North Holland.Attanasio, O., J. Banks and S. Tanner (2002), ‘Asset holding and consumption

volatility’, Journal of Political Economy, 110 (4), 771–92.Barsky, R.B., F.T. Juster, M.S. Kimball and M.D. Shapiro (1997), ‘Preference

parameters and behavioral heterogeneity: an experimental approach in the health and retirement study’, Quarterly Journal of Economics, 112 (2), 537–79.

Beetsma, R.M. and P.C. Schotman (2001), ‘Measuring risk attitudes in a natural experiment: data from the television game show Lingo’, Economic Journal, 111 (474), 821–48.

Binswanger, H.P. (1981), ‘Attitudes toward risk: theoretical implications of an experiment in rural India’, Economic Journal, 91 (364), 867–90.

Blackorby, C., D. Donaldson and J.A. Weymark (1984), ‘Social choice theory with interpersonal utility comparisons: a diagrammatic introduction’, International Economic Review, 25 (2), 327–56.

Boadway, R. and N. Bruce (1984), Welfare Economics, London: Blackwell.Calvo, C. and S. Dercon (2013), ‘Vulnerability of individual and aggregate

poverty’, Social Choice and Welfare, 41 (4), 721–40.

* The Asian Development Bank recognizes Vietnam as Viet Nam.



Chakravarty, S.R. (1983), ‘A new index of poverty’, Mathematical Social Sciences, 6 (3), 307–13.

Chakravarty, S.R. (2013), An Outline of Financial Economics, New York: Anthem Press.

Chakravarty, S.R., N. Chattopadhyay and L. Qingbin (2015), ‘Vulnerability order-ings for expected poverty indices’, Japanese Economic Review, 66 (3), 300–310.

Chaudhuri, S., J. Jalan and A. Suryahadi (2002), ‘Assessing household vul-nerability to poverty from cross- section data: a methodology and estimates from Indonesia,’ Discussion Paper Series 0102- 52, Department of Economics, Columbia University.

Chetty, R. (2003), ‘A new method of estimating risk aversion’, NBER Working Paper 9988, National Bureau of Economic Research, Cambridge. MA.

Chiappori, P.- A. and M. Paiella (2011), ‘Relative risk aversion is constant: evidence from panel data’, Journal of the European Economic Association, 9 (6), 1021–52.

Christiaensen, L. and K. Subbarao (2005), ‘Towards an understanding of house-hold vulnerability in rural Kenya’, Journal of African Economies, 14 (4), 520–58.

Coudouel, A. and J. Hentschel (2000), ‘Poverty data and measurement’, draft for the Sourcebook on Poverty Reduction Strategies, World Bank, Washington, DC.

Dang, H.H. and P.F. Lanjouw (2014), ‘Two definitions of a vulnerability line and their empirical application’, Policy Research Working Paper No. 6944, World Bank, Washington, DC.

Demange, G. and G. Laroque (2006), Finance and the Economics of Uncertainty, London: Blackwell.

Dercon, S. (2002), ‘Income risk, coping strategies and safety nets’, World Bank Research Observer, 17 (2), 141–66.

Dutta, I., J. Foster and A. Mishra (2011), ‘On measuring vulnerability to poverty’, Social Choice and Welfare, 37 (4), 743–61.

Elbers, C. and J. W. Gunning (2003), ‘Vulnerability in a stochastic dynamic model’, mimeo., Tinbergen Institute, Amsterdam.


Fujii, T. (2013), ‘Vulnerability: a review of literature’, paper presented at the Workshop on Poverty Reduction in Asia: Drivers, Best Practices and Policy Initiatives, organized by the Asian Development Bank and Chongqing University.

Fullenkamp, C., R. Tenorio and R. Battalio (2003), ‘Assessing individual risk atti-tudes using field data from lottery games’, Review of Economics and Statistics, 85 (1), 218–26.

Gaiha, R. and K. Imai (2009), ‘Measuring vulnerability and poverty: estimates for rural India’, in W. Naude, A. Santos- Paulino and M. McGillivray (eds), Vulnerability in Developing Countries, Helsinki: United Nations University Press, pp. 13–54.

Gándelman, N. and R. Hernández- Murillo (2011), ‘What do happiness and health satisfaction data tell us about relative risk aversion’, Working Paper 2011- 039, Federal Reserve Bank of St. Louis.

Glewwe, P. and G. Hall (1998), ‘Are some groups more vulnerable to macroeco-nomic shocks than others? Hypothesis tests based on panel data from Peru’, Journal of Development Economics, 56 (1), 181–206.

Gravelle, H. and R. Rees (2004), Microeconomics, Upper Saddle River, NJ: Prentice Hall.

Günther, I. and K. Harttgen (2009), Estimating Households Vulnerability to



Idiosyncratic and Covariate Shocks: A Novel Method Applied in Madagascar. World Development 37 (7), 1222–1234.

Hardeweg, B., A. Wagner and H. Waibel (2013), ‘A distributional approach to comparing vulnerability, applied to rural provinces in Thailand and Vietnam’, Journal of Asian Economics, 25 (April), 53–65.

Hartley, R., G. Lanot and I. Walker (2013), ‘Who really wants to be a mil-lionaire: estimates of risk aversion from game show data’, Journal of Applied Econometrics, online version, 18 September, DOI: 10.1002/jae.2353.

Hersch, P.L. and G.S. McDougall (1997), ‘Decision making under uncertainty when the stakes are high: evidence from a lottery game show’, Southern Economic Journal, 64 (1), 75–84.

Hoddinott, J. and A. Quisumbing (2003), ‘Methods for micro- econometric risk and vulnerability assessments’, Social Protection Discussion Paper 0324, World Bank, Washington, DC.

Holzmann R. and S. Jorgensen (1999), ‘Social protection as social risk manage-ment: conceptual underpinnings for social protection sector strategy paper’, Social Protection Discussion Paper 9904, World Bank, Washington, DC.

Hoogeveen, J., E. Tesliuc, R. Vakis and S. Dercon (2004), ‘A guide to analysis of risk, vulnerability and vulnerable groups’, Policy Research Working Paper, World Bank, Washington, DC.

Jianakoplos, N.A. and A. Bernasek (1998), ‘Are women more risk averse?’, Economic Inquiry, 36 (4), 620–30.

Kamanou, G. and J. Morduch (2004), ‘Measuring vulnerability to poverty’, in S. Dercon (ed.), Insurance against Poverty, New York: Oxford University Press, pp. 155–75.

Klasen, S. and F. Povel (2013), ‘Defining and measuring vulnerability: state of the art and new proposals’, in S. Klasen and H. Waibel (eds), Vulnerability to Poverty: Theory, Measurement and Determinants, New York: Palgrave- Macmillan.

Ligon, E. and L. Schechter (2003), ‘Measuring vulnerability’, Economic Journal, 113 (486), C95–C102.

Ligon, E. and L. Schechter (2004), ‘Evaluating different approaches to estimat-ing vulnerability’, Social Protection Discussion Paper 0410, World Bank, Washington, DC.

Pratt, J.W. (1964), ‘Risk aversion in the small and the large’, Econometrica, 32 (1/2), 122–36.

Pritchett, L., A. Suryahadi and S. Sumarto (2000), ‘Quantifying vulnerability to poverty: a proposed measure, with application to Indonesia’, Social Monitoring and Early Response Unit Working Paper, World Bank, Washington, DC.

Ravallion, M. (1988), ‘Expected poverty under risk- induced welfare variability’, Economic Journal, 98 (393), 1171–82.

Rothschild, M. and J.E. Stiglitz (1970), ‘Increasing risk: I. A definition’, Journal of Economic Theory, 2 (3), 225–43.

Ryu, H.K. and D.J. Slottje (1999), ‘Parametric approximations to the Lorenz curve’, in J. Silber (ed.), Handbook of Income Inequality Measurement, Dordrecht: Kluwer Academic, pp. 291–314.

Sen, A.K. (1977), ‘Social choice theory: a re- examination’, Econometrica, 45 (1), 53–88.

Sen, A.K. (1999), ‘A Plan for Asia’s growth: build on much that is good in the “Eastern Strategy”’, Asia Week, 25 (40), accessed 1 April 2016 at http://edition.cnn.com/ASIANOW/asiaweek/magazine/99/1008/viewpoint.html.



Shorrocks, A. and G. Wan (2009), ‘Ungrouping Income Distributions: Synthesizing Samples for Inequality and Poverty Analysis’, in K. Basu and R. Kanbur (eds), Arguments for a Better World. Essays in Honor of Amartya Sen, vol. 1, Ethics, Welfare and Measurement, Oxford: Oxford University Press, pp. 414–34.

Skoufias, E. and A.R. Quisumbing (2005), ‘Consumption insurance and vulner-ability to poverty: a synthesis of evidence from Bangladesh, Ethiopia, Mali, Mexico and Russia’, European Journal of Development Research, 17 (1), 24–58.

Suryahadi, A. and S. Sumarto (2003), ‘Poverty and vulnerability in Indonesia before and after the economic crisis’, Asian Economic Journal, 17 (1), 45–64.

Szpiro, G.G. (1986), ‘Measuring risk aversion: an alternative approach’, Review of Economics and Statistics, 68 (1), 156–9.

Townsend, R. (1994), ‘Risk and insurance in village India’, Econometrica, 62 (3), 539–91.

Watts, H.W. (1968), ‘An economic definition of poverty’, in D.P. Moynihan (ed.), On Understanding of Poverty, New York: Basic Books, pp. 316–19.

World Bank (1997), ‘Identifying the vulnerable: new evidence from Peru’, Poverty Lines, A Joint Publication by the Policy Research and Social Policy Departments, World Bank, No. 6, accessed 1 April 2016 at http://web.worldbank.org/archive/website00002/WEB/PDF/PL_N06.PDF.

World Bank (2000), The World Development Report 2000/2001: Attacking Poverty, New York: Oxford University Press.



APPENDIX 4A.1 ON SHORROCKS AND WAN’S (2009) ‘UNGROUPING INCOME DISTRIBUTIONS’

Assume a Lorenz curve with (m + 1) coordinates (p*k, L*k) where p*k and L*k (k = 1, . . ., m) refer respectively to the cumulative shares in the total popu-lation and in total income of income classes 1 to k, while p*0 5 L*0 5 0. These Lorenz coordinates can, for example, refer to decile shares published on a given country. Because the corresponding average income is often not available, it is assumed to be equal to 1 so that the mean income m*k of class k is expressed as

m*k 5L*k 2 L*k21

p*k 2 p*k21, k = 1, . . ., m. (4A.1)

The goal is to obtain a synthetic sample of n equally weighted observations whose mean value is 1 and which conform to the original data. These n observations are therefore partitioned into m non- overlapping and ordered groups having each mk 5n(p*k 2 p*k21) observations. Call xki the ith obser-vation in class k, the sample mean of this class being mk.

The algorithm proposed by Shorrocks and Wan (2009) includes two stages. The first consists of building an initial sample with unit mean which is generated from a parametric form fitted to the grouped data (see, for example, Ryu and Slottje 1999 for a survey of various parametrizations of the Lorenz curve1).

In the second stage, the algorithm adjusts the observations generated in the initial sample to the true values available from the grouped data. More precisely, the initial sample value, xj, assumed to belong to class k, is trans-formed into an intermediate value xjˆ via the following rule:

xjˆ 2 m*km*k11 2 m*k

5xj 2 mk

mk11 2 mk

(4A.2)

For the first class, we write

xjˆ

m*15

xj

m1 for xj ≤ m1 (4A.3)

while, for the last class, we have:



xjˆ

m*m5

xj

mm for xj ≥ mm. (4A.4)

In the next iteration, the intermediate values xjˆ are themselves transformed into new values until the algorithm produces an ordered sample that exactly replicates the properties of the original grouped data. Convergence is, in fact, very quickly obtained.

Note

1. Shorrocks and Wan chose to generate the initial sample on the basis of a lognormal distribution. For more details, see, Shorrocks and Wan (2009).


110

APP

EN

DIX

4A

.2

VU

LN

ER

AB

ILIT

Y- A

DJU

STE

D H

EA

DC

OU

NT

RA

TIO

S A

ND

N

UM

BE

RS

OF

PO

OR

: DE

TAIL

ED

RE

SULT

S BY

CO

UN

TRY

Tabl

e 4A

.1

The

ben

chm

ark

case

(w

ithou

t vul

nera

bilit

y ad

just

men

t): h

eadc

ount

ratio

s and

num

ber o

f po

or w

hen

the

pove

rty

line

is $

1.25

Sub-

regi

on/c

ount

ryH

eadc

ount

ratio

(%

) 200

5H

eadc

ount

ratio

(%

) 200

8H

eadc

ount

ratio

(%

) 201

0N

umbe

r of

poor

(m

illio

n) 2

005

Num

ber o

f po

or

(mill

ion)

200

8N

umbe

r of

poor

(m

illio

n) 2

010

Cen

tral

and

Wes

t Asi

a18

.817

.111

.238

.79

37.0

525

.14

Arm

enia

4.0

1.3

2.5

0.12

0.04

0.08

Aze

rbai

jan

1.5

0.4

0.3

0.13

0.04

0.03

Geo

rgia

16.0

15.3

18.0

0.70

0.67

0.80

Kaz

akhs

tan

0.8

0.1

0.3

0.12

0.02

0.04

Kyr

gyz

Rep

ublic

22.9

6.4

6.7

1.18

0.34

0.36

Paki

stan

22.3

21.0

13.5

35.3

835

.23

23.3

8Ta

jikist

an17

.710

.76.

61.

140.

720.

45Tu

rkm

enist

an0.

50.

10.

10.

020.

010.

00E

ast A

sia

(PR

C)

16.3

13.1

11.6

211.

8517

3.00

155.

51So

uth

Asi

a41

.537

.833

.255

2.03

523.

8547

2.72

Ban

glad

esh

50.5

46.6

43.3

70.9

667

.82

64.3

1B

huta

n18

.99.

34.

40.

120.

070.

03In

dia

40.8

37.4

32.7

466.

3044

5.02

400.

08M

aldi

ves

2.3

0.2

0.2

0.01

0.00

0.00

Nep

al46

.333

.924

.812

.64

9.80

7.44

Sri L

anka

10.1

5.6

4.1

2.00

1.14

0.86


111

Sout

heas

t Asi

a18

.917

.214

.295

.87

90.4

776

.59

Cam

bodi

a33

.822

.814

.74.

513.

142.

08In

done

sia21

.422

.618

.248

.73

53.1

943

.32

Lao

PD

R39

.533

.926

.02.

272.

041.

61M

alay

sia0.

4–

–0.

11–

–Ph

ilipp

ines

22.2

19.4

18.4

19.0

217

.49

17.1

8T

haila

nd1.

00.

40.

40.

680.

250.

26V

iet N

am24

.916

.914

.020

.55

14.3

412

.14

Paci

fic43

.037

.834

.93.

423.

213.

10Fe

dera

ted

Stat

es o

f

Mic

rone

sia (u

rban

)30

.632

.132

.00.

010.

010.

01

Fiji

17.9

5.0

6.1

0.15

0.04

0.05

Papu

a N

ew G

uine

a46

.642

.538

.62.

842.

782.

65T

imor

- Les

te42

.034

.734

.70.

420.

370.

39D

evel

opin

g A

sia

26.9

23.9

20.7

901.

9682

7.57

733.

06

Not

e:

The

term

‘Dev

elop

ing

Asia

’ cov

ers a

ll th

e co

untr

ies t

hat a

ppea

r in

the

tabl

e.


112

Tabl

e 4A

.2

The

coe

ffici

ent o

f re

lativ

e ri

sk a

vers

ion

(CR

RA

) is

equ

al to

1.8

Sub-

regi

on/c

ount

ryH

eadc

ount

ratio

(%

) 200

5H

eadc

ount

ratio

(%

) 200

8H

eadc

ount

ratio

(%

) 201

0N

umbe

r of

poor

(m

illio

n) 2

005

Num

ber o

f po

or

(mill

ion)

200

8N

umbe

r of

poor

(m

illio

n) 2

010

Cen

tral

and

Wes

t Asi

a21

.820

.015

.644

.98

43.2

935

.01

Arm

enia

5.0

1.9

3.1

0.15

0.06

0.10

Aze

rbai

jan

2.0

0.6

0.4

0.17

0.05

0.04

Geo

rgia

18.8

18.2

20.4

0.82

0.80

0.91

Kaz

akhs

tan

1.0

0.1

0.3

0.15

0.02

0.05

Kyr

gyz

Rep

ublic

24.5

8.9

9.0

1.26

0.47

0.48

Paki

stan

25.9

24.5

18.8

41.0

540

.97

32.6

7Ta

jikist

an20

.613

.710

.91.

330.

920.

75Tu

rkm

enist

an0.

90.

20.

10.

040.

010.

01E

ast A

sia

(PR

C)

24.9

24.1

23.2

324.

1231

8.93

310.

71So

uth

Asi

a44

.641

.337

.159

3.72

573.

3352

9.30

Ban

glad

esh

53.0

49.7

46.6

74.5

672

.25

69.3

1B

huta

n20

.611

.95.

90.

140.

080.

04In

dia

44.0

41.0

36.7

502.

7448

7.99

449.

30M

aldi

ves

3.1

0.4

0.5

0.01

0.00

0.00

Nep

al50

.139

.831

.913

.66

11.5

09.

55Sr

i Lan

ka13

.27.

35.

32.

621.

501.

10


113

Sout

heas

t Asi

a22

.020

.417

.711

1.43

107.

2395

.05

Cam

bodi

a36

.326

.418

.84.

843.

642.

66In

done

sia24

.726

.421

.856

.24

61.9

252

.34

Lao

PD

R42

.937

.530

.82.

472.

261.

91M

alay

sia0.

60.

50.

50.

150.

130.

15Ph

ilipp

ines

25.9

22.5

21.8

22.1

520

.33

20.3

3T

haila

nd1.

80.

80.

81.

190.

530.

55V

iet N

am29

.621

.619

.724

.39

18.4

217

.09

Paci

fic45

.040

.037

.43.

583.

403.

32Fe

dera

ted

Stat

es o

f

Mic

rone

sia (u

rban

)31

.833

.433

.40.

010.

010.

01

Fiji

19.5

6.2

8.6

0.16

0.05

0.07

Papu

a N

ew G

uine

a48

.644

.941

.12.

962.

942.

82T

imor

- Les

te44

.537

.237

.10.

450.

400.

42D

evel

opin

g A

sia

32.1

30.2

27.5

1077

.82

1046

.17

973.

39

Not

e:

The

term

‘Dev

elop

ing

Asia

’ cov

ers a

ll th

e co

untr

ies t

hat a

ppea

r in

the

tabl

e.


114

Tabl

e 4A

.3

The

coe

ffici

ent o

f re

lativ

e ri

sk a

vers

ion

(CR

RA

) is

equ

al to

5

Sub-

regi

on/c

ount

ryH

eadc

ount

ratio

(%

) 200

5H

eadc

ount

ratio

(%

) 200

8H

eadc

ount

ratio

(%

) 201

0N

umbe

r of

poor

(m

illio

n) 2

005

Num

ber o

f po

or

(mill

ion)

200

8N

umbe

r of

poor

(m

illio

n) 2

010

Cen

tral

and

Wes

t Asi

a28

.426

.423

.258

.57

57.1

451

.87

Arm

enia

7.8

3.5

5.0

0.24

0.11

0.16

Aze

rbai

jan

3.3

1.1

0.7

0.27

0.10

0.06

Geo

rgia

23.3

22.6

25.3

1.02

0.99

1.13

Kaz

akhs

tan

1.7

0.1

0.6

0.26

0.02

0.09

Kyr

gyz

Rep

ublic

28.1

13.6

13.8

1.44

0.72

0.74

Paki

stan

33.7

32.2

27.9

53.4

853

.84

48.4

2Ta

jikist

an27

.520

.118

.21.

771.

341.

25Tu

rkm

enist

an1.

90.

50.

30.

090.

030.

01E

ast A

sia

(PR

C)

32.0

28.9

26.7

417.

1338

2.31

357.

09So

uth

Asi

a51

.449

.045

.368

3.74

679.

2564

4.90

Ban

glad

esh

58.8

56.4

53.8

82.7

281

.98

79.9

8B

huta

n24

.417

.010

.90.

160.

120.

08In

dia

50.9

48.7

44.9

581.

5757

9.98

549.

84M

aldi

ves

4.9

0.8

1.2

0.01

0.00

0.00

Nep

al56

.849

.642

.515

.49

14.3

312

.73

Sri L

anka

19.1

13.9

10.8

3.80

2.84

2.26


115

Sout

heas

t Asi

a28

.226

.624

.014

2.97

139.

6212

9.08

Cam

bodi

a41

.633

.526

.05.

564.

633.

68In

done

sia32

.433

.629

.573

.57

79.0

670

.85

Lao

PD

R50

.244

.839

.42.

882.

702.

44M

alay

sia1.

11.

21.

30.

270.

340.

37Ph

ilipp

ines

32.0

29.1

28.6

27.3

526

.25

26.6

4T

haila

nd3.

41.

71.

72.

241.

181.

20V

iet N

am37

.729

.927

.531

.10

25.4

623

.89

Paci

fic49

.444

.942

.33.

933.

813.

76Fe

dera

ted

Stat

es o

f

Mic

rone

sia (u

rban

)34

.236

.336

.30.

010.

010.

01

Fiji

23.2

11.1

14.1

0.19

0.09

0.12

Papu

a N

ew G

uine

a52

.849

.545

.83.

223.

253.

14T

imor

- Les

te50

.643

.142

.80.

510.

470.

48D

evel

opin

g A

sia

38.9

36.5

33.6

1306

.34

1262

.14

1186

.69

Not

e:

The

term

‘Dev

elop

ing

Asia

’ cov

ers a

ll th

e co

untr

ies t

hat a

ppea

r in

the

tabl

e.


116

Tabl

e 4A

.4

The

coe

ffici

ent o

f re

lativ

e ri

sk a

vers

ion

(CR

RA

) is

equ

al to

10

Sub-

regi

on/c

ount

ryH

eadc

ount

ratio

(%

) 200

5H

eadc

ount

ratio

(%

) 200

8H

eadc

ount

ratio

(%

) 201

0N

umbe

r of

poor

(m

illio

n) 2

005

Num

ber o

f po

or

(mill

ion)

200

8N

umbe

r of

poor

(m

illio

n) 2

010

Cen

tral

and

Wes

t Asi

a29

.827

.923

.561

.34

60.4

152

.68

Arm

enia

8.6

3.6

5.8

0.26

0.11

0.18

Aze

rbai

jan

3.4

1.0

0.6

0.28

0.09

0.06

Geo

rgia

23.2

22.4

25.2

1.01

0.98

1.12

Kaz

akhs

tan

1.9

0.2

0.6

0.29

0.02

0.10

Kyr

gyz

Rep

ublic

29.3

13.1

13.8

1.51

0.69

0.74

Paki

stan

35.4

34.2

28.4

56.0

957

.18

49.2

8Ta

jikist

an28

.019

.517

.31.

811.

311.

19Tu

rkm

enist

an1.

80.

50.

30.

080.

020.

01E

ast A

sia

(PR

C)

28.5

24.1

22.0

371.

6131

9.71

294.

00So

uth

Asi

a53

.450

.746

.971

0.77

703.

4366

7.90

Ban

glad

esh

60.9

58.4

55.7

85.6

284

.95

82.8

7B

huta

n25

.617

.510

.90.

170.

120.

08In

dia

53.0

50.5

46.6

605.

2660

1.07

570.

17M

aldi

ves

5.1

1.0

1.3

0.02

0.00

0.00

Nep

al57

.649

.541

.415

.72

14.3

212

.41

Sri L

anka

20.1

14.5

11.4

3.98

2.97

2.37


117

Sout

heas

t Asi

a28

.726

.923

.814

5.53

141.

6112

8.25

Cam

bodi

a43

.334

.526

.55.

784.

773.

75In

done

sia33

.334

.729

.775

.74

81.5

071

.22

Lao

PD

R52

.046

.139

.72.

992.

782.

46M

alay

sia1.

00.

91.

00.

270.

250.

27Ph

ilipp

ines

32.2

29.3

28.6

27.5

326

.43

26.6

8T

haila

nd3.

21.

61.

62.

121.

121.

09V

iet N

am37

.729

.126

.231

.10

24.7

722

.78

Paci

fic50

.645

.943

.34.

033.

903.

84Fe

dera

ted

Stat

es o

f

Mic

rone

sia (u

rban

)34

.936

.836

.80.

010.

010.

01

Fiji

23.6

11.6

14.4

0.19

0.10

0.12

Papu

a N

ew G

uine

a53

.950

.446

.63.

293.

303.

20T

imor

- Les

te53

.045

.545

.20.

530.

490.

51D

evel

opin

g A

sia

38.5

35.5

32.4

1293

.27

1229

.06

1146

.67

Not

e:

The

term

‘Dev

elop

ing

Asia

’ cov

ers a

ll th

e co

untr

ies t

hat a

ppea

r in

the

tabl

e.


118

5. Climate change and vulnerability to poverty: an empirical investigation in rural IndonesiaTomoki Fujii1

1 INTRODUCTION

The impacts of climate change are multifarious and heterogeneous across the globe. Scientists now widely agree that climate change is likely to affect not only the average temperature of the earth’s surface but also various other dimensions, including agriculture, water resources, ecosystems, and prevalence of diseases. Climate change is also expected to affect frequency and magnitude of extreme weather and climate events, which, in turn, may alter the pattern of disasters such as floods and droughts.

The way people are affected by these disasters may be different, even within relatively small areas, because some people are more resilient or adaptive. Those who are not resilient or adaptive may fall into poverty as a result of the negative shocks that disasters bring about. It is, therefore, important to understand who are vulnerable to extreme weather and climate events so that appropriate measures can be taken to minimize the negative shocks that these events bring about. However, despite the poten-tial importance of these events, there is a dearth of research on climate- driven vulnerability to poverty.

There are a few reasons for this. First, although there are some indica-tions that the pattern of some extreme events has changed as a result of anthropogenic influences, including increases in atmospheric concentra-tions of greenhouse gases, there is a lack of clear scientific evidence that quality and quantity of extreme events have changed on regional and global scales for certain specific events. For example, the available instru-mental records of floods at gauge stations are limited in space and time for a complete assessment of the climate- driven observed changes in the magnitude and frequency of floods at regional scales (IPCC 2012).

This is also an important issue in Indonesia. Although the National Disaster Management Agency (Badan Nasional Penanggulangan Bencana)


Climate change and vulnerability to poverty 119

collects and maintains disaster information in Indonesia, the data are not directly comparable over time. For example, the number of recorded flood events is less than 15 each year between 1985 and 1997. However, the number of events after 2002 is over 100 every year between 2003 and 2013.2 This massive increase in the number of recorded flood events may be partly due to the actual increase in flood events, but it is most likely due to the better data collection in recent years.

Second, the physical impact of extreme events may translate into differ-ent economic shocks to different households, even within the same town or village. Various factors, including the occupation of the household head, the household assets, the access to credit and insurance, and the local infrastructure development, are all likely to matter. However, socioeco-nomic surveys, from which poverty statistics are usually derived, typically contain no or very limited information about disasters and extreme events. Therefore, it is difficult to directly link poverty with extreme events.

Despite these difficulties, given the observed increases in extreme events across the world, the topic is more relevant than ever before. The timeliness and increased importance of the climate- driven vulnerability to poverty can also be seen from the fact that the Fifth Assessment Report by the Intergovernmental Panel on Climate Change (IPCC) Working Group II, which traditionally focuses on adaptation and vulnerability, has a new chapter on ‘Livelihoods and poverty’ (IPCC 2014).

Because of the data availability and relevance, we focus on two common types of disasters in Indonesia, floods and droughts. We evaluate how these two types of disasters affect the vulnerability of households to poverty and simulate the impact of climate change on vulnerability to poverty under some plausible scenarios.

This chapter is organized as follows. In section 2, we briefly present an overview of the situation of floods and droughts in Indonesia. In section 3, we describe the data used, followed by a discussion of the method in section 4. Section 5 presents the results and Section 6 offers some discussion.

2 CLIMATE CHANGE AND DISASTERS IN INDONESIA

In Indonesia, various impacts of climate change have already been observed and are expected to take place. For example, modest temperature increase has already occurred and it is expected to continue. The rainy season is expected to shorten with more intense rainfall during the rainy season which, in turn, leads to a significant increase in the risk of flood-ing.3 Sea- level rise will inundate productive coastal zones and the warming



of ocean water will affect the marine biodiversity. Climate change will also intensify water- and vector- borne diseases and threaten food security (PEACE 2007).

Indonesia is among the first countries to experience the ‘climate depar-ture’, which is the moment when the average temperature becomes so impacted by climate change that the old climate is left behind. It can be considered a tipping point such that the average temperature of the coolest year from then on is projected to be warmer than the average temperature of the hottest year between 1960 and 2005. Mora et al. (2013) estimate that Manokwari, Indonesia, is going to experience climate departure as early as 2020. Jakarta is estimated to have climate departure in 2029. These are sub-stantially earlier than the world average of 2047 reported in the same study.

The climate departure potentially will have a significant impact on the lives of people in Indonesia, the poor in particular, because there remain a sizable fraction of people who are either still under the poverty line or only slightly above the poverty line. For these people, the threat of poverty is far from over. If they are hit by a negative shock due to climate change, they may fall (further) below the poverty line. Therefore, Indonesia is a particularly important country to study in the context of climate- driven vulnerability to poverty.

As mentioned earlier, we choose to focus on floods and droughts. We make this choice for two reasons. First, they are two of the most important impacts of climate change in Indonesia. Future climate change is likely to increase their frequency and severity in Indonesia. Second, floods and droughts are among the most commonly observed disasters and, therefore, we have an accumulation of data on these types of disasters. Hence, we can arguably better predict whether climate change alters their frequency or incidence.

In contrast, it is generally much more difficult to predict the impact of events that have never happened before. Just for the sake of comparison, consider coastal erosion induced by climate change. A substantial frac-tion of the population live close to the coast in Indonesia and they are sure to be negatively affected by sea- level rise; their lives as well as homes, lands, and other assets may become more vulnerable as a result of climate change. However, this impact is difficult to predict because we have little information on how people would cope with coastal erosion.

2.1 Droughts

Droughts are common disasters in Indonesia, affecting some parts of Indonesia every year. Droughts negatively affect agricultural output and water supply. They are also associated with an increased incidence of forest



fires. The incidence and magnitude of drought tend to be particularly higher during the phase of El Niño–Southern Oscillation (ENSO), which refers to the variations in the surface temperature of the tropical eastern Pacific Ocean and in air surface pressure in the tropical western Pacific.

These variations occur because the trade winds, which carry wet and warm air from the west, tend to be weaker and, thus, dry and cold air tend to blow from the east during the El Niño years in Indonesia. This, in turn, tends to push back the onset of the rainy season as much as two months. As a result, ENSO tends to lead to droughts at the end of dry season. El Niño–Southern Oscillation also tends to lead to floods during the rainy season because the rain tends to intensify during the rainy season.4

Using a model linking ENSO- based climate variability to Indonesian cereal production, Naylor et al. (2002) find, among others, that Indonesia’s paddy production varies, on average, by 1.4 million tonnes for every 1 degree centigrade change in sea- surface temperature anomalies – the deviation in temperature from a long- term monthly mean sea- surface temperature – for August.

Droughts affect agricultural outputs because water is a key input for most agricultural outputs including rice, the main food crop grown in Indonesia. During El Niño years, widespread droughts affected 1–3 million hectares under paddy cultivation. Even during La Niña years, in which rainfall tends to be higher than average, localized droughts affect 30 000 to 80 000 hectares. On average, 280 000 hectares under paddy cultivation, which is much more than 2 percent of the total paddy area, are affected annually by drought to varying degrees. This means that nearly 160 000 farm house-holds are vulnerable to these periodic droughts (Kishore et al. 2000).

Droughts affect those farmers whose lives are dependent on their farm-land. Based on regression analysis with cross- sectional data, Skoufias et al. (2012) report a negative welfare impact of a significant shortfall in rain for farm households. Korkeala et al. (2009) find that a delayed onset of the monsoon season is associated with a 13 percent decline in per capita consumption for poor households but the delayed onset two years previ-ously was positively correlated with consumption. This means that poor households experience greater volatility, but no lasting reduction in con-sumption, following delayed onset of the monsoon season.

The findings of these studies indicate that drought mitigation measures may be useful. For example, Pattanayak and Kramer (2001a) measure the willingness to pay for drought mitigation from watershed protection in Ruteng Park in Indonesia by the contingent valuation method. They find that farmers are willing to pay up to $2–3 per year, which is about 10 percent of annual agricultural cost, 75 percent of the annual irrigation fees, and 3 percent of annual food expenditures. Pattanayak and Kramer



(2001b) also reports a sizable benefit of drought mitigation based on a separate household model.

2.2 Floods

Floods are also common in Indonesia. For example, Jakarta has a long history of floods because of its geomorphology and intense seasonal rainfall. This problem has been exacerbated by rapid population growth, land- use change, waterways being clogged with household wastes and sediment from upstream. Massive floods were recorded in January 2002 and February 2007; there were, respectively, 57 and 70 deaths and 365 000 and 150 000 evacuees in these events. In January 2014, 17.4 percent of Jakarta across 89 districts had been affected by a flood, with 23 deaths and over 65 000 evacuees, according to the Jakarta Province Regional Disaster Mitigation Agency (Badan Penanggulangan Bencana Daerah Provinsi DKI Jakarta).5

Floods also affect agricultural output. The order of magnitude of the impact of floods is comparable with that of droughts. For example, Hadi et al. (2000), cited by Pasaribu (2010), estimate that the sizes of paddy harvest failures due to floods and droughts are, respectively, 0.21 and 0.50 percent of the planted area during 1980–98. According to the estimates by the Directorate General of Crop Protection, Ministry of Agriculture cited by Pasaribu (2010), the actual rice areas affected by floods and droughts are 333 000 and 319 000 hectares, respectively, in 2008.

3 DATA AND SUMMARY STATISTICS

The main data source for this study is the Indonesian Family Life Survey (IFLS), an ongoing panel survey in Indonesia. The original sample frame covered 13 of the 27 provinces in Indonesia in 1993. Within each of these 13 provinces, enumeration areas were randomly drawn from a nationally representative sample frame used in the 1993 National Socio- Economic Survey (SUSENAS) designed by the Indonesian Central Bureau of Statistics (BPS). The sample was representative of about 83 percent of the Indonesian population in 1993.

The first round of the IFLS (IFLS 1) was conducted in 1993–94 by the RAND Corporation, in collaboration with Lembaga Demografi, University of Indonesia. The IFLS 2 was conducted in 1997, by the RAND Corporation, in collaboration with the University of California at Los Angeles and Lembaga Demografi, University of Indonesia.6 The IFLS 3 was completed in 2000 and conducted by the RAND Corporation,



in collaboration with the Population Research Center, University of Gadjah Mada. The IFLS 4 took place in 2007–08 and it was conducted by the RAND Corporation, the Center for Population and Policy Studies of the University of Gadjah Mada, and Survey METRE.

In the IFLS 1, a total of 7224 households were interviewed and detailed individual- level data were collected from over 22 000 individuals. In the IFLS 2, 94 percent of the IFLS 1 households and 91 percent of the IFLS 1 target individuals were re- interviewed. In the IFLS 3, 95.3 percent of IFLS 1 households were re- contacted. In the IFLS 4, the re- contact rate was 93.6 percent. Among IFLS 1 dynasty households (any part of the original IFLS 1 households, 90.3 percent were either interviewed in all four waves or died, and 87.6 percent were actually interviewed in all four waves). These re- contact rates are as high as or higher than most panel surveys in the United States and Europe. High re- interview rates were obtained, in part, because the data collection team was committed to tracking and interview-ing individuals who had moved or split off from the original IFLS 1 house-holds. High re- interview rates contribute significantly to improve the data quality in a longitudinal survey because they lessen the risk of bias due to nonrandom attrition.7

In each round of the IFLS, there was also an associated community- level survey, in which questions about the characteristics of the commu-nity were asked. We use the climate component of these data. Because the survey format has changed over rounds and because a complete history of extreme events that households have experienced is not available, we only use the indicator variable for whether the community has experi-enced each of flood and drought over the final five years for our main analysis.

In this study, we choose to use only those rural households that appear in all rounds of the survey and did not move across villages.8 Removing the records with missing values in key variables, we are left with a total of 4680 observations across four rounds, or 1170 households, to be used for our main analysis. The difference between our sample and the whole sample is briefly discussed later.

Table 5.1 provides some summary statistics for our sample. All the reported statistics in the table are weighted by the sample weight that takes into account attrition.

The first row in Table 5.1 shows that the average age of the household heads increases as expected. However, even though we track the same set of households, the average age of the household heads does not increase exactly by the number of years between the surveys because the original head may die or disappear from the household for other reasons. Similarly, household size tends to get smaller over time. Table 5.1 also shows that



the housing condition has generally improved over time. For example, the proportion of households that have a toilet within their premises has increased from 12.9 percent to 50.7 percent over the four rounds. The last and first row from the last, respectively, show the proportions of house-holds that have experienced droughts and floods within the last five years before the survey. As Table 5.1 shows, there are substantial fluctuations in the incidence of droughts and floods across rounds.

Table 5.2 shows the distribution of households that experienced floods and droughts over the four rounds of the IFLS surveys. Owing to the limitations of the data discussed earlier, we use the indicator that the com-munity has experienced floods and/or droughts over the past five years. Therefore, a caution must be exercised when interpreting Table 5.2. The table shows, for example, that 18 households in our sample experienced at least one drought within a period of five years before an IFLS survey for three rounds but no floods within a period of five years within any round

Table 5.1 Sample means of key variables by the IFLS rounds

Description IFLS 1 IFLS 2 IFLS 3 IFLS 4

Head’s age 46.5 49.2 50.9 53.3Household size 4.5 4.4 4.4 3.9Toilet in premise (%) 12.9 23.3 26.2 50.7Single- level single unit (%) 92.3 80.3 81.0 86.4Roof is tile (%) 80.8 81.2 80.1 80.3Roof is foliage/leaves (%) 3.3 2.3 1.2 0.6Wall is masonry (%) 36.9 47.7 54.3 63.6Flood in past 5 years (%) 14.0 3.9 10.8 18.8Drought in past 5 years (%) 3.0 8.6 16.7 13.9

Table 5.2 The numbers of households that have experienced floods and droughts in the IFLS rounds

Flood Drought

Total0 1 2 3

0 486 127 67 18 6981 241 25 71 0 3372 38 74 23 0 135

Total 765 226 161 18 1170



of the IFLS surveys. Note that these households may have experienced droughts more than three times in our study period, because, for example, they may have experienced multiple droughts within five years before a particular round of the IFLS.

Because the floods and droughts are reported by the survey respondents and the way floods and droughts are reported across communities may not be strictly comparable, it is desirable to have an alternative measure of climate variations. To this end, we have compiled daily rainfall data at the provincial level.9 We then computed for each household the standard deviation in daily rainfall in the province the household belongs to over the past 365 days from the first interview for the household consumption module. We took this measure as a convenient measure of climate variabil-ity. This measure also has an advantage that the reference period is shorter than the flood and drought indicators taken from the IFLS data. However, because the rainfall are available only from 1997, the standard deviation over the last 365 days can be computed only from 1998.

In Figure 5.1, we plot the standard deviation of provincial- level daily rainfall averaged over all the provinces for each year between 1997 and 2013. The dashed line represents the linear trend in the standard deviation of daily rainfall. We can see from this figure, that there is an increasing trend in the standard deviation of daily rainfall over the years involved.

In this chapter, we follow the standard consumption- based definition of poverty. To this end, we first define poverty lines. We consider the

7.0

7.5

8.0

8.5

9.0

9.5

1997 2001 2005 2009 2013

Stan

dard

dev

iatio

n of

dai

lyra

infa

ll in

a y

ear (

mill

imet

ers p

er d

ay)

Year

Figure 5.1 Standard deviation of daily rainfall between 1997 and 2013



following three alternative sets of poverty lines: (1) the official poverty lines, which are defined at the level of urban and rural areas annually;10 (2) the US$1.25- a- day international poverty line; and (3) the US$2- a- day international poverty line. For (2) and (3), we use the purchasing power parity conversion factor for private consumption in 2005 published in the World Development Indicators (USD 1 = INR 4192.8) and adjust for the spatial price difference and inflation using the Consumer Price Index (CPI) also available from the BPS website.11 Because the CPI data are only avail-able for major cities, we use the CPI for the capital of the province in which the household was located.

To measure poverty at the household level, we compare the total monthly consumption expenditure per capita, or the total monthly house-hold expenditure divided by the household size, with the poverty line. If the consumption per capita of the household that the individual belonged to fell below the poverty line, the individual is deemed poor.

4 METHO DOLOGY

4.1 Measures of Vulnerability

As with most other studies in the literature, we define vulnerability to poverty V as expected poverty. We denote the consumption per capita by c, the poverty line by z, their ratio by q| ; c/z. Further, we denote the cen-sored ratio byq 5 min(1,q|) . We consider the following four vulnerability measures:

V∞ = E[Ind(q < 1)] V1 = E[1 − q]

V1/2 = E[1 − q1/2] V0 = −E[lnq],

where Ind(·) is an indicator function that is equal to one when the argu-ment is true and zero otherwise.

The first measure is simply the expected headcount index and the most widely used measure in the literature including Chaudhuri et al. (2002). The second measure is the expected poverty- gap index. The third measure is the expected Chakravarty index with parameter 1/2. The fourth measure is the expected Watts measure. All these measures are an unscaled version of the measure proposed by Calvo and Dercon (2013).12 Although their parameter restriction would exclude V∞ and V1, we include them in this study because they have intuitive interpretations (they are respectively



expected poverty rate and expected poverty gap). The former is also closely related to other vulnerability measures (see also Klasen and Povel 2013 and Fujii 2015a for reviews of various vulnerability measures).

To operationalize the expectations given above, we assume the following model of the ratio for each individual i at time t:

lnq|it5XitT b 1eit, (5.1)

where Xit is a column vector of values of covariates for ln q|it; and the idi-osyncratic error term eit is assumed to be normally distributed with a zero mean but may be correlated across time or individuals. The error term eit is allowed to be heteroskedastic and its standard deviation is given by:

sit ; "Var [eit ] 5 "exp(ZTitq) ,

where sit is a column vector of covariates for the variance of the idiosyn-cratic term. Although we set Zit = Xit in our empirical application as with various other empirical studies in the literature, Zit and Xit can be different, in general, and we maintain this difference in this section. Note that there are 1170 individuals and 4 time periods. Hereafter, we focus on a particular individual in a particular period and drop subscripts i and t for most of the remainder of this section to keep the presentation simple.

Given these assumptions, the vulnerability measures can be rewritten with the probability density function and the cumulative distribution func-tion of normal distribution as the following proposition shows:

Proposition 1: Given the assumptions above, V∞, V1, V1/2, and V0 can be written as follows:

V` 5 Fa2XTb

s b (5.2)

V1 5 Fa2XTb

s b 2 expaXTb 1s2

2bFa2XTb

s 2 sb (5.3)

V1/2 5 Fa2XTb

s b 2 expaXTb

21

s2

8bFa2XTb

s 2s2b (5.4)

V0 5 2XTbFa2XTb

s b 1 s�a2XTb

s b (5.5)



Proof: It is convenient to define the normalized error term by v ; e/s. Then, equations (5.2)–(5.5) follow from below:

V` 5 Pr(e , 2XTb) 5 Prav ,2XTb

s b V1 5 E [max(0,1 2 exp(XTb 1 sv)) ]

5 Fa2XTb

s b 2 exp(XTb)32XT b

s2`

exp(sv)v�(v)dv

5 Fa2XTb

s b 2 expaXTb 1s2

2b32XTb

s

2`

expa2(v 2 s) 2

2b"2p

dv

V1/2 5 E [max(0,1 2 "exp(XTb 1 sv)) ]

5 Fa2XTb

s b 2 expaXTb

2b32XTb

s

2` expasv

2b�(v)dv

5 Fa2XTb

s b 2 expaXTb

21

s2

8b32XTb

s

2`

expa2(v 2 s/2) 2

2b"2p

dv

V0 = E[max(0, −(XTb + sv))]

5 2XTbFa2XTb

s b 2 s32XTb

s2`

v�(v)dv,

where we use F9(v) = −vF(v) to obtain equation (5.5).As it can be seen from proposition 1, both V1 and V1/2 have a very similar

form. Their first terms are the same and represent the expected change in the extensive margin (that is., whether the individual is below the poverty line). The differences in the second terms essentially come from the way the two measures treat the left tail in the consumption distribution.

To estimate these measures, we first obtain an estimate b of the coefficient b by ordinary least squares (OLS) regression. We then compute a logarithmic squared residual u ; ln((lnq| 2 XTb) 2) . By an OLS regression of u on Z,



we obtain an estimate q. Then, we obtain an estimate sit of sit for each combination of (i, t) as follows:

sit 5 "exp(ZTitq)

Replacing b and s by b and ŝ in equations (5.2)–(5.5), we can estimate the vulnerability measures for each individual and each time period. In our empirical application, we assume that the vulnerability is the same for every member in the household. Therefore, we will aggregate household- level vulnerability by taking the average across households weighted by the population expansion factor, or the product of the household size and the household weight.

Note here that we run a linear regression of the logarithmic household consumption per capita over the poverty line on its covariates. This point is different from various other methods, including that of Chaudhuri et al. (2002), which often involve estimating a binary regression of poverty status on its covariates. We chose a linear model because we can analyti-cally derive various vulnerability measures in a coherent manner. This, in turn, has an added advantage that we are able to verify how our results are (in)sensitive to the choice of vulnerability measures.

4.2 Future Climate Scenarios and Simulations

Using the measures introduced above, we simulate the impact of climate change on vulnerability to poverty by changing the value of covariates. To operationalize this idea, we need some future climate scenarios. The main challenge here is that we do not yet know exactly how climate change would affect the lives of people through the channels of floods and droughts. In particular, scientists do not yet have enough evidence to establish a clear causal relationship between climate change and floods, even though they generally agree that anthropogenic climate change has increased and is likely to continue to increase the incidence of droughts and change the fre-quency and pattern of ENSO events. Therefore, we choose to adopt a few simple scenarios to present the possible order of magnitude of the impacts that future climate change may bring about.

Our first scenario is the doubling incidence of floods and/or droughts from the 2007 (IFLS 4) level. This scenario is motivated by Cai et al. (2014), who predict that the frequency of major El Niño events may double in this century. Because El Niño events are related to floods and droughts, the doubling incidence of floods and droughts would not be completely unrealistic. However, because doubling incidence may appear extreme and the time horizon involved is very long, we also consider



the case where the incidence of floods and/or droughts increases by 50 percent.

As discussed in detail in the next section, we consider two cases under this scenario. In the first case (scenario 1(a)), we treat all the floods and droughts observed in the IFLS data equally. In the second case (scenario 1(b)), we assume that the droughts and floods in 1997 were different, because the ENSO event in 1997 is considered one of the largest in the observation history. As we shall show, there is some evidence that the ENSO event in 1997 was indeed different.

Our second scenario (scenario 2) is that the standard deviation of daily rainfall in a year at the provincial level changes linearly over time. In this exercise, we are, essentially, using the linear trend line similar to the line drawn in Figure 5.1 to predict the future standard deviation except that the trend line is defined for each province. Using a linear extrapolation to year 2030 for each province, we obtain the predicted standard deviation of daily rainfall. We then use this predicted value to compute the vulnerability to poverty under climate change. Although these scenarios are admittedly naïve, the results we present in the next section provide a plausible order of magnitude of the impact of climate change on vulnerability to poverty.

5 EMPIRICAL RESULTS

5.1 Baseline Results

To compute the vulnerability measures, we first run regressions to estimate b and q. Because the dependent variable in equation (5.1) is lnq|, which is the logarithm of consumption per capita normalized by the poverty line, the estimates depend not only on the consumption per capita but also on the poverty line.

In Table 5.3, we report the baseline regression results when international poverty lines are used. In these regressions, we include the household- level fixed- effects terms to capture the unobserved heterogeneity across households. We also include IFLS- round- specific fixed- effects to absorb the aggregate shock to rural Indonesia in each round of the survey so that the changing macroeconomic environment is appropriately controlled for. In addition, we control for demographic characteristics of households as well as our main variables of interest, indicator variables for floods and droughts experienced over the last five years in the community of residence.

Note that the results presented in Table 5.3 are independent of whether we use $1.25 poverty line or $2 poverty line, because the constant term will



absorb the difference. However, when the national poverty lines are used, the regression results are slightly different. This is because the national poverty lines are uniform within the rural areas each year, whereas we adjust for the spatial price differences for the international poverty lines. In this section, we present the regression results based on international poverty lines only. The corresponding regression results based on national poverty lines are reported in table A.1 in the appendix to Fujii (2015b).

As Table 5.3 shows, the flood variable has a negative b- coefficient, indi-cating that a flood tends to decrease the expected logarithmic consump-tion, though this coefficient is not significant. The q- coefficient on floods and droughts are both positive, suggesting that they tend to increase the variance of consumption, though the coefficient for floods is the only one that is significant.

Table 5.4 presents various poverty and vulnerability measures for each round of the IFLS survey. All the results are weighted by the population expansion factor. In the first three rows, we report the Foster–Greer–Thorbecke (FGT) poverty measures (Foster et al. 1984) with parameters a = 0, a = 1, and a = 2 for each round of IFLS survey, where the FGT measure with parameter a is defined as follows:

FGTa 51Nai

Ind(qi , 1) (1 2 qi)a.

FGT0 is simply the proportion of people who are under the poverty line and is often called the poverty rate or headcount index. Therefore, Table 5.4 shows, for example, that 53.8 percent of people in the IFLS 4

Table 5.3 Regression estimates of b and q for Scenario 1(a)

Variable b q

Est. (s.e.) Est. (s.e.)

Head’s age 0.018*** (0.0046) 0.00026 (0.022)Head’s age squared/100 −0.020*** (0.0043) −0.0063 (0.021)Household size −0.15*** (0.0066) −0.011 (0.031)Flood last five years −0.025 (0.024) 0.25** (0.12)Drought last five years 0.0067 (0.026) 0.14 (0.12)R2 0.6643 0.2893N 4680 4680

Note: Household- specific and IFLS- round- specific fixed- effects terms are included in the model. International poverty lines are used for the calculation of q|. *, **, and ***, respectively, represent statistical significance at 10, 5, and 1 percent levels.


132

Tabl

e 5.

4 Po

vert

y m

easu

res a

nd v

ulne

rabi

lity

mea

sure

s bas

ed o

n th

e re

gres

sion

repo

rted

in T

able

5.3

and

in ta

ble

A.1

in

the

appe

ndix

to F

ujii

(201

5b)

Pove

rty

line

Nat

iona

l pov

erty

line

Inte

rnat

iona

l pov

erty

line

$1.

25In

tern

atio

nal p

over

ty li

ne $

2

Rou

ndIF

LS

1IF

LS

2IF

LS

3IF

LS

4IF

LS

1IF

LS

2IF

LS

3IF

LS

4IF

LS

1IF

LS

2IF

LS

3IL

FS

4

FGT

00.

182

0.17

00.

192

0.09

50.

606

0.42

30.

349

0.21

40.

838

0.69

90.

693

0.53

8FG

T1

0.05

60.

050

0.04

60.

019

0.23

40.

138

0.10

30.

051

0.42

40.

302

0.26

70.

178

FGT

20.

026

0.02

20.

017

0.00

60.

121

0.06

60.

042

0.01

80.

256

0.16

50.

133

0.07

9W

0.07

90.

068

0.05

80.

024

0.35

10.

198

0.13

60.

065

0.69

40.

463

0.38

40.

242

C1/

20.

033

0.02

90.

026

0.01

10.

141

0.08

20.

059

0.02

90.

265

0.18

40.

158

0.10

3V

∞0.

145

0.13

50.

173

0.09

70.

619

0.39

70.

354

0.21

80.

857

0.71

80.

686

0.53

7V

10.

038

0.03

40.

044

0.02

20.

218

0.11

70.

099

0.05

50.

422

0.28

90.

264

0.17

8V

1/2

0.02

10.

019

0.02

50.

013

0.12

80.

067

0.05

70.

031

0.26

00.

172

0.15

70.

104

V0

0.04

90.

044

0.05

80.

028

0.30

80.

158

0.13

30.

072

0.66

10.

423

0.38

00.

247

Not

e:

Popu

latio

n ex

pans

ion

fact

or is

app

lied.



sample was living in a household whose consumption per capita was below the $2- a- day international poverty line. FGT1 is also called the poverty gap, which measures the average shortfall from the poverty line. FGT2 is called the poverty severity or the squared poverty gap and puts higher weights on the poorest of the poor.

In the fourth and fifth rows of Table 5.4, we respectively report the Watts poverty measure (Watts 1968) and the Chakravarty poverty measure (Chakravarty 1983) with parameter w = 1/2, which are defined as follows:

W 5 21Na

lnqi,Cw 51Nai

(1 2 qwi ) .

The Watts measure is the average logarithmic shortfall from the poverty line. As with FGT2, both the Watts and the Chakravarty measure put higher weight on the poorest of the poor.

In all these measures, poverty has generally dropped over the four rounds of IFLS surveys, except that the poverty rate under the national poverty line has slightly increased between IFLS 2 and IFLS 3. Regardless of the poverty measure used, there is a substantial drop in poverty between IFLS 3 and IFLS 4, during which Indonesia achieved a healthy economic growth of around 4 percent per year in per capita income.

The fifth to ninth rows in Table 5.4 are our vulnerability measures. To compute these, we plug the parameter values reported in Table 5.3, or table A.1 in the appendix to Fuiji (2015b), as well as the estimate of V into equations (5.2)–(5.6). Because we have V∞ = E[FGT0], V1 = E[FGT1], V1/2 = E[C1/2], and V0 = E[W] by definition, we expect to have V∞ FGT0, V1 . FGT1, V1/2 C1/2, and V0 W, which indeed holds, as shown in Table 5.4. As expected, the changes in our vulnerability measures have been similar to those of poverty measures.

5.2 Scenario 1(a): Doubling Incidence of Flood and Drought from IFLS 4

We now simulate how the vulnerability measures change as a result of future climate change. As discussed in section 4, our first scenario is where the incidence of floods and droughts double from the 2007 level observed in IFLS 4. More precisely, 17.7 percent and 14.5 percent of the sample households experienced floods and droughts no more than five years from the IFLS 4 survey, respectively. We consider the effect of doubling these proportions. Because doubling may appear extreme and involves a long time horizon, we also consider 50 percent increase as a plausible change in the middle run.

A problem in this exercise is which households should bear the impact



of floods and droughts in the future. Although it would not be impossible to estimate the floods and droughts risk for each household, we choose to assign floods and droughts randomly with an equal probability. We do this repeatedly under the assumption of independence between floods and droughts.13 That is, in each round of simulation, we randomly pick a pre-determined number of households that are affected by floods or droughts. For these households, we change the values of X and Z corresponding to floods or droughts in the computation of vulnerability while keeping all the other covariates and fixed- effects terms constant at the baseline level in 2007. We repeat this 1000 times and take an average over all the rounds of simulation.

The random assignment carried out in this way is not without prob-lems. For the sake of argument, consider a situation in which only those households that are well above the poverty line are affected by floods and droughts. In this case, floods and droughts would not increase the vulnerability measures much, because the households that are hit by the disasters are likely to remain well above the poverty line. If we randomly assign floods and droughts without taking this pattern into consideration, the vulnerability would unambiguously increase, because the vulnerability measures for those household that are close or below the poverty line – assuming that we have such households – would worsen. In other words, the random assignment would increase the vulnerability measure. Hence, random assignment is not an innocuous exercise in general.

It turns out that the pure effect of the random assignment is small in our data. The second column (‘IFLS 4’) of Table 5.5 refers to the vulnerability measures for the IFLS 4 survey (they are the same as those reported in Table 5.4), which serve as our baseline measurement. In the third column (‘Randomize’), we compute vulnerability measures by randomly and independently assigning floods and droughts without changing the total number of households that are affected by each of these disasters. Since there is little difference in these two columns, the random assignment has only negligible effect on the resulting vulnerability measure.

The fourth column (1.5 × Fl) of Table 5.5 shows the effect of increasing the incidence of floods by 50 percent. Compared with the third column, the vulnerability measure increases by around 2–3 percent (for example, (0.100–0.098)/0.098 . 2% for V∞) when the national poverty line is used. The increase is even smaller when an international poverty line, especially the $2- a- day poverty line, is adopted. The fifth column (1.5 × Dr) gives the effect of increasing the incidence of droughts by 50 percent. The change in vulnerability is generally smaller than those found for floods. The sixth column (1.5 × Fl & Dr) gives the combined effect of the increase of inci-dence of both floods and droughts by 50 percent.


135

Tabl

e 5.

5 Si

mul

ated

effe

cts o

f in

crea

sing

the

inci

denc

e of

floo

d an

d dr

ough

t by

50 p

erce

nt (

1.5×

) an

d 10

0 pe

rcen

t (2

×)

for v

ario

us p

over

ty li

nes u

nder

scen

ario

1(a

)

Scen

ario

IFL

S 4

Ran

dom

ize

1.5×

Fl

1.5×

Dr

1.5×

Fl &

Dr

2× F

l2×

Dr

2× F

l & D

r

Nat

iona

l pov

erty

line

V∞

0.09

70.

098

0.10

00.

099

0.10

00.

102

0.09

90.

102

V1

0.02

20.

022

0.02

30.

023

0.02

30.

023

0.02

30.

024

V1/

20.

013

0.01

30.

013

0.01

30.

013

0.01

30.

013

0.01

3V

00.

028

0.02

90.

029

0.02

90.

030

0.03

00.

029

0.03

0

Inte

rnat

iona

l pov

erty

line

$1.

25V

∞0.

218

0.21

70.

219

0.21

70.

219

0.22

10.

217

0.22

1V

10.

055

0.05

60.

056

0.05

60.

056

0.05

70.

056

0.05

7V

1/2

0.03

10.

032

0.03

20.

032

0.03

20.

033

0.03

20.

033

V0

0.07

20.

073

0.07

40.

073

0.07

40.

075

0.07

30.

075

Inte

rnat

iona

l pov

erty

line

$2

V∞

0.53

70.

537

0.53

80.

536

0.53

80.

540

0.53

60.

540

V1

0.17

80.

178

0.17

90.

178

0.17

90.

181

0.17

80.

181

V1/

20.

104

0.10

40.

105

0.10

40.

105

0.10

50.

104

0.10

5V

00.

247

0.24

70.

249

0.24

70.

249

0.25

10.

247

0.25

1

Not

e:

Popu

latio

n ex

pans

ion

fact

or is

app

lied.



The seventh, eighth, and ninth columns in Table 5.5 give the vulner-ability measures when the incidence of flood, drought and both flood and drought double, respectively. As can be seen from Table 5.5, the impact of doubling the incidence is also small. The biggest relative change is seen in V0 under the national poverty line, but even in this case, the increase is only around 6 percent. Therefore, Table 5.5 shows that the combined impact of increased incidence of flood and drought is relatively small. The impact simulated here should be considered a long- run average and not a one- off impact as the flood and drought indicators used in this study are based on the incidence over the last five years.

5.3 Scenario 1(b): Special Treatment of Major ENSO Events

Although an up to 7 percent increase in vulnerability (expected poverty) is not negligible, it may give a misleading impression about the impor-tance of the impacts of flood and drought as the short- run effects may be severer. Hence, to simulate the possible magnitude of the short- run effects of major ENSO events, we utilize the fact that there was a major ENSO event right before the data collection of the IFLS2 survey. Because this was clearly a major event, it is reasonable to treat floods and droughts separately from those in other years.

Table 5.6 reports the regression results under international poverty lines14 when the flood and drought effects are assumed to be different between IFLS 2 and other rounds of IFLS surveys. The table clearly shows that the order of magnitude of the effects of floods and droughts are different between IFLS 2 and other rounds. Unlike Table 5.3, the b- coefficients are statistically significant for both floods and droughts for IFLS 2, but not for other rounds of IFLS. Furthermore, we find that the major drought significantly increased the variance of consumption.

It should be noted here that the vulnerability measures are generally model dependent. Therefore, the vulnerability measures reported in Table 5.4 are generally different from those calculated from the regression results reported in Table 5.6. However, because the models are similar, the vulner-ability measures are generally very close.15 As with Table 5.5, we report, in Table 5.7, the simulated effects of increased incidence of floods and droughts from the IFLS 4 level by 50 or 100 percent. However, unlike sce-nario 1(a), the impacts of floods and droughts considered in scenario 1(b) are those associated with a major ENSO event. Hence, we first replace the effects of flood and drought in IFLS 4 with those effects for 1997 (IFLS 2) without changing the flood or drought status in the IFLS 4 records.

It should be noted here that the vulnerability measures are generally model dependent. Therefore, the vulnerability measures reported in



Table 5.4 are generally different from those calculated from the regression results reported in Table 5.6. However, because the models are similar, the vulnerability measures are generally very close.16 As with Table 5.5, we report, in Table 5.7, the simulated effects of increased incidence of floods and droughts from the IFLS 4 level by 50 or 100 percent. However, unlike scenario 1(a), the impacts of floods and droughts considered in scenario 1(b) are those associated with a major ENSO event. Hence, we first replace the effects of flood and drought in IFLS 4 with those effects for 1997 (IFLS 2) without changing the flood or drought status in the IFLS 4 records.

By comparing the baseline vulnerability in the second column (IFLS 4) with the third column (1997- effect), it can be seen that simply replacing the effects of floods and droughts in 2007 (or non- IFLS 2) with those in 1997 (or IFLS 2) have a substantial impact on vulnerability. When the national poverty lines are used, there is about 40 percent increase in vulnerability, whereas the increase is around 20 and 10 percent when $1.25- a- day and $2- a- day poverty lines are used, respectively.

The fourth column (Randomize) reports vulnerability measures when the assignment of floods and droughts are randomized. As with Table 5.5,

Table 5.6 Regression estimates of b and q for scenario 1(b)

Variable b q


Head’s age 0.019*** (0.005) 0.0082 (0.022)Head’s age squared/100

−0.020*** (0.004) −0.015 (0.021)

Household size −0.15*** (0.007) −0.0050 (0.031)Flood last five years (non- IFLS2)

−0.0015 (0.026) 0.32** (0.13)

Drought last five years (non- IFLS2)

0.041 (0.030) 0.14 (0.14)

Flood last five years (IFLS2)

−0.21*** (0.071) −0.46 (0.34)

Drought last five years (IFLS2)

−0.073* (0.044) 0.36* (0.21)

R2 0.6386 0.3014N 4680 5584

Note: Household- specific and IFLS- round- specific fixed- effects terms are included in the model. International poverty lines are used for the calculation of q|. *, **, and *** respectively represent statistical significance at 10, 5, and 1 percent levels.


138

Tabl

e 5.

7 Si

mul

ated

effe

cts o

f in

crea

sing

the

inci

denc

e of

floo

d an

d dr

ough

t by

50 p

erce

nt (

1.5×

) an

d 10

0 pe

rcen

t (2

×)

for v

ario

us p

over

ty li

nes u

nder

scen

ario

1(b

)

Scen

ario

IFL

S 4

1997

- effe

ctR

ando

miz

e1.

5× F

l1.

5× D

r1.

5× F

l & D

r2×

Fl

2× D

r2×

Fl &

Dr

Nat

iona

l pov

erty

line

V∞

0.09

80.

131

0.13

30.

148

0.13

80.

152

0.16

30.

142

0.17

2V

10.

022

0.03

10.

032

0.03

60.

033

0.03

70.

040

0.03

40.

042

V1/

20.

013

0.01

70.

018

0.02

00.

019

0.02

10.

022

0.01

90.

024

V0

0.02

90.

040

0.04

10.

046

0.04

30.

048

0.05

10.

044

0.05

5

Inte

rnat

iona

l pov

erty

line

$1.

25V

∞0.

219

0.25

80.

254

0.26

80.

258

0.27

20.

282

0.26

10.

290

V1

0.05

50.

065

0.06

60.

070

0.06

70.

072

0.07

50.

069

0.07

8V

1/2

0.03

10.

037

0.03

80.

040

0.03

80.

041

0.04

20.

039

0.04

4V

00.

072

0.08

60.

087

0.09

20.

089

0.09

40.

098

0.09

10.

102

Inte

rnat

iona

l pov

erty

line

$2

V∞

0.53

80.

579

0.58

10.

597

0.58

40.

600

0.61

20.

588

0.61

9V

10.

179

0.20

10.

201

0.21

00.

204

0.21

20.

219

0.20

60.

223

V1/

20.

104

0.11

80.

118

0.12

30.

119

0.12

40.

128

0.12

10.

131

V0

0.24

70.

281

0.28

10.

294

0.28

50.

298

0.30

80.

289

0.31

5

Not

e:

Popu

latio

n ex

pans

ion

fact

or is

app

lied.



the randomization has very little impact on the resulting vulnerability measures.

The fifth column (1.5 × Fl) reports the simulated vulnerability measures when the incidence of flood increases by 50 percent, where the impact of flood is equivalent to that observed in IFLS 2. Compared with the baseline vulnerability, the vulnerability has increased by well more than 50 percent in this case under the national poverty lines. Under international poverty lines, the relative change is about 11–28 percent, depending on the poverty line and vulnerability measure used. The impact of drought is less substan-tial than flood as shown in the sixth column (1.5 × Dr) but the impact is still sizable. The combined effect is even more substantial as shown in the seventh column (1.5 × Fl & Dr).

Obviously, the impact is even larger when the incidence increases by 100 percent instead of 50 percent. The eighth to tenth columns report the vul-nerability measures under the doubling incidence scenario. The combined effect of doubling the incidence of both floods and droughts is particularly large with the increase in vulnerability from the IFLS- 4 baseline reaching as high as 91 percent.

5.4 Scenario 2: Using Linearly Extrapolated Standard Deviation of Daily Rainfall

In our second scenario, instead of the floods and droughts over the last five years, we use the standard deviation of daily provincial- level rainfall over the past 365 days counting from the first interview for the consump-tion component of the survey for each household. Table 5.8 reports the

Table 5.8 Regression estimates of b and q for scenario 2

Variable b q


Head’s age 0.017** (0.0074) −0.044* (0.024)Head’s age squared/100 −0.017** (0.0070) 0.041* (0.022)Household size −0.17*** (0.011) −0.0025 (0.028)SD of daily rainfall over the past 365days

−0.032* (0.017) −0.064 (0.044)

R2 0.7675 0.0023N 2340 2340

Note: Household- specific and IFLS- round- specific fixed- effects terms are included in the model. International poverty lines are used for the calculation of q|. *, **, and *** respectively represent statistical significance at 10, 5, and 1 percent levels.



regression results with international poverty lines.17 Note that the number of observations in this table is smaller because we can compute the stand-ard deviation of daily rainfall only for IFLS 3 and IFLS 4 records.

Table 5.8 shows that the b- coefficient on the standard deviation of daily rainfall over the past 365 days is negative and significant. The q- coefficient is also negative but it is not significant.

To simulate the impact of climate change, we extrapolate the linear trend of provincial- level standard deviation in the annual rainfall to year 2030. To predict the future vulnerability in 2030, we replace the current standard deviation for IFLS 4 records with those extrapolated standard deviations. The results obtained in this way are provided in Table 5.9. For each set of poverty lines, we report the baseline vulnerability at IFLS 4 and the predicted vulnerability in 2030. We observe about 2, 15, and 10 percent increase in vulnerability measures, respectively, when the national, $1.25- a- day, and $2- a- day poverty lines are used.

6 DISCU SSION

In this study, we consider the impact of climate change on vulnerability to poverty, defined as expected poverty, in rural Indonesia. We have consid-ered two main scenarios. In the first scenario, we consider the case where future climate change doubles the incidence of floods and droughts. As an intermediary case, we also computed vulnerability when the incidence increases by 50 percent. Under this scenario, we computed the change in vulnerability for two cases, one case where the impact is estimated from flood and drought records over all the four rounds of the survey and the other case where the impact is derived essentially from the cross- sectional variations in year 1997, when a major ENSO event took place.

Table 5.9 Vulnerability measures under scenario 2

Poverty line National International $1.25 International $2

Scenario IFLS 4 2030 IFLS 4 2030 IFLS 4 2030

V∞ 0.105 0.107 0.247 0.285 0.579 0.623V1 0.020 0.021 0.057 0.068 0.194 0.218V1/2 0.011 0.011 0.032 0.038 0.112 0.126V0 0.025 0.025 0.071 0.086 0.263 0.299

Note: Population expansion factor is applied.



Based on the former case, the increase in the vulnerability is modest and no greater than 7 percent for all the poverty lines and vulnerability measures considered in this study. However, for the latter case, doubling the flood and the drought incidences had a major impact on vulnerability. The increase in vulnerability was at least 15 percent and reached as high as 91 percent, depending on the vulnerability measure and poverty line used.

Because the measurement of floods and droughts may not be strictly comparable, we also used rainfall data. By linearly extrapolating the stand-ard deviation of daily rainfall over the last 365 days to year 2030, we pre-dicted the vulnerability for year 2030. We found that there was a relatively large increase in vulnerability when $1.25- a- day international poverty line was used. The order of magnitude in this case is comparable to the 50 percent increase in the incidence of both floods and droughts associated with a major ENSO event (scenario 1(b)).

There are a few important limitations in this study. First, our climate sce-narios and simulation method are admittedly rudimentary. For example, we chose a random assignment for the sake of simplicity and tractability. Because the effect of random assignment is small in our sample, we do not have any evidence to indicate that our prediction is seriously biased due to the random assignment. However, this does not exclude the possibility that the future climate change systematically affects certain types of people more than others.

Second, we only consider the impacts of floods and droughts. Other important changes such as sea- level rises are ignored. Therefore, our esti-mates are likely to underestimate the overall effect of climate change on vulnerability to poverty.

Third, our measures of vulnerability are all individual- level vulnerability averaged over the sample. That is, our vulnerability measures are additively separable across individuals. However, it could be argued that the society is more vulnerable if a bad shock simultaneously affects everyone once it happens.

To take this perspective into consideration, it is possible, for example, to use the social vulnerability measure proposed by Calvo and Dercon (2013). A practical difficulty, though, is that we need to know the current and future correlation of floods and droughts across households. Because our understanding of the impact of climate change through floods and droughts, especially flood, is limited, we chose to leave this as an exercise for future research.

Fourth, the current analysis ignores the general equilibrium effects. To see this issue, suppose that various parts of Indonesia or even various parts of the world including Indonesia are hit simultaneously by correlated climate shocks (not necessary hit by the same flood or drought). Then,



the impact on the household would be different from what it would be without such correlated climate shocks. This is because, for example, such correlated shocks would affect the relative prices, whereas an idiosyncratic shock for a particular household or a community would have a negligible impact on relative prices.

Fifth, we do not take into account the possibility of nonlinearity of the impact, even though it is potentially important. For example, once climate departure occurs, the nature of the impacts of an ENSO event may change systematically and non- linearly. Similarly, when we extrapolate the impact with the standard deviation of rainfall, we assumed that the impact would increase linearly with the standard deviation but this may not hold even in approximation. Further scientific research will be needed to fully address these issues.

Finally, the estimates we provide are based on the condition that the households stay in the same village throughout our observation periods. This is a stringent restriction especially given that those households which can no longer survive in the same village will have to move. However, we chose to restrict our sample to control for a variety of unobservable factors that are specific to the location of residence with fixed- effects terms.

Although we cannot draw strong conclusions, we can find the nature of households we used by comparing summary statistics in tables A.5 and A.6 in the appendix to Fuiji (2015b) with Tables 5.1 and 5.2. The comparison appears to indicate that the general housing conditions at the beginning of the survey in our sample were slightly better than the average for the whole sample, which may be because those living in a poorly- built house are more likely to move when they are hit by a disaster. On the other hand, the incidences of floods and droughts do not appear to be drastically different.

Given the limitations above, it appears likely that our estimates provide a plausible lower- bound of the impact of future climate change on vulner-ability. Although the long- run effects of floods and droughts appear rather limited, the short- run effects are sizable.

Besides providing some plausible estimates of the impact of future climate change on vulnerability to poverty in Indonesia, this study con-tributes to the existing literature in several ways. First, to the best of our knowledge, this is the first study to directly link climate change with vulner-ability to poverty using panel data. This is an important first step because most of the existing studies on the impact of climate change rely heavily on global climate models and do not take into account the standards of living observed in household surveys. Although there have been a few exceptions, such as Adger (1999), they are based on cross- sectional evidence and thus require much stronger assumptions than ours. Further, they do not provide any estimates on the possible impact of future climate change.



Second, we also make a methodological contribution by proposing a variant of the expected poverty approach that bridges the popular measure of expected poverty rate by Chaudhuri et al. (2002) and the axiomatically derived vulnerability measure by Calvo and Dercon (2013). We offer a practical method in which various vulnerability measures can be computed in a coherent manner. The methodology used in this study can be applied easily to other countries if a relevant panel dataset is available.

This study also underscores the importance of monitoring the eco-nomic situation of households in developing countries that are likely to be affected by climate change because current global climate models do not tell us how climate change affects vulnerability to poverty. By linking global climate models with household- based observations, we will be able to make more meaningful prediction about the possible impacts of future climate changes on households including vulnerability to poverty.

NOTES

1. I have benefited from useful discussion with Madhav Aney, Indranil Dutta, Carlos Gradin, Christine Ho, A.Q.M. Golam Mawla, Jacques Silber, Kala Sridhar, Anthony Tay, and Guanghua Wan. Orlee Velarde provided research assistance.

2. See also, http://dibi.bnpb.go.id/DesInventar/simple_data.jsp (accessed 5 September 2015).

3. The increasing trend in the standard deviation of daily rainfall presented in Figure 5.1 is also indicative of the heightened risk of droughts and floods in the future.

4. See, for example, Garrison (2010) for a general introductory discussion on ENSO events.

5. ‘Your letters: Flooding in Jakarta–the facts’, Jakarta Post, 20 January 2014. 6. Additionally, IFLS 2+ was conducted in 1998, which covered a 25 percent subsample

of the IFLS households. IFLS 2+ is not used in this study. 7. See the following IFLS website for further details: http://www.rand.org/labor/FLS/

IFLS.html (accessed 5 September 2015). 8. We retain a small number of households that moved within the village. 9. We first obtain the provincial- level geographical coordinates from MyGeoPosition

(http://mygeoposition.com/) and use these coordinates to obtain daily rainfall data from the agroclimatology data website by the Prediction of World Energy Resource, the National Aeronautics and Space Administration (http://power.larc.nasa.gov/cgi- bin/cgiwrap/solar/[email protected]). (Both websites accessed 5 September 2015.)

10. They are available from the following website: http://www.bps.go.id/eng/tab_sub/view.php?kat=1&tabel=1&daftar=1&id_subyek=23 notab=7 (accessed 5 September 2015).

11. Obtained from http://www.bps.go.id/eng/aboutus.php?inflasi=1 (accessed 5 September 2015). Because the base year for the CPI changed over time, we link them by the CPI for the two contiguous months and the inflation rate reported in this website to cover our study period.

12. Their measure is VrCD 5 E [ (1 2 qr) /r ] for r < 1 and r ≠ 0 and V0

CD 5 2E [lnq ].13. It is also possible to assign floods and droughts jointly. However, we chose to maintain

the independence assumption because the correlation between the flood and drought incidence is very small in our sample.



14. The regression results under the national poverty lines are reported in table A.2 in the appendix to Fuiji (2015b). As with Table 5.3, the regression results for $2 and $1.25 international poverty lines are identical except for the constant term.

15. Round- by- round vulnerability measures for scenario 1(b) are reported in table A.4 in the appendix to Fuiji (2015b).

16. Round- by- round vulnerability measures for scenario 1(b) are reported in table A.4 in the appendix to Fuiji (2015b).

17. The regression results under national poverty lines are reported in table A.3 in the appendix to Fuiji (2015b).

REFERENCES44*

Adger, N. (1999), ‘Social vulnerability to climate change and extremes in coastal Vietnam’, World Development, 27 (2), 249–69.

Cai, W., S. Borlace, M. Lengaigne, P. van Rensch, M. Collins, G. Vecchi, et al. (2014), ‘Increasing frequency of extreme El Niño events due to greenhouse warming’, Nature Climate Change, 4 (February), 111–16.

Calvo, C. and S. Dercon (2013), ‘Vulnerability of individual and aggregate poverty’, Social Choice and Welfare, 41 (4), 721–40.

Chakravarty, S.R. (1983), ‘A new index of poverty’, Mathematical Social Sciences, 6 (3), 307–13.

Chaudhuri, S., J. Jyotsna and A. Suryahadi (2002), ‘Assessing household vulner-ability to poverty from cross- sectional data: a methodology and estimates from Indonesia’, Discussion Paper Series 0102- 52, Department of Economics, Columbia University.


Fujii, T. (2015a), ‘Concepts and measurement of vulnerability to poverty and other issues: a review of literature’, ADBI Working Papers (to appear).

Fujii, T. (2015b), ‘Climate change and vulnerability to poverty: an empirical inves-tigation in rural Indonesia’, ADBI Working Papers (to appear).

Garrison, T.S. (2010), Oceanography: An Invitation to Marine Science, 7th edn, Belmont, CA: Cengage Learning.

Hadi, P.U., C. Saleh, A.S. Bagyo, Hendayana R., Y. Marisa and I. Sadikin (2000), ‘Studi kebutuhan asuransi pertanian pada pertanian rakyat’, research report, Indonesian Center for Agricultural Socio- economic Research, Bogor, Indonesia.

Intergovernmental Panel on Climate Change (IPCC) (2012), Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaption: A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change, eds CB. Field, V. Barros, T.F. Stocker, D. Qin, D.J. Dokken, K.L. Ebi, et al., Cambridge and New York: Cambridge University Press.

Intergovernmental Panel on Climate Change (IPCC) (2014), Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, eds C.B. Field, V.R. Barros, D.J.




Dokken, K.J. Mach, M.D. Mastrandrea, T.E. Bilir, et al., Cambridge and New York: Cambridge University Press.

Kishore, K., A.R. Subbiah, T. Sribimawati, I.S. Dihart, S. Alimoeso, P. Rogers and D. Setiana (2000), ‘Indonesia country study’, in M.H. Glantz (2001), ‘Executive summary: reducing the impact of environmental emergencies through early warning and preparedness: the case of the 1997- 98 El Niño’, a UNEP/NCAR/UNU/WMO/ISDR Assessment, United Nations Environment Programme, National Center for Atmospheric Research, World Meteorological Organization, United Nations University, and International Strategy for Disaster Reduction, Asian Disaster Preparedness Center, January, pp. 103–9.

Klasen, S. and F. Povel (2013), ‘Defining and measuring vulnerability: state of the art and new proposals’, in S. Klasen and H. Waibel (eds), Vulnerability to Poverty: Theory, Measurement and Determinants, New York: Palgrave Macmillan.

Korkeala, O., D. Newhouse and M. Duarte (2009), ‘Distributional impact analy-sis of past climate variability in rural Indonesia’, World Bank Policy Research Working Paper 5070, World Bank, Washington, DC.

Mora, C., A.G. Frazier, R.J. Longman, R.S. Dacks, M.M. Walton, E.J. Tong, et al. (2013), ‘The projected timing of climate departure from recent variability’, Nature, 502 (7470), 183–7.

Naylor, R., W. Falcon, N. Wada and D. Rochberg (2002), ‘Using El Niño- Southern Oscillation climate data to improve food policy planning in Indonesia’, Bulletin of Indonesian Economic Studies, 38 (1), 75–91.

Pasaribu, S.M. (2010), ‘Developing rice farm insurance in Indonesia’, Agriculture and Agricultural Science Procedia, 1 (1), 33–41.

Pattanayak, S.K. and R.A. Kramer (2001a), ‘Pricing ecological services: willingness to pay for drought mitigation from watershed protection in eastern Indonesia’, Water Resources Research, 37 (3), 771–8.

Pattanayak, S.K. and R.A. Kramer (2001b), ‘Worth of watersheds: a pro-ducer surplus approach for valuing drought mitigation in eastern Indonesia’, Environmental and Development Economics 6 (1), 123–46.

PT Pelangi Energi Abadi Citra Enviro (PEACE) (2007), Indonesia and Climate Change: Current Status and Policies, Jakarta: World Bank, Department for International Development, and PT Pelangi Energi Abadi Citra Enviro.

Skoufias, E., R.S. Katayama and B. Essama- Nssah (2012), ‘Too little too late: welfare impacts of rainfall shocks in rural Indonesia’, Bulletin of Indonesian Economic Studies, 48 (3), 351–68.

Watts, H.W. (1968), ‘An economic definition of poverty’, in D.P. Moynihan (ed.), On Understanding Poverty, New York: Basic Books, pp. 316–29.


PART III

The Multidimensionality of Poverty in Asia


149

6. Measuring multidimensional poverty in three Southeast Asian countries using ordinal variablesValérie Bérenger

1 INTRODUCTION

There is a broad consensus among academic and institutional organiza-tions that poverty can no longer be defined only as a lack of monetary resources but reflects many factors that act as constraints on the achieve-ment of the capabilities of the population and affect its well- being. The enlargement of the framework traditionally used to address poverty and well- being is at the origin of methodologies that attempt to capture the essence of the multidimensionality of poverty.

The recent use by the United Nations Development Programme (UNDP) of the so- called Multidimensional Poverty Index (MPI) is an illustration of the importance of taking into account and addressing the multiple dimensional aspects of poverty. The MPI draws on the counting approach developed by Alkire and Foster (2008) and assesses poverty along the same dimensions as the Human Development Index (HDI). It includes ten indicators that affect the well- being of the population and does not only count the percentage of the population that suffers from at least 30 percent of multiple deprivations but also provides a snapshot of the breadth of poverty in assessing the proportion of the total number of dimensions of well- being in which multidimensional poor are really poor.

As was the case with the inception of the HDI in 1990, the MPI led to a renewal of the discussion among researchers regarding the issues to be addressed when adopting a multidimensional approach to poverty and, in fact, recent literature points out some weaknesses of the MPI.

One of these points concerns the choice of the dimensions included in the index, whereas others emphasize the arbitrariness of the cutoff used to identify the multidimensional poor across the dimensions. There are also those who question the sensitivity of the MPI to inequality in deprivation across individuals. In particular, owing to the counting nature



of the approach to the MPI, traditional indices of poverty that are based on continuous variables cannot be applied. Indeed, most of the indica-tors included in the MPI, or, more generally, in survey data that are used to capture direct achievements of individuals, are of an ordinal nature. Several recent studies suggested alternative ways of defining multidimen-sional poverty indices that comply with the basic axiomatic properties of multidimensional poverty indices using continuous variables (Bossert et al. 2013). Recently, Silber and Yalonetzky (2013) proposed a general framework that measures multidimensional poverty with ordinal variables. They address concerns regarding the identification and aggregation steps involved in constructing any poverty measure. In particular, they make a distinction between an individual poverty function and a social poverty function. At the individual level, they review the properties of such individ-ual functions that take into account the identification as well as the breadth of poverty. At the aggregation step, they highlight several ways to generate social poverty indices that deal with the issues of inequality in the distri-bution of deprivations. In particular, they suggested an extension of the approach of Aaberge and Peluso (2012) that makes it possible to address the issue of inequality in the distribution of deprivations and to capture additional information on poverty that is not well addressed by the MPI.

The main goal of this chapter is to highlight the contribution of the meth-odological refinements of poverty measures based on counting approaches using ordinal variables to the understanding of multidimensional poverty in three Southeast Asian countries, namely, Cambodia, Indonesia and the Philippines. More precisely, we compare results obtained from poverty measures defined as the summation of individual deprivation functions such as the MPI (an index based on the approach of Alkire and Foster 2011) and others suggested by Chakravarty and D’Ambrosio (2006) and Rippin (2010) and those based on the extended approach of Aaberge and Peluso (2012), as suggested by Silber and Yalonetzky (2013).

The adoption of such an approach is of particular interest in the context of these countries. Indeed, Southeast Asia experienced rapid socioeconomic changes during the past two decades which translated into high- growth performance and poverty reduction. Although the Asian financial crisis in the late 1990s severely affected the well- being of the pop-ulation, human development achievements continued to show progress. However, these achievements have not been uniform. Indonesia was one of the fastest growing economies before the onset of the late 1990s crisis that generated drastic improvements in average incomes and in access to human development opportunities (Sumner et al. 2012). As the economy slowly recovers and welfare gains stabilize, Indonesia is on track to meet Millennium Development Goals (MDG) targets. In contrast, in the past


Measuring multidimensional poverty in three Southeast Asian countries 151

decades, economic growth in the Philippines was low by the standards of the Southeast Asia and poverty estimates showed a lack of response of the incidence of income poverty to growth during the 2000s (Habito 2009; Balicasan 2011). Finally, Cambodia is one of the least developed countries of the region. However, since 1998, owing to macroeconomic and political stability, Cambodia has experienced high and sustained economic growth. As a result, it has registered higher gains in human development between 1990 and 2012 in comparison with those that would have been predicted by its previous performances (UNDP 2013). Nevertheless, Cambodia is lagging in terms of human development relative to its neighbours.

Despite the fact that the multidimensional nature of poverty is now well recognized, studies of poverty in these countries are still dominated by the absolute monetary approach. Apart from the latest published statistics on the MPI in UNDP (2010), there does not seem to have been much work on these countries that take a multidimensional approach to poverty. To our knowledge, the only exceptions are Casimiro et al. (2013) and Balicasan (2011) on the Philippines and a report on child deprivations and multi-dimensional poverty by the United Nations Children’s Fund (UNICEF 2011) in seven countries in East Asia and the Pacific.

The present chapter is organized as follows. Section 2 presents a review of recent methodological refinements suggested in the literature dealing with counting approaches to multidimensional poverty measurement using ordinal indicators. Section 3 illustrates results obtained from the applica-tion of some multidimensional poverty indices presented in section 2 using data from the Demographic Health Surveys for Cambodia (2000, 2005 and 2010), Indonesia (1997, 2003 and 2007) and the Philippines (1997, 2003 and 2008). An analysis of trends over time in multidimensional poverty is also provided for each country.

Given the availability of data and, for comparison purposes, our multidimensional measures include indicators relating to the same three dimensions included in the MPI, namely, standard of living, health and education. Concluding comments are given in section 4.

2 REVIEW OF COUNTING APPROACHES TO MULTIDIMENSIONAL POVERTY

During the past three decades, studies of poverty have moved from a tra-ditional approach that relies on a single indicator of well- being, namely, income or consumption, to an approach that increases the number of admissible attributes when measuring living conditions. The major advan-tage of such a wider concept of poverty is that it allows the researcher to



go beyond the inclusion of only material conditions when attempting to capture the essential aspects of household living conditions.

The main areas of criticism of the traditional income and consump-tion data- based definition are now well known, whether they concern the limitations of income as a proxy for well- being (or its dual aspect, poverty) or the arbitrariness inherent in identifying poor individuals on the basis of a poverty line defined with reference to a whole population’s mean or median income or a predefined consumption basket.

New definitions of poverty have emerged since the seminal works of Townsend (1979) and Sen (1985), each sharing the idea that income can only serve as an indirect indicator when assessing well- being. However, selecting a multidimensional approach to poverty implies addressing issues that need not be faced when taking a unidimensional approach. Thus, several approaches have been proposed in the literature to operationalize the multidimensionality of poverty. At the same time, there is a lack of consensus concerning the best methodology to derive multidimensional poverty measures. According to Thorbecke (2007), the first approach involves aggregating several attributes of well- being into a single index via sophisticated techniques of aggregation and deriving a poverty measure on the basis of this aggregated index. Such an approach, however, de facto amounts to using a unidimensional view of poverty. Several studies have followed this route using methodologies borrowed from efficiency analysis (Lovell et al. 1994) and information theory (Maasoumi 1986, 1999), as well as inertia approaches (Klasen 2000; Sahn and Stifel 2000; Booysen et al. 2008).

As a whole these attempts may be criticized. Sen (1985: 33), for example, believes that: ‘The passion for aggregation makes good sense in many contexts but it can be futile or pointless in others. . . . When we hear of variety, we need not invariably reach for an aggregator.’ Another possibility is to analyze separately each dimension of poverty. The advan-tage of such an analysis lies in its simplicity, but, at the same time, it lacks synthesis, making it difficult to draw a clear picture of multidimensional poverty.

Finally, between these two extremes exists another strategy that preserves the essence of the multiple facets of poverty. This strategy first defines poverty as a combination of shortfalls in each dimension of an individual’s well- being and then derives a multidimensional measure. This is precisely the route adopted by the axiomatic approach to multidimensional poverty. Unlike the one- at- a- time analysis, this approach provides a comprehensive picture of poverty by revealing complexities and ambiguities arising from the interaction between various dimensions and their correlation in the sampled population. The launch of the MPI, popularized by the work of



Alkire and Foster (2011), provides an illustration of such an approach. Naturally there are situations where the choices of the researcher can be severely constrained if the data sets available include solely binary indica-tors of well- being, providing information only on the presence or absence of a deprivation but not on its extent.

Among the shortcomings of the MPI, one may stress its lack of sensi-tivity to the inequality in the distribution of deprivations. This aspect has been recently considered by Alkire and Seth (2014). The counting nature of the approach to the MPI prevents one from using traditional indices of poverty based on continuous variables. Recent contributions have indeed suggested alternative ways of defining poverty indices based on counting but having the same properties as multidimensional poverty indices using continuous variables (Bossert et al. 2013). The main idea of these studies is to provide multidimensional poverty measures that go beyond evaluating some headcount ratio.

Although the axiomatic approach has largely been developed for the unidimensional case, a few studies have attempted to axiomatically derive multidimensional indices of poverty.1 Since the seminal study by Chakravarty et al. (1998), additional extensions and multidimensional classes of poverty have been proposed by Bourguignon and Chakravarty (2003), Alkire and Foster (2011) and Chakravarty and Silber (2008). Studies considering the case of ordinal variables are even more limited and have been recently reviewed by Silber and Yalonetzky (2012) whose contri-bution we now shortly summarize.

Suppose that the relevant population consists of n individuals. Let z = (z1,. . ., zm) be the m- vector of poverty lines and xi = (xi1,. . ., xim) the vector of achievements (ordinal indicators). Let X be an n×m matrix of these achievements so that xij denotes the level of the jth attribute for individual i. More precisely, in the case of ordinal variables, that level might be related to the possession of a given good, the access to some basic services or con-cerns health status or education attainment. Because some attributes may be more important than others, we define a vector of indicator- specific weights: w = (w1,. . ., wm) such that:

wj > 0 and am

j51wj 5 1.

The identification step raises some issues that are more acute in the multi-dimensional case based on ordinal variables. As suggested by Rippin (2012) and Silber and Yalonetzky (2013), it includes several stages rather than the one- step identification that occurs in the unidimensional approach to poverty.



2.1 The Individual Poverty Function

In counting approaches, the first step consists of defining for each dimen-sion a dichotomous function which takes the value of 1 or 0 depending on whether the individual is deprived or not in that dimension.

Let x(xij, zj) be such a dichotomous function that is equal to 1 if the value xij of the attribute j falls short of the poverty line zj, and to 0 otherwise, that is:

x(xij, zj) 5 e1 if xij # zj

0 if xij . zj. (6.1)

In a second step, using this simple dichotomous function, it is possible to define a counting function for each individual which is then used to gen-erate an individual poverty function that might reflect different ways of identifying the poor.

Let ci be the deprivation vector of individual i that consist of values of 0 or 1 on each attribute according to (6.1). The counting function is then defined as follows:

ci(xi, z,w) 5 am

j51x(xij,zj) wj (6.2)

which provides the individual deprivation score as the weighted sum of dichotomous functions defined by (6.1) with wj being the weight assigned to attribute j.

At this stage, the question is to decide when a given individual is classi-fied as poor. Three main approaches have been suggested in the literature: the union, the intersection and an intermediate definition.

Under the intersection approach, individuals are deemed poor if and only if they are deprived in every dimension. In that case, attributes or dimensions of poverty act as substitutes because the absence of depriva-tion in a single dimension is sufficient to classify the person as not being poor. This approach can be regarded as a very conservative way of think-ing about poverty, but it is interesting because it helps place the focus on the ‘extremely poor’.

On the contrary, the union approach states that individuals are poor if they are deprived with respect to at least one attribute. Because every dimension or deprivation counts and is considered as essential, this approach corresponds to an extensive way to identify the poor. It is exten-sively used in the literature on social exclusion measurement theoretically founded on an axiomatic approach.

Between these two extremes, Alkire and Foster (2011) introduce an



innovative approach called the ‘intermediate approach’ which, unlike multidimensional deprivation or social exclusion measures, uses a cross- dimensional cutoff to define the poor. Let k the minimum number of dimensions in which an individual should be deprived to be considered as poor. Because k corresponds to a number of weighted dimensions, k lies between 0 and 1.

This approach can be summarized using the following identification function yAF that returns 1 when an individual is deemed poor relative to the set of poverty lines z and the threshold k:

yAF (xi, z, w, k) 5 µ 1 if am

j51x(xij, zj) wj $ k

0 if am

j51x(xij, zj) wj , k

. (6.3)

This approach is quite flexible and includes as special cases the two tradi-tional identification functions, namely the intersection (that corresponds to k = 1) and the union approach (with k = min(w1,. . ., wm)).

As argued by Alkire and Foster (2011), such an approach is more helpful than the union approach in focusing on deprivations that are reflective of poverty and also for distinguishing and targeting the most extensively deprived. However, as for the choice of the poverty line in unidimensional poverty measurement, the choice of the dimensional cutoff is rather arbi-trary (Ray and Kompal 2011). Indeed, it amounts to ignoring the depriva-tions of those who are deprived in less than k dimensions. In addition, a cross- dimensional cutoff that is reasonable in a given society might not be so in another. As pointed out by Datt (2013), the use of k cannot, by itself, be an adequate solution for the need to identify a target group.

To avoid high poverty rates yielded by the union approach, Rippin (2012) suggested another identification function that can take the follow-ing specific functional form:

yRI(xi,z,w) 5 e cgi if ci 2 0

0 if ci 5 0 (6.4)

with g ≥ 0 and ci, according to expression (2), being the number of weighted deprived attributes experienced by individual i.

This function differentiates between the poor and the non- poor on one hand but takes into account the degree of poverty severity on the other hand. As mentioned by Silber and Yalonetzky (2013), it can be considered as a fuzzy identification function of the poor whose shape is dependent on the value of g, which can be interpreted as a parameter of aversion



to interpersonal inequality that takes into account association between attributes. Thus, yRI is concave for any value of g smaller than one. In that case, attributes are imperfect complements. They act as perfect comple-ments when g = 0 which coincides with the union view of the identifica-tion procedure. On the contrary, the function yRI is convex if g is greater than one which implies that attributes are substitutes. In the extreme case where g → ∞, emphasis is put on those individuals that suffer deprivation in every dimension according to the intersection approach. As pointed out by Rippin (2010), under the intermediate approach introduced by Alkire and Foster (2011), attributes are supposed to be substitutes below the threshold value k and then to act as complements above that value of k.

As mentioned by Silber and Yalonetzky (2013), it is possible to trans-form any fuzzy identification function into a dichotomized identification function. Because yRI increases monotonically with ci, then the choice of a cutoff value d [ [0,1] implicitly defines a threshold k for the individual count ci that solves the implicit function: yRI(k) = d. Therefore, the new dichotomized function works like yAF.

Although the identification step gives an answer on who is poor and what their number is, the measure obtained, at this stage, is rather restric-tive. It is also possible to take into account to what extent those individuals classified as poor are poor. One way to proceed is to consider the poverty gap that identifies the distance between each dimension cutoff and the achievement of the poor in the dimensions they are deprived of. However, unlike in the case of continuous indicators, it is not easy to tell something about the depth of poverty owing to the arbitrariness of any scaling of an ordinal variable. The only thing that can be done in that case is to make the individual poverty function sensitive to the breadth of poverty which may be captured by the number of dimensions in which the individual is deprived.

Hence, the individual poverty function of the counting approach has the form:

pi(xi, z, w, k) = y(xi, z, w, k) g(xi, z, w) (6.5)

which is the product of an identification function y and a function g that measures the breadth of poverty, and may be defined as a function of the deprivation score ci. More generally, g is a real- valued function that maps into the interval [0,1] and is non- decreasing when deprivation increases in any one dimension.

For instance, in the case of the adjusted headcount ratio or the MPI from the Alkire and Foster family of poverty measures, gAF = ci. In the case



of the family of social exclusion measures, defined by Chakravarty and D’Ambrosio (2006), gCD = h(ci) where h is increasing at a non- decreasing rate. In other words, the effect of an additional deprivation in any one dimension is more burdensome for an individual if it is accompanied by deprivation in other dimensions. The function h takes into account the compounding negative effects of multiple deprivations on the overall well- being of the individual. We note that concave breadth functions are never considered in the literature because otherwise whenever inequality among the poor increases, poverty would not decrease as is expected from any poverty measure (see Sen 1976).2

It is easy to show that pi fulfills the following properties drawn from a broader set of properties discussed by Alkire and Foster (2011):

● Normalization: pi reaches a minimum value of 0 if and only if the person is not poor, that is, y = 0 and a maximum value of 1 if indi-vidual i is deprived in every dimension, that is, g = 1.

● Scale invariance: pi is not affected by a scale transformation of the ordinal attributes and thresholds.

● Individual deprivation focus: If an individual i, not deprived in dimension j, receives a transfer, then pi does not change.

● Individual weak monotonicity: pi does not increase where individual i receives a transfer.

● Individual dimensional monotonicity: pi decreases when individual i receives a transfer which makes him or her non- deprived in that dimension.

Having defined the individual poverty functions, the next step is to con-sider the different ways of aggregating individual poverty characteristics to derive poverty measures. Following Silber and Yalonetzky (2013), the aggregation procedure yields what they call a social poverty function.

2.2 The Social Poverty Function

There are two different ways of performing the aggregation of indi-vidual poverty functions to derive poverty measures with the counting approaches. The first, which is extensively used in the literature, derives a class of poverty measures as an average of the individual poverty func-tions. The second, suggested by Aaberge and Peluso (2012), defines the social poverty function directly as a function of the distribution of depri-vation among the poor.



Averaging the individual poverty functions and additional axiomatic propertiesThe social poverty function P is then defined as:

P(X,z,w,k) 51n

an

i51pi(xi,z,w,k) (6.6)

We note that P has all the properties of the pis and also satisfies the follow-ing properties:

● Anonymity or symmetry: It ensures that if two individuals switch their deprivation vectors, the poverty measure P remains unaffected. It implies equal treatment of the equals.

● Principle of population: If each individual is replicated p > 0 times then P does not change. This property allows for comparisons across different sized populations.

● Poverty focus: Changes in the well- being of the non- poor that do not change their poverty status do not affect P.

● Additive decomposability: It states that overall poverty is a weighted average of the shares of the subgroup poverty levels. This axiom enables the identification of those groups that are the most afflicted by poverty.

● Subgroup consistency: If the population is partitioned in G non- overlapping groups of people, and poverty increases/decreases in one group, but does not change in others, then the overall poverty P should increase/decrease. This property is implied by additive decomposability.

● Factor decomposability: This property allows the poverty index to be broken down by dimensions and enables the evaluation of the contribution of each dimension to overall poverty (Chakravarty et al. 1998; Alkire and Foster, 2011). This property is particularly suitable for policy targeting. However, it requires the individual poverty index to be additive across dimensions; this could prevent the fulfilment of some desirable transfer axioms.

Furthermore, the literature on multidimensional poverty measurement with ordinal variables has recently expressed some concern about inequal-ity among the poor. Alkire and Seth (2014) suggested using a separate decomposable measure of inequality – a positive multiple variance – to analyze inequality in deprivation counts among the poor and disparity in poverty across population subgroups. However, drawing from the litera-ture on one- dimensional poverty and on multidimensional poverty based



on continuous attributes, a common approach to account for inequal-ity among the poor has been to adjust the poverty function through the introduction of a parameter of aversion to inequality. Because, unlike the approaches using continuous indicators, it is not possible to capture inequality within each dimension, the only way to address inequality is to consider the distribution of deprivation scores among the poor.

Indeed, following Sen (1976), although changes of poverty can be ana-lyzed considering the evolution of its incidence and intensity, it is also important to analyze whether the changes have been equitable among the poor.

As pointed out by Silber and Yalonetzky (2013), three definitions of reductions in inequality among the poor have been proposed. In all three cases, the social poverty indices are required to increase when inequality increases, or at least not to decrease (in a weak form).

The first definition of change in inequality in deprivations among the poor, which is analogous to a Pigou–Dalton transfer, is the rank- preserving transfer of a deprivation from the poorer to the less- poor person, in which the degree of poverty corresponds to the weighted number of depriva-tions. A measure that is sensitive to inequality among the poor is sup-posed to decrease in the presence of such a transfer. Rippin (2010) used this definition and defined an axiom called the ‘non- decreasingness under inequality- increasing switch’ (NDS). Under this axiom, a transfer of one deprivation from a less poor individual to a poorer individual should not decrease poverty. As shown by Rippin (2010), this property makes it possible to consider situations that are not covered by the Pigou–Dalton transfer principle.

Chakravarty and d’Ambrosio (2006) proposed a similar axiom called ‘non- decreasingness of marginal social exclusion’ (NMS). This axiom states that an increment of deprivation in a poorer person induces a higher, or at least as high, poverty than the same deprivation increase in a less- poor person. The fulfilment of this property requires the individual poverty function to be quasiconvex.

As demonstrated by Silber and Yalonetzky (2013), the social poverty function, P, satisfies NMS if and only if it satisfies NDS. A strong version of this property, also called ‘cross- dimensional convexity’ by Datt (2013), is particularly appealing because it takes into account the fact that the impact of multiple disadvantages on an individual’s well- being cannot be reduced to the sum of their individual effects. In other words, the effect of an increase in deprivation in a given dimension increases with the level of deprivation in other dimensions.

The second definition has been generalized in the multidimen-sional context following the study of Kolm (1977) of inequality in



a multidimensional context. The multidimensional transfer principle (MTP) states that poverty should not increase if it is obtained by a redis-tribution of attributes among individuals according to a bistochastic transformation.3 In other words, MTP requires that the post- transfer dis-tribution of attributes should be more even than the initial distribution.4 This definition has been considered both by Alkire and Foster (2011) and Bossert et al. (2013) but is more suitable for continuous variables. Bossert et al. (2013) propose a property called ‘S- convexity’, whereas Alkire and Foster (2011) call it ‘weak transfers’. Because Alkire and Foster (2011) use a more general approach to poverty identification, they modify the bistochastic matrix so that the averaging of deprivation counts only takes place among the poor.

Finally, the third definition was proposed by Alkire and Foster (2011) and is called the ‘association decreasing rearrangement among the poor’. Under this property, any rearrangement of attributes between two poor individuals i and i9 that breaks the dominance of the initial distribution of deprivation counts between i and i9 (individual i being initially poorer than individual i9) implies that poverty should not increase. The fulfil-ment of this property requires quasiconvex individual poverty functions. However, Alkire and Foster (2011) propose a weaker version of this axiom.

Based on some of these properties, five classes of counting poverty measures can be found in the literature. Some of them are explicit counting measures of multidimensional poverty, as those introduced by Alkire and Foster (2011) and Rippin (2010). Others are implicit measures of poverty because they have been introduced as a class of social exclusion measures by Chakravarty and D’Ambrosio (2006) and Bossert et al. (2013) or are subgroups of this family (Jayaraj and Subramanian 2010).

We consider the class of Alkire and Foster poverty measures that are ‘dimension- adjusted’ multidimensional poverty measures based on the traditional Foster–Greer–Thorbecke measures of poverty.

Alkire and Foster ‘dimension- adjusted’ multidimensional poverty measuresThis class of measures satisfies an array of desirable axioms, including decomposability and dimensional monotonicity, and is defined by:

PAFa (X, z) 5

1na

n

i51yAF (xi, z, c)a

m

j51x(xij, zj) wj c1 2

xij

zjd aj

.

In situations where attributes of poverty are represented by dichot-omized variables, this class of measures is restricted to the case with a = 0. The social poverty function is then:



P0AF 5

1na

n

i51yAF (xi, z, k) ci (6.7)

where ci is given by (2).This measure is the adjusted headcount ratio used for the MPI, and

designated as M0 by Alkire and Foster (2011). As is well known, M0 can be expressed as M0 5 PAF

0 5 H 3 A, that is, the product of the percentage of the multidimensional poor (H) times the average deprivation share across the poor (A).

It is easy to show that PAF0 violates the NDS axiom. Indeed, at best, PAF

0 remains unaffected when a transfer does not change the poverty status of people involved. This is the case of a progressive transfer of deprivations. However, it would be easy to find examples when a regressive transfer in a single dimension implies a decrease of poverty for certain values of k. This occurs because the transfer to the less- poor not only eliminates a particular deprivation for that individual but can also render the individual non- poor. It is also possible to show that PAF

0 does not satisfy the rearrangement axiom in cases of increasing association switches of attributes among the poor for certain values of k. In that sense, PAF

0 is insensitive to how a given set of deprivations is distributed across individuals.

The multidimensional Rippin (2010) class of ordinal poverty measures

PRIg 5

1na

n

i51yRI(xi, z,w)a

m

j51wj x(xij, zj) (6.8)

Replacing yRI by its expression given by (6.4) and rearranging the terms of the summations, it is easy to show that PRI

g can be equivalently expressed as:

PRIg 5

1na

n

i51cg11

i . (6.9)

This class of measures provides poverty measures that are sensitive to the concentration of deprivations because it satisfies NDS and NMS for g ≥ 0. We note that strong versions of NDS and NMS are satisfied for g > 0 even when the adopting identification approaches are based on a cross- dimensional cutoff. Moreover, the identification function has been made to take into account the association between attributes while preserving an additive structure of (6.8) so that this class of poverty measures satisfies not only subgroup decomposability but also factor decomposability.

The multidimensional Chakravarty and D’Ambrosio class of poverty measures

PCD 51na

n

i51y(xi, z, w, k 5 min(w1,. . . ,wm)) h(ci)



with h increasing at a non- decreasing rate, whereas k corresponds to the union approach to identification.

We consider the following specific functional form of h(ci ):

PCDa 5

1na

n

i51ca

i (6.10)

Taking an implicit union approach, this class of measures complies with NDS and NMS if a ≥ 1 Strong versions of these axioms require that a > 1, which is also satisfied even for more general identification approaches. We note that PCD

a becomes more sensitive to the higher deprivation scores as a increases from 2 to infinity. For a = 1, PCD

a becomes the average deprivation score of the society (designated as A by Chakravarty and D’Ambrosio) which corresponds to PAFunion

0 in the case of the union approach. It is easy to show that, a = 2, PCD

a can be rewritten as the sum of the average depri-vation score squared (PAFunion

0 ) 2 and the variance of the society deprivation scores s2:

PCD2 5 (PAFunion

0 ) 2 1 s2. (6.11)

Thus, given PAFunion0 , a reduction in s2 reduces poverty measures in (6.11).

However, unlike the Rippin class of measures, PCDa does not satisfy factor

decomposability.A subgroup of this family of measures has been also derived by Jayaraj

and Subramanian (2010).5

As mentioned earlier, an alternative aggregation method of individual poverty functions has been suggested by Aaberge and Peluso (2012) and extended by Silber and Yalonetzky (2013). This is dealt with in the section immediately below.

The Aaberge and Peluso (2012) approachDrawing on the rank- dependent framework introduced by Sen (1974) and Yaari (1988), Aaberge and Peluso (2012) introduced summary measures of deprivation that are derived from an alternative aggregation method. Indeed, the social poverty function is directly a function of the distribu-tion of deprivation counts because it takes into account the proportions of individuals with j deprivations, j = 1,. . ., m. More precisely, for a number of deprivations h, let F(h) = Pr(ci ≤ h) be the cumulative probability of individuals with up to h deprivations. Then, applying the theorem on the dual theory of choice under risk due to Yaari (1987), and using axioms similar to those defined by Yaari (1988), Aaberge and Peluso (2012) con-cluded that a cumulative distribution F1 is preferable to distribution F2 if and only if:



am21

j50G(F1 (j)) $ a

m21

j50G(F2 (j))

where G is a continuous and non- decreasing real function defined on the unit interval and subscripts 1 and 2 refer to the two distributions, F1 and F2. It is important to note that the function G acts as a weight func-tion used to distort probabilities in the rank- dependent framework. The shape of G reflects whether the preference of the social evaluator is turned towards those people suffering deprivation over many dimensions or those suffering from at least one dimension.

Aaberge and Peluso (2012) then defined the social deprivation measure DG:

DG (F) 5 m 2 am21

j50G(F(j)) . (6.12)

It is easy to understand that DG(F) is equal to 0 if no one in the population has any deprivation. Then F(j) = 1 5 j = 1,.., m − 1 so that:

am21

j50G(F(j)) 5 m.

If, on the contrary, everyone has the maximal number of deprivation, then F(j) = 0 5 j = 1,.., m − 1 and F(m) = 1 so that DG(F) is equal to m.

As demonstrated by Aaberge and Peluso (2012), DG(F) may be decom-posed into the extent of and dispersion in multiple deprivations. In addi-tion, DG(F) satisfies all the properties mentioned earlier except subgroup consistency. The fulfilment of the inequality axioms requires the shape of G to be convex. An extension of the approach of Aaberge and Peluso (2012) was proposed by Silber and Yalonetzky (2013).

Extension proposed by Silber and Yalonetzky (2013)Drawing on the same framework as Aaberge and Peluso (2012), Silber and Yalonetzky (2013) develop a social poverty function that can be manipulated to account for different methods of identification of the poor. Unlike Aaberge and Peluso (2012), Silber and Yalonetzky (2013) work with the concept of the ‘survival function’ or the ‘decumulative distribution function’.6 More precisely, for a number of deprivations h, they consider S(h) 5 Pr(ci $ h) . Then, they suggest the following social poverty function:

PSY (x,z) 51

m 2 k 1 1am

h5kG(S(h)) (6.13)



where G is a non- negative, non- decreasing, real- valued function mapping from, and into, the real interval [0,1] and taking the values G(0) = 0 and G(1) = 1. The first and second derivatives satisfy G9 > 0 and G0 ≤ 0.

The class of measures defined by (6.13) corresponds to a union approach to poverty whenever k = 1. However, by manipulating the choice of k, it is possible to produce measures that identify the poor using the intersection or any other intermediate approach, as in the case of Alkire and Foster (2011).

For empirical purposes, this class of measures has to be adjusted for general weighting. Because the underlying aggregation procedure is con-cerned with the interrelationship between given population proportions and the weighted average of the corresponding number of deprivations, there is only one vector of possible values of deprivation scores for a par-ticular choice of weights.

Suppose we have m dimensions whose weights are given by the vector w = (w1,. . ., wm) with am

j51wj 5 1. In this case, the maximum number of

non- zero deprivation scores m9 will be higher than the given number of dimensions m.

Suppose that deprivation scores are ranked by increasing order of dep-rivation and that we define c = (c0, c1,. . ., ch, . . ., cm9). We therefore let ch [ [0,1] with h = 0,1. . ., m9 all possible values of deprivation scores. The case where cm9 = 1 denotes the deprivation score of an individual deprived in every dimension. It should be mentioned that the deprivation score ch does not give a number of dimensions but a percentage of the overall dimen-sions in which the individual suffers from deprivation.

Hence, because the deprivation scores are ranked by increasing order, the cutoff value k means that we consider as multidimensional poor those individuals with deprivation scores at least equal to ck. The identification and the counting of the poor are now based on (m9 − k + 1) values of all possible non- zero values of c.

In that case, the class PSY can be expressed as follows:

PSY 5m r

m r 2 k 1 1amr

h5kwhG(S(h)) (6.14)

where wh = ch − ch−1 acts as a weight associated with G, which is a func-tion of the proportion of individuals who have at least a deprivation score equal to ch. If all dimensions are equally weighted, then wj = 1/m for all j = 1,. . ., m. In this case, m = m9 and c = (1/m,. . .,h/m,. . .1) with ch − ch−1 = 1/m for all h, it is easy to recover (6.13).

In addition, as for the Aaberge and Peluso (2012) measures, we can prove that the family of indices defined in (6.14) may be broken down into



components reflecting the impact of the mean and dispersion of the distri-bution of deprivation counts.

Let m be the mean of the deprivation counts which is defined by:

m 5 amr

h51chqh

where qh is the proportion of individuals with a deprivation score equal to ch. To supplement information provided by PSY and m, it is useful to intro-duce a measure of dispersion:

DG (S) 5 camrh51

cG(S(h)) 2 amr

j5hqj d d .

We note that, in the case of the union approach, the mean of the distribu-tion coincides with index A of Chakarvarty and D’Ambrosio (2006) and with the M0 of Alkire and Foster (2011). By using (6.14), it is then possible to identify the contribution to PSY of the average number of deprivations, m, as well as of the dispersion of deprivations across the population.7

3 EMPIRICAL APPLICATION TO THREE ASIAN COUNTRIES

Despite of the fact that the multidimensional nature of poverty is now well- recognized in the academic community as well as in international development institutions, studies of poverty in these countries are still dominated by the absolute monetary approach. Thus, it is instructive to begin the analysis by providing comparative evidence on monetary poverty rates along with the economic performances captured by the gross domes-tic product (GDP) growth in the three countries under study. Table 6.1 relates income poverty reduction figures, as measured by the World Bank’s $1.25 a day, and income growth performance within periods chosen as being as close as possible to those associated to the databases available to investigate trends in multidimensional poverty. The results show a wide variation in poverty reduction experience among the three countries. Over the first period, Indonesia emerges as the best performer because poverty decreases from 43.4 percent in 2006 to 29.3 percent in 2002, whereas the annual average growth rate was only 0.68 percent during that period.

By contrast, the Philippines show an increase of poverty within the period 1997–2003 although the economy grew. Also, in Cambodia high performance in the GDP growth rate was accompanied by a slow reduc-tion in monetary poverty during the period 1994–2004. However, over the



second period, Cambodia experienced the strongest poverty reduction with economic growth, whereas, in comparison, the increase of the GDP translates into more moderate declines of poverty in Indonesia and the Philippines.

However, at this stage, it is necessary to supplement the analysis of the trends in the well- being of the population in these countries taking a multi-dimensional approach to poverty. As we show in the following subsections, multidimensional poverty comparisons over time provide useful informa-tion to assess whether income growth translates into social gains.

3.1 Data Description

The use of Demographic and Health Surveys (DHS) initiated by the US Agency for International Development (USAID) offers an alternative instrument to the lack of available data to perform poverty analysis. This is also one of the main sources of data used by the UNDP (2010) for measuring the MPI in several countries. Although these surveys do not include data on income and expenditure, they contain significant informa-tion on the living conditions of the populations in Cambodia (2000, 2005 and 2010), Indonesia (1997, 2003 and 2007) and the Philippines (1997,

Table 6.1 GDP growth rates and poverty changes for three Asian countries

Country Period GDP growth in the period

Poverty rate $1.25 initial

% poverty change

Cambodia 1994–2004 109.5(7.7)

44.5 −15.3(−1.4)

2004–09 47.6(8.1)

37.7 −50.7(−8.5)

Indonesia 1996–2002 4.2(0.7)

43.4 −32.5(−4.8)

2002–08 38.3(5.6)

29.3 −22.9(−3.5)

Philippines 1997–2003 19.8(3.1)

21.6 1.9(0.3)

2003–09 32.2(7.5)

22.0 −16.4(−2.6)

Note: Numbers in brackets refer to annual average growth (or reduction) rates.

Sources: Data on poverty and on GDP growth are from the World Bank (World Development Indicators) and from the ADB (ADB Key Indicators of Developing, Asian and Pacific Countries).



2003 and 2008). In these databases, two main sources of information are available: a list of characteristics of the households and an individual questionnaire for women of reproductive age (15–49), which can be com-bined to extract dimensions of interest. Following the methodology used in the UNDP 2010 report, poverty estimates are performed along the same dimensions as the HDI, namely, education, health and standard of living, and are based on eight attributes available for each country and each year considered. The list of these indicators is presented in Table 6.2.

In addition, because one of our goals is to make poverty comparisons over time and across countries, poverty is estimated for three different years for each country: 2000, 2005 and 2010 for Cambodia; 1997, 2003 and 2007 for Indonesia; and 1997, 2003 and 2008 for the Philippines. Following the methodology of the 2010 UNDP report, a nested- weight structure is adopted where each of the three dimensions mentioned previously has the same weight and each indicator for a given dimension also has the same weight.8

3.2 Empirical Results Based on the Methodology of Alkire and Foster (2011)

We begin this section by analyzing the results obtained from the multi-dimensional poverty measures based on the methodology of Alkire and Foster (2011). Hence, poverty measures are calculated using different values of the cutoff k which corresponds to the minimum weighted sum of indicators in which a household should be deprived to be identified as poor.

Tables 6.3a, 6.3b and 6.3c present the results obtained using the Alkire and Foster multidimensional poverty measures for different values of the cross- dimensional cutoff values of k for Cambodia, Indonesia and the Philippines, respectively. In particular, we consider the union approach and the intermediate approach using the threshold value of k = 33 percent chosen in UNDP (2010) and the value of k = 50 percent capturing house-holds affected by severe poverty. As expected, poverty incidence (H) decreases with the dimensional cutoff value of k, indicating that higher poverty thresholds provide lower levels of poverty and the values of H are higher than the corresponding values of the adjusted headcount ratio (M0) because poor individuals are rarely deprived in all dimensions.9

Comparisons across countries show that the incidence of multidi-mensional poverty is lower in the Philippines (13.8 percent in 2008) and Indonesia (18.9 percent in 2007) than in Cambodia where 33 percent of people are multidimensional poor in at least 33 percent of dimensions in 2010. This ranking remains the same over time for whatever the chosen



value of k. As is evident from Figures 6.1, 6.2 and 6.3, the incidence of poverty (H) and the adjusted headcount ratios (M0) become close to 0 with k = 93 percent for Cambodia in 2010; k = 82 percent for Indonesia in 2007; and k = 77 percent, for the Philippines in 2008. This suggests that, in Cambodia, and to a lesser extent in Indonesia, the dimensions of poverty seem to be more correlated than in the Philippines.

All countries reduced their multidimensional poverty over time at the national level, irrespective of the approach adopted for the identification of the poor. In particular, taking k = 33 percent, the incidence of poverty decreased from 64.5 percent in 2000 to 33 percent in 2010 for Cambodia; from 30.0 percent in 1997 to 18.9 percent in 2007 for Indonesia; and from 22.0 percent in 1997 to 13.8 percent in 2008 for the Philippines. More inter-esting are the results obtained once the headcount ratio (H) is adjusted by the share of deprivations of the poor (A), which provides the adjusted headcount ratio, M0. In particular, declines in M0 are larger in relative terms than those in H, in particular for lower values of k (Tables 6.3a, 6.3b and 6.3c). This is owing to the fact that there are fewer deprived people but those who are deprived experienced fewer deprivations, on average. However, the proportional variation in each component of M0 differs also

Table 6.2 List of dimensions and variables used to compute poverty measures

Dimension Indicator Cut- off Relative weight

Education Child enrollment

Any school- aged child (6–14) is not attending school

1/6

Years of schooling

No household member has completed 5 years of schooling

1/6

Health Mortality Any child has died in the household

1/3

Standard of living

Water Household does not have access to clean drinking water

according to MDG guidelines

1/15

Electricity Household has no electricity 1/15Sanitation Household’s sanitation facility

is not improved1/15

Floor Household has rudimentary floor 1/15Assets Household does not own more

than one of radio, television, telephone, bike or motorbike and does not own a car

1/15


169

Tabl

e 6.

3a

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: C

ambo

dia

Cam

bodi

aH

eadc

ount

M0

A in

%

2000

2005

2010

2000

2005

2010

2000

2005

2010

k =

uni

onN

atio

nal

0.96

60.

917

0.85

80.

434

0.32

60.

252

44.9

35.5

29.4

Urb

an0.

801

0.63

10.

449

0.27

90.

193

0.09

134

.830

.620

.2R

ural

0.99

70.

967

0.94

70.

463

0.34

90.

287

46.4

36.1

30.3

Gap

ratio

1.24

51.

532

2.10

71.

660

1.80

43.

164

1.3

1.2

1.5

k =

33%

Nat

iona

l0.

645

0.45

60.

330

0.37

10.

243

0.16

557

.653

.350

.1U

rban

0.40

10.

272

0.11

80.

219

0.13

90.

052

54.4

51.2

44.2

Rur

al0.

690

0.48

80.

375

0.39

90.

261

0.19

057

.953

.550

.5G

ap ra

tio1.

718

1.79

43.

171

1.82

71.

875

3.62

31.

064

1.04

51.

1

k =

50%

Nat

iona

l0.

397

0.23

00.

136

0.27

00.

152

0.08

768

.165

.964

.0U

rban

0.22

00.

122

0.03

10.

146

0.08

10.

020

66.5

66.4

63.7

Rur

al0.

429

0.24

90.

158

0.29

30.

164

0.10

168

.265

.964

.0G

ap ra

tio1.

953

2.03

55.

117

2.00

42.

018

5.14

11.

026

0.99

21.

005


170

Tabl

e 6.

3a

(con

tinue

d)

Varia

tion

in %

of

HA

nnua

l rat

e of

cha

nge

of H

Varia

tion

in %

of

M0

Ann

ual r

ate

of c

hang

e of

M0

2000

–200

520

05–1

020

00–2

005

2005

–10

2000

–200

520

05–1

020

00–2

005

2005

–10

k =

uni

onN

atio

nal

−5.

1−

6.4

−1.

0−

1.3

−25

.0−

22.6

−5.

6−

5.0

Urb

an−

21.2

−28

.8−

4.6

−6.

6−

30.7

−53

.1−

7.1

−14

.0R

ural

−3.

009

−2.

066

−0.

609

−0.

417

−24

.6−

17.7

−5.

5−

3.8

k =

33%

Nat

iona

l−

29.3

−27

.7−

6.7

−6.

3−

34.5

−32

.0−

8.1

−7.

4U

rban

−32

.23

−56

.5−

7.5

−15

.3−

36.2

−62

.4−

8.6

−17

.8R

ural

−29

.2−

23.1

−6.

7−

5.1

−34

.6−

27.4

−8.

1−

6.2

k =

50%

Nat

iona

l−

41.9

−41

.1−

10.3

−10

.0−

43.8

−42

.8−

10.9

−10

.6U

rban

−44

.3−

74.7

−11

.0−

24.0

−44

.3−

75.8

−11

.0−

24.7

Rur

al−

41.9

−36

.5−

10.3

−8.

7−

43.9

−38

.3−

10.9

−9.

2


171

Tabl

e 6.

3b

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: In

done

sia

Indo

nesia

Hea

dcou

ntM

0A

in %

1997

2003

2007

1997

2003

2007

1997

2003

2007

k =

uni

onN

atio

nal

0.83

80.

774

0.75

50.

230

0.18

90.

160

27.4

24.5

21.2

Urb

an0.

647

0.63

30.

639

0.12

60.

129

0.10

919

.420

.417

.0R

ural

0.91

80.

900

0.84

00.

274

0.24

30.

198

29.8

27.0

23.6

Gap

ratio

1.41

91.

423

1.31

62.

179

1.88

61.

826

1.53

61.

325

1.38

8

k =

33%

Nat

iona

l0.

300

0.23

80.

189

0.14

80.

113

0.08

749

.247

.445

.9U

rban

0.15

50.

169

0.12

70.

067

0.07

40.

054

43.3

43.8

42.7

Rur

al0.

361

0.30

00.

235

0.18

10.

148

0.11

150

.249

.247

.2G

ap ra

tio2.

329

1.77

81.

841

2.70

11.

998

2.03

41.

159

1.12

41.

105

k =

50%

Nat

iona

l0.

120

0.08

20.

056

0.07

60.

051

0.03

462

.962

.260

.9U

rban

0.03

00.

040

0.02

40.

018

0.02

40.

014

59.5

60.6

59.0

Rur

al0.

158

0.11

90.

080

0.10

00.

075

0.04

963

.262

.761

.4G

ap ra

tio5.

238

3.00

33.

271

5.56

93.

103

3.40

11.

063

1.03

31.

040


172

Tabl

e 6.

3b

(con

tinue

d)

Varia

tion

in %

of

HA

nnua

l rat

e of

cha

nge

Varia

tion

in %

of

M0

Ann

ual r

ate

of c

hang

e

1997

–200

320

03–0

719

97–2

003

2003

–07

1997

–200

320

03–0

719

97–2

003

2003

–07

k =

uni

onN

atio

nal

−7.

6−

2.6

−1.

3−

0.6

−17

.6−

15.5

−3.

2−

4.1

Urb

an−

2.2

0.9

−0.

40.

22.

8−

15.8

0.5

−4.

2R

ural

−1.

9−

6.68

2−

0.3

−1.

7−

11.1

−18

.5−

1.9

−5.

0

k =

33%

Nat

iona

l−

20.7

−20

.6−

3.8

−5.

6−

23.6

−23

.1−

4.4

−6.

4U

rban

8.9

−24

.51.

4−

6.8

9.9

−26

.31.

6−

7.3

Rur

al−

16.9

−21

.8−

3.0

−6.

0−

18.7

−25

.0−

3.4

−6.

9

k =

50%

Nat

iona

l−

32.0

−31

.0−

6.2

−8.

9−

32.8

−32

.5−

6.4

−9.

4U

rban

31.6

−38

.54.

7−

11.4

34.3

−40

.15.

0−

12.0

Rur

al−

24.5

−32

.9−

4.6

−9.

5−

25.2

−34

.3−

4.7

−10

.0


173

Tabl

e 6.

3c

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: th

e P

hilip

pine

s

Phili

ppin

esH

eadc

ount

M0

A in

%

1997

2003

2008

1997

2003

2008

1997

2003

2008

k =

uni

onN

atio

nal

0.62

90.

617

0.65

00.

166

0.14

50.

120

26.4

23.5

18.5

Urb

an0.

481

0.50

70.

632

0.10

10.

093

0.09

121

.118

.314

.3R

ural

0.77

60.

735

0.66

80.

231

0.20

10.

150

29.7

27.4

22.5

Gap

ratio

1.61

31.

450

1.05

72.

276

2.17

71.

660

k =

33%

Nat

iona

l0.

220

0.18

10.

138

0.10

60.

085

0.06

148

.047

.244

.1U

rban

0.14

00.

112

0.09

60.

059

0.04

70.

040

42.5

42.2

41.5

Rur

al0.

300

0.25

40.

181

0.15

20.

126

0.08

250

.649

.645

.5G

ap ra

tio2.

148

2.27

11.

874

2.55

72.

671

2.05

7

k =

50%

Nat

iona

l0.

089

0.06

30.

036

0.05

50.

040

0.02

262

.462

.860

.9U

rban

0.03

70.

022

0.01

70.

022

0.01

30.

010

58.3

60.3

60.5

Rur

al0.

141

0.10

70.

055

0.08

90.

068

0.03

363

.563

.461

.0G

ap ra

tio3.

762

4.85

43.

188

4.09

35.

106

3.21

71.

088

1.05

21.

009


174

Tabl

e 6.

3c

(con

tinue

d)

Varia

tion

in %

of

HA

nnua

l rat

e of

cha

nge

of H

Varia

tion

in %

of

M0

Ann

ual r

ate

of c

hang

e of

M0

1997

–200

320

03–0

819

97–2

003

2003

–08

1997

–200

320

03–0

819

97–2

003

2003

–08

k =

uni

onN

atio

nal

−1.

95.

3−

0.3

1.0

−12

.6−

17.0

−2.

2−

3.7

Urb

an5.

324

.70.

94.

5−

8.8

−2.

0−

1.5

−0.

4R

ural

−5.

3−

9.1

−0.

9−

1.9

−12

.7−

25.3

−2.

2−

5.7

k =

33%

Nat

iona

l−

18.0

−23

.3−

3.3

−5.

2−

19.4

−28

.4−

3.5

−6.

5U

rban

−20

.0−

13.8

−3.

7−

2.9

−20

.7−

15.3

−3.

8−

3.3

Rur

al−

15.5

−28

.9−

2.8

−6.

6−

17.1

−34

.7−

3.1

−8.

2

k =

50%

Nat

iona

l−

28.9

−43

.4−

5.5

−10

.8−

28.4

−45

.1−

5.4

−11

.3U

rban

−40

.89

−22

.5−

8.4

−5.

0−

39.0

−22

.2−

7.9

−5.

0R

ural

−23

.7−

49.1

−4.

4−

12.6

−23

.8−

51.0

−4.

4−

13.3



00

10

20

30

40

50

60

70

80

90

100

10 20 30 40 50Cut-off values of k

Cambodia 2000

Cambodia 2005

Cambodia 2010

Perc

enta

ge o

f m

ultid

imen

siona

l poo

r (H

)

60 70 80 90 100

Figure 6.1 Multidimensional poverty headcount ratios in Cambodia

00

10

20

30

40

50

60

70

80

90

100


Indonesia 1997

Indonesia 2003

Indonesia 2007

Perc

enta

ge o

f m

ultid

imen

siona

l poo

r (H

)

60 70 80 90 100

Figure 6.2 Multidimensional poverty headcount ratios in Indonesia



according to the values of k. We observe that the contribution of the vari-ation of H increases with k, implying a more significant contribution of the share of deprived dimensions to the variation of M0 when moving to a more extensive identification approach.

More interesting are the results obtained from the analysis of poverty trends over time for each country and by area of residence.

For the case of Cambodia, the reduction of poverty has been larger in relative terms between 2000 and 2005 than between 2005 and 2010, whatever the value of k. As is evident from Table 6.3a, multidimen-sional poverty is higher in rural areas where most of the population lives (84.48 percent, 85.10 percent and 82.20 percent in 2000, 2005 and 2010, respectively) than in urban areas: the incidence of poverty (H) as well as the intensity of poverty (A) are higher among the poor living in rural areas than in urban areas for every year analyzed, implying higher values of M0 in rural areas. However, poverty decreased in both urban and rural areas over the whole period. More precisely, the alleviation of multidimensional poverty has been higher in urban than in rural areas over the first sub- period 2000–2005 despite of the fact that the poor in rural areas benefited from higher reduction in the intensity of poverty (A) than those living in urban areas, whatever the k values. For instance, we note that in 2005 the

00

10

20

30

40

50

60

70

80

90

100


Philippines 1997

Philippines 2003

Philippines 2008

Perc

enta

ge o

f m

ultid

imen

siona

l poo

r (H

)

60 70 80 90 100

Figure 6.3 Multidimensional poverty headcount ratios in the Philippines



intensity of poverty is lower in rural than in urban areas when considering the most severely deprived (k = 50 percent). The rate of decrease has also been reinforced over the second sub- period 2005–10 in urban areas due to the compounding effect of higher rates of decrease in the incidence of poverty, as well as in the share of deprived dimensions among the poor (A) than in rural area. Despite the fact that the magnitude of the rural/urban gap decreases significantly between 2000 and 2005, it increases or remains roughly stable between 2005 and 2010 depending on the values of k chosen. However, when examining the ratio between rural and urban areas, the results show that the gap ratio for H and M0 increased over the whole period and more significantly between 2005 and 2010 whatever the value of k. This implies that rural populations have benefited less from improve-ments in dimensions of well- being than urban populations. In addition, we note too that the rural/urban gap ratios exhibit higher values when moving to a more restrictive view of poverty.

Decomposition by region of residence provides interesting insights. Tables 6A.1 and 6A.2 in the Appendix to this chapter display poverty esti-mates based on the Alkire and Foster’s approach by region of residence. First, we note that the Plains and Tonle Sap regions concentrate the highest proportion of the population (roughly 70 percent over the whole period: 39.8 percent and 30.5 percent in 2010 in Plains and Tonle Sap, respectively) followed by Mountains (13 percent). In comparison, Phnom Penh and Coastal concentrates 9.3 percent and 7 percent of the whole population, respectively. Our results indicate that the Phnom Penh region registers the lowest levels of multidimensional poverty in comparison with the remain-ing regions where poverty incidence has values from 65.3 percent in Plains to 76.5 percent in Mountain for k = 33 percent. We also observe that all poverty measures (H and M0) decrease over time for each region whatever the value of k. Plains region registers the fastest reduction in poverty over the first sub- period when adopting a restrictive view of poverty, but it is not the case according to the union approach. Indeed, in that case, it is Phnom Penh that exhibits the highest performance as the decrease of the adjusted headcount (M0) is mainly due to the high value of the rate of decrease of H, whereas the intensity of poverty (A) increases. Further, the intensity of poverty increases in Phnom Penh when emphasis is put on the severely poor (k = 50 percent). Over the second sub- period 2005–10, pop-ulation living in Phnom Penh benefited the most from social interventions than people living in other regions whatever the values of k. However, despite the fact that individuals cumulating simultaneously more than 50 percent of deprivations experienced the fastest improvements, they are also deprived in a higher number of dimensions than in 2005 and also than the poor living in the Plains and Coastal regions during the year 2010.



The rates of decrease of poverty were roughly the same values for the other regions except for the Mountains region that records, after Phnom Penh, the highest decrease between 2005 and 2010 although this rate has been among the lowest ones over the first sub- period. However, the results show a slowdown of poverty decrease in the Plains region. By contrast, the Mountains region, where the population is worse off than in other regions, is experiencing a catch- up process because the decrease of poverty over the period 2005–2010 has been among the fastest ones.

Overall, the results provide evidence of a widening gap between urban and rural areas and also between Phnom Penh and the remaining regions. They reveal that the development process has been uneven regarding the areas and the regions of residence.

In Indonesia, poverty also declined at the national level over the two sub- periods (Table 6.3b). However, this trend conceals a non- monotonic evolution of poverty according to the areas of residence and the identifica-tion approach to the poor selected. Indeed, according to the union view of poverty, the incidence of poverty decreased in urban areas between 1997 and 2003 but the poor were poorer in 2003 than in 1997 because they suf-fered from a higher number of deprivations implying an increase of M0. In contrast, the second sub- period (2003–07) registers a higher percentage of the poor, compensated by a higher decrease of the intensity of poverty (A). Similarly, the non- monotonic evolution of poverty in urban areas is confirmed when moving to a more restrictive identification approach. The results indicate an increase of poverty between 1997 and 2003 fol-lowed by a decline over the second sub- period 2003–07. In contrast, there are clear continued reductions in multidimensional poverty rates in rural areas where around 57 percent of the population lives in 2007. In addition, the estimates tend to show an acceleration of poverty reduction between 2003 and 2007 though being lower than the urban ones during the same period, except for the union approach. Therefore, as shown in Table 6.3b, the magnitude of the gap between rural and urban poverty rates highlights a continued decrease over time irrespective of the dimensional cutoff chosen. However, the rural/urban gap ratio provides a somewhat different picture regarding the second sub- period. Although a clear decline of the disparities between rural and urban areas is observed between 1997 and 2003, largely due to the increase of urban poverty, the trends within the second sub- period seem to be highly dependent on the choice of the cutoff value of k. The slight decrease of the rural/urban gap ratio, according to the union approach, should be contrasted with the slightly widening gap between rural and urban areas that emerges from using k = 50 percent and k = 33 percent.

Analysis by region of residence shows that Java which concentrates



about 60 percent of the population is better off in comparisons with other regions (see Appendix Tables 6A.3 and 6A.4). By contrast, Bali, which only accounts for less than 6 percent of the whole population, has the highest rates of multidimensional poor. We note that the ranking of the other regions in terms of incidence of poverty is dependent on the values of k. The same is true when we attempt to identify the region that benefited the most from poverty reduction over the two sub- periods. Indeed, following the union approach, the highest decline in the percentage of multidimen-sional poor (H) is found in Kalimantan over the first sub- period, whereas Java followed by the Sulawesi region shows higher performance in adjusted headcount values (M0) because the intensity of poverty declines at a faster rate especially in the Java region. However using a more restrictive view of poverty, it is the Java region that exhibits the fastest decline in poverty rates and Kalimantan the slowest over the first sub- period. In contrast, over the second sub- period, a catching- up process can be observed as Bali records the highest progress. Moreover, we point out the case of Sulawesi and Sumatera, which show very similar poverty rates in 1997, apart from the union approach, but their poverty trends over time are rather different. Indeed, Sumatera outperforms Sulawesi over the first period and witnesses an acceleration of the poverty decline over the second sub- period widening the gap with Sulawesi where the weakening of poverty decline is particu-larly evident according to k = 33 percent.

Turning now to the case of the Philippines, a decline of poverty is observed at the national level over the two sub- periods but this trend is more ambiguous when adopting an extensive view of poverty (see Table 6.3c). Indeed, the percentage increase in the multidimensional poor between 2003 and 2008 is compensated by a higher decline of the share of deprivations among the poor. In addition, the second sub- period witnesses an acceleration of poverty reduction, except for the union view of poverty.

As for Cambodia and Indonesia, poverty remains a rural phenomenon where about 50 percent of the population lives. Remarkably, as for Indonesia, conclusions drawn from comparisons of percentage rates of change of poverty over the two sub- periods between urban and rural areas are highly dependent on the identification approach adopted. The union approach shows that rural population benefited from a faster reduction of poverty than urban population over time, implying a decrease of the rural/urban gap, as shown in Table 6.3c. This is largely driven by the increase of the poverty headcount ratio in urban areas although reductions in the intensity of poverty tend to be more in favor of the urban population. By contrast, conflicting trends may be observed when we move to more restrictive views of poverty. Hence, according to k = 33 percent, it may be observed that urban population benefited more from social progress than



rural population within the period 1997–2003, whereas the reverse is true over the second sub- period because the percentage rate of decline in rural areas is higher than in urban areas. However, when emphasis is put on the more severely deprived (k = 50 percent), the results indicate inverted trends characterized by an increase, followed by a decline, of the rural/urban gap ratio over the two sub- periods, respectively. Remarkably, it is important to mention that the intensity of poverty increases over time in urban areas counteracting the ameliorating effect of changes to the poverty headcount ratio.

Consequently, it is not easy to obtain conclusive results regarding changes in the rural/urban gap for the whole period under study. The same conclusion holds when we look at poverty changes over the period by region of residence (see Appendix Tables 6A.5 and 6A.6). We see that the Luzon region where more than 56 percent of the population is living seems to be the least- deprived region, according to H and M0, whatever the iden-tification approach adopted. In addition, whereas Mindanao was worse off than Visayas in 1997 (according to H and M0) whatever the values of k, it registers the highest performance in poverty reduction between 1997 and 2003, explaining the very similar poverty rates between the two regions in 2003. However, this no longer holds over the second sub- period because Visayas exhibits higher percentage poverty changes than Mindanao, except for k = 50 percent. It is not easy to identify which region has benefited from the highest drop in poverty over the period 2003–08 because the results differ according to the dimension cutoff value of k.

The previous discussion is largely based on poverty measures where an individual’s contribution to aggregate poverty depends on his/her own achievement vector. As emphasized in section 2, the implicit assumption underlying the Alkire and Foster approach is that the overall effect of multiple deprivations is summarized by the sum of their individual effects. Thus, these measures are completely insensitive to the distribution of a given set of deprivations.

3.3 Empirical Results from Poverty Measures Sensitive to Inequality

We now proceed to look at poverty measures that place a greater empha-size on the compounding effect of multiple deprivations and, thus, are sensitive to the impact of the spread of the deprivations across individuals.

Tables 6.4a, 6.4b and 6.4c present for the three countries, respectively, the measures based on the family of Rippin’s indices for values of g = 1.5 and g = 2. These tables also display poverty measures obtained from a subgroup of the Chakravarty and D’Ambrosio family of social exclu-sion indices (a = 2) and from the extension of the Aaberge and Peluso


181

Tabl

e 6.

4a

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: Cam

bodi

a

Cam

bodi

aM

easu

res

Varia

tion

in %

Ann

ual r

ate

of c

hang

e

2000

2005

2010

2000

–05

2005

–10

2000

–200

520

05–1

0

Rip

pin

g =

1.5

Nat

iona

l0.

197

0.11

90.

076

−39

.5−

36.4

5−

9.6

−8.

7U

rban

0.10

90.

067

0.02

1−

39.1

−68

.8−

9.4

−20

.8R

ural

0.21

30.

129

0.08

8−

39.7

−31

.7−

9.6

−7.

3G

ap ra

tio1.

949

1.92

74.

216

Rip

pin

g =

2N

atio

nal

0.16

20.

093

0.05

6−

42.4

−39

.6−

10.4

−9.

6U

rban

0.08

70.

052

0.01

5−

40.6

−71

.8−

9.9

−22

.3R

ural

0.17

60.

100

0.06

5−

42.8

−35

.0−

10.6

−8.

3G

ap ra

tio2.

010

1.93

64.

456

Cha

krav

arty

and

D’A

mbr

osio

for a

= 2

Nat

iona

l0.

247

0.15

80.

107

−35

.9−

32.7

−8.

5−

7.6

Urb

an0.

142

0.08

90.

031

−37

.0−

65.0

−8.

8−

19.0

Rur

al0.

266

0.17

00.

123

−36

.0−

27.9

−8.

5−

6.3

Gap

ratio

1.87

61.

907

3.93

4

Ext

ensio

n of

Aab

erge

and

Pel

uso

Nat

iona

l0.

572

0.45

40.

366

−20

.6−

19.3

−4.

5−

4.2

Urb

an0.

419

0.31

20.

158

−25

.5−

49.4

−5.

7−

12.7

Rur

al0.

593

0.47

20.

398

−20

.3−

15.7

−4.

4−

3.4

Gap

ratio

1.41

31.

512

2.51

8


182

Tabl

e 6.

4b

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: Ind

ones

ia

Indo

nesia

Mea

sure

sVa

riatio

n in

%A

nnua

l rat

e of

cha

nge

1997

2003

2007

1997

–200

320

03–0

719

97–2

003

2003

–07

Rip

pin

g =

1.5

Nat

iona

l0.

066

0.04

90.

036

−26

.4−

25.7

−5.

0−

7.2

Urb

an0.

025

0.02

80.

020

11.9

−29

.21.

9−

8.3

Rur

al0.

083

0.06

70.

048

−19

.6−

28.1

−3.

6−

7.9

Gap

ratio

3.32

42.

389

2.42

7

Rip

pin

g =

2N

atio

nal

0.04

90.

035

0.02

5−

28.5

−28

.0−

5.4

−7.

9U

rban

0.01

70.

019

0.01

315

.5−

32.2

2.4

−9.

2R

ural

0.06

20.

049

0.03

4−

21.6

−30

.3−

4.0

−8.

6G

ap ra

tio3.

742

2.53

92.

610

Cha

krav

arty

and

D’A

mbr

osio

for a

= 2

Nat

iona

l0.

094

0.07

20.

055

−24

.0−

23.1

−4.

5−

6.4

Urb

an0.

040

0.04

30.

032

8.6

−25

.91.

4−

7.2

Rur

al0.

117

0.09

70.

072

−17

.3−

25.6

−3.

1−

7.1

Gap

ratio

2.93

82.

237

2.24

7

Ext

ensio

n of

Aab

erge

and

Pel

uso

Nat

iona

l0.

341

0.29

00.

249

−14

.8−

14.2

−2.

6−

3.7

Urb

an0.

203

0.21

00.

177

3.5

−15

.90.

6−

4.2

Rur

al0.

387

0.34

90.

295

−9.

8−

15.5

−1.

7−

4.1

Gap

ratio

1.90

41.

660

1.66

8


183

Tabl

e 6.

4c

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: the

Phi

lippi

nes

Phili

ppin

esM

easu

res

Varia

tion

in %

Ann

ual r

ate

of c

hang

e

1997

2003

2008

1997

–200

320

03–0

819

97–2

003

2003

–08

Rip

pin

g =

1.5

Nat

iona

l0.

046

0.03

70.

024

−19

.9−

34.3

−3.

6−

8.0

Urb

an0.

022

0.01

80.

015

−18

.89

−17

.0−

3.4

−3.

7R

ural

0.07

10.

058

0.03

4−

18.3

−40

.9−

3.3

−10

.0G

ap R

atio

3.27

93.

300

2.34

9

Rip

pin

g =

2N

atio

nal

0.03

40.

027

0.01

7−

21.1

−38

.0−

3.9

−9.

1U

rban

0.01

50.

012

0.01

0−

20.2

−18

.1−

3.7

−3.

9R

ural

0.05

30.

043

0.02

4−

19.4

−44

.8−

3.5

−11

.2G

ap R

atio

3.64

93.

685

2.48

4

Cha

krav

arty

and

D’A

mbr

osio

for a

= 2

Nat

iona

l0.

066

0.05

40.

038

−18

.2−

29.9

−3.

3−

6.9

Urb

an0.

034

0.02

80.

024

−16

.7−

14.7

−3.

0−

3.1

Rur

al0.

099

0.08

20.

052

−16

.9−

36.5

−3.

0−

8.7

Gap

Rat

io2.

934

2.92

72.

178

Ext

ensio

n of

Aab

erge

and

Pel

uso

Nat

iona

l0.

268

0.23

70.

196

−11

.6−

17.5

−2.

0−

3.8

Urb

an0.

174

0.15

80.

147

−9.

3−

6.4

−1.

6−

1.3

Rur

al0.

348

0.31

10.

240

−10

.6−

22.9

−1.

8−

5.1

Gap

Rat

io2.

003

1.97

51.

627



approach suggested by Silber and Yalonetzky. All of these measures take implicitly or explicitly a union approach to poverty. At first sight, these measures are not directly comparable with M0 because they consider deprivation for the whole population and not just the poor (as is the case for k ≠ min(w1,. . ., wm)). Even in the case of the union approach, the comparison could be misleading because these measures involve the choice of an inequality aversion parameter. With higher values of the parameter measuring the degree of aversion to inequality, higher weights are assigned to larger deprivation scores. In that case, the concern of the social evaluator tends to be more in favor of the intersection approach. In particular, as mentioned in Section 2, it is easy to observe an equivalence between PCD and PRI for a = g + 1. Hence, looking at the Chakravarty and D’Ambrosio measures and the Rippin measures (see Tables 6.4a–6.4c), we see that poverty estimates decreases as a(g) increases. These measures seem to be more related to those obtained when taking a more restrictive view of poverty, following Alkire and Foster. In contrast, the estimates provided by extension of the Aaberge and Peluso approach are completely different and their range of variation is, to some extent, closer to the one obtained for M0 using the union approach. It is also interesting to note that for the three countries the rural/urban gap ratio becomes more important with higher values of a or g. This suggests that when more weight is assigned to the most deprived populations in rural areas they are more likely to suffer simultaneously from multiple disadvantages than those living in urban areas. In other words, deprivations are more related in rural than in urban areas for the most deprived.

For Cambodia, generally speaking, accounting for the dispersion of dep-rivation counts does not seem to change the overall picture obtained with the Alkire and Foster poverty measures. Poverty declines over the whole period but we observe that these poverty measures provide ambiguous con-clusions regarding the trends in the rural/urban gap. Although the Rippin measures indicate that poverty declines at a faster rate in rural than in urban areas between 2000 and 2005, implying a slight decrease of the rural/urban gap ratio, the reverse is true following the Chakravarty and D’Ambrosio measure and the extension of the Aaberge and Peluso measure. Also, as shown in Appendix Table 6A.7a, the trends observed by area of residence are consistent with those obtained when adopting a more restrictive iden-tification approach to poverty, as for k = 33 percent and k = 50 percent in the case of the Alkire and Foster measures. Overall, accounting for the concentration of deprivation counts across individuals makes it possible to highlight the strong performance of Phnom Penh between 2005 and 2010 in comparison with the other regions. Similarly, the Mountains region that was the worse off over the first sub- period was playing a catching- up



process over the second sub- period. Hence, its rates of poverty decline accelerated within 2005 and 2010 and accounted for about half the rate of Phnom Penh. The results also confirm the slowdown in the process of poverty reduction pointed out previously for the Plains region.

As for Indonesia, unlike the ambiguous conclusions derived from the Alkire and Foster measures regarding the trends in urban poverty, Table 6.4b emphasizes a marked increase of urban poverty between 1997 and 2003, followed by a significant decline at a rate very close to that of the rural one over 2003–07. In addition, it is interesting to note that with the exception of Java and Bali where the population is better- off and worse- off, respectively, poverty was very similar in the other regions in 1997. However, social progress has not favored equally every region, con-firming the trends underlined especially for the case of Kalimantan and Sulawesi, according to the Alkire and Foster measures (see Appendix Table 6A.7b)

Finally, as for the Philippines, Table 6.4c shows that poverty measures provide results that support findings drawn from the restrictive approach to poverty. Poverty decreases over the whole period and rural population have benefited more from social improvements than the urban popula-tion especially during the second sub- period (2003–08). The results in Appendix Table 6A.7c are also consistent with the trends observed by region of residence over time derived from the Alkire and Foster measures, except for the union approach to the identification of the poor.

3.4 Decompositions into the Mean and Dispersion of Deprivation Counts

One of the advantages of the measures based on Chakravarty and D’Ambrosio and those derived from the extension of the Aaberge and Peluso approach is that they allow decomposition into the mean and dispersion of the deprivation counts. Tables 6.5a and 6.5b display the results of such decomposition for each country. We remind the reader that the Chakravarty and D’Ambrosio measure for a = 2 can be expressed by summing- up the square of the average deprivation score and the variance of deprivation scores. Similarly, the measure derived from the extension of the Aaberge and Peluso measure can also be defined as the sum of M0 and a dispersion measure of the distribution of deprivation counts. This is the reason why the values obtained from the latter measure are of the same order of magnitude as those related to M0.

First, we may observe that there is a positive relationship between poverty levels and inequality in deprivations. However, the reverse is true because the inequality component is expressed as a percentage of the poverty measure. It is also interesting to observe that the dispersion in


186

Tabl

e 6.

5a

Dec

ompo

sitio

n of

pov

erty

mea

sure

s with

rega

rd to

the

mea

n an

d di

sper

sion

of

the

depr

ivat

ion

coun

ts

dist

ribu

tion

Cam

bodi

a

Pove

rty

mea

sure

s of

Cha

krav

arty

and

D

’Am

bros

io fo

r a =

2In

equa

lity

com

pone

nt

2000

2005

2010

2000

–05

2005

–10

2000

2005

2010

2000

–05

2005

–10

Nat

iona

l0.

247

0.15

80.

107

−35

.9−

32.7

0.05

80.

052

0.04

3−

2.5

−5.

9U

rban

0.14

20.

089

0.03

1−

37.0

−65

.00.

064

0.05

20.

023

−8.

6−

32.4

Rur

al0.

266

0.17

00.

123

−36

.0−

27.9

0.05

20.

049

0.04

1−

1.2

−4.

9Ph

nom

Pen

h0.

074

0.05

20.

014

−29

.4−

73.4

0.03

90.

037

0.01

1−

2.5

−49

.8Pl

ains

0.24

30.

134

0.10

3−

44.7

−23

.30.

050

0.04

10.

038

−3.

8−

2.2

Tonl

e Sa

p0.

279

0.18

60.

120

−33

.4−

35.5

0.05

70.

053

0.04

3−

1.3

−5.

7C

oast

al0.

259

0.17

00.

110

−34

.4−

35.3

0.05

70.

049

0.03

9−

3.0

−5.

8M

ount

ains

0.30

40.

239

0.15

0−

21.5

−37

.20.

055

0.05

80.

050

1.1

−3.

7

Indo

nesia

1997

2003

2007

1997

–200

320

03–0

719

9720

0320

0719

97–2

003

2003

–07

Nat

iona

l0.

094

0.07

20.

055

−24

.0−

23.1

0.04

10.

036

0.02

9−

6.0

−8.

8U

rban

0.04

00.

043

0.03

28.

6−

25.9

0.02

40.

027

0.02

06.

4−

14.7

Rur

al0.

117

0.09

70.

072

−17

.3−

25.6

0.04

20.

038

0.03

3−

3.9

−5.

0Su

mat

era

0.10

00.

077

0.05

8−

23.0

−24

.20.

044

0.03

80.

031

−6.

5−

9.0

Java

0.08

50.

062

0.04

7−

26.7

−24

.20.

038

0.03

20.

026

−8.

1−

8.8

Bal

i0.

146

0.12

20.

081

−16

.4−

33.9

0.05

40.

049

0.03

8−

3.3

−9.

7K

alim

anta

n0.

103

0.08

80.

063

−14

.6−

28.4

0.03

80.

041

0.03

23.

2−

10.9

Sula

wes

i0.

102

0.08

10.

075

−20

.3−

7.5

0.04

10.

041

0.03

6−

0.4

−6.

2


187

Phili

ppin

es19

9720

0320

0819

97–2

003

2003

–08

1997

2003

2008

1997

–200

320

03–0

8

Nat

iona

l0.

066

0.05

40.

038

−18

.2−

29.9

0.03

90.

033

0.02

4−

8.4

−17

.8U

rban

0.03

40.

028

0.02

4−

16.7

−14

.70.

023

0.02

00.

016

−11

.6−

13.5

Rur

al0.

099

0.08

20.

052

−16

.9−

36.5

0.04

60.

042

0.03

0−

4.1

−14

.8L

uzon

0.04

50.

037

0.02

8−

18.3

−25

.50.

029

0.02

50.

018

−10

.2−

18.1

Visa

yas

0.08

60.

075

0.04

8−

12.3

−35

.80.

043

0.04

00.

027

−3.

5−

17.1

Min

dana

o0.

101

0.07

80.

056

−23

.1−

28.6

0.04

80.

041

0.03

0−

7.8

−13

.0

Not

e:

Col

umns

200

0–05

and

200

5–10

(199

7–20

03, 2

003–

07; 1

997–

2003

, 200

3–08

) disp

lay

chan

ges o

f th

e co

mpo

nent

s exp

ress

ed a

s per

cent

ages

of

the

valu

e of

the

initi

al y

ear o

f th

e pe

riod.


188

Tabl

e 6.

5b

Dec

ompo

sitio

n of

pov

erty

mea

sure

s with

rega

rd to

the

mea

n an

d di

sper

sion

of

the

depr

ivat

ion

coun

ts

dist

ribu

tion

Cam

bodi

a

Mea

sure

s bas

ed o

n th

e ex

tens

ion

of A

aber

ge a

nd P

elus

oIn

equa

lity

com

pone

nt o

f PSY

2000

2005

2010

2000

–05

2005

–10

2000

2005

2010

2000

–05

2005

–10

Nat

iona

l0.

572

0.45

40.

366

−20

.6−

19.3

0.13

80.

128

0.11

4−

1.6

−3.

1U

rban

0.41

90.

312

0.15

8−

25.5

−49

.40.

141

0.11

90.

067

−5.

2−

16.5

Rur

al0.

593

0.47

20.

398

−20

.3−

15.7

0.13

00.

124

0.11

1−

1.0

−2.

7Ph

nom

Pen

h0.

291

0.21

10.

090

−27

.5−

57.5

0.10

50.

090

0.04

1−

5.4

−23

.3Pl

ains

0.56

60.

418

0.36

3−

26.2

−13

.20.

127

0.11

10.

107

−2.

7−

1.1

Tonl

e Sa

p0.

607

0.49

40.

392

−18

.6−

20.6

0.13

60.

130

0.11

5−

0.96

−3.

2C

oast

al0.

585

0.47

20.

375

−19

.3−

20.6

0.13

60.

125

0.10

9−

1.9

−3.

3M

ount

ains

0.63

20.

562

0.44

0−

11.1

−21

.60.

133

0.13

70.

124

0.7

−2.

40

Indo

nesia

1997

2003

2007

1997

–200

320

03–0

719

9720

0320

0719

97–2

003

2003

–07

Nat

iona

l0.

341

0.29

00.

249

−14

.8−

14.2

0.11

10.

101

0.08

9−

3.0

−4.

1U

rban

0.20

30.

210

0.17

73.

5−

15.9

0.07

80.

081

0.06

81.

8−

6.2

Rur

al0.

387

0.34

90.

295

−9.

8−

15.5

0.11

30.

106

0.09

7−

1.9

−2.

6Su

mat

era

0.35

20.

303

0.25

8−

14.0

−14

.80.

115

0.10

40.

092

−3.

2−

4.2

Java

0.32

20.

270

0.22

8−

16.2

−15

.40.

106

0.09

40.

083

−3.

7−

4.4

Bal

i0.

435

0.39

40.

313

−9.

4−

20.6

0.13

10.

124

0.10

5−

1.7

−4.

8K

alim

anta

n0.

360

0.32

40.

270

−9.

9−

16.7

0.10

50.

108

0.09

00.

9−

5.6

Sula

wes

i0.

357

0.30

80.

299

−13

.7−

2.9

0.11

20.

109

0.10

2−

1.0

−2.

0


189

Phili

ppin

es19

9720

0320

0819

97–2

003

2003

–08

1997

2003

2008

1997

–200

320

03–0

8

Nat

iona

l0.

268

0.23

70.

196

−11

.6−

17.5

0.10

20.

092

0.07

5−

3.8

−7.

1U

rban

0.17

40.

158

0.14

7−

9.3

−6.

40.

072

0.06

50.

057

−4.

2−

5.2

Rur

al0.

348

0.31

10.

240

−10

.6−

22.9

0.11

70.

110

0.08

9−

2.1

−6.

6L

uzon

0.21

00.

185

0.16

0−

12.0

−13

.80.

085

0.07

50.

063

−4.

6−

6.7

Visa

yas

0.31

90.

294

0.22

8−

7.7

−22

.40.

112

0.10

70.

084

−1.

6−

7.7

Min

dana

o0.

350

0.30

00.

250

−14

.2−

16.7

0.12

00.

107

0.09

1−

3.8

−5.

3

Not

e:

Col

umns

200

0–05

and

200

5–10

(199

7–20

03, 2

003–

07; 1

997–

2003

, 200

3–08

) disp

lay

chan

ges o

f th

e co

mpo

nent

s exp

ress

ed a

s per

cent

ages

of

the

valu

e of

the

initi

al y

ear o

f th

e pe

riod.



deprivation counts is higher in rural than in urban areas in Indonesia and in the Philippines over the whole period. We now proceed to analyze trends over time in the inequality components for each country to see whether the decomposition of the measure brings something new to our understanding of poverty.

Tables 6.5a and 6.5b show that inequality decreases over time in the urban and rural areas. In particular, the contribution of the inequality component to the poverty reduction increases over time. The decrease in the dispersion in deprivations accounts for about 50 percent and 34 percent of the rate of decrease of urban poverty between 2005 and 2010 according to the Chakravarty and D’Ambrosio (CDA) and the exten-sion of Aaberge and Peluso (Ext. AP) measures, respectively. At the regional level, the decrease of poverty in Phnom Penh is largely driven by the inequality component (about 68 percent and 41 percent following CDA and Ext. AP) over the second- period. In addition, we see that the Mountain region registers an increase of inequality over the first period (2000–2005) that may explain the low rate of decrease of poverty reduc-tion compared with other regions.

As for the case of Indonesia, the increase of urban poverty between 1997 and 2003 seems to be largely driven by an increase of the inequal-ity component. It contributes to more than 74 percent (and 51 percent in Table 6.5b) to the rate of poverty increase. By contrast, inequality decreases in rural areas but its contribution to the percentage variation of poverty slows down over the second sub- period. At the regional level, we see that Kalimantan exhibits a slight increase of the inequality component during the first period that hampers about 21 percent (about 33 percent in Table 6.5b) of the reduction of poverty. By contrast, in spite of the fact that Sulawesi registers very low performance in poverty reduction between 2003 and 2007, more than 80 percent (96 percent in Table 6.5b) comes from the decline in the dispersion of deprivations. Finally, decrease of inequal-ity goes hand in hand with poverty decrease over the whole period in the Philippines. In particular, it is in the Luzon region, which is better off in comparison with the other regions, that the contribution of inequality to the variation of poverty has been the highest.

3.5 Decompositions by Dimension

Finally, as mentioned in section 2, a significant advantage of the Alkire and Foster family of measures is that once multidimensional poor have been identified, the aggregate poverty measure may be broken down into the sum of the contributions of the different dimensions. This provides interesting information that is particularly suitable for policy targeting.



However, most of the measures that are sensitive to the breadth of the dis-tribution of deprivation counts do not allow dimensional decomposability except the Rippin family of measures. Here, it is instructive to compare the results obtained with those based on the decomposition of the Alkire and Foster and the Rippin measures.

Tables 6.6a, 6.6b and 6.6c report the percentage contributions of each of three dimensions10 to the overall poverty for the Alkire and Foster measure using the cutoff value of k = 33 percent and for the Rippin measure with g = 1.5. The final columns of these tables also display deprivations by dimension. In addition, because the Alkire and Foster approach makes it possible to interpret the censored headcount ratios with respect to the per-centage of people who are poor, Table 6.7 presents the intensity of poverty in each dimension among the poor.

In Tables 6.6a, 6.6b and 6.6c considering first the decomposition of the Alkire and Foster measure, we see that for every country the main contributor to overall poverty is deprivation in health dimension which refers to children mortality. Note that health contribution is particularly disproportionally high in comparison with the contributions of other dimensions in urban areas. We also find that the contribution of health dimension increases over time in all three countries, regardless of the area of residence. Indeed, the percentage change over time in health depriva-tion has been lower than changes in education and standard of living dimensions.

As shown in Table 6.7, we also find that health registers an increase of deprivations among the poor in each country. By contrast, it is educa-tion that shows the highest progress at the national level and regardless of the area of residence, except for Indonesia between 1997 and 2003 in urban areas. It is observed that the increase of urban poverty in Indonesia between 1997 and 2003 was mainly due to an increase of deprivations in education and health dimensions.

Finally, turning to the Rippin decomposition that takes into account the dispersion in the distribution of deprivation counts, the results do not differ significantly. The contributions in indicators related to health and standard of living remain the highest contributors to overall poverty. However, some differences with the previous decomposition are worth mentioning. For Cambodia, results drawn from the Rippin approach indicate a slight decrease in the contribution of health deprivation over the period 2000–2005. This suggests that people experiencing cumula-tive deprivations benefited from higher progress in the health dimen-sion in rural areas than is suggested by the Alkire and Foster measures. Although decompositions by areas of residence are not reported here, the lowest performance of the Mountains region between 2000 and 2005


192

Tabl

e 6.

6a

Con

trib

utio

n of

eac

h di

men

sion

to o

vera

ll po

vert

y fo

llow

ing

Alk

ire a

nd F

oste

r’s m

etho

dolo

gy w

ith k

= 3

3 pe

rcen

t and

Rip

pin

mea

sure

Cam

bodi

aC

ontr

ibut

ions

in %

% c

hang

e in

co

ntrib

utio

nD

epriv

atio

ns%

cha

nge

in

depr

ivat

ion

2000

2005

2010

2000

–05

2005

–10

2000

2005

2010

2000

–05

2005

–10

Alk

ire a

nd F

oste

r for

k =

33%

Nat

iona

lE

duca

tion

32.5

28.7

25.6

−11

.7−

10.6

36.2

20.9

12.7

−42

.2−

39.3

Mor

talit

y28

.234

.236

.121

.55.

731

.424

.917

.9−

20.5

−28

.2St

d of

livi

ng39

.437

.138

.2−

5.7

3.0

43.9

27.1

19.0

−38

.2−

30.0

Urb

anE

duca

tion

28.9

26.7

17.7

−7.

6−

33.5

18.9

11.2

2.8

−41

.1−

75.0

Mor

talit

y37

.241

.960

.312

.843

.824

.417

.59.

5−

28.0

−46

.0St

d of

livi

ng33

.931

.421

.9−

7.6

−30

.122

.313

.13.

4−

41.1

−73

.7R

ural

Edu

catio

n32

.828

.826

.1−

12.1

−9.

639

.322

.614

.8−

42.5

−34

.4M

orta

lity

27.2

33.5

34.7

22.8

3.6

32.6

26.2

19.7

−19

.6−

24.8

Std

of li

ving

39.9

37.7

39.2

−5.

64.

147

.829

.522

.3−

38.3

−24

.4


193

Rip

pin

(g =

1.5

)N

atio

nal

Edu

catio

n31

.628

.425

.3−

10.3

−10

.60.

187

0.10

20.

058

−45

.7−

43.3

Mor

talit

y31

.134

.634

.511

.5−

0.4

0.18

40.

124

0.07

9−

32.5

−36

.7St

d of

livi

ng37

.337

. 040

.2−

0.9

8.6

0.22

10.

132

0.09

1−

40.0

−30

.9U

rban

Edu

catio

n30

.129

.023

.7−

3.7

−18

.20.

099

0.05

80.

015

−41

.3−

74.5

Mor

talit

y35

.037

.947

.58.

125

.60.

115

0.07

60.

030

−34

.1−

60.8

Std

of li

ving

34.9

33.2

28.7

−4.

9−

13.4

0.11

50.

066

0.01

8−

42.1

−72

.9R

ural

Edu

catio

n31

.828

.325

.4−

10.9

−10

.30.

203

0.10

90.

067

−46

.3−

38.7

Mor

talit

y30

.734

.433

.811

.9−

1.5

0.19

60.

132

0.08

9−

32.5

52−

32.7

Std

of li

ving

37.5

37.3

40.8

−0.

69.

20.

240

0.14

40.

107

−40

.1−

25.4


194

Tabl

e 6.

6b

Con

trib

utio

n of

eac

h di

men

sion

to o

vera

ll po

vert

y fo

llow

ing

Alk

ire a

nd F

oste

r met

hodo

logy

with

k

= 3

3 pe

rcen

t and

Rip

pin

mea

sure

Indo

nesia

Con

trib

utio

ns in

%%

cha

nge

in

cont

ribut

ion

Dep

rivat

ions

% c

hang

e in

dep

rivat

ion

1997

2003

2007

1997

–200

320

03–0

719

9720

0320

0719

97–2

003

2003

–07

Alk

ire a

nd F

oste

r for

k =

33%

Nat

iona

lE

duca

tion

22.2

20.7

17.7

−6.

7−

14.7

9.8

7.0

4.6

−28

.7−

34.4

Mor

talit

y41

.146

.851

.513

.910

.018

.215

.813

.4−

12.9

−15

.4St

d of

livi

ng36

.732

.530

.9−

11.5

−5.

016

.311

.08.

0−

32.4

−26

.9U

rban

Edu

catio

n11

.814

.312

.120

.8−

15.2

2.4

3.2

2.0

32.8

−37

.5M

orta

lity

65.6

65.9

67.1

0.5

1.8

13.2

14.6

11.0

10.4

−25

.0St

d of

livi

ng22

.519

.820

.7−

12.3

4.9

4.5

4.4

3.4

−3.

6−

22.7

Rur

alE

duca

tion

23.8

23.6

19.7

−1.

0−

16.5

13.0

10.5

6.5

−19

.4−

37.4

Mor

talit

y37

.238

.245

.72.

719

.620

.316

.915

.2−

16.5

−10

.3St

d of

livi

ng38

.938

.134

.5−

2.0

−9.

521

.216

.911

.5−

20.2

−32

.1


195

Rip

pin

(g =

1.5

)N

atio

nal

Edu

catio

n23

.622

.820

.5−

3.4

−9.

90.

047

0.03

30.

022

−28

.9−

33.1

Mor

talit

y37

.139

.741

.87.

05.

30.

074

0.05

80.

045

−21

.3−

21.8

Std

of li

ving

39.3

37.5

37.6

−4.

50.

30.

078

0.05

50.

041

−29

.7−

25.5

Urb

anE

duca

tion

17.6

20.3

17.7

15.3

−12

.60.

013

0.01

70.

011

29.0

−38

.2M

orta

lity

52.1

52.4

53.0

0.6

1.1

0.03

90.

044

0.03

212

.5−

28.4

Std

of li

ving

30.3

27.3

29.3

−9.

97.

20.

023

0.02

30.

017

0.8

−24

.1R

ural

Edu

catio

n24

.323

.721

.4−

2.6

−9.

80.

061

0.04

80.

031

−21

.7−

35.2

Mor

talit

y35

.235

.038

.4−

0.7

9.9

0.08

80.

070

0.05

6−

20.2

−21

.0St

d of

livi

ng40

.441

.302

40.1

752.

2−

2.7

0.10

20.

083

0.05

8−

17.8

−30

.1


196

Tabl

e 6.

6c

Con

trib

utio

n of

eac

h di

men

sion

to o

vera

ll po

vert

y fo

llow

ing

Alk

ire a

nd F

oste

r met

hodo

logy

with

k

= 3

3 pe

rcen

t and

Rip

pin

mea

sure

Phili

ppin

esC

ontr

ibut

ions

in %

% c

hang

e in

co

ntrib

utio

nD

epriv

atio

ns%

cha

nge

in d

epriv

atio

n

1997

2003

2008

1997

–200

320

03–0

819

9720

0320

0819

97–2

003

2003

–08

Alk

ire a

nd F

oste

r for

k =

33%

Nat

iona

lE

duca

tion

20.3

18.9

14.7

−7.

1−

21.9

6.4

4.8

2.7

−25

.1−

44.1

Mor

talit

y49

.350

.158

.91.

617

.515

.612

.810

.8−

18.1

−15

.9St

d of

livi

ng30

.431

.026

.42.

1−

15. 0

9.6

7.9

4.8

−17

.7−

39.1

Urb

anE

duca

tion

12.8

11.2

8.7

−12

.4−

22.6

2.3

1.6

1.0

−30

.5−

34.4

Mor

talit

y69

.667

.973

.7−

2.5

8.5

12.4

9.6

8.8

−22

.6−

8.0

Std

of li

ving

17.6

20.9

17.7

18.8

−15

.63.

13.

02.

1−

5.8

−28

.5R

ural

Edu

catio

n23

.322

. 017

.7−

5.7

−19

.410

.68.

34.

4−

21.8

−47

.4M

orta

lity

41.4

43.0

51.7

3.9

20.2

18.9

16.2

12.7

−13

.9−

21.6

Std

of li

ving

35.4

35.1

30.6

−0.

9−

12.6

16.1

13.3

7.6

−17

.9−

43.0


197

Rip

pin

(g =

1.5

)N

atio

nal

Edu

catio

n23

.021

.818

.1−

5.2

−17

.20.

032

0.02

40.

013

−24

.1−

45.5

Mor

talit

y41

.641

.045

.9−

1.4

11.9

0.05

80.

046

0.03

4−

21.0

−26

.4St

d of

livi

ng35

.437

.236

.05.

0−

3.1

0.04

90.

041

0.02

6−

15.8

−36

.3U

rban

Edu

catio

n19

.617

.515

.1−

10.5

−13

.50.

013

0.00

90.

007

−27

.4−

28.2

Mor

talit

y54

.851

.855

.5−

5.6

7.2

0.03

60.

027

0.02

4−

23.3

−11

.1St

d of

livi

ng25

.630

.729

.320

.0−

4.5

0.01

70.

016

0.01

3−

2.6

−20

.7R

ural

Edu

catio

n24

.123

.219

.3−

3.6

−16

.80.

051

0.04

00.

020

−21

.2−

50.8

Mor

talit

y37

.537

.541

.8−

0.0

11.3

0.08

00.

065

0.04

3−

18.3

−34

.2St

d of

livi

ng38

.439

.338

.92.

3−

0.9

0.08

20.

068

0.04

0−

16.4

−41

.5


198

Tabl

e 6.

7 In

tens

ity o

f po

vert

y by

dim

ensi

on fo

llow

ing

Alk

ire a

nd F

oste

r app

roac

h fo

r k =

33%

Cam

bodi

a

% o

f M

PI p

oor-

depr

ived

% c

hang

e

Indo

nesia

% o

f M

PI p

oor-

depr

ived

% c

hang

e

2000

2005

2010

2000

–05

2005

–10

1997

2003

2007

1997

–200

320

03–0

7

Nat

iona

lN

atio

nal

Edu

catio

nM

orta

lity

Std

of li

ving

Urb

anE

duca

tion

Mor

talit

ySt

d of

livi

ngR

ural

Edu

catio

nM

orta

lity

Std

of li

ving

56.1

48.6

68.0

47.2

60.7

55.4

57.0

47.3

69.4

45.8

54.7

59.4

41.0

64.4

48.2

46.3

53.7

60.5

38.5

54.3

57.5

23.5

80.0

29.1

39.5

52.6

59.4

−18

.312

.4−

12.7

−13

.1 6.1

−13

.1

−18

.813

.5−

12.8

−16

.0−

0.6

−3.

2

−42

.624

.2−

39.6

−14

.6−

2.2

−1.

7

Edu

catio

nM

orta

lity

Std

of L

ivin

gU

rban

Edu

catio

nM

orta

lity

Std

of L

ivin

gR

ural

Edu

catio

nM

orta

lity

Std

of L

ivin

g

32.8

60.6

54.2

15.4

85.3

29.3

35.9

56.1

58.6

29.5

66.5

46.2

18.8

86.5

5925

.945

34.9

56.4

56.3

24.4

70.9

42.5

15.5

86.0

2526

.559

27.9

64.8

48.9

−10

.1 9.7

−14

.8

22.0 1.45

7−

11.4

12

−3.

00.

5−

4.0

−17

.4 6.6

−8.

0

−17

.2−

0.6

2.4

−19

.914

.8−

13.1

Phili

ppin

es19

9720

0320

0819

97–2

003

2003

–08

Nat

iona

lE

duca

tion

Mor

talit

ySt

d of

livi

ngU

rban

Edu

catio

nM

orta

lity

Std

of li

ving

Rur

alE

duca

tion

Mor

talit

ySt

d of

livi

ng

29.3

71.1

43.8

16.3

88.8

22.5

35.3

62.8

53.7

26.7

71.0

44.0

14.1

4885

.948

26.4

85

32.7

64. 0

52.2

19.5

77.9

34.9

10.8

91.7

22.0

24.2

70.5

41.8

−8.

7−

0.1

0.4

−13

.1−

3.3

17.8

−7.

552

1.87

7−

2.83

9

−27

.1 9.7

−20

.6

−23

.9 6.7

−17

.1

−26

.010

.3−

19.8



is mainly due a disproportional increase in the health contribution. In Indonesia, deprivations increase in every dimension, albeit slightly in the standard of living in urban areas over the period 1997–2003 according to the decomposition drawn from the Rippin measures. As for Cambodia, the results suggest that progress has been higher in the health dimension than in the standard of living in rural areas between 1997 and 2003. This explains why the percentage changes of contribution of these dimen-sions have opposite signs in comparison with those obtained from the Alkire and Foster approach. Moreover, although the results are not reported, Sulawesi registers an increase in health deprivation over the period 2003–07 which is only replicated by the Alkire and Foster decom-position of the intensity of poverty among the poor. Note that depriva-tions among the poor are also recorded in Sulawesi within the period 1997–2003, whereas the reverse is true according to the decomposition following the Rippin measures.

4 CONCLUSION

Using the general framework proposed by Silber and Yalonetzky (2013), this chapter compares poverty measures based on the approach of Alkire and Foster (2011) and used for the construction of the MPI by the UNDP with measures which are sensitive to the distribution of the distribution counts. Among such measures are those introduced by Chakravarty and D’Ambrosio (2006) and Rippin (2010) and those based on the extension of the approach of Aaberge and Peluso (2012) as suggested by Silber and Yalonetzky (2013).

Poverty was estimated using Demographic and Health Surveys in three Asian countries for three different years: for Cambodia in 2000, 2005 and 2010; for Indonesia in 1997, 2003 and 2007; and for the Philippines in 1997, 2003 and 2008.

Our findings indicate that Cambodia shows the highest level of poverty, followed by Indonesia and Philippines, irrespective of the poverty meas-ures used and the identification approach adopted for the Alkire and Foster measures. At the national level, all countries reduced their multi-dimensional poverty over time according to the poverty measures based on the Alkire and Foster approach and those that are sensitive to the concentration of deprivations across individuals. As in most Asian devel-oping countries, poverty is largely a rural phenomenon. However, when examining the evolution of poverty over time for each country, conclu-sions drawn from the use of the various poverty measures differ regarding



trends in poverty over time, by area of residence, as well as by region of residence.

Broadly speaking, our results highlight that trends in poverty (espe-cially in urban areas in the Philippines and Indonesia) could be highly dependent on the identification approach selected when adopting the Alkire and Foster approach. Note that poverty measures sensitive to inequality in deprivations provided results consistent with those obtained using a restrictive approach to the identification of the poor. Moreover, this study provided an illustration of some attractive features of the various measures, namely the dimensional decomposability of the Alkire and Foster and the Rippin measures and the decomposition into the mean and dispersion of the deprivation counts of the Chakravarty and D’Ambrosio measures as well as the extension of the Aaberge and Peluso measures.

The conclusions obtained for each country are especially interesting, as outlined below.

For Cambodia, multidimensional poverty declined over the two sub- periods, 2000–2005 and 2005–10, but at a faster rate over the first period whatever the poverty measures used. However, poverty reduction has been overly biased toward urban areas that registered an impressive accelera-tion of the rate of poverty decline between 2005 and 2010. Analysis by region of residence also made clear the growing tendency of an uneven spread of progress across the regions. Among the regions, it is the Phnom Penh region, initially the least deprived, that was the best performer. Although the Plains region (the least deprived after Phnom Penh in 2000) registered a significant decline of poverty between 2000 and 2005, a slowdown of poverty decrease was observed. By contrast, the Mountains region, where the population is worse off than in other regions, seems to be experiencing a catch- up process because the decrease of poverty over the period 2005–10 has been among the fastest. Interestingly, the decrease of poverty in Phnom Penh seems to have been largely driven by a decline in the concentration of deprivations. By contrast, the Mountains region registered an increase of inequality over the first period that may explain the low rate of decrease of poverty reduction compared with other regions. In addition, decomposition by dimensions emphasized that percentage changes over time in health deprivation have been lower than changes in education and the standard of living dimensions. Finally, these results showed that progress in poverty alleviation has been faster in the most prosperous parts of the country. In addition, keeping in mind that the growth rates of GDP over the two periods, 2000–2005 and 2005–10, were 56.25 percent and 38.24 percent, respectively, these findings may question the inclusiveness of the growth process.



As for Cambodia, poverty also declined in Indonesia at the national level over the two sub- periods, 1997–2003 and 2003–07, with an average annual decrease higher over the period 2003–07 than over 1997–2003. Note that during the period 1997–2003, the annual average GDP growth rate was only 1.2 percent owing to the impact of the Asian financial crisis, and the share of the urban population also increased. In particular, our results indicated a non- monotonic evolution of poverty by areas of residence over time. Unlike the ambiguous conclusions derived from the Alkire and Foster measures, regarding the trends in urban poverty, poverty sensi-tive to the dispersion of deprivations emphasized a marked increase of urban poverty between 1997 and 2003, followed by a significant decline at a rate very close to that of the rural one over 2003–07. In particular, decompositions of poverty measures by dimension showed that depriva-tions increased in every dimension, albeit slightly in the standard of living in urban areas over the period 1997–2003, according to the decomposi-tion drawn from the Rippin measures. In addition, the increase of urban poverty between 1997 and 2003 seems to have been largely driven by an increase of the inequality component.

Thus, whereas disparities between urban and rural areas weakened during 1997–2003, the trend is ambiguous between 2003 and 2007, even according to poverty measures sensitive to the dispersion of depriva-tion counts. In addition, with the exception of Java and Bali where the population was the better off and the worse off, respectively, poverty was very similar in the other regions in 1997. However, social progress has not favored equally every region over the period. The Java region exhibited the fastest decline in poverty rates and Kalimantan the slowest over the first sub- period. In contrast, over the second sub- period, a catching- up process was observed as Bali recorded the highest progress.

Being the least deprived in comparison with Cambodia and Indonesia, the Philippines registered the lowest rate of poverty decline over the first sub- period but exhibited a higher percentage rate of decline than Indonesia between 2003 and 2008. As for Indonesia, trends of poverty over time by areas of residence were dependent on the identification approach used. Nevertheless, the results revealed that when emphasis is put on the most deprived, disparities between urban and rural areas, albeit being relatively stable within the period 1997–2003, clearly exhibited a downward trend between 2003 and 2007. In addition, the reduction of poverty went hand- in- hand with a decrease in the concentration of dep-rivations. Note also that health dimension should be prioritized because its contribution remained disproportionally high and increased over time. These findings mitigate the conclusions drawn from those based on monetary measures that showed a lack of response of income poverty



incidence to growth during the 2000s (Habito 2009; Balicasan 2011). Poverty incidence at $1.25 had even increased between 2003 and 2006 when the economy grew.

NOTES

1. We note the contributions of Tsui (1995, 1999, 2002) on axiomatic derivations of multi-dimensional inequality and poverty indices.

2. In that case, it would minimize the additional welfare cost of deprivation in an addi-tional dimension. It would be easy to find examples to illustrate that the poverty meas-ures derived from a concave g function would correspond to interpersonal inequality preference (Rippin 2012).

3. A bistochastic matrix is a square matrix with the sum of each column and row equal to one. 4. PDP and MTP both require that the individual poverty function to be convex. 5. In particular, in cases where attributes are equally weighted (weight for each attribute is

equal to 1/m) the authors define the corresponding class of headcount measures:

PJRb 5 a

m

j51a j

mbb

Hj

where Hj is the proportion of individuals deprived in exactly j dimensions. This family satisfies the properties mentioned previously. In particular, range sensitivity, which is similar to the Pigou–Dalton transfer principle, is verified for all b > 1, whereas strong- range sensitivity is fulfilled for all b > 2. Jayaraj and Subramanian (2010) show that PJR

b is identical to PCD

a in the special case where each dimension receives an equal weight. 6. For more details, see Silber and Yalonetzky (2013). 7. For more details regarding the decomposition of the extension of the approach of

Aaberge and Peluso (2012) for any intermediate approach to the identification of the poor, see Bérenger (2015).

8. Because our main goal is to highlight empirically the contribution of methodological refinements of counting approaches to poverty measurement, the issue of sensitivity of poverty estimates to the choice of weighting schemes is not addressed here. Note that there is no consensus in the literature on weighting that should be used. For more details, see Decancq and Lugo (2013) who identify three types of methods to assign weights. Here, we adopt a normative approach assuming that each dimension is equally important in terms of well- being. The advantage is that weights remain constant and are thus relevant for comparisons over time and across countries.

9. Of course, the two measures are equal when adopting the intersection approach because poor individuals are then, by definition, systematically deprived with respect to all attributes.

10. For an easier presentation, the contributions of the eight indicators of poverty have been grouped into three dimensions: education, health and standard of living.



REFERENCES*55

Aaberge, R. and E. Peluso (2012), ‘A counting approach for measuring multi-dimensional deprivation’, Discussion Paper No. 700, Research Department, Statistics Norway.

Alkire, S. and J. Foster (2011), ‘Counting and multidimensional poverty measurement’, Journal of Public Economics, 95 (7–8), 476–87.

Alkire, S. and S. Seth (2014), ‘Analysis of inequality across multidimensionally poor and population subgroups for counting approaches’, Working Paper 68, Oxford Poverty & Human Development Initiative, University of Oxford.

Balicasan, A. (2011), ‘What has really happened to poverty in the Philippines? New measures, evidence and policy implications’, UP School of Economics Discussion Papers, Discussion Paper No. 2011- 14, University of the Philippines School of Economics, Quezon City.

Bérenger, V. (2015), ‘Using ordinal variables to measure multidimensional poverty in two south Mediterranean countries’, GREDEG Working Paper, Group for Research in Law, Economics and Management, University of Nice Sophia Antipolis, Valbonne.

Bossert, W., S. Chakravarty and C. D’Ambrosio (2013), ‘Multidimensional poverty and material deprivation with discrete data’, Review of Income and Wealth, 59 (1), 29–43.

Booysen, F., S. Van Der Berg, R. Burger, M. Von Maltitz and G. Durand (2008), ‘Using an asset index to assess trends in poverty in seven sub- Saharan African countries’, World Development, 36 (6), 1113–30.

Bourguignon, F. and S. Chakravarty (2003), ‘The Measurement of Multidimensional Poverty’, Journal of Economic Inequality, 1 (1), 25–49.

Casimiro, G.G., R.E. Ballester and M.N. Garingalao (2013), ‘A multidimensional approach to child poverty in the Philippines’, paper presented at the Twelfth National Convention on Statistics, EDSA Shangri- La Hotel, Mandaluyong City, 1–2 October.

Chakravarty, S. and C. D’Ambrosio (2006), ‘The measurement of social exclusion’, Review of Income and Wealth, 52 (3), 377–98.

Chakravarty, S. and J. Silber (2008), ‘Measuring multidimensional poverty: the axiomatic approach’, in N. Kakwani and J. Silber (eds), Quantitative Approaches to Multidimensional Poverty Measurement, New York: Palgrave Macmillan, pp. 192–209.

Chakravarty, S., D. Mukherjee and R. Ranade (1998), ‘On the Family of Subgroup and Factor Decomposable Measures of Multidimensional Poverty’, in D. Slottje (ed.), Research on Economic Inequality, vol. 8, Stanford, CT and London: JAI Press, pp. 175–94.

Datt, G. (2013), ‘Making every dimension count: multidimensional without the dual cutoff’, Discussion Paper 32/13, Monash University, Melbourne.

Decancq, K. and M.A. Lugo (2013), ‘Weights in multidimensional indices of well- being: an overview’, Econometric Reviews, 32(1), 7–34.

Habito, C. (2009), ‘Patterns of inclusive growth in developing Asia: insights from

* The Asian Development Bank recognizes China as the People’s Republic of China and Vietnam as Viet Nam.



enhanced growth poverty elasticity analysis’, ADBI Working Paper Series, No. 145, Tokyo.

Jayaraj, D. and S. Subramanian (2010), ‘A Chakravarty- D’Ambrosio view of mul-tidimensional deprivations: some estimates for India’, Economic and Political Weekly, 45 (6), 53–65.

Klasen, S. (2000), ‘Measuring poverty and deprivation in South Africa’, Review of Income and Wealth, 46 (1), 33–58.

Kolm, S.C. (1977), ‘Multidimensional egalitarianisms’, Quarterly Journal of Economics, 91 (1), 1–13.

Lovell, C.A.K, S. Richardson, P. Travers and L. Wood (1994), ‘Resources and func-tionings: a new view of inequality in Australia’, in W. Eichhorn (ed.), Models and Measurement of Welfare and Inequality, Heidelberg: Springer- Verlag.

Maasoumi, E. (1986), ‘The measurement and decomposition of multidimensional inequality’, Econometrica, 54 (4), 991–7.

Maasoumi, E. (1999), ‘Multidimensioned approaches to welfare analysis’, in J. Silber (ed.), Handbook of Income Inequality Measurement, Boston, MA: Kluwer Academic.

Ray, R. and S. Kompal (2011), ‘Multidimensional deprivation in China, India and Vietnam: a comparative study on micro data’, Discussion Paper 06/11, Monash University, Melbourne.

Rippin, N. (2010), ‘Poverty severity in a multidimensional framework: the issue of inequality between dimensions’, Courant Research Center, Discussion Paper No. 47, University of Göttingen.

Rippin, N. (2012), ‘Integrating inter- and intra- personal inequality in additive poverty indices’, paper presented at the Thirty- second International Association for Research in Income and Wealth (IARIW) Conference, Boston, MA.

Sahn, D. and D. Stifel (2000), ‘Poverty comparisons over time and across countries in Africa’, World Development, 28 (12), 2123–55.

Sen, A.K. (1974), ‘Informational bases of alternative welfare approaches: aggregation and income distribution’, Journal of Public Economics, 3 (4), 387–403.

Sen, A.K. (1976), ‘Poverty: an ordinal approach to measurement’, Econometrica, 44 (2), 219–32.

Sen, A.K. (1985), Commodities and Capabilities, Amsterdam: North Holland.Silber, J. and G. Yalonetzky (2013), ‘Measuring Multidimensional Deprivation

with Dichotomized and Ordinal Variables’, in G. Betti and A. Lemmi (eds), Poverty and Social Exclusion: New Methods of Analysis, Routledge Frontiers of Political Economy, London and New York: Routledge.

Sumner, A., A. Suryahadi and N. Thang (2012), ‘Poverty and inequalities in middle- income Southeast Asia’, Working Paper, Institute of Development Studies, University of Sussex, Brighton.

Thorbecke, E. (2007), ‘Multidimensional poverty: conceptual and measurement issues’, in N. Kakwani and J. Silber (eds), The Many Dimensions of Poverty, New York: Palgrave Macmillan.

Townsend, P. (1979), Poverty in the United Kingdom, Hardsmonsworth: Penguin Books.

Tsui, K.- Y. (1995), ‘Multidimensional generalizations of the relative and absolute inequality indices: the Atkinson- Kolm- Sen Approach’, Journal of Economic Theory, 67 (1), 251–65.

Tsui, K.- Y. (1999), ‘Multidimensional inequality and multidimensional generalized



entropy measures: an axiomatic derivation’, Social Choice and Welfare, 16 (1), 145–57.

Tsui, K.- Y. (2002), ‘Multidimensional poverty indices’, Social Choice and Welfare, 19 (1), 69–93.

United Nations Children’s Fund (UNICEF) (2011), Child Poverty in East Asia and the Pacific: Deprivation and Disparities. A Study on Seven Countries, Bangkok: UNICEF East Asia and Pacific.

United Nations Development Programme (UNDP) (2010), Human Development Report 2010. The Real Wealth of Nations: Pathways to Human Development, New York: Palgrave Macmillan.

United Nations Development Programme (UNDP) (2013), Human Development Report 2013: The Rise of the South: Analysis on Cambodia, Phnom Penh: United Nations Development Programme.

Yaari, M.E. (1987), ‘The dual theory of choice under risk’, Econometrica, 55 (1), 95–115.

Yaari, M.E. (1988), ‘A controversial proposal concerning inequality measurement’, Journal of Economic Theory, 44 (2), 381–97.


206

APP

EN

DIX

6A

.1

Tabl

e 6A

.1

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: C

ambo

dia

by re

gion

Cam

bodi

aH

eadc

ount

M0

A in

%

2000

2005

2010

2000

2005

2010

2000

2005

2010

k =

uni

onPh

nom

Pen

h0.

711

0.43

60.

296

0.18

60.

121

0.04

926

.227

.816

.6Pl

ains

0.99

30.

962

0.91

10.

440

0.30

60.

256

44.3

31.9

28.1

Tonl

e Sa

p0.

988

0.95

70.

916

0.47

10.

364

0.27

847

.738

.030

.3C

oast

al0.

986

0.94

00.

904

0.44

90.

348

0.26

645

.637

.029

.4M

ount

ains

0.99

40.

981

0.93

70.

499

0.42

40.

317

50.2

43.3

33.8

k =

33%

Phno

m P

enh

0.25

00.

199

0.06

40.

119

0.09

30.

025

47.8

46.6

39.1

Plai

ns0.

653

0.40

90.

325

0.37

00.

208

0.16

056

.650

.949

.2To

nle

Sap

0.70

00.

519

0.36

90.

414

0.28

30.

186

59.1

54.6

50.4

Coa

stal

0.65

40.

492

0.34

50.

383

0.26

20.

171

58.6

53.2

49.5

Mou

ntai

ns0.

765

0.61

70.

430

0.45

30.

355

0.22

859

.157

.653

.0

k =

50%

Phno

m P

enh

0.09

20.

068

0.00

80.

057

0.04

30.

005

61.4

63.9

64.9

Plai

ns0.

393

0.18

20.

124

0.26

40.

116

0.07

867

.163

.862

.6To

nle

Sap

0.45

00.

278

0.15

40.

310

0.18

50.

099

69.0

66.7

64.5

Coa

stal

0.42

60.

254

0.14

10.

290

0.16

60.

089

68.0

65.4

63.0

Mou

ntai

ns0.

484

0.37

40.

213

0.33

70.

257

0.14

169

.768

.766

.1


207

Tabl

e 6A

.2

Perc

enta

ge c

hang

es in

mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h:

Cam

bodi

a by

regi

on

Cam

bodi

aVa

riatio

n in

% o

f H

Ann

ual r

ate

of c

hang

e of

HVa

riatio

n in

% o

f M

0A

nnua

l rat

e of

cha

nge

of M

0

2000

–05

2005

–10

2000

–05

2005

–10

2000

–05

2005

–10

2000

–05

2005

–10

k =

uni

onPh

nom

Pen

h−

38.7

−32

.1−

9.3

−7.

5−

34.7

−59

.5−

8.2

−16

.5Pl

ains

−3.

2−

5.3

−0.

6−

1.1

−30

.3−

16.5

−7.

0−

3.5

Tonl

e Sa

p−

3.1

−4.

3−

0.6

−0.

9−

22.8

−23

.7−

5.0

−5.

3C

oast

al−

4.7

−3.

9−

1.0

−0.

8−

22.7

−23

.5−

5.0

−5.

2M

ount

ains

−1.

3−

4.5

−0.

3−

0.9

−14

.9−

25.4

−3.

2−

5.7

k =

33%

Phno

m P

enh

−20

.3−

67.8

−4.

4−

20.3

−22

.3−

73.0

−4.

9−

23.0

Plai

ns−

37.4

−20

.5−

8.9

−4.

5−

43.7

−23

.2−

10.8

−5.

1To

nle

Sap

−25

.9−

28.9

−5.

8−

6.6

−31

.5−

34.3

−7.

3−

8.1

Coa

stal

−24

.8−

29.9

−5.

5−

6.9

−31

.6−

34.9

−7.

3−

8.2

Mou

ntai

ns−

19.4

−30

.3−

4.2

−7.

0−

21.5

−35

.8−

4.7

−8.

5

k =

50%

Phno

m P

enh

−26

.4−

88.8

−6.

0−

35.4

−23

.4−

88.6

−5.

2−

35.2

Plai

ns−

53.7

−31

.7−

14.3

−7.

3−

56.0

−32

.9−

15.2

−7.

7To

nle

Sap

−38

.2−

44.6

−9.

2−

11.2

−40

.3−

46.5

−9.

8−

11.8

Coa

stal

−40

.3−

44.6

−9.

8−

11.1

−42

.7−

46.6

−10

.5−

11.8

Mou

ntai

ns−

22.7

−42

.9−

5.0

−10

.6−

23.9

−45

.1−

5.3

−11

.3


208

Tabl

e 6A

.3

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: In

done

sia

by re

gion

Indo

nesia

Hea

dcou

ntM

0A

in %

1997

2003

2007

1997

2003

2007

1997

2003

2007

k =

uni

onSu

mat

era

0.82

20.

780

0.76

00.

237

0.19

90.

166

28.8

25.4

21.9

Java

0.82

50.

764

0.73

70.

216

0.17

50.

146

26.1

23.0

19.8

Bal

i0.

872

0.84

20.

813

0.30

30.

270

0.20

834

.832

.125

.6K

alim

anta

n0.

975

0.82

80.

797

0.25

50.

216

0.17

726

.126

.122

.2Su

law

esi

0.85

70.

751

0.79

20.

245

0.20

00.

197

28.6

26.6

24.8

k =

33%

Sum

ater

a0.

313

0.24

90.

201

0.15

70.

121

0.09

350

.148

.546

.3Ja

va0.

277

0.21

80.

167

0.13

30.

100

0.07

548

.245

.844

.8B

ali

0.45

30.

383

0.26

80.

234

0.19

30.

128

51.7

50.5

47.9

Kal

iman

tan

0.31

70.

261

0.19

70.

159

0.13

20.

094

50.3

50.6

47.7

Sula

wes

i0.

316

0.25

20.

248

0.15

70.

125

0.11

849

.649

.647

.6

k =

50%

Sum

ater

a0.

135

0.09

70.

065

0.08

50.

060

0.03

963

.162

.060

.3Ja

va0.

103

0.06

20.

042

0.06

40.

038

0.02

562

.461

.560

.8B

ali

0.22

10.

177

0.10

20.

143

0.11

30.

062

64.5

63.7

61.4

Kal

iman

tan

0.12

20.

105

0.06

40.

078

0.06

80.

040

63.4

65.0

63.0

Sula

wes

i0.

131

0.11

30.

094

0.08

30.

070

0.05

763

.362

.261

.0


209

Tabl

e 6A

.4

Perc

enta

ge c

hang

es in

mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h:

Indo

nesi

a by

regi

on

Indo

nesia

Varia

tion

in %

of

HA

nnua

l rat

e of

cha

nge

of H

Varia

tion

in %

of

M0

Ann

ual r

ate

of c

hang

e of

M0

1997

–200

320

03–0

719

97–2

003

2003

–07

1997

–200

320

03–0

719

97–2

003

2003

–07

k =

uni

onSu

mat

era

−5.

1−

2.7

−0.

9−

0.7

−16

.1−

16.2

−2.

9−

4.3

Java

−7.

4−

3.6

−1.

3−

0.9

−18

.7−

16.9

−3.

4−

4.5

Bal

i−

3.4

−3.

4−

0.6

−0.

9−

11.0

−23

.0−

1.9

−6.

3K

alim

anta

n−

15.1

−3.

8−

2.7

−1.

0−

15.2

−18

.1−

2.7

−4.

9Su

law

esi

−12

.35.

5−

2.2

1.3

−18

.6−

1.4

−3.

4−

0.3

k =

33%

Sum

ater

a−

20.5

−19

.4−

3.6

−5.

2−

23.1

−23

.0−

4.3

−6.

3Ja

va−

21.3

−23

.4−

3.9

−6.

4−

25.4

−25

. 0−

4.8

−6.

9B

ali

−15

.6−

30.0

−2.

8−

8.5

−17

.5−

33.7

−3.

2−

9.8

Kal

iman

tan

−17

.5−

24.4

−3.

2−

6.8

−16

.9−

28.8

−3.

0−

8.1

Sula

wes

i−

20.1

−1.

7−

3.7

−0.

4−

20.1

−5.

6−

3.7

−1.

4

k =

50%

Sum

ater

a−

28.1

−32

.5−

5.3

−9.

4−

29.3

−34

.3−

5.6

−10

.0Ja

va−

40.2

−32

.2−

8.2

−9.

3−

41.1

−33

.0−

8.4

−9.

5B

ali

−19

.9−

42.7

−3.

6−

13.0

−20

.9−

44.7

−3.

8−

13.8

Kal

iman

tan

−13

.9−

39.2

−2.

5−

11.7

−11

.7−

41.1

−2.

0−

12.4

Sula

wes

i−

13.9

−16

.8−

2.5

−4.

5−

15.4

−18

.3−

2.7

−4.

9


210

Tabl

e 6A

.5

Mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: th

e P

hilip

pine

s by

regi

on

Phili

ppin

esH

eadc

ount

M0

A in

%

1997

2003

2008

1997

2003

2008

1997

2003

2008

k =

uni

onL

uzon

0.53

50.

543

0.60

30.

126

0.11

00.

097

23.5

20.2

16.0

Visa

yas

0.73

40.

699

0.72

80.

207

0.18

70.

145

28.2

26.8

19.9

Min

dana

o0.

763

0.72

20.

699

0.22

90.

193

0.15

930

.126

.722

.7

k =

33%

Luz

on0.

167

0.13

40.

108

0.07

50.

060

0.04

544

.844

.542

.0V

isaya

s0.

264

0.24

40.

159

0.13

10.

119

0.07

449

.948

.746

.3M

inda

nao

0.31

10.

237

0.19

50.

159

0.11

80.

089

51.0

49.7

45.5

k =

50%

Luz

on0.

053

0.03

60.

022

0.03

20.

022

0.01

360

.762

.060

.5V

isaya

s0.

117

0.09

80.

055

0.07

50.

062

0.03

363

.662

.760

.9M

inda

nao

0.15

20.

098

0.05

50.

096

0.06

20.

034

63.1

63.6

61.3


211

Tabl

e 6A

.6

Perc

enta

ge c

hang

e in

mul

tidim

ensi

onal

pov

erty

mea

sure

s fol

low

ing

the

Alk

ire a

nd F

oste

r app

roac

h: th

e P

hilip

pine

s by

regi

on

Phili

ppin

esVa

riatio

n in

% o

f H

Ann

ual r

ate

of c

hang

e of

HVa

riatio

n in

% o

f M

0A

nnua

l rat

e of

cha

nge

of M

0

1997

–200

320

03–0

819

97–2

003

2003

–08

1997

–200

320

03–0

819

97–2

003

2003

–08

k =

uni

onL

uzon

1.5

11.0

0.2

2.1

−12

.5−

12.0

−2.

2−

2.5

Visa

yas

−4.

74.

2−

0.8

0.8

−9.

4−

22.8

−1.

6−

5.0

Min

dana

o−

5.4

−3.

2−

0.9

−0.

6−

15.9

−17

.8−

2.8

−3.

8

k =

33%

Luz

on−

19.9

−19

.4−

3.6

−4.

2−

20.5

−23

.9−

3.7

−5.

3V

isaya

s−

7.3

−34

.8−

1.3

−8.

2−

9.5

−38

.1−

1.7

−9.

1M

inda

nao

−23

.9−

17.4

−4.

4−

3.8

−25

.9−

24.3

−4.

9−

5.4

k =

50%

Luz

on−

31.9

−40

.1−

6.2

−9.

8−

30.4

−41

.6−

5.9

−10

.2V

isaya

s−

16.3

−44

.2−

2.9

−11

.0−

17.4

−45

.8−

3.1

−11

.5M

inda

nao

−35

.5−

44.1

−7.

1−

11.0

−35

.0−

46.1

−6.

9−

11.6


212

Tabl

e 6A

.7a

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: Cam

bodi

a by

regi

on

Cam

bodi

aM

easu

res

Varia

tion

in %

Ann

ual r

ate

of c

hang

e

2000

2005

2010

2000

–05

2005

–10

2000

–05

2005

–10

Rip

pin

g =

1.5

Phno

m P

enh

0.05

10.

037

0.00

8−

27.0

−77

.3−

6.1

−25

.7Pl

ains

0.19

10.

097

0.07

2−

49.5

−26

.0−

12.8

−5.

9To

nle

Sap

0.22

60.

142

0.08

6−

36.9

−39

.7−

8.8

−9.

6C

oast

al0.

208

0.12

80.

077

−38

.6−

39.5

−9.

3−

9.5

Mou

ntai

ns0.

249

0.19

10.

112

−23

.6−

41.2

−5.

2−

10.1

Rip

pin

g =

2Ph

nom

Pen

h0.

037

0.02

80.

005

−24

.7−

80.2

−5.

5−

27.7

Plai

ns0.

155

0.07

20.

052

−53

.3−

28.4

−14

.1−

6.5

Tonl

e Sa

p0.

188

0.11

30.

064

−39

.8−

43.3

−9.

7−

10.7

Coa

stal

0.17

20.

099

0.05

7−

42.1

−42

.9−

10.4

−10

.6M

ount

ains

0.21

00.

157

0.08

7−

25.2

−44

.4−

5.7

−11

.1

Cha

krav

arty

and

D’A

mbr

osio

for a

= 2

Phno

m P

enh

0.07

40.

052

0.01

4−

29.4

−73

.4−

6.7

−23

.3Pl

ains

0.24

30.

134

0.10

3−

44.7

−23

.4−

11.2

−5.

2To

nle

Sap

0.27

90.

186

0.12

0−

33.4

−35

.5−

7.8

−8.

4C

oast

al0.

259

0.17

00.

110

−34

.4−

35.3

−8.

1−

8.3

Mou

ntai

ns0.

304

0.23

90.

150

−21

.5−

37.2

−4.

7−

8.9

Ext

ensio

n of

Aab

erge

and

Pel

uso

Phno

m P

enh

0.29

10.

211

0.09

0−

27.5

−57

.5−

6.2

−15

.7Pl

ains

0.56

60.

418

0.36

3−

26.2

−13

.2−

5.9

−2.

8To

nle

Sap

0.60

70.

494

0.39

2−

18.6

−20

.6−

4.0

−4.

5C

oast

al0.

585

0.47

20.

375

−19

.3−

20.6

−4.

2−

4.5

Mou

ntai

ns0.

632

0.56

20.

440

−11

.1−

21.6

−2.

3−

4.8


213

Tabl

e 6A

.7b

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: Ind

ones

ia b

y re

gion

Indo

nesia

Mea

sure

sVa

riatio

n in

%A

nnua

l rat

e of

cha

nge

1997

2003

2007

1997

–200

320

03–0

719

97–2

003

2003

–07

Rip

pin

g =

1.5

Sum

ater

a0.

071

0.05

30.

039

−25

.5−

27.2

−4.

8−

7.6

Java

0.05

90.

041

0.03

0−

29.9

−26

.3−

5.7

−7.

3B

ali

0.10

90.

089

0.05

5−

18.3

−37

.8−

3.3

−11

.2K

alim

anta

n0.

072

0.06

20.

042

−13

.6−

32.6

−2.

4−

9.4

Sula

wes

i0.

072

0.05

70.

051

−20

.6−

10.3

−3.

8−

2.7

Rip

pin

g =

2Su

mat

era

0.05

30.

038

0.02

7−

27.7

−29

.7−

5.3

−8.

4Ja

va0.

042

0.02

90.

021

−32

.8−

28.0

−6.

4−

7.9

Bal

i0.

084

0.06

80.

040

−19

.8−

41.1

−3.

6−

12.4

Kal

iman

tan

0.05

30.

046

0.02

9−

12.1

−36

.4−

2.1

−10

.7Su

law

esi

0.05

30.

042

0.03

6−

20.8

−13

.0−

3.8

−3.

4

Cha

krav

arty

and

D’A

mbr

osio

for g

= 2

Sum

ater

a0.

100

0.07

70.

058

−23

.0−

24.2

−4.

3−

6.7

Java

0.08

50.

062

0.04

7−

26.7

−24

.2−

5.0

−6.

7B

ali

0.14

60.

122

0.08

1−

16.4

−33

.9−

2.9

−9.

8K

alim

anta

n0.

103

0.08

80.

063

−14

.6−

28.4

−2.

6−

8.0

Sula

wes

i0.

102

0.08

10.

075

−20

.3−

7.5

−3.

7−

1.9

Ext

ensio

n of

Aab

erge

and

Pel

uso

Sum

ater

a0.

352

0.30

30.

258

−14

.0−

14.8

−2.

5−

3.9

Java

0.32

20.

270

0.22

8−

16.2

−15

.4−

2.9

−4.

1B

ali

0.43

50.

394

0.31

3−

9.4

−20

.6−

1.6

−5.

6K

alim

anta

n0.

360

0.32

40.

270

−9.

9−

16.7

−1.

7−

4.5

Sula

wes

i0.

357

0.30

80.

299

−13

.7−

2.9

−2.

4−

0.7


214

Tabl

e 6A

.7c

Pove

rty

mea

sure

s sen

sitiv

e to

the

disp

ersi

on o

f de

priv

atio

ns a

cros

s ind

ivid

uals

: the

Phi

lippi

nes b

y re

gion

Phili

ppin

esM

easu

res

Varia

tion

in %

Ann

ual r

ate

of c

hang

e

1997

2003

2008

1997

–200

320

03–0

819

97–2

003

2003

–08

Rip

pin

g =

1.5

Luz

on0.

030

0.02

40.

017

−19

.5−

29.9

−3.

5−

6.9

Visa

yas

0.06

10.

053

0.03

2−

13.8

−39

.6−

2.4

−9.

6M

inda

nao

0.07

30.

055

0.03

7−

25.4

−33

.2−

4.8

−7.

7

Rip

pin

g =

2L

uzon

0.02

10.

017

0.01

1−

19.9

−33

.7−

3.6

−7.

9V

isaya

s0.

045

0.03

80.

022

−15

.2−

42.4

−2.

7−

10.4

Min

dana

o0.

055

0.04

00.

025

−27

.1−

37.4

−5.

1−

8.9

Cha

krav

arty

and

D’A

mbr

osio

for a

= 2

Luz

on0.

045

0.03

70.

028

−18

.3−

25.5

−3.

3−

5.7

Visa

yas

0.08

60.

075

0.04

8−

12.3

−35

.8−

2.2

−8.

5M

inda

nao

0.10

10.

078

0.05

6−

23.1

−28

.6−

4.3

−6.

5

Ext

ensio

n of

Aab

erge

and

Pel

uso

Luz

on0.

210

0.18

50.

160

−12

.0−

13.8

−2.

1−

2.9

Visa

yas

0.31

90.

294

0.22

8−

7.7

−22

.4−

1.3

−5.

0M

inda

nao

0.35

00.

300

0.25

0−

14.2

−16

.7−

2.5

−3.

6


215

7. Poverty and nutrition: a case study of rural households in Thailand and Viet NamHermann Waibel and Lena Hohfeld

1 INTRODUCTION

Asian countries have made significant progress in poverty reduction in the past 20 years. This has been largely due to economic growth and direct measures for poverty reduction. The optimistic view is that poverty in Asia may soon come to an end. There are at least two reasons to be more careful in this prediction. First, the headcount ratio as a static poverty measure does not allow any conclusion about the risk of people falling back into poverty, that is, their vulnerability to poverty (Klasen and Waibel 2013). In the past, economic, ecological and political shocks have been responsible for many people falling back into poverty. Examples are the financial, economic and food- price crisis that hit Asian countries in 2008. Second, it is perhaps much too early to declare victory on the poverty front in Asia because monetary poverty is only one of several dimensions of poverty. Education, health and nutrition, for example, are other poverty dimen-sions that need to be taken into account (Sen 2000; Tsui 2002; Carter and Barrett 2006; Clark and Hulme 2010). Several studies have demonstrated that the correlation between monetary and nonmonetary poverty is low (Baulch and Masset 2003; Mckay and Lawson 2003; Günther and Klasen 2009).

Clearly one of these dimensions is nutrition. The global food- price crisis reminded the development community that food security remains a global concern. The number of undernourished people in the world has passed beyond 1 billion, the majority of whom are in Asia. In this chapter, we analyze the link between nutrition and poverty in two Asian countries where monetary- based poverty reduction was especially suc-cessful, namely, Thailand and Viet Nam, two emerging market economies where poverty rates are now below 10 percent and are declining further. Despite this success, it is not clear to what extent this success has translated



into similar improvements in the nutritional situation of the people, espe-cially of children. The analysis in this chapter is concentrated on the rural population in these two countries. We have panel data on all aspects of household livelihoods, including food consumption, and we have a set of anthropometric data for all household members, including mothers and their children.

Specifically, we address the following questions:

1. Is there still a nutrition problem in Thailand and Viet Nam in spite of the progress made in poverty reduction?

2. What are the factors that condition the nutritional status of children and adults in rural areas of these two countries?

3. What are the factors that influence nutrition outcomes as households depart from the monetary poverty line?

4. What is the time horizon to reach the end of malnutrition under diff er-ent income growth scenarios?

The chapter proceeds as follows. In section 2, the conceptual framework is introduced which outlines the measures and the econometric model used in this chapter. In section 3, a description of the data is presented. In section 4, the empirical results and a prediction of nutrition outcomes are shown. Finally, section 5 summarizes and concludes.

2 CONCEPTUAL FRAMEWORK

In this section, we establish the conceptual basis for this study. We intro-duce three aspects necessary to analyze the relationship between nutrition and poverty. First, we define the most common measures of nutrition to identify the nutrition outcome variables. Second, we discuss the direction of influence between wealth and nutrition by reviewing relevant literature. Third, we identify the main variables that have been used in models that aim to explain the change in the nutritional status of people in developing countries.

The nutritional status of a population is often measured using anthro-pometric indicators, mostly for children below the age of five. For example, in the Millennium Development Goals underweight children is one of the indicators for hunger. Stunting and wasting of children are indicators for the World Health Organization’s (WHO) Global Targets 2025. Also, the largest share of scientific publications on malnutrition concentrates on children below 5 years of age. There are several reasons for this choice of indicators. First, for children, even short periods of undernutrition can


Poverty and nutrition: Thailand and Viet Nam 217

cause long- lasting and irreversible damage. Child malnutrition can lead to low cognitive outcomes and, therefore, to lower productivity even when they become adults. Second, children’s bodies react faster to changes in the food supply, and food shortages manifest faster in their weight and height than in those of adults. Therefore, the nutritional status of children aged below 5 years of age is a good proxy for the current nutritional situ-ation of a population. However, only a share of households has children below 5 years of age. For a complete picture of the nutritional status of a population, indicators for adults should be included, even if adults are less vulnerable to short- term food shortages.

The most commonly used anthropometric measures to describe the nutritional status of a population are weight and height. For children, the parameters are related to age (see (a)–(d)), whereas, for adults, the Body Mass Index (e) is the only measure of those listed below:

(a) weight- for- age (WFA);(b) height- for- age (HFA);(c) weight- for- height (WFH);(d) Body Mass Index (BMI) for age (for children); and(e) Body Mass Index (BMI) for adults.

Weight- for- age is an indicator of underweight; HFA is an expression of stunting; WFH is called wasting; and the BMI is a measure for under-weight, generally used for adults, but also for children. All five indicators are used as proxies of undernutrition relative to defined threshold values. The most commonly- used measure is weight- for- age (WHO Working Group 1986), which is used, for example, in the Millennium Development Goals, because it reveals both, acute and chronic, malnutrition (de Onis and Blössner 2003). Stunted growth, which means low height relative to age (HFA), is an indicator for chronic malnutrition and early childhood illnesses. Weight- for- height is regarded as an indicator for acute under-nutrition because weight can drop rapidly in cases of acute food short-ages, whereas height is unaffected by short- time changes in food supply. For adults, BMI is the most widely used indicator, measuring the current nutritional status; for children, reference standards and cutoffs for BMI- for- age have only recently been developed and are not yet that widely used (Cole et al. 2007).

Statistically, child undernutrition is measured using growth data in com-parison with an international healthy reference population of the same age (height), based on WHO standards (de Onis et al. 2009). Malnutrition of populations using the indicators (a) to (d) is usually done by calculating Z- scores, defined as: ‘observed value minus the median value of a reference



population divided by the standard deviation of that reference population’ (WHO 2015). For indicators (a) to (d) a Z- score of −2 is used; that is, for WFA, if children are more than two standard deviations below the median (or mean) of their reference group they would be called underweight. The BMI is calculated as weight in kilograms divided by the square of height, measured in meters. For children, Z- scores for BMI for age are used, whereas for adults a fixed BMI cutoff value of below 18.5 is considered to be underweight.

The second issue that must be dealt with when analyzing undernu-trition problems in developing countries is how to integrate nutrition into economic models. The theoretical foundation to establish causality between nutrition outcomes and the physical and socioeconomic condi-tions of a target population in developing countries is household theory (Becker 1965; Strauss and Thomas 1995). Aside from income, health and nutrition can be considered as components of a household’s utility func-tion, given its production choices and resource constraints. However, as pointed out by Alderman (2012), the explanatory power of income- based indicators is poor and, referring to Almond and Currie (2011), it is increas-ingly recognized that the health and nutritional status of children is not only subject to postnatal but to prenatal conditions as well. This suggests that information about the mother’s health prior to childbirth is important to assess the nutritional status of children. Modelling nutrition outcomes (N), therefore, can be formulated as a function of household income, household and village characteristics, and the child’s and the mother’s characteristics. Following Kabubo- Mariara et al. (2009), we specify a model for the nutritional status of children below the age of five as follows:

Nit = f(Yjt, Cit, Mit, Xjt, Zkt, eit) (7.1)

where Nit is the nutritional outcome of child i at time t; Y is income of household j; C includes child, mother and adult characteristics of person i; X describes household characteristics; and Z is a vector of characteristics of village k; all variables being measured at time t; and e is a randomly dis-tributed error term. In our models, we use Z- scores of nutrition outcomes, that is, WFA as dependent variable.

The choice of explanatory variables follows the general framework devel-oped by UNICEF (Menon 2012). The framework distinguishes between immediate, underlying and basic causes of undernutrition, whereby imme-diate causes are (1) lack of food and nutrition intake; and (2) poor health status. Underlying factors are the sanitary conditions of a household and the provision of basic health services. Food and nutrition intakes are subject to food access which is determined by the market infrastructure



and the general state of agricultural development. As a basic cause of undernutrition, maternal and childcare practices are hypothesized to influ-ence the health and nutrition of children. In the following, we describe the choice of variables in detail.

Most of the literature suggests that higher income and reduction in poverty have positive effects on nutrition and health (for example, Anand and Ravallion 1993; Strauss and Thomas 1995) but this relationship can vary across countries and within households (Haddad, et al. 2003). This difference can be attributed to inequality and the extent to which public goods are directed towards nutrition (Anand and Ravallion 1993).

Further, we include for child characteristics, the three variables, age, gender and a dummy variable, to reflect whether the child was sick in the reference period. Because the risk of malnutrition has been shown to differ with the age of children (Alderman et al. 2006; Menon 2012), we include age dummies. A slower growth of girls/boys might occur if intra- household allocation discriminates for gender (Belitz et al. 2010). The nutritional status of a child will suffer in times of illness, but with good health care, effects will be less strong (Menon 2012). For mother’s characteristics, her height is generally believed to predetermine the child’s nutritional status, which underlines intergenerational transmission of undernutrition through genes and economic status (Belitz et al. 2010). Mother’s education (Smith et al. 2003) is used as a proxy for childcare practices and mothers who migrated might have more childcare knowledge, in addition to the remittances that may benefit a child’s nutritional status. Adult character-istics include similar variables: gender, education, age and a dummy vari-able for sickness. For household characteristics, we include household size and dependency ratio, which may influence the resource situation of the household and the degree of childcare (Belitz et al. 2010). Migration of other household members, measured in months absent per year, is included as a proxy for the amount of remittances sent to the rural household. To measure the influence of sanitation facilities in the household, we include dummy variables on having running water and whether the household has a private water toilet. For village characteristics, health infrastructure is included, proxied by the percentage of households with sanitation, and the availability of public water (Haddad et al. 2003). We control for the relative wealth of the village by including the average income of the village. In Viet Nam, we also include a dummy variable for ethnic minorities and control for different agro- ecological zones, that is, whether the household is located in a mountainous region.

Most studies on child undernutrition use Demographic and Health Survey data (for example, Kabubo- Mariara et al. 2009), which are rich in terms of health information on child and mother, but do not always



provide income or consumption data. In our panel data set, which is described in more detail in the next section, we have available direct meas-ures, therefore we include (log) income per capita as Y. We take WFA as the nutrition- outcome indicator (N) as a continuous variable in Z- scores for children below 5 years of age. First, we estimate the model of equa-tion (7.1) using ordinary least- squares (OLS) regression over the entire sample. In accordance with our objective to explore the relationship between poverty reduction and nutritional status of the rural population in Viet Nam and Thailand, we establish four groups, namely: (1) children who live in poor households based on a poverty line of $2 income per capita and are underweight based on a Z- score cutoff of −2 for WFA; (2) children who live in poor households but are not underweight; (3) children who are from non- poor households but are underweight; and (4) children who are from non- poor households but are not underweight (as expected).

As the dependent variable, we use Z- scores of WFA. The dependent var-iable is truncated at the respective cut off points for per capita income and WFA. With this approach, we are able to identify whether the factors that condition nutritional status of rural children in the two emerging market economies change as households move away from the poverty line. The comparison also shows the importance of income as a factor for undernu-trition because we look at those households which are income poor but do not have underweight children. Because households tend to shift income shares to food when resources become scarce, an increase in income might have different influences for households below than above the poverty line. With our methodology, we therefore identify different influencing factors on nutritional outcome below and above income and nutrition thresholds. To correct for the thresholds, we use a truncated Tobit model (Wooldridge 2010) with an underlying latent variable. Because we have pooled panel data, we use cluster robust standard errors.

We estimate two kinds of models on the four groups of children as explained above. First, we pool data for both countries in order to see the overall pattern of nutrition poverty. Second, we estimate the model for the Viet Nam data set separately.1 As pointed out by several authors (for example, Alderman et al. 2006; Haddad et al. 2003) income measures can be subject to endogeneity, for example, owing to measurement errors. A possible solution is to use asset value as an instrumental variable. We tested for endogeneity using the Durban–Wu–Hausman test for the OLS models and the Smith–Blundell test for the Tobit models (Wooldridge 2010). In most of our models, we cannot reject exogeneity of the income measure and, therefore, prefer OLS and Tobit variants to instrumental variable approaches. Where we detected endogeneity (Viet Nam data, full model on



all groups), we additionally reported an instrumental variables, two- stage least- squares regression.

3 DESCRIPTIVE ANALYSIS

In this section, we describe the background of our data which were col-lected among some rural 4000 households in both countries in 2007, 2008 and 2010. The data originate from a household and village survey adminis-tered in the context of a research project on vulnerability to poverty.2 In this project a comprehensive survey with four panel waves was carried out in six provinces of the two countries. The provinces were selected purposively based on criteria such as low per capita income, importance of agriculture, generally risky conditions because of remoteness and poor infrastructure. In Thailand, the three provinces are Nakhon Phanom, Ubon Ratchathani and Buri Ram; all belonging to the northeastern part of the country which has a long history of poverty and underdevelopment. In Viet Nam, the three provinces involved include two that belong to the Central Highlands, namely, Hat Tinh and Dak Lak, and the landlocked province of Dak Lak in the southern part of the country. The sampling procedure differed between the two countries due to difference in ecological conditions (see Hardeweg et al. 2012). In Thailand, the primary sampling unit was the sub- districts of the selected provinces and systematic random sampling was applied. In the second stage, two villages per sub- district were sampled with probability proportional to size of the population. At the third stage, ten households per village were selected systematically from a list of house-holds ordered by household size. In Viet Nam, the sampling procedure was different at the first sampling stage owing to high diversity in natural con-ditions of the three provinces. Here provincial agro- ecological zones were defined with a minimum of 160 households per strata. Within these strata two communes (equivalent to a sub- district in Thailand) were sampled according to population density and subsequently the procedure followed the one from Thailand. For the analysis in Viet Nam, the use of sample weights was necessary, whereas the sample in Thailand was self- weighting by design.

Summary statistics for all variables included in the model are presented in Table 7.1, showing their means and standard deviation over the entire sample by country.

Table 7.2 shows the poverty headcount ratios for 2007, 2008 and 2010 for the $1.25 and the $2 income per capita per day poverty lines for each of the provinces involved. The data show that, whereas absolute poverty is relatively low (that is, using the $1.25 line), a large number of the rural


222

Tabl

e 7.

1 D

efin

ition

and

sum

mar

y st

atis

tics o

f pa

nel d

ata

for T

haila

nd a

nd V

iet N

am, c

hild

ren

belo

w 5

yea

rs o

f ag

e

Nut

ritio

n ou

tcom

eT

haila

ndV

iet N

am

Mea

nSd

Mea

nSd

Z_s

core

WFA

Z- s

core

of

wei

ght-

for-

age

0.33

2.49

−0.

622.

34

Inco

me

Inco

me

PCIn

com

e pe

r cap

ita a

nd m

onth

, $PP

P12

1.70

177.

9788

.68

123.

71

Chi

ldsic

kC

hild

was

sick

(yes

= 1

, no

= 0

)0.

040.

200.

050.

22ch

ildG

irlG

irl (y

es =

1, n

o =

0)

0.45

0.50

0.50

0.50

Mot

her

m_h

eigh

tH

eigh

t of

the

mot

her (

cm)

156.

966.

6015

3.98

7.87

m_e

duye

ars

Edu

catio

n of

the

mot

her (

year

s)8.

473.

566.

623.

98m

_mig

rant

Dum

my:

mot

her m

igra

ted

(yes

= 1

, no

= 0

)0.

200.

400.

010.

10

Hou

seho

ldH

Hsiz

eN

ucle

ar fa

mily

hou

seho

ld si

ze5.

381.

985.

281.

80de

p.ra

tioD

epen

denc

y ra

tio (H

H si

ze/d

epen

dent

s)?

0.77

2.14

0.74

mig

mon

th_o

ther

No.

of

mon

ths o

ther

HH

mem

. mig

rate

d3.

217.

281.

111.

44et

hnic

Min

HH

is e

thni

c m

inor

ity (y

es =

1, n

o =

0)

0.31

0.46

Priv

Toile

tPr

ivat

e w

ater

toile

t (ye

s = 1

, no

= 0

)0.

960.

190.

180.

39Ta

pwat

erH

H h

as ta

p w

ater

(yes

= 1

, no

= 0

)0.

280.

450.

110.

31

Vill

age

VPs

anita

tion

% o

f H

Hs w

ith sa

nita

tion

faci

litie

s76

.51

40.0

314

.30

28.7

9V

pubW

ater

Acc

ess t

o pu

blic

wat

er (y

es =

1, n

o =

0)

0.89

0.32

0.22

0.42

DIS

Tto

wn

Dist

ance

to to

wn

(min

utes

)153

.56

30.4

446

.47

53.7

4V

ILL

inc

Ave

rage

inco

me

in th

e vi

llage

(pc/

mon

th16

4.93

152.

9910

6.02

74.0

8V

Smou

ntV

illag

e in

mou

ntai

n re

gion

(yes

= 1

, no

= 0

)0.

390.

49N

1947

1960

Not

e:

1 Dist

ance

was

mea

sure

d in

min

utes

usin

g th

e m

ost c

omm

on fo

rm o

f tr

ansp

ort (

for e

xam

ple,

bus o

r mot

orcy

cle)

.

Sour

ce:

Hou

seho

ld su

rvey

s 200

7, 2

008

and

2010

.



population in both countries is just above the poverty line. Increasing the threshold to $2 per day in 2007 puts between 36 percent and 45 percent of the Thai households and almost 70 percent of the Vietnamese households below the poverty line. In both countries, poverty increases by 10 percent to 20 percent when the poverty line is increased from $1.25 per day to $2 per day. Variation between provinces is small but increased in 2010, after the food price and economic crisis, suggesting that provinces have been coping differently with the crisis. It is also interesting to note that poverty in 2010 decreased more in Thailand than in Viet Nam which suggests that Thailand recovered better from the crisis and social protection measures may have been effective in favor of the poor.

This is further illustrated in Figure 7.1, which shows the cumulative empirical distribution functions of consumption expenditures in 2010 for both countries. The probability of a rural household with a consumption level below the poverty line is very low for Thailand and, even at a level of $120 per capita per month ($4 per capita per day), some 60 percent of households are above this level. In the Vietnamese provinces, consumption poverty is much higher (see right- hand panel of Figure 7.1) and less than 20 percent of them would surpass a level of $4 per day, which could be considered a ‘middle- class threshold’. Consumption is more evenly spread

Table 7.2 Poverty headcount ratios in Thailand and Viet Nam based on per capita income in 2007, 2008 and 2010

$1.25 poverty line $2 poverty line

2007 2008 2010 2007 2008 2010

ThailandBuriram 30.2 33.0 07.7 44.8 44.2 17.5Ubon Ratchathani

21.8 21.7 12.3 36.3 36.3 21.8

Nakhon Phanom

23.5 31.2 14.3 41.2 48.1 29.3

Viet NamHa Thin 55.7 18.5 16.6 69.9 36.3 31.5Thua Thien Hue

38.1 27.5 16.6 57.4 46.9 31.7

Dak Lak 29.9 23.2 23.3 45.4 37.2 36.6

Note: Poverty based on income measure, VN adjusted for survey weights.

Source: Household surveys 2007, 2008 and 2010.



among different levels, whereas differences among provinces are more pro-nounced in Viet Nam.

In Figure 7.2, the effect of food prices on the distributions of food con-sumption shares are shown for both countries aggregating the data for the three provinces in the respective countries. It can be seen that, in 2010, that is, after the economic crisis but at a time when food prices were still higher than in 2007, the distributions shifted to the right for both countries. This indicates that the majority of rural households had to allocate a much higher share of their consumption expenditures to food. The effect was stronger in Viet Nam where the mode shifted to about 80 percent, whereas it increased to above 60 percent in Thailand. Relating these observations

0

0.2

0.4

0.6

0.8

1.0

0 100 200 300

Consumption per capita per month, 2010

Buriram

Ubon Ratchathani

Nakhon Phanom

0

0.2

0.4

0.6

0.8

1.0

0 100 200 300

Consumption per capita per month, 2010

Ha Thin

HueDak Lak

Viet Nam, 2010

Thailand, 2010

Note: Poverty lines at $1.25, $2 and $4 per day.

Source: Household survey 2010.

Figure 7.1 Distribution of consumption expenditures for three provinces in Thailand and Viet Nam, 2010


225

Tha

iland

, 200

7

Vie

t Nam

, 201

0 V

iet N

am, 2

007

Tha

iland

, 201

0

050100

150

Frequency

0.2

0.4

0.6

0.8

1.0

Sha

re o

f foo

d in

tota

l con

sum

ptio

n, V

N, 2

007

050100

150

200

Frequency

0.2

0.4

0.6

0.8

1.0

Sha

re o

f foo

d in

tota

l con

sum

ptio

n, V

N, 2

010

050100

150

Frequency

00.

20.

40.

60.

81.

0S

hare

of f

ood

in to

tal c

onsu

mpt

ion,

TH

, 200

7

050100

150

200

Frequency

00.

20.

40.

60.

81.

0S

hare

of f

ood

in to

tal c

onsu

mpt

ion,

TH

, 201

0

Sour

ce:

Hou

seho

ld su

rvey

s 200

7 an

d 20

10.

Figu

re 7

.2

Shar

e of

food

in to

tal c

onsu

mpt

ion,

200

7 an

d 20

10, T

haila

nd a

nd V

iet N

am



to the data on poverty and consumption shows that, in spite of a decline in poverty, adjustments in food consumption became necessary and, there-fore, consequences for nutrition are likely.

We now assess nutritional outcomes of the households in our sample. Figure 7.3 shows the distribution of the weight- for- age Z- scores for chil-dren below 5 years of age in 2010. Referring to the −2 Z- score threshold, it is noted that about 19 percent of the pre- school children in Thailand and 27 percent of the children in Viet Nam are considered undernourished based on WFA. Only around one- third of households have children below the age of 5 – and only these can be included in any child- nutrition meas-ures. Again, it is worth noting that the rate of undernutrition of children is similar or even higher than the rate of poverty both for the $1.25 and the

0

20

40

60

80

100

Fre

quen

cy

–10 –5 0 5Z_scoreWFA

Thailand

0

20

40

60

80

100

Fre

quen

cy

–10 –5 0 5

Z_scoreWFA

Viet Nam


Figure 7.3 Distributions of the weight- for- age Z- scores for children below 5 years of age: Thailand and Viet Nam, 2010



$2 poverty lines. This supports the notion that the reduction of monetary poverty is not a sufficient condition for the elimination of undernutrition.

A complete overview of nutrition indicators, pooled over three years, is presented in Table 7.3. We calculated the means in nutrition outcomes for the pooled data set of three years on average and across different intervals of per capita income for Thailand and Viet Nam separately. We observe that values for stunting are much higher than for those for underweight and wasting, which is consistent with the standards defined by WHO (2014). As expected, undernutrition rates are still higher in Viet Nam. For the comparison across income groups, we start with a per capita income of below $2 per day until above $10 per day as the upper range. For nutrition indicators, we take the respective shares based on WFA, BMI, HFA, and WFH for children and BMI for adults.

In the next step of the descriptive analysis, we establish four groups based on the criteria, poverty and nutrition. Group 1 consists of children living in households below the $2 poverty line and who are underweight, according to the WFA indicator. Group 2 represents children from poor households who are not underweight. Groups 3 and 4 involve children from non- poor households who are underweight or not, respectively.

From Table 7.4, we note differences in parameters on individual, house-hold and village levels among the four groups in Thailand. First, poor households with underweight children have lower per capita food con-sumption although they may have the same level of income as compared to poor households with no underweight children. Also the former have a lower share of agricultural income and rely relatively more on food from natural resources which tend to be more erratic in supply. Such difference can no longer be observed for non- poor households. Another difference is migration of the child’s mother. Poor households with normal weighted children have an eight percent higher share of mothers working outside the village. This is also reflected in the time that mothers spend outside the household, that is, mothers from poor households with normal weighted children spend almost thrice the time away. In non- poor households such differences are smaller.

A major factor seems to be assets. Poor households with underweight children (Table 7.4) have only about half the assets in value terms com-pared with their counterfactual group. Again this difference is smaller in absolute and relative terms for the non- poor groups.

The prenatal condition of children, as indicated by the mother’s height, shows some differences in the poor household group while the mother’s education is considerably higher in the non- poor groups.

Interestingly, no difference can be observed in the food consumption expenditure shares among the four groups which suggest that differences


228

Tabl

e 7.

3 M

ean

diffe

renc

es in

nut

ritio

n ou

tcom

es o

f ho

useh

olds

in T

haila

nd a

nd V

iet N

am, 2

007,

200

8 an

d 20

10 (

pool

ed)

Inco

me

($ P

PP p

er c

apita

per

day

)To

tal

0 –

<2

2 –

<3

3 –

<5

5 –

<7

7 –

<10

10

Tha

iland

Shar

e of

chi

ldre

n un

derw

eigh

t (W

FA)

0.12

0.12

0.13

0.13

0.10

0.07

0.11

Shar

e of

chi

ldre

n un

derw

eigh

t (B

MI)

0.13

0.14

0.13

0.12

0.13

0.15

0.15

Shar

e of

chi

ldre

n st

unte

d (H

FA)

0.42

0.43

0.45

0.45

0.41

0.33

0.37

Shar

e of

chi

ldre

n w

aste

d (W

FH

)0.

120.

120.

120.

120.

110.

150.

11Sh

are

of a

dults

und

erw

eigh

t (B

MI)

0.12

0.13

0.13

0.12

0.12

0.12

0.11

Vie

t Nam

Shar

e of

chi

ldre

n un

derw

eigh

t (W

FA)

0.27

0.33

0.27

0.18

0.21

0.16

0.14

Shar

e of

chi

ldre

n un

derw

eigh

t (B

MI)

0.14

0.14

0.14

0.15

0.14

0.09

0.13

Shar

e of

chi

ldre

n st

unte

d (H

FA)

0.50

0.52

0.50

0.49

0.45

0.50

0.53

Shar

e of

chi

ldre

n w

aste

d (W

FH

)0.

130.

140.

130.

150.

160.

080.

12Sh

are

of a

dults

und

erw

eigh

t (B

MI)

0.25

0.29

0.26

0.24

0.22

0.20

0.20

Sour

ce:

Hou

seho

ld su

rvey

s 200

7, 2

008

and

2010

.



Table 7.4 Comparison of children by poverty ($2 poverty line) and nutritional status (WFA), Thailand 2007–2010

Groups (1)Poor and

underweight

(2)Poor

and no underweight

(3)Non- poor

and underweight

(4)Non- poor

and no underweight

IncomeIncome per capita and

month (PPP $)

22.53 22.91 165.63 185.63

Share agricultural

income*

0.41 0.53 0.21 0.22

Share natural resources

income*

0.09 0.07 0.04 0.03

Food consumption per capita and

month (PPP$)

41.88 49.09 72.90 68.83

Share food of total

consumption

0.60 0.62 0.63 0.60

Share households

with small scale business

0.19 0.20 0.33 0.34

ChildShare of children sick

0.05 0.05 0.01 0.04

Share of girls 0.44 0.46 0.42 0.45

MotherM_height (cm) 153.89 156.07 156.76 157.82M_edu (years) 7.02 7.30 8.60 9.31Share M_migrant

0.15 0.23 0.17 0.18

HouseholdHHsize 5.28 5.32 5.11 5.27Dependency ratio

2.18 2.05 2.01 1.89

Migmonth_other 1.03 2.70 0.96 2.06Share agricultural

worker

0.59 0.62 0.52 0.49



may exist in the quality of food assuming positive income elasticity for food expenditures, that is, as households become better off, their absolute expenditures on food increase. Also, no difference can be observed in sani-tation parameters, neither on household nor village level.

The respective comparison for Viet Nam also shows differences among the four groups (Table 7.5). Among the poor households differences in food consumption are relatively higher for households with normal weighted children than differences in per capita income although absolute differences are small. Unlike in Thailand, this difference is higher in the two non- poor groups. Poor households with undernourished children are less likely to have a small- scale business and must rely on own agriculture, food from natural resources and wage employment. A marked differ-ence can be observed in health with a higher share of children who were

Table 7.4 (continued)

Groups (1)Poor and

underweight

(2)Poor

and no underweight

(3)Non- poor

and underweight

(4)Non- poor

and no underweight

Share wage worker 0.05 0.05 0.10 0.10Share business worker

0.36 0.33 0.38 0.41

Share PrivToilet 0.91 0.95 0.97 0. 97Share Tapwater 0.25 0.27 0.23 0.30Value assets per capita (PPP $)

717.04 1364.20 1648.86 2014.91

Value livestock per capita (PPP

$)

195.47 179.28 201.43 241.39

Land per capita (ha)

0.67 0.57 0.73 0.76

VillageVPsanitation 75.69 75.2 80.34 77.25VpubWater 0.88 0.88 0.91 0.90Distance market 17.73 19.96 18.27 16.79Distance hospital

23.43 23.70 21.05 21.54

N 97 586 132 831

Notes: * Negative crop/natural resource incomes excluded.

Source: Household surveys 2007–2010.



Table 7.5 Comparison of households by poverty ($2 poverty line) and nutritional status of children (WFA), Viet Nam 2007–2010

Groups (1)Poor and

underweight

(2)Poor

and no underweight

(3)Non- poor

and underweight

(4)Non poor

and no underweight

IncomeIncome per capita (PPP $)

24.55 26.93 147.67 162.50

Share of agricultural

income*

0.51 0.65 0.36 0.33

Share natural resources

income*

0.07 0.07 0.02 0.03

Food consumption per

capita/month (PPP $)

33.01 34.93 47.5 52.92

Share food in total

consumption

0.70 0.69 0.68 0.66

Share households

with small- scale business

0.36 0.41

ChildShare of children sick

0.09 0.04 0.04 0.02

Share of girls 0.56 0.49 0.49 0.50

MotherM_height (cm) 154.69 154.62 155.12 154.99M_edu (years) 5.43 6.12 6.80 8.27M_migrant 0.00 0.01 0.02 0.01

HouseholdHHsize 5.50 5.56 5.15 5.03Dependency ratio

2.27 2.26 1.97 2.03

Migmonth_other 0.04 0.08 0.15 0.14Share ethnic minority

0.39 0.30 0.21 0.10



reported sick. Also undernourished children tend to have mothers with fewer years of education. A marked difference is also in ethnicity. The share of ethnic minorities is highest among poor households with under-nourished children.

Marked differences exist between poor and non- poor households, for example, in labor allocation, poor households are more agriculturally based and non- poor households have a higher share of wage employment and small- scale business. Furthermore differences also exist in sanitary

Table 7.5 (continued)

Groups (1)Poor and

underweight

(2)Poor

and no underweight

(3)Non- poor

and underweight

(4)Non poor

and no underweight

Share agricultural

worker

0.82 0.78 0.66 0.54

Share wage worker

0.04 0.05 0.16 0.18

Share business worker

0.15 0.17 0.19 0.28

Share PrivToilet 0.09 0.12 0.23 0.31Share Tapwater 0.05 0.07 0.10 0.15Value assets per capita (PPP $)

379.06 382.59 791.10 1049.20

Value livestock per capita (PPP $)

160.50 140.28 332.96 223.60

Land per capita (ha)

0.13 0.14 0.22 0.17

Percent HH with no land

8.10 6.53 4.36 12.56

VillageVPsanitation 11.37 12.00 22.69 18.96VpubWater 0.13 0.17 0.13 0.19Distance market 20.06 18.38 17.97 15.69Distance hospital 37.07 35.90 30.65 31.58Share households in

mountain region

0.31 0.29 0.24 0.16

N 297 658 144 632

Source: Household surveys 2007–2010.



conditions, for example, non- poor households have better access to water and better hygienic conditions. Furthermore, poor households tend to live in remote mountainous areas. However differences between households with underweight and normal weighted children for both income groups (poor and non- poor) are small. Nevertheless, the share of undernourished children among poor households is about 1:2, whereas it is about 1:5 in non- poor households.

In summary, our descriptive and explorative analyses for some 4000 rural households consisting of over 22 000 individuals, including adults and children, allow us to draw some lessons that provide some initial answers to the questions asked in section 1 of the chapter. These findings also form the basis for the establishment of some hypotheses to be further explored in the econometric analysis below.

First, we observe that, whereas poverty reduction has been quite success-ful in both countries, this success is subject to the choice of the poverty line. Clearly, extreme poverty is now negligible in both countries but, by increas-ing the poverty line to $2 or $4 per day, headcount ratios increase. This suggests that poverty does not end when a household surpasses the official poverty line and that vulnerability to poverty continues to be a problem.

Second, we note that nutrition problems persist in both countries in spite of their success in poverty reduction. Again, the problem is bigger in Viet Nam than in Thailand. For children, HFA and WFA the respective shares below the critical levels are 42 percent and 12 percent for Thailand and 50 percent and 27 percent for Viet Nam, which suggests that particu-larly underweight is still a problem, especially in Viet Nam. The latter value corresponds well with Haddad et al. (2003) who predicted, on the basis of their cross- country nutrition model, underweight (WFA) for pre- school children in Viet Nam to be at around 28 percent in 2015.

Third, as suggested in the literature, income is a poor predictor for success in reducing undernutrition. Increasing the poverty line from $2 per capita income and going beyond $10 per capita income shows that under-nutrition of children declines only slightly in Thailand, but it more rapidly declines in Viet Nam, starting at a higher level, but clearly with a declining rate above $5 per day. This underlines the role of non- income factors for governments wanting to improve the nutritional status of their population

Exploring the relationship between consumption levels and nutrition suggests that the poverty line is not a strong indicator for the disappear-ance of nutrition problems because the share of individuals who fall out of the norm values for nutrition outcomes only gradually decline with higher incomes. This lends some support to the hypothesis that reducing or elimi-nating monetary poverty does not automatically reduce other forms of poverty to the same extent. Although there are some differences between



poor and non- poor when comparing nutrition indicators, nutrition prob-lems do exist beyond the poverty line. This suggests that the factors respon-sible for income poverty are not necessarily the same as those for nutrition and other forms of poverty and therefore additional exploration using the econometric model described above is warranted.

Finally, by establishing four different household categories based on poverty and nutrition, it is observed that households with undernour-ished children have some common characteristics that are independent of monetary wealth. Undernourished children live in settlements where sanitation is generally poorer than in other villages. The comparison across household types also suggests that nonmonetary factors are important for reducing undernutrition of children.

4 ECONOMETRIC ANALYSIS

To further explore the hypotheses derived from the literature and the find-ings of our descriptive and explorative analysis, the econometric model outlined in section 2 is applied for different nutrition variables. The first dependent variable for this model is the WFA Z- score; hence a positive significant sign of any explanatory variable suggests improvement of the nutritional status of a child. An OLS regression was estimated for the pooled data set for the years, 2007, 2008 and 2010, including data for both countries and capturing the country effects by a dummy variable. Also, separate models were estimated for the four household groups based on poverty and nutritional status. We first estimate the models combined for Thailand and Viet Nam, and supplement it with a version only including data from Viet Nam, where the nutritional problem is more severe.

In column 1 of Table 7.6, the results of the OLS regression are shown. As expected, log income positively influences the nutrition outcome, but with a relatively low coefficient of 0.162, which is in line with previous estimates in the literature (for example, Haddad et al. 2003; Alderman et al. 2006). Child characteristics also have a significant influence, that is, if a child was sick in the previous period its nutrition outcome is negatively affected. On average, a sickness event decreases Z- scores by 0.3. The gender variable is significant but with a sign contrary to expectations. On average, girls seem to be better nourished, which does not seem to be in line with usual gender discrimination patterns against girls, but this finding has also been reported by some authors (Svedberg 1990; Belitz et al. 2010). However, the variable mother’s height is positive which is consistent with findings in the literature and suggests that prenatal conditions influence the nutritional status of the child. On the other hand, we do not find a significant effect


235

Tabl

e 7.

6 E

stim

ates

for p

oole

d O

LS

and

Tobi

t mod

els f

or W

FA Z

- sco

res f

or fo

ur d

iffer

ent g

roup

s hou

seho

lds b

ased

on

pove

rty

and

nutr

ition

al st

atus

Varia

bles

AL

L (O

LS)

Poor

and

un

derw

eigh

tPo

or a

nd n

o un

derw

eigh

tN

on- p

oor a

nd

unde

rwei

ght

Non

poo

r and

no

unde

rwei

ght

Inco

me

Inco

me

PC0.

162*

**0.

038

0.20

6***

−0.

037

0.10

5(0

.036

)(0

.036

)(0

.061

)(0

.092

)(0

.091

)

Chi

ldSi

ck−

0.31

2*−

0.29

2*−

0.14

2−

0.19

0−

0.23

5(0

.163

)(0

.161

)(0

.210

)(0

.252

)(0

.242

)C

hild

Girl

0.17

0*0.

257*

**0.

220*

0.45

9***

0.23

2**

(0.0

88)

(0.0

93)

(0.1

18)

(0.1

25)

(0.1

06)

Mot

her

m_h

eigh

t0.

018*

**−

0.00

30.

012

−0.

016

0.02

0**

(0.0

07)

(0.0

07)

(0.0

10)

(0.0

14)

(0.0

08)

med

uyea

rs0.

012

0.03

0**

−0.

032*

−0.

001

0.02

6*(0

.012

)(0

.014

)(0

.018

)(0

.020

)(0

.015

)m

_mig

rant

0.27

00.

459*

*0.

598*

*−

0.43

10.

354

(0.2

21)

(0.2

13)

(0.2

74)

(0.4

70)

(0.2

57)

Hou

seho

ldH

Hsiz

e0.

024

0.00

30.

039

0.06

0*0.

003

(0.0

24)

(0.0

23)

(0.0

27)

(0.0

32)

(0.0

35)

Dep

.ratio

0.05

10.

023

0.06

40.

162

0.07

1(0

.059

)(0

.052

)(0

.079

)(0

.104

)(0

.070

)M

igm

onth

_oth

0.01

7−

0.01

0−

0.00

70.

040

0.00

7(0

.012

)(0

.009

)(0

.017

)(0

.040

)(0

.013

)


236

Tabl

e 7.

6 (c

ontin

ued)

Varia

bles

AL

L (O

LS)

Poor

and

un

derw

eigh

tPo

or a

nd n

o un

derw

eigh

tN

on- p

oor a

nd

unde

rwei

ght

Non

poo

r and

no

unde

rwei

ght

Eth

nicM

in−

0.29

5**

0.03

9−

0.34

8**

−0.

009

0.35

6*(0

.126

)(0

.113

)(0

.155

)(0

.204

)(0

.206

)Pr

ivTo

ilet

0.33

9***

−0.

259

0.43

0**

0.08

60.

186

(0.1

23)

(0.2

63)

(0.1

93)

(0.1

84)

(0.1

32)

Tapw

ater

0.00

30.

045

−0.

116

0.01

10.

168

(0.0

93)

(0.1

10)

(0.1

41)

(0.1

34)

(0.1

15)

Vill

age

VPs

anita

tion

0.00

3**

0.00

30.

005*

**−

0.00

00.

002

(0.0

01)

(0.0

02)

(0.0

02)

(0.0

02)

(0.0

02)

DIS

Tto

wn

−0.

001

−0.

001

−0.

000

0.00

1**

−0.

001

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

VIL

Lin

c−

0.00

0−

0.00

0−

0.00

1−

0.00

10.

000

(0.0

00)

(0.0

01)

(0.0

01)

(0.0

00)

(0.0

00)

Tha

iland

0.25

6*−

0.11

2−

0.10

9−

0.17

00.

411*

*(0

.146

)(0

.263

)(0

.222

)(0

.203

)(0

.161

)_c

ons

−0.

743

−2.

539*

*0.

877

−0.

709

−0.

988

(1.1

07)

(1.1

45)

(1.5

24)

(2.0

23)

(1.3

39)

N28

7336

595

926

412

85

Not

es:

* ⇒

p <

0.1

; ** ⇒

p <

0.0

5; *

** ⇒

p <

0.0

1. S

tand

ard

erro

rs a

re c

lust

ered

on

indi

vidu

al le

vel.

Hou

seho

lds w

ith n

egat

ive

inco

mes

are

ex

clud

ed. Y

ear i

s con

trol

led

for.

Age

is c

ontr

olle

d fo

r and

sign

ifica

nt. M

igra

nt m

onth

s nor

mal

ized

(+1)

.

Sour

ce:

Hou

seho

ld su

rvey

s 200

7–20

10.



of education nor the migrant status of the mother. The same is true for a range of household characteristics including size and dependency ratio. Other household and village characteristics, however, are significant. Sanitary conditions of a household and the sanitation infrastructure at the village level significantly increase nutritional outcomes. Being born in an ethnic minority decreases the nutritional outcome, which is plausible because, in Viet Nam, many ethnic minorities belong to economically disadvantaged, and often marginalized, population groups. The observa-tion from the descriptive analysis that the nutritional status in Thailand is better than in Viet Nam is reflected in the significant country dummy.

In columns 2–5 of Table 7.6, regression results of the truncated Tobit regressions on Z- scores of WFA for the four different groups are presented. The income variable is only significant in one of the four groups, namely, for no underweight children that live in poor households (column 3). A higher income, that is, an income nearer to the $2 threshold is positively correlated with a better child nutritional outcome above the underweight cutoff point.

A similar observation can be made for child sickness. Generally, sick-ness leads to decreasing nutrition Z- scores; the effect is larger, if good health care is not available or not used. Although in the regression with the complete data set, sickness is correlated with lower nutrition Z- scores, this effect can only be observed for the poor and undernourished children. Whereas well- nourished children might be less often those with sickness, richer households might have access to better health care to invalidate the effect. The presence of girls in the household does have significantly higher Z- scores in all groups. Mother characteristics are differently correlated with Z- scores over the groups. Mother’s height is positively correlated with nutrition only for those children who are non- poor and well- nourished. For those above the poverty line, a long- term economic and nutritional well- being of the household improves nutritional status of the children. Education, as measure for childcare, is, as expected, positively correlated with nutrition for poor and undernourished children as well as for non- poor and well- nourished. Interestingly, better nourished poor children have less educated mothers. A possible explanation might be overweight; some of those children classified as well- nourished might even suffer from overweight, which might be favored by low nutritional knowledge. For stronger explanations, more research on this topic is necessary. Migration status of the mother gives a clearer picture because it is correlated posi-tively with nutrition for those children below the poverty line. Although the effect appears quite clear, channels are less clear. On the one hand, migrated mothers can spent less time with their children, which might be especially negative for very young children because of breastfeeding. For older children, especially in Thailand, grandparents, who are experienced



in childcare, take over responsibility. Additionally, mothers might gain knowledge on childcare in the cities where they work, and send back remittances, which might be directed at their children’s well- being. For the children below the poverty line, positive effects dominate.

As in the model using the complete sample, household size, dependency ratio and migration of other household members do not significantly affect nutrition, except for household size which is positively correlated with nutrition for those underweight children above the poverty line. Ethnicity of a child has a different effect depending on the group. For poor and well- nourished children, belonging to an ethnic minority is correlated with lower nutrition Z- scores. This might be interpreted as children below the poverty line and belonging to ethnic minorities are rather those close to the cutoff than at the upper tail of the distribution of Z- scores. For non- poor and no underweight children, a positive correlation is observed, hinting at the possibility of over- nourished ethnic minority children.

For household and village- level effects, our results show that sanitation is important. Having a private toilet as well as the percentage of house-holds with sanitation in the village has a positive effect on nutrition for those well- nourished children below the poverty line. Good sanitary facili-ties and hygiene is one way to improve nutrition and overcome undernutri-tion for the poor. Distance to town (exception: non- poor underweight) and the average income level in the village seem to not influence nutrition to a measurable extent. The Thailand dummy is positive and significant for the complete sample and in the non- poor group with normal weight children. This confirms that the overall nutritional conditions are better in Thailand. However, in all other equations the coefficient is not significant. Therefore, additional models were estimated for the Thailand data set separately and results are reported in the Appendix to this chapter.

Not reported are the control variables for the year, which shows negative significant effect for 2010, and the age of children, which shows, as expected from the literature, significantly worse nutrition values for older children.

Empirical results obtained by applying the regression model to the data for Viet Nam only are presented in Table 7.7. Because we cannot reject exogeneity in the case of the full model for Viet Nam, we also estimated an instrumental variable (IV) model, using value of assets as an instrument (following Haddad et al. 2003). Results from the OLS and IV estimations are quite robust. The coefficient of the income variable is larger from the IV regression, but both are positive. As in the regression for both countries, sickness of a child has a negative effect on nutrition, but gender discrimina-tion is not significant in the general model. The influence of household char-acteristics differs slightly for the two estimations. In the IV case, household size and dependency ratio positively influence nutrition, but, in the OLS


239

Tabl

e 7.

7 C

ontin

uous

out

com

e va

riab

le fo

r chi

ldre

n: p

oole

d O

LS/

Tobi

t VN

, ind

icat

or W

FA, p

over

ty li

ne $

2 (P

PP

$),

V

iet N

am

AL

L (O

LS)

All

(IV

)Po

or a

nd

unde

rwei

ght

Poor

and

no

unde

rwei

ght

Non

- poo

r and

un

derw

eigh

tN

on- p

oor a

nd n

o un

derw

eigh

t

Inco

me

inco

me

PC0.

123*

**0.

420*

**0.

032

0.05

3−

0.07

10.

022

(0.0

46)

(2.7

5)(0

.039

)(0

.074

)(0

.145

)(0

.127

)

Chi

ldsic

k−

0.54

1**

−0.

441*

−0.

356*

*−

0.02

0−

0.30

30.

334

(0.2

39)

(−1.

84)

(0.1

80)

(0.2

72)

(0.2

71)

(0.3

85)

child

Girl

0.16

10.

163

0.34

2***

0.31

2**

0.61

1***

0.20

2(0

.101

)(1

.62)

(0.1

04)

(0.1

32)

(0.1

61)

(0.1

27)

Mot

her

m_h

eigh

t0.

014

0.01

3−

0.00

80.

010

−0.

027

0.02

8**

(0.0

09)

(1.3

8)(0

.008

)(0

.010

)(0

.017

)(0

.012

)m

eduy

ears

−0.

005

−0.

015

0.04

3***

−0.

069*

**0.

041*

−0.

004

(0.0

15)

(−0.

99)

(0.0

15)

(0.0

22)

(0.0

23)

(0.0

18)

Hou

seho

ldH

Hsiz

e0.

034

0.05

6*−

0.00

30.

041

0.06

6**

0.05

8(0

.032

)(1

.67)

(0.0

24)

(0.0

40)

(0.0

32)

(0.0

45)

dep.

ratio

0.10

60.

143*

*0.

054

0.03

40.

485*

**0.

064

(0.0

69)

(2.0

4)(0

.052

)(0

.092

)(0

.129

)(0

.095

)m

igm

onth

_oth

0.02

8*0.

027*

0.06

7***

0.01

70.

057*

0.03

2***

(0.0

17)

(1.6

9)(0

.025

)(0

.022

)(0

.029

)(0

.012

)E

thni

cMin

−0.

364*

*−

0.28

2*−

0.13

1−

0.45

1**

−0.

064

0.14

8(0

.147

)(−

1.85

)(0

.130

)(0

.184

)(0

.195

)(0

.218

)


240

Tabl

e 7.

7 (c

ontin

ued) AL

L (O

LS)

All

(IV

)Po

or a

nd

unde

rwei

ght

Poor

and

no

unde

rwei

ght

Non

- poo

r and

un

derw

eigh

tN

on- p

oor a

nd n

o un

derw

eigh

t

Priv

Toile

t0.

290*

*0.

170

−0.

196

0.62

1***

−0.

199

0.10

8(0

.142

)(1

.21)

(0.3

44)

(0.2

12)

(0.1

92)

(0.1

53)

Tapw

ater

−0.

007

0.00

10.

063

−0.

055

−0.

113

0.28

9(0

.139

)(0

.01)

(0.1

53)

(0.1

89)

(0.1

76)

(0.1

80)

Vill

age

VPs

anita

tion

0.00

0−

0.00

0−

0.00

00.

004

−0.

000

−0.

001

(0.0

02)

(−0.

20)

(0.0

03)

(0.0

03)

(0.0

03)

(0.0

03)

DIS

Tto

wn

−0.

001

−0.

000

−0.

000

−0.

001

0.00

1***

0.00

1(0

.001

)(−

0.98

)(0

.002

)(0

.001

)(0

.001

)(0

.000

)V

ILL

inc

0.00

1−

0.00

0−

0.00

10.

001

−0.

000

−0.

000

(0.0

01)

(−0.

58)

(0.0

01)

(0.0

01)

(0.0

01)

(0.0

01)

Hue

0.05

00.

013

0.13

9−

0.41

9*0.

653*

*−

0.19

3(0

.174

)(0

.07)

(0.1

98)

(0.2

25)

(0.2

66)

(0.2

15)

Dak

Lak

0.22

90.

086

0.34

2**

−0.

096

0.64

2**

−0.

133

(0.1

69)

(0.4

7)(0

.149

)(0

.204

)(0

.278

)(0

.207

)V

Smou

nt−

0.20

6−

0.19

90.

024

0.00

7−

0.09

7−

0.16

6(0

.137

)(−

1.45

)(0

.134

)(0

.180

)(0

.147

)(0

.154

)_c

ons

0.32

2−

0.48

2−

2.15

5*2.

076

0.35

5−

1.26

2(1

.376

)(−

0.34

)(1

.236

)(1

.643

)(2

.430

)(2

.001

)N

1586

1586

292

547

151

596

Not

es:

* ⇒

p <

0.1

, ** ⇒

p <

0.0

5, *

** ⇒

p <

0.0

1. S

tand

ard

erro

rs a

re c

lust

ered

on

indi

vidu

al le

vel.

Hou

seho

lds w

ith n

egat

ive

inco

mes

are

ex

clud

ed. Y

ear i

s con

trol

led

for.

Age

is c

ontr

olle

d fo

r and

sign

ifica

nt. M

igra

ntm

onth

nor

mal

ized

(+1)

. IV:

Ass

et v

alue

.

Sour

ce:

Hou

seho

ld su

rvey

s 200

7–20

10.



case, a private toilet has a positive influence. In both equations, belonging to an ethnic minority decreases nutrition, whereas migration of other house-hold members increases nutritional outcomes. The effect of migration was not significant in the combined Thailand/Viet Nam model. Village charac-teristics do not have significant influences in this model.

In the truncated models, we find no significant income effect which might be a result of the lower variance owing to the income restriction, whereas in the OLS and IV models this is not the case. Sickness has mainly a negative effect in the ‘poor and underweight’ group. Although no gender discrimi-nation was observed in the complete models, girls seem to be better off in all groups, except in the ‘non- poor/no underweight’ group where, however, mother’s height is significant and positive. Although in both underweight groups, education of the mother increases nutrition outcomes, it is nega-tive for the ‘poor and no underweight’ children. Belonging to an ethnic minority has a negative effect on nutrition; having a private toilet increases nutritional outcomes in the ‘poor and no underweight’ group. We also find positive provincial effects for Dak Lak, a more commercialized province with much better infrastructure than Ha Thin (the poorest among the three provinces in Viet Nam) in both the poor and non- poor underweight groups. Also, for Hue province, Z- scores decrease for ‘poor and no under-weight’ group and increases for the ‘non- poor underweight’ group.

In summary, we find different variables to be correlated with nutrition outcomes, depending on whether the child is undernourished and poor or not. This supports our assumption of nonlinearity in factors influenc-ing nutrition outcomes depending on income and nutritional status. In general, income has an influence, but only for parts of the population. Child and mother characteristics show a correlation, whereas household characteristics, except for ethnic minority, are less important. However, quite consistently, sanitation has been found to be important, especially in the poor but no underweight group.

To further illustrate the implications of our findings, we establish a pre-diction for child nutrition outcome by the year 2030, that is, the year when based on some income projections (ADB, UNDP and UN ESCAP 2013) income poverty will have disappeared in almost all Asian countries when using the $1.25 poverty line. For our prediction, we follow the approach of Haddad et al. (2003). The authors made a prediction for WFA in several countries including Viet Nam for the year 2015. Based on the assumption of an average annual income growth of 2.5 percent, they predicted the WFA to decline from some 40 percent in the 1990s to 27 percent in 2015. Interestingly, the latter value is close to what we find with our data set for rural children in the three Vietnamese provinces.

In Table 7.8, predicted shares of underweight children are reported



separately for Thailand and Viet Nam and cover WFA and HFA indica-tors, to reflect developments in current and chronic malnutrition. The pre-dictions are made using income coefficients from regressions on Z- scores, for the two countries separately. For Thailand, OLS is used but, for Viet Nam, we use an IV regression. Coefficients are higher for Viet Nam than for Thailand. The prediction for each country is based on the distribution of incomes and Z- scores from the 2010 data. Predictions are made for different rates of average annual income growth ranging from a modest 2 percent to an overoptimistic 8 percent and assuming that growth is equally distributed. As a reference point, projected shares can be compared with the WHO thresholds for situations of low severity (WHO 2014). The WHO defined low severity when less than 10 percent of children are under-weight (WFA) and less than 20 percent are stunted (HFA), respectively.

Results in Table 7.7 show the shares of underweight children, in the 2030 target year. For Thailand, the decline is modest; for the 2 percent scenario underweight, shares decline by less than two percentage points and, even for the (unrealistically) high income scenario, underweight in 2030 is still above 10 percent (see Table 7.6). In Viet Nam, the income effect is some-what stronger, which is to be expected because the point of departure is much higher with an almost 30 percent underweight share. A 2 percent income growth would bring down underweight to 21 percent, whereas the high income scenario would, however, result in a low- severity situation, based on the WHO definition. The picture is similar for the HFA indica-tor, which reflects chronic malnutrition. The difference in the income effect

Table 7.8 Predicted values of child nutritional outcomes in 2030 for different levels of average income growth by country

Income growth

Thailand Viet Nam

WFA HFA WFA HFA

Base year (2010)

18.9 47.0 29.8 59.3

2% 17.2 46.7 21.9 47.44% 15.8 45.1 18.1 45.46% 14.5 39.5 13.4 35.78% 11.2 36.4 6.5 17.9

Notes: Income growth assumes average annual income growth. Coefficients are from OLS in Thailand (WFA: 0.144; HFA: 0.123) and IV in Viet Nam (WFA: 0.442; HFA: 1.048).




between Thailand and Viet Nam is even higher for this indicator. Thailand, starting with 47 percent of children being stunted in 2010, even under optimistic growth assumptions, will not be able to reach a level of low severity. In Viet Nam, where almost 60 percent of children in our sample were stunted in 2010, a stronger influence of income on nutrition leads to a faster reduction of stunting rates than in Thailand. Hence, with 6 percent growth, Viet Nam will achieve lower stunting rates than Thailand and, with 8 percent, will even be able to reach a level of low severity. However, results must be treated with care because it is not clear if the underlying pattern of nutritional improvement will continue.

5 SUMMARY AND CONCLUSIONS

In this chapter, we investigate the relationship between poverty and nutri-tion of rural households in the context of two emerging Asian market economies, namely, Thailand and Viet Nam. We start out by asking four questions. First, we examine to what extent the problem of undernutrition continues to exist in spite of the enormous progress that these two countries have made in poverty reduction. Second, we try to identify the characteris-tics of households that have children with undernutrition problems. Third, we assess the relationship between monetary wealth and nutrition by ana-lyzing the factors that influence the nutritional status of children in rural households as these households move out of poverty. Fourth, and building on the results of the third point, we speculate about the future of undernu-trition by setting 2030 as the target because this is believed to be the period when poverty will have come to an end in most Asian countries.

The answer to the first question is a clear yes. As expected, there are dif-ferences between the two countries. The rate of undernutrition based on WFA Z- scores from our 2010 data set is clearly lower in Thailand with just about 19 percent of children below the WHO- defined threshold and some 30 percent in Viet Nam. The latter figure is quite close to the one predicted by Haddad et al. (2003) for 2015.

As regards a typology of households with undernourished children, we conclude that, as expected, socioeconomic conditions matter. For example, undernourished children live in households with less migrant members and, thus, fewer remittances which limit their possibility to buy higher quality food. They also tend to have mothers who are less educated than children who are beyond the nutrition threshold. However, it is not merely wealth status that matters. There seem to be distinct environments of undernourishment, especially related to poor sanitation. The comparison across household types suggests that nonmonetary factors are important



for reducing undernutrition of children and, therefore, monetary poverty reduction is unlikely to be a sufficient condition for solving the nutrition problem of rural populations in emerging market economies.

For the third question, we developed a model linking nutritional out-comes for children with income and a set of other control variables. The results are similar to findings in the literature (for example, Alderman et al. 2006). Our four categories based on poverty and nutritional status show that the factors that condition a child’s nutritional outcome differ by poverty status. As expected, growth in income helps to improve nutritional outcomes, but the effect is weak, which is in line with the conclusions in the literature. However, child and mother characteristics have effects as well. For example, education matters regardless of whether the household is below or above the poverty line. In addition, there is a significant child- gender effect that is consistent across all four groups, with girls having better nutrition Z- scores. Migration and remittances are important for poor households. Similarly, ethnicity matters in Viet Nam because children without nutritional problems who live in poor households tend to belong to the ethnic majority. The models also reveal differences between the two countries, as shown by a positive country effect for Thailand.

Regarding the fourth question, we conclude that our predictions show that undernutrition is likely to exceed the period after which most Asian countries might be out of poverty. Even when using quite optimistic assumptions for growth in income, undernutrition is predicted to persist beyond 2030 so that it cannot be expected that the WHO threshold of 10 percent will be achieved for Viet Nam, which starts at a much lower level in the base year 2010, and, even for Thailand, this may not be the case.

Some caution is necessary when interpreting our results. Although the panel data set with some 4000 rural households and 22 000 individuals, including adults and children, is suitable to conduct such an analysis, the sample size for children under 5 years of age is not very large compared with other nutrition studies.

Overall, our results give some evidence that reducing or eliminating monetary poverty does not directly translate into reduction of nonmonetary poverty. Further advancing the econometric analysis, however, will help to establish stronger evidence for the persistence of nutritional poverty beyond income poverty.

NOTES

1. We also estimated a separate model for Thailand but the results were not very conclusive. This is perhaps because the food security situation is much better in Thailand and the



extent of both poverty and malnutrition is less severe. Therefore, the number of cases in the group of poor households with undernourished children is too small for estimating meaningful nutrition equations. However, we estimated an OLS and IV model for the full sample (and across different income intervals) which gave some plausible results. These results are reported in the Appendix.

2. See http://www.vulnerability- asia.uni- hannover.de/overview.html (accessed 1 April 2016).

REFERENCES66*

Alderman, H. (2012), ‘The response of child nutrition to changes in income: linking biology with economics’, CESifo Economic Studies, 58 (2), 256–73, DOI: 10.1093/cesifo/ifs012.

Alderman, H., H. Hoogeveen and M. Rossi (2006), ‘Reducing child mal-nutrition in Tanzania: combined effects of income growth and program interventions’, Economics and Human Biology, 4 (1), 1–23, DOI: 10.1016/j.ehb.2005.07.001.

Almond, D. and J. Currie (2011), ‘Killing me softly: the fetal origins hypothesis’, Journal of Economic Perspectives, 25 (3), 153–72, DOI: 10.1257/jep.25.3.153.

Anand, S. and M. Ravallion (1993), ‘Human development in poor countries: on the role of private incomes and public services’, Journal of Economic Perspectives, 7 (1), 133–50.

Asian Development Bank, United Nations Development Programme and United Nations Economic and Social Commission for Asia and the Pacific (ADB, UNDP and UN ESCAP) (2013), Asia- Pacific Aspirations: Perspectives for a Post- 2015 Development Agenda, Bangkok: ADB.

Baulch, B. and E. Masset (2003), ‘Do monetary and nonmonetary indicators tell the same story about chronic poverty? A study of Vietnam in the 1990s’, World Development, 31 (3), 441–53, DOI: 10.1016/S0305- 750X(02)00215- 2.

Becker, G.S. (1965), ‘Towards a theory of time allocation’, Economic Journal, 76, 493–517.

Belitz, C., J. Hübner and S. Klasen (2010), ‘Determinants of the socioeconomic and spatial pattern of undernutrition by sex in India : a geoadditive semiparametric regression approach’, in T. Kneib and G. Tutz (eds), Statistical Modelling and Regression Structures: Festschrift in Honour of Ludwig Fahrmeir, Physica- Verlag, pp. 155–79.

Carter, M.R. and C.B. Barrett (2006), ‘The economics of poverty traps and persist-ent poverty: an asset- based approach’, Journal of Development Studies, 42 (2), 178–99, DOI: 10.1080/00220380500405261.

Clark, D. and D. Hulme (2010), ‘Poverty, time and vagueness: integrating the core poverty and chronic poverty frameworks’, Cambridge Journal of Economics, 34 (2), 347–66, DOI: 10.1093/cje/ben046.

Cole, T.J., K.M. Flegal, D. Nicholls and A.A. Jackson (2007), ‘Body Mass Index cut offs to define thinness in children and adolescents: international survey’, British Medical Journal, 335 (7612), 194–7.

De Onis, M. and M. Blössner (2003), ‘The World Health Organization global




database on child growth and malnutrition: methodology and applications’, International Journal of Epidemiology, 32 (4), 518–26.

De Onis, M., C. Garza, A.W. Onyango, M.- F. Rolland- Cachera and le Comité de nutrition de la Société française de pédiatrie (2009), ‘WHO growth standards for infants and young children’, Archives de pédiatrie: organe officiel de la Sociéte française de pédiatrie, 16 (1), 47–53, DOI: 10.1016/j.arcped.2008.10.010.

Günther, I. and S. Klasen (2009), ‘Measuring chronic non- income poverty’, in T. Addison, D. Hulme and S.M.R. Kanbur (eds), Poverty Dynamics: Interdisciplinary Perspectives, Oxford: Oxford University Press, pp. 77–101.

Haddad, L., H. Alderman, S. Appleton, L. Song and Y. Yohannes (2003), ‘Reducing child malnutrition: how far does income growth take us?’, World Bank Economic Review, 17 (1), 107–31, DOI: 10.1093/wber/lhg012.

Hardeweg, B., S. Klasen and H. Waibel (2013), ‘Establishing a database for vulner-ability assessment’, in S. Klasen and H. Waibel (eds), Vulnerability to Poverty: Theory, Measurement and Determinants, London: Palgrave Macmillan, pp. 50–79.

Kabubo- Mariara, J., G.K. Ndenge and D.K. Mwabu (2009), ‘Determinants of children’s nutritional status in Kenya: evidence from demographic and health surveys’, Journal of African Economies, 18 (3), 363–87, DOI: 10.1093/jae/ejn024.

Klasen, S. and H. Waibel (eds) (2013), Vulnerability to Poverty: Theory, Measurement, and Determinants, London: Palgrave Macmillan.

Mckay, A. and D. Lawson (2003), ‘Assessing the extent and nature of chronic poverty in low income countries: issues and evidence’, World Development, 31 (3), 425–39, DOI: 10.1016/S0305- 750X(02)00221- 8.

Menon, P. (2012), ‘Childhood undernutrition in South Asia: perspectives from the field of nutrition’, CESifo Economic Studies, 58 (2), 274–95, DOI: 10.1093/cesifo/ifs015.

Sen, A. (2000), ‘Social exclusion: concept, application and scrutiny’, Social Development Paper No. 1, Manila: Asian Development Bank.

Smith, L.C., U. Ramakrishnan, A. Ndiaye, L. Haddad and R. Martorell (2003), ‘The importance of women’s status for child nutrition in developing countries’, IFPRI Research Report 131, Washington, DC, accessed 10 April 2014 at http://www.ifpri.org/sites/default/files/publications/rr131.pdf.

Strauss, J. and D. Thomas (1995), ‘Human resources: empirical modeling of house-hold and family decisions’, in J. Behrman and T. Srinivasan (eds), Handbook of Development Economics, Amsterdam: Elsevier, pp. 1883–2023.

Svedberg, P. (1990), ‘Undernutrition in sub- Saharan Africa: is there a gender bias?’, Journal of Development Studies, 26 (3), 469–86, DOI: 10.1080/00220389008422165.

Tsui, K. (2002), ‘Multidimensional poverty indices’, Social Choice and Welfare, 19 (1), 69–93, DOI: 10.1007/s355- 002- 8326- 3.

Wooldridge, J.M. (2010), Econometric Analysis of Cross Section and Panel Data, 2nd edn, Cambridge, MA: MIT Press.

World Health Organization (WHO) (2014), ‘Global database on child growth and malnutrition’, accessed 24 September 2015 at http://www.who.int/nutgrowthdb/about/introduction/en/index5.html.

World Health Organization (WHO) (2015), ‘Global database on child growth and malnutrition’, accessed 14 October 2015 at http://www.who.int/nutgrowthdb/about/introduction/en/index4.html.

World Health Organization (WHO) Working Group (1986), ‘Use and interpreta-tion of anthropometric indicators of nutritional status’, Bulletin of the World Health Organization, 64 (6), 929–41.


247

APP

EN

DIX

7A

.1

010203040

BMI

010

020

030

0C

onsu

mpt

ion

per

capi

ta p

er m

onth

Tha

iland

, 200

7

010203040

BMI

010

020

030

0C

onsu

mpt

ion

per

capi

ta p

er m

onth

Tha

iland

, 201

0

Vie

t Nam

, 200

7 V

iet N

am, 2

010

010203040

BMI

010

020

030

0C

onsu

mpt

ion

per

capi

ta p

er m

onth

010203040 BMI

010

020

030

0C

onsu

mpt

ion

per

capi

ta p

er m

onth

Sour

ce:

Hou

seho

ld su

rvey

s 200

7 an

d 20

10.

Figu

re 7

A.1

C

onsu

mpt

ion

and

BM

I of

adul

ts, T

H a

nd V

N, 2

007

and

2010


248

Tha

iland

, 200

7T

haila

nd, 2

010

Vie

t Nam

, 200

7V

iet N

am, 2

010

–50510

Z_scoreWFAX

010

020

030

0

Con

sum

ptio

n pe

r ca

pita

per

mon

th

–10–50510

Z_scoreWFAX

010

020

030

0

Con

sum

ptio

n pe

r ca

pita

per

mon

th

–10–50510

Z_scoreWFAX

010

020

030

0

Con

sum

ptio

n pe

r ca

pita

per

mon

th

–10–50510

Z_scoreWFAX

010

020

030

0

Con

sum

ptio

n pe

r ca

pita

per

mon

th

Sour

ce:

Hou

seho

ld su

rvey

s 200

7 an

d 20

10.

Figu

re 7

A.2

C

onsu

mpt

ion

and

weig

ht fo

r age

of

child

ren

belo

w 5

yea

rs o

f ag

e, T

H a

nd V

N, 2

007

and

2010


249

Tabl

e 7A

.1

Nut

ritio

n m

odel

s for

chi

ldre

n, T

haila

nd

Dep

ende

nt v

aria

ble:

Z- s

core

WFA

AL

L (O

LS)

AL

L (I

V)

0–2$

>2–

4$>

4$

Inco

me

Inco

me

per c

apita

(log

)0.

503

2.96

8*(0

.319

)(1

.660

)

Chi

ldC

hild

was

sick

0.11

70.

181

0.12

71.

214*

*−

0.74

6***

(0.2

27)

(0.2

52)

(0.3

60)

(0.4

76)

(0.2

85)

Chi

ld is

a g

irl0.

156

0.21

00.

033

0.14

20.

533*

*(0

.140

)(0

.148

)(0

.205

)(0

.268

)(0

.212

)

Mot

her

Mot

her h

eigh

t0.

018*

0.01

40.

023*

−0.

002

0.01

7(0

.010

)(0

.010

)(0

.013

)(0

.017

)(0

.017

)M

othe

r edu

catio

n0.

039*

0.00

10.

022

0.04

80.

040

(0.0

21)

(0.0

32)

(0.0

36)

(0.0

40)

(0.0

29)

Mot

her i

s a m

igra

nt0.

295

0.36

30.

461*

−0.

228

0.50

1(0

.213

)(0

.223

)(0

.272

)(0

.392

)(0

.462

)

Hou

seho

ldH

ouse

hold

size

−0.

006

0.02

80.

040

0.07

6−

0.11

4**

(0.0

34)

(0.0

40)

(0.0

41)

(0.0

72)

(0.0

56)

Dep

ende

ncy

ratio

−0.

066

−0.

037

−0.

037

−0.

131

−0.

033

(0.0

96)

(0.0

98)

(0.1

37)

(0.1

50)

(0.1

61)

Mig

ratio

n of

HH

mem

bers

0.01

00.

012

0.00

10.

035

−0.

003

(0.0

11)

(0.0

11)

(0.0

14)

(0.0

24)

(0.0

21)

Priv

ate

Toile

t0.

213

0.21

00.

093

0.26

2−

0.09

7(0

.251

)(0

.254

)(0

.387

)(0

.314

)(0

.494

)


250

Tabl

e 7A

.1

(con

tinue

d)

Dep

ende

nt v

aria

ble:

Z- s

core

WFA

AL

L (O

LS)

AL

L (I

V)

0–2$

>2–

4$>

4$

Hou

seho

ldTa

p w

ater

−0.

005

−0.

036

0.03

1−

0.44

0*0.

225

(0.1

38)

(0.1

41)

(0.2

28)

(0.2

54)

(0.2

09)

Vill

age

% H

Hs w

ith sa

nita

tion

0.00

4**

0.00

4**

0.00

7**

0.00

30.

003

(0.0

02)

(0.0

02)

(0.0

03)

(0.0

03)

(0.0

03)

Dist

ance

to to

wn

−0.

003

−0.

003

0.00

0−

0.00

2−

0.00

7*(0

.002

)(0

.002

)(0

.004

)(0

.004

)(0

.004

)V

illag

e in

com

e m

ean

(log)

−0.

386

−1.

064

−0.

105

−0.

865

0.06

5(0

.473

)(0

.655

)(0

.789

)(0

.755

)(0

.558

)U

bon

Prov

ince

dum

my

0.25

20.

216

−0.

064

0.58

2*0.

410*

(0.1

68)

(0.1

73)

(0.2

74)

(0.3

12)

(0.2

46)

Nak

hon

Phan

om P

rovi

nce

dum

my

0.14

00.

132

0.10

00.

515

0.24

5(0

.212

)(0

.212

)(0

.333

)(0

.377

)(0

.313

)20

08 y

ear d

umm

y−

0.07

5−

0.08

5−

0.20

00.

015

−0.

068

(0.1

26)

(0.1

31)

(0.2

13)

(0.2

82)

(0.2

41)

2010

yea

r dum

my

−0.

557*

**−

0.66

4***

−0.

908*

**−

0.72

4**

−0.

165

(0.1

64)

(0.1

86)

(0.2

93)

(0.3

02)

(0.2

52)

_con

s0.

365

−10

.929

−1.

453

7.43

30.

036

(3.5

79)

(8.2

96)

(5.2

88)

(5.6

45)

(4.7

10)

N13

7613

7656

736

344

6R

20.

200.

160.

180.

250.

27

Not

es:

* p

< 0

.1, *

* p

< 0

.05,

***

p <

0.0

1. S

tand

ard

erro

rs a

re c

lust

ered

on

indi

vidu

al le

vel.

Age

is c

ontr

olle

d fo

r and

sign

ifica

nt. I

V: a

sset

val

ue.

Sour

ce:

Ow

n ca

lcul

atio

ns b

ased

on

Hou

seho

ld su

rvey

s 200

7–20

10.


PART IV

Poverty and Inequity


253

8. Poverty and ethnicity in Asian countriesCarlos Gradín

1 INTRODUCTION

A large number of Asian countries have recently experienced significant economic growth that has led to an unprecedented reduction in poverty levels and to generally improved living conditions. In this context, however, it is crucial to investigate whether or not the benefits of this higher well- being have reached the entire population. An important issue is to know whether economic opportunities depend on given characteristics such as race or ethnicity because of the large history of economic and social disadvantage that many ethnic or indigenous groups face in many socie-ties, of which Asian countries have remarkable examples. These groups are more likely overrepresented in those segments of the population that might not be reached by economic growth if, for example, they lack the most demanded skills or live in inaccessible remote areas. This could be the consequence of them being historically denied access to the proper edu-cation and basic infrastructure that would allow them to take advantage of the greater economic opportunities, or it could be the consequence of segregation and wage discrimination in the labor market. Identifying the extent of the ethnic differentials in poverty is of extraordinary importance for implementing policies aimed at reducing this gap. Understanding its nature helps to evaluate what types of policies are expected to be more effective in closing the gap in each country.

The existence of ethnic and racial inequalities in well- being has long been an issue of concern all over the world but it has recently attracted considerable attention. This is the result of the combination of greater public concern about the situation of disadvantaged ethnic groups, and the growing availability of data and adequate methodologies for its research. Outstanding examples of this growing interest in the Asian and Pacific Region are the series of reports from the Asian Development Bank (ADB) (see ADB 2002), including analysis of the situation of ethnic groups in Cambodia, Indonesia, the Philippines, Viet Nam and the Pacific region,



or the books recently edited by Hall and Patrinos (2012) – including analysis for the People’s Republic of China (PRC), India, the Lao People’s Democratic Republic and Viet Nam – and Bhalla and Luo (2013) about India and the PRC. A number of papers have also been published analyz-ing the situation of particular groups, areas and countries or focusing on specific dimensions such as labor market performance or educational gaps. The introduction, in recent years, of regression- based decomposition analysis, previously developed in labor economics, has allowed a more in- depth investigation of the nature of those inequalities. Ethnic inequalities have already been documented in some Asian countries using any of these regression- based techniques.1 However, there has been very little compara-tive research so far on both the extent and the nature of ethnic inequalities in Asia to identify common and country- specific patterns.2 Very often, the analysis has focused on the mean gap only, ignoring the existence of pos-sible distributional patterns that make the disadvantaged poor differ from those of the most affluent.

For this reason, the aim of this chapter is to investigate the extent and the nature of the gap in poverty across ethnic groups in a selection of Asian countries. The emphasis on the comparative perspective and its focus on the poor are the main contributions of this chapter. Data come from a highly comparable demographic dataset that uses similar surveys across many developing countries. Individual economic status is approximated with a synthetic index of wealth defined as the weighted average of a series of indicators of assets, utilities and housing conditions and equipment. For the sake of greater comparability, we use the same indicators and esti-mate a common set of weights across the selected countries using multiple correspondence analysis (MCA).3 We undertake the analysis in two steps.

In a first stage, we measure cross- country variability in the ethnic gap in absolute and relative poverty rates. For that, we compute the ethnic poverty gap as the difference in poverty rates between two ethnic groups (comparison and reference) in each country along all possible poverty lines. When the poverty lines are the same levels of wealth in all countries, we call this the absolute ethnic poverty gap curve. When the poverty lines are wealth percentiles of the reference group in each country, in line with the interdistributional inequality approach (Butler and McDonald 1987; Le Breton et al. 2012), we call it the relative ethnic poverty gap curve.

In a second stage, we investigate the main factors determining the ethnic poverty gap in four countries among the possible competing explanations, using regression- based counterfactual analyses. By comparing the actual differential in poverty with that remaining when the comparison group is given the same distribution of characteristics of the reference group, we estimate the characteristics and coefficients effects of the ethnic poverty


Poverty and ethnicity in Asian countries 255

gap. The characteristics effect provides an idea of how much of a given poverty differential is explained by the disadvantaged group having greater prevalence of those attributes associated with lower wealth, what might be the result of discrimination, historical and cultural factors, and so on. For example, because their members generally have lower attained educa-tion, their households tend to have more children, or they live in the least developed rural areas. The coefficients effect quantifies to what extent these factors have a stronger association with wealth in some groups. That is, one ethnicity might be obtaining lower returns to education owing to prevailing wage discrimination in the labor market or because of the lower quality of the schools they attend. Similarly, one ethnic group might be more harmed by living in rural areas because of their poorer access to pro-ductive assets. This analysis is undertaken using the Gradín (2013, 2014) approach, based on the reweighting technique proposed by DiNardo et al. (1996) in the context of wage differentials.4

The rest of the chapter is organized as follows. Section 2 describes the data; section 3 outlines the methodology; and in sections 4 and 5, we report the empirical results on measuring the extent of the ethnic poverty gap and explaining this poverty gap, respectively. Finally, section 6 summarizes our main findings and presents conclusions.

2 DATA

For the empirical analysis, we use data from the Demographic and Health Survey (DHS). This is a standardized nationally representative household- based survey that collects a wide range of data on population, health and nutrition in many different developing countries in the world. The DHS is implemented under the Monitoring and Evaluation to Assess and Use Results Demographic and Health Surveys (MEASURE DHS) project, funded by the US Agency for International Development (USAID) and other international agencies. Since 1984, it has been implemented in over-lapping five- year phases (for example, DHS VI during 2008–13).5 We use the most recent data for those Asian countries with information on ethnic-ity. These include Azerbaijan (2006, DHS V), India (1998/99 DHS IV and 2005/06 DHS V), Nepal (2011 DHS VI), Pakistan (2006/07 DHS V), the Philippines (2003 DHS IV and 2008 DHS V), and Viet Nam (2005 DHS V).6

Demographic and Health Surveys are generally representative of the whole population for which they provide basic demographic and socioeconomic information.7 However, detailed information on other aspects, including ethnicity and labor market performance, is usually



only provided for eligible subsamples. The common eligible subsample in all datasets used in the analysis is ever- married women between 15 and 49 years old. Thus, this is the target group for the study, but using also information reported about their partners and other members of their households.

The study uses the ethnic groups defined according to the information available in DHSs for each country. The reference group is the wealthiest among the outstanding groups in each country and the comparison groups are the rest of the population, except some advantaged minorities. Given that sample sizes for individual groups are generally small, for most of the analysis, we pool disadvantaged ethnicities into one group in each country, but, in some cases, whenever the sample sizes allow, we analyze the situ-ation of outstanding groups. Sample sizes of the eligible subsamples are reported in Table 8A.3 in the Appendix to this chapter.

In Azerbaijan and Viet Nam, we distinguish the reference group as the majoritarian ethnicity (Azerbaijani and Vietnamese), whereas the comparison groups are the rest of the population, except Russian in Azerbaijan or Chinese in Viet Nam. In India and Nepal, ethnicity refers to caste or tribe. In India, the reference group refers to people not clas-sifying themselves as any of the traditionally disadvantaged groups rec-ognized as such by the Indian Constitution and protected by affirmative action policies: Scheduled Castes (SC), Scheduled Tribes (ST), and Other Backward Class (OBC). The latter groups, ST, SC and OBC, make up the comparison groups. In Nepal, the reference group is Hill Brahmin, a traditional elite caste in Hinduism, whereas the comparison groups are the rest of the castes: Hill Chhetri, Hill Dalit, Hill Janajati and Other. In Pakistan, ethnicity refers to the mother tongue, and the Urdu speak-ers are taken as the reference, with Punjabi, Pushto, Sindhi, Siraiki and Other being the comparison groups. In the Philippines, the reference is the major ethnic group, Tagalog, whereas the comparison groups include the major other ethnicities such as Cebuano, Ilocano, Ilonggo, and Other. Ethnicities in each country are listed in Table 8.1 reporting their shares of the eligible population. The shares of disadvantaged ethnic groups over the eligible population vary greatly across countries. Disadvantaged groups altogether are a minority of the population only in Azerbaijan (6 percent) and Viet Nam (16 percent), but make up the majority in the other countries, ranging from about 70 percent in India or the Philippines, to 92 percent in Pakistan.

In the literature on poverty, there are different ways to approximate indi-vidual well- being. The most common approach is to use income or expend-iture, although more multidimensional approaches have been gaining popularity in recent years. The DHSs do not include information on


257

Tabl

e 8.

1 E

thni

c gr

oups

by

coun

try

Cou

ntry

Ref

eren

ce g

roup

% e

ligib

leC

ompa

rison

gro

ups

% e

ligib

le

Aze

rbai

jan,

200

6A

zerb

aija

ni93

.9O

ther

(exc

ept R

ussia

n): T

allis

h, L

esgi

n, O

ther

.5.

6In

dia,

200

5/06

Non

e of

them

30.8

Sche

dule

d C

aste

(SC

)19

.6Sc

hedu

led

Trib

e (S

T)

8.6

Oth

er B

ackw

ard

Cla

ss (O

BC

)41

.1A

ll co

mpa

rison

gro

ups

69.2

Nep

al, 2

011

Hill

Bra

hmin

14.1

Hill

Chh

etri

19.5

Hill

Dal

it10

.0H

ill Ja

naja

ti24

.0O

ther

(Ter

ai B

rahm

in/C

hhet

ri, T

erai

Dal

it, T

erai

Ja

naja

ti, O

ther

Ter

ai C

aste

, New

ar, M

uslim

, Oth

er)

32.5

All

com

paris

on g

roup

s85

.9Pa

kist

an, 2

006/

07U

rdu

8.0

Punj

abi

41.3

Sind

hi10

.5Pu

shto

13.7

Sira

iki

15.6

Oth

er (H

indk

o, K

ashm

iri, P

ahar

i, Po

tow

ari,

Fars

i, B

aloc

hi, B

arau

hi, M

arw

ari,

Oth

er)

10.9

All

com

paris

on g

roup

s92

.0T

he P

hilip

pine

s,

2008

Taga

log

28.8

Ceb

uano

22.4

Iloca

no8.

4Ilo

nggo

9.7


258

Tabl

e 8.

1 (c

ontin

ued)

Cou

ntry

Ref

eren

ce g

roup

% e

ligib

leC

ompa

rison

gro

ups

% e

ligib

le

The

Phi

lippi

nes,

20

08O

ther

(Bic

olan

o, W

aray

, Kap

ampa

ngan

, A

klan

on, C

hava

kano

, Kan

kana

ey, P

angg

asin

ense

, Z

ambo

angu

eno,

Ant

ique

no, B

ohol

ano,

Ifug

ao, I

raya

, M

agui

ndan

awon

, Man

daya

, Mar

anao

, Mas

bate

no,

Taus

og, O

ther

)

30.7

All

com

paris

on g

roup

s71

.2V

iet N

am, 2

005

Vie

tnam

ese

85.3

Oth

er (e

xcep

t Chi

nese

): Ta

y, T

hai,

Khm

er, M

uong

, N

ung,

Phu

La,

E D

e, D

ao, C

ham

, Hm

ong,

Gia

Rai

, B

a N

a, X

o D

ang,

san

chay

(cao

lan

– sa

n ch

i), S

an

Diu

, Mno

ng, M

a, T

a O

i.

14.0

Not

e:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

.



income or expenditure, or on the market value of assets. The primary vari-able usually taken to capture people’s economic status is the DHS Wealth Index. This index is estimated using principal components analysis (PCA), based on all variables available in each sample describing household assets and utility services, plus whether there is a domestic servant and whether the household owns agricultural land. That is, wealth is computed as a weighted average of a number of categories, with weights obtained using the first dimension from the PCA. This approach has several advantages (Rutstein and Johnson 2004: 4): ‘It represents a more permanent status than does either income or consumption. In the form that it is used, wealth is more easily measured (with only a single respondent needed in most cases) and requires far fewer questions than either consumption expendi-tures or income.’ The authors also point out some evidence showing that the wealth index actually performed better than the traditional consump-tion expenditure index in explaining differences in educational attainment and attendance or in health outcomes (Filmer and Pritchett 2001; Rutstein and Johnson 2004).

In the presence of categorical variables, MCA is more appropriate to estimate economic status because PCA is designed for continuous vari-ables. Furthermore, the set of variables used to estimate DHS wealth are sample- specific, and so are the weights estimated separately for each sample. For the sake of cross- country comparability, we prefer an index estimated using a common set of variables. This necessarily means restricting the information used to construct the index to only those variables available in all datasets although the loss of information is small. Furthermore, we believe that using the same weights for all countries has the advantage of making cross- country comparisons of wealth and poverty easier to interpret. The use of country- specific weights, although raising comparability issues, is an appealing alternative but this choice turned out to be of little empirical relevance because the overall correla-tion is about 0.94, with also high correlation within countries.8 Thus, by using common weights we gain comparability and pay only a small price in terms of loss of information and how meaningful the weights are in each country.

For all these reasons, we estimate a new wealth index using MCA based on a common set of variables reflecting economic status in all countries (using the most recent sample data) involving common weights. Despite all these differences, the new index is highly correlated within countries with the DHS index: 0.80 in Azerbaijan, 0.88 in India, and around 0.96–0.97 in the other countries. However, in our view, the values of the new index better reflect cross- country differences in wealth. Instead of normalizing the index to have a mean of zero and standard deviation of one in each



sample (as in the DHS index), we normalize it to have a value between zero and one, reflecting the lowest and highest possible wealth profiles, respectively. The next section explains, in more detail, how we constructed this new index.

3 METHODOLOGY

3.1 The Composite Index of Wealth

In this subsection, we explain how we construct the wealth index using a set of categorical variables that associate with the economic status of a household. Note that we do not aim to construct an index of multidi-mensional poverty, which would call for using additional dimensions of well- being, but to estimate a proxy for the unobserved wealth (or economic status). Thus, the weights have no normative values, they just reflect the extent to which each category is associated with the latent economic status. For that, we use 17 variables that account for the conditions of dwellings (materials used in the roof, floor, and walls; and the number of people per room used for sleeping), basic assets owned by the household (such as vehicles and domestic appliances), cooking fuel, and type of access to water and sanitation. All the variables are categorical. The only originally non- categorical variable (the number of household members per sleeping room) has been discretized in different intervals. Given that this informa-tion refers to basic items, we expect the index to discriminate better among the poor than among the rich, which is consistent with our focus on poverty. These categorical variables are listed in Table 8.2, whereas Table

Table 8.2 Variables used to construct the wealth index

Source of drinking water

Has television Main floor material

Type of toilet facility Has refrigerator Main wall materialShare toilet with other households

Has bicycle Main roof material

Has electricity Has motorcycle/scooter

Household members / Rooms used for sleeping

Has telephone Has car/truck Type of cooking fuelHas radio Has an animal- drawn

cart

Note: See Table 8A.1 in the Appendix for more details.



8A.1 in the Appendix reports the distribution by country and comparison/reference groups across the different categories.

We estimate the index using data from the most recent sample for all six countries, in which each country has the same weight (1/6). This allows us to interpret differences in wealth values across countries as reflecting dif-ferences in their economic status using a common framework (an average of the selected countries).9

Let c1, . . ., cQ be the set of categorical variables associated with the eco-nomic status of a population of size N, where cq is coded with consecutive integers, 1, . . ., nq. Let Zq be the N × nq binary indicator matrix associated with cq, where Zq

ij 5 1 if and only if the qth categorical variable for the ith individual ciq = j. Let Z = (Z1, . . ., ZQ) be the N × J indicator matrix of the set of variables, where J = n1 + . . . +nQ is the total number of categories.

For each variable cq, we estimate scores (coordinates) sq1,. . .,sq

nq using the first extracted dimension with MCA. Let s 5 s1,. . .,sQ and s 5 s1,. . ., sQ be, respectively, the vectors with the highest and lowest scores associated with the Q categorical variables. Given that higher scores are associated with higher wealth, s and s represent the worst and best possible wealth profiles.

We define yi to be a wealth composite index that summarizes the eco-nomic status profile for the ith person as a weighted average of the cat-egories for this individual. The index is normalized to range between 0 and 1, the values corresponding to the worst and the best possible profiles, respectively.10 Thus, the weights represent the relative marginal contribu-tions to the individual wealth of being in each category, compared with being in the worst category, expressed as a proportion of the maximum possible contribution:

yi 5 aQ

q51anq

j51Zq

ijwqj , i = 1,.., N; with wq

j 5sq

j 2 sq

aQ

q51

(sq 2 sq). (8.1)

In particular, this means that the weights attached to the worst categories of each variable are all zero, whereas the weights attached to the best categories sum to one. Table 8A.1 in the Appendix reports the estimated scores and the corresponding normalized weights.11 Given that all categori-cal variables refer to the household, all individuals within a household will share the same wealth.

To analyze the evolution of poverty among ethnicities in the Philippines and in India, we also construct two new wealth indices based on the two- year pool for each of these countries. We do so because the information of earlier samples is more restrictive, so we cannot reproduce the same set of



variables used for the other samples. Thus, given that we are only interested in the time trend, we estimate country- specific indices using the common information in both years (see Table 8A.2 in the Appendix for the list of variables used).12

3.2 The Ethnic Poverty Gap Curves

To measure the extent to which disadvantaged ethnic groups tend to have more poverty than advantaged groups across countries, we first estimate (nonparametrically) their corresponding cumulative distribution functions (CDFs).

We call F0(y) and F1(y) the CDFs of wealth y [ [0,1] for the refer-ence (advantaged) and comparison (disadvantaged) groups.13 We define the absolute ethnic poverty gap curve, g(y) = F1(y) − F0(y), as the dif-ference in the cumulative distribution (headcount ratio) between the comparison and the reference group for each possible wealth level used as a poverty line. For example, g(.25) indicates the differential in poverty rates between both groups when the wealth poverty level is fixed at y = 0.25. We interpret g(y) as the ethnic differential in absolute poverty levels because the poverty threshold used is the same wealth level for all samples (across countries or over time). This curve is not invariant to changes in the scale of wealth. Then comparisons across samples would be influenced by differences in average wealth, for example, the differ-ential in poverty at lower wealth levels would tend to be higher in the poorest countries.

Similarly, we define the relative ethnic poverty gap curve, f(t) 5 F1 (F21

0 (t)) 2 t, where F210 (t) is the tth quantile of the reference

distribution, t [ [0,1] with F−1 denoting the quantile (right- inverse) func-tion attached to the distribution F. In other words, f(t) is the differential between the observed proportion of poor people in the comparison group for each quantile of the reference group taken as the poverty line, and the value one would expect if both groups had a similar distribution (that is, the proportion corresponding to the quantile). For example, f(.5) is the difference between the proportion of people in the comparison group below the median of the reference group and 50 percent (the value expected if both distributions were identical). This provides an idea of the differential in relative poverty because the wealth threshold used as a poverty line is country specific. It is indexed to the percentiles of the reference group in each country. Similarly, it is year specific in compari-sons over time.14 This makes the curve, and thus the comparison across samples, invariant to changes in the scale of wealth for all individuals in each sample.



The construction of f(t) is, in the spirit of the interdistributional Lorenz curve of the first type proposed by Butler and McDonald (1987), also called the first- order discrimination curve in the extended approach of Le Breton et al. (2012). This curve is a representation of the CDFs of the reference and comparison groups, �1 (t) 5 F0 (F21

1 (t)) , where the vertical distance between the 45° line and the interdistributional Lorenz curve, t − f1(t), is a measure of the economic disadvantage of members of the comparison group.15 In our context, we prefer the poverty line to be indexed to the reference group because then the wealth threshold used as a poverty line is the same for the various ethnicities in the country.

Note that, by construction, f(t) 5 g(F210 (t)) . For example,

f(.5) 5 g(p500 ) if p50

0 is the corresponding median of the reference group (see Figure 8.1). The difference between both curves is that in the cross- sample comparisons the differential is associated either with a common wealth threshold (absolute comparison) or with a sample- specific wealth threshold (a percentile of the corresponding reference group, relative com-parison). Both curves would be constructed by joining the points estimated nonparametrically at several values of respectively y and t.

Whenever the ethnic poverty gap is always non- negative, this means that F0 dominates F1 at the first- order of stochastic dominance. This has strong implications because it implies dominance in higher orders (cf. Foster and Shorrocks 1988a, 1988b) and means that, for whatever poverty line and index, among the Foster–Greer–Thorbecke (FGT) class, the reference group has more poverty than the comparison group. The FGT class of indices, in this context, can be written for any group and for any t [ [0,1], as:16

FGT(t;a) 51Na

N

i51cmax e F21

0 (t) 2 yi

F210 (t)

,0 f d a, a $ 0. (8.2)

.5

01

1

)

y

t

Figure 8.1 Illustration of g(y) and w(t)



3.3 Decomposing the Ethnic Poverty Gap

After estimating the poverty rates by ethnic group for different thresholds, we provide an aggregate decomposition of these gaps into the explained (characteristics effect) and unexplained (coefficients effect) parts. For this, we estimate a counterfactual distribution, in which we give members of the comparison group the same distribution of the relevant characteristics of the reference group, using the adaptation of a propensity- score technique (DiNardo et al. 1996) in Gradín (2014). This procedure also produces a detailed decomposition of the characteristics effect by quantifying the contribution to the gap by the different potential explicative factors men-tioned above (such as region, area, demographic structure, labor market performance and education).

We assume that each individual observation is drawn from some joint density function f over (y, x, g), where y indicates the vector of wealth, x is a vector of observed characteristics, and g identifies whether the individual is in the reference group (g = 0) or the comparison (g = 1) group.

The marginal distribution of wealth for group g is given by the density:

fg (y) ; f (y 0g) 5 3x

f (y, x 0g)dx 5 3x

f (y 0x, g) # f (x 0g)dx. (8.3)

This can be obtained as the product of two conditional distributions, where:

f (x 0g) ; 3y

f (y,x 0g)dy. (8.4)

In other words, each wealth density is determined by the marginal wealth density of members of the group having each combination of character-istics (a high level of education, living in the poorest regions, and so on) times the proportion of group members having this set of characteristics.

Then, we define the counterfactual wealth distribution fc(y) as the dis-tribution of y that would prevail if the comparison group kept their own conditional wealth distribution (the probability of having a certain wealth given their characteristics) but had the same characteristics (marginal distribution of x) of the reference group. We produce this counterfactual distribution by properly reweighting the actual wealth distribution of the comparison group:

fc(y) 5 3x

f (y 0x,g 5 1) # f (x 0g 5 0)dx



5 3x

f (y 0x,g 5 1) # yx# f (x 0g 5 1)dx 5 3

x

yxf (y,x 0g 5 1)dx (8.5)

where Fc is the corresponding CDF. Based on Bayes’s theorem, the reweighting scheme yx can be expressed as the product of two ratios:

yx 5f (x 0g 5 0)f (x 0g 5 1) 5

Prob(g 5 1)Prob(g 5 0)

Prob(g 5 0 0x)Prob(g 5 1 0x) (8.6)

where the ratio

Prob(g 5 1)Prob(g 5 0)

is constant and indicates the share of people that belongs to each group in the pooled sample with individuals from both groups. We estimate the ratio

Prob(g 5 0 0x)Prob(g 5 1 0x)

using the predictions from a logit model of the probability of belonging to the reference group, conditional on x, in the pooled sample.

In parallel with the conventional Oaxaca–Blinder procedure (see Blinder 1973; Oaxaca 1973), widely used in labor economics to estimate wage discrimination, we add and subtract the counterfactual distribution to produce the following decomposition of the relative ethnic poverty gap:

f(t)5F1 (F210 (t))2t5[F1 (F21

0 (t))2Fc(F210 (t))]1[Fc(F21

0 (t))2t ]. (8.7)

The first term in the last expression represents the part of the poverty dif-ferential by ethnicity that is explained by characteristics (or characteristics effect) because it measures the change in poverty owing to shifting the dis-tribution of characteristics (after reweighting the comparison group). The second part is the unexplained part (or coefficients effect) because it is the gap that remains when both the comparison and the reference group have the same distribution of characteristics but differ in the conditional wealth distributions. Given the correspondence between f(t) and g(y) discussed above, the same decomposition applies to the latter.

In the detailed decomposition, we quantify the impact on the poverty differential of changes in a single covariate (or set of covariates) xj instead of the whole vector. For that, we use the Shapley decomposition that



results from averaging over all possible sequences of factors (Chantreuil and Trannoy 2013; Shorrocks 2013).

For example, to compute the contribution of education, we have to estimate first the impact of education when it is the only factor equalized between both groups. That is, we estimate the gap between the comparison group and the counterfactual when the latter is estimated using only the coefficients of education- related variables in the logit regression (while the rest of the coefficients are replaced by zeros). To estimate the contribution of education when it is the second equalized factor, we need to measure the gap between the counterfactual in which we only use the coefficients of education jointly with another factor (for example, region), and the coun-terfactual using only the coefficients of this other factor. Then, we repeat the same exercise replacing region by each of the other three factors (area, demographics, and labor variables). Similarly, we estimate the contribution of education when it is the third, fourth and fifth factor equalized between both groups. The overall contribution of education is the average of all these estimated contributions. Using this same procedure, we compute the contributions of each of the five factors. The resulting individual effects are path independent and add up to the overall effect.17

4 ESTIMATION OF ETHNIC POVERTY GAPS

4.1 Ethnic Differences in Mean Wealth

First of all, Table 8.3 reports the estimates and standard errors of the mean and median values of the wealth index (ranging between 0 and 1) in each country for the entire population. Table 8.3 also shows the values for the eligible subpopulation (15–49- year- old ever- married women), which are very similar to the estimates for the population. There is a large difference between Azerbaijan, where the population is, on average, at 0.76, and the rest of the countries. Among them, India and Nepal are the poorest, below 0.4, whereas Viet Nam and the Philippines are richer, about 0.56, with Pakistan in the middle, 0.49. Table 8.3 also reports average and median wealth values for the comparison and the reference groups within the eligible subpopulations. In all six countries, the mean values of wealth for the disadvantaged groups are less than those of the corresponding reference groups, although the magnitude of the ethnic wealth gap differs across countries. It is just 0.044 in Azerbaijan, but rises to 0.262 in Pakistan or 0.205 in Viet Nam. With intermediate levels of this gap, we find the Philippines, India and Nepal, respectively, at 0.121, 0.154 and 0.170. It is interesting to note that the Pakistani Urdu report a


267

Tabl

e 8.

3 M

ean

and

med

ian

weal

th b

y co

untr

y an

d gr

oup

Cou

ntry

Mea

n w

ealth

Med

ian

wea

lth

All

Elig

ible

popu

latio

nC

ompa

rison

grou

pR

efer

ence

grou

pA

llE

ligib

lepo

pula

tion

Com

paris

onG

roup

Ref

eren

cegr

oup

Aze

rbai

jan,

200

60.

762

0.76

50.

722

0.76

70.

772

0.77

40.

721

0.77

7(0

.001

)(0

.002

)(0

.006

)(0

.002

)(0

.001

)(0

.003

)(0

.007

)(0

.003

)In

dia,

200

5/06

0.38

880.

394

0.34

60.

501

0.34

20.

350

0.29

60.

517

(0.0

004)

(0.0

01)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

02)

(0.0

01)

(0.0

03)

Nep

al, 2

011

0.38

80.

396

0.37

20.

542

0.36

60.

376

0.34

70.

535

(0.0

01)

(0.0

02)

(0.0

03)

(0.0

05)

(0.0

01)

(0.0

03)

(0.0

04)

(0.0

08)

Paki

stan

, 200

6/07

0.49

40.

497

0.47

60.

738

0.50

60.

511

0.48

00.

772

(0.0

01)

(0.0

03)

(0.0

03)

(0.0

05)

(0.0

02)

(0.0

05)

(0.0

05)

(0.0

05)

The

Phi

lippi

nes,

2008

0.56

50.

568

0.53

30.

654

0.58

80.

591

0.55

00.

673

(0.0

01)

(0.0

02)

(0.0

02)

(0.0

03)

(0.0

01)

(0.0

03)

(0.0

03)

(0.0

04)

Vie

t Nam

, 200

50.

561

0.56

50.

388

0.59

30.

553

0.55

60.

377

0.58

3(0

.001

)(0

.003

)(0

.006

)(0

.003

)(0

.001

)(0

.003

)(0

.008

)(0

.005

)

Indi

an- s

peci

fic w

ealth

inde

x19

98/9

90.

2989

0.30

00.

250

0.37

90.

244

0.24

50.

195

0.35

3(0

.000

3)(0

.001

)(0

.001

)(0

.001

)(0

.001

)(0

.001

)(0

.001

)(0

.002

)20

05/0

60.

3381

0.34

40.

297

0.45

00.

288

0.29

60.

245

0.45

3(0

.000

4)(0

.001

)(0

.001

)(0

.002

)(0

.001

)(0

.001

)(0

.001

)(0

.003

)

Phi

lippi

ne- s

peci

fic w

ealth

inde

x20

030.

440

0.44

30.

390

0.56

80.

438

0.44

30.

366

0.58

0(0

.001

)(0

.003

)(0

.003

)(0

.004

)(0

.001

)(0

.004

)(0

.004

)(0

.007

)20

080.

493

0.49

60.

453

0.60

20.

507

0.51

10.

463

0.61

3(0

.001

)(0

.003

)(0

.003

)(0

.004

)(0

.001

)(0

.004

)(0

.004

)(0

.006

)

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1. S

tand

ard

erro

rs in

par

enth

eses

.



median wealth similar to that of the Azerbaijani, despite the large mean differential in wealth between their two countries.

The lower panel of Table 8.3 reports the average and median values using country- specific wealth indices for India and the Philippines to analyze the trends over time. It shows that there was a large improvement in the average and median wealth of people living in both countries, compared with the level in the previous survey. The increase in the average wealth was larger for the reference group in India and for the comparison group in the Philippines. As a consequence, the ethnic gap in average wealth increased in the former (from 0.129 to 0.152), whereas it decreased in the latter (from 0.178 to 0.150).

4.2 The Distribution of Wealth by Groups

The inter- ethnic difference in average wealth is a summary measure of the economic disadvantage of one group over another. But the information it provides is limited because it does not take into consideration how wealth is concentrated within the two populations. In this context, it is much more informative to consider the entire distribution of the comparison and ref-erence groups. This is what we do in this subsection.

Figure 8.2 displays the nonparametric densities of wealth estimated separately for the reference and comparison groups in each country. It is clear that, in every country, there is an unequal distribution of wealth along ethnic lines, with the reference group being generally overrepresented at the upper end of the wealth index. In some cases, the distributions are very different, as if they were obtained from two different countries. Disadvantaged ethnic groups tend to be overrepresented at the lowest levels of wealth. The exception to this is Azerbaijan, where both groups are concentrated at the upper end of the wealth index. The distribution of the reference groups are generally to the left of the comparison groups, although in India, there is a high within- group heterogeneity, with a large proportion of the reference group (those not ST/SC/OBC) having low wealth levels as well.

Figure 8.3 displays the corresponding CDFs. It shows the headcount ratios (the share of each group’s poor population) for every possible poverty line. In every country, the cumulative distribution of the compari-son group tends to be above that of the reference group.18 This means that there generally is first- order stochastic dominance that, as we know, also implies higher- order stochastic dominance. As a result, for a large range of poverty lines and all members of the FGT class of poverty indices, poverty will be systematically higher among disadvantaged ethnic groups, although the intensity at which this happens varies across countries. We analyze this in more detail in the next subsection.



4.3 The Absolute Ethnic Poverty Gap Curves in Six Countries

The comparison of inter- ethnic poverty across countries is better sum-marized in Figure 8.4, which displays each country’s absolute and relative

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5

Den

sity

Azerbaijan, 2006

Wealth index

Other ethnicities Azerbaijani

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5India, 2005/06

Wealth index

ST/SC/OBC None of them

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5

Den

sity

Nepal, 2011

Wealth index

Other ethnicities Hill Brahmin

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5Pakistan, 2006/07

Wealth index

Other ethnicities Urdu

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5

Den

sity

Philippines, 2008

Wealth index

Other ethnicities Tagalog

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5Viet Nam, 2005

Wealth index

Other ethnicities Vietnamese

Notes: Eligible population: ever- married women, 15–49 years old. Ethnic groups as listed in Table 8.1. Nonparametric densities with adaptive optimal bandwidth and Gaussian kernels.

Figure 8.2 Wealth densities by ethnic groups in six countries



poverty gap curves, g(y) and f(t). On the left graph, the absolute ethnic poverty gap curve g(y) projects the differential between the poverty rates of the comparison and reference groups for each possible wealth cutoff. Which country shows the largest ethnic poverty gap depends on the specific

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0C

DF

Azerbaijan, 2006

Wealth index

0.1

0.3

0.5

0.7

0.9

Other ethnicities Azerbaijani ST/SC/OBC None of them

Other ethnicities Urdu

Other ethnicities Vietnamese

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0India, 2005/06

Wealth index

0.1

0.3

0.5

0.7

0.9

Other ethnicities Hill Brahmin

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0

CD

F

Nepal, 2011

Wealth index

0.1

0.3

0.5

0.7

0.9

Other ethnicities Tagalog

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0

CD

F

CD

FC

DF

CD

F

Philippines, 2008 Viet Nam, 2005

Wealth index

0.1

0.3

0.5

0.7

0.9

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0Pakistan, 2006/07

Wealth index

0.1

0.3

0.5

0.7

0.9

00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

0.8

1.0

Wealth index

0.1

0.3

0.5

0.7

0.9


Figure 8.3 Wealth CDFs by ethnic groups in six countries



threshold used. Nepal shows the largest ethnic gap in severe poverty, up to a level of wealth of about 0.370. Then, the differential becomes largest in Viet Nam (until wealth is about 0.545), and after that level in Pakistan (up to 0.849). Azerbaijan joins Pakistan with the largest poverty differential only when the threshold is fixed at the very top of the wealth distribution, which does not seem very reasonable for a poverty line.

The largest differential in poverty rates is as much as 50 percent-age points in Viet Nam and Pakistan, about 40 in Nepal and 30 in the

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

Eth

nic

pove

rty

gap

Wealth index

0.1

0.3

0.5

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.2

0.4

0.6

Eth

nic

pove

rty

gap

CDF reference group

0.1

0.3

0.5

Azerbaijan, 2006Pakistan, 2006/07

India, 2005/06Philippines, 2008

Nepal, 2011Viet Nam, 2005


Figure 8.4 Ethnic poverty gap curves g(y) and �(t) across six countries



Philippines and India. The lowest, 20 percentage points, is the maximum achieved in Azerbaijan.

4.4 The Relative Ethnic Poverty Gap Curves across Countries

The previous comparison of absolute poverty differentials across the six countries is obviously influenced by their differences in average wealth. For that reason, the graph on the right in Figure 8.4 displays the relative ethnic poverty gap curve, f(t), the same ethnic poverty gap as a function of each percentile of the reference group. This is a better measure of rela-tive poverty or how well the comparison groups in each country perform relative to their reference groups. We can distinguish basically three clubs of countries in terms of the level of the relative ethnic poverty gap. Azerbaijan stands out for having the smallest differential among the six countries all over the distribution of the corresponding reference group. Thus, this country shows the smallest ethnic differential in both absolute and relative poverty. Below the median of the reference group, the relative ethnic gap in poverty is the largest in Pakistan, Viet Nam and Nepal. India and the Philippines show intermediate levels. Above the median, Nepal tends to converge with the latter countries.

Poverty indices of the FGT family (for a = 0, 1, and 2) computed using the tenth, twenty- fifth and fiftieth percentiles of the reference group (t =0.1, 0.25, 0.5) as poverty lines, are reported for all groups in Table 8.4. By definition, the FGT(0) or headcount ratio is 10 percent, 25 percent and 50 percent, respectively, for the reference group in each country. Thus, the gap depends on by how much the comparison groups deviate from those figures. The FGT(1), poverty gap ratio, additionally takes into account the average gap in wealth between the poor and non- poor in each case. The FGT(2) also incorporates inequality among the poor. However, both indices provide a significantly similar picture of the gap (in some cases exacerbating the inter- ethnic differentials). For simplicity, we concentrate on the gap in the headcount ratio from now on.

4.5 The Ethnic Poverty Gap Curves for Outstanding Groups in Selected Countries

The situation described above in the selection of countries conceals a high degree of heterogeneity within disadvantaged ethnic groups in each country, which is explored in Figure 8.5, displaying the CDFs and the corresponding ethnic poverty gaps for outstanding groups in India, Nepal, Pakistan and the Philippines. In these four countries, the ethnic poverty gaps tend to be systematically higher for some groups. India is


273

Tabl

e 8.

4 Po

vert

y m

easu

res b

y co

untr

y an

d et

hnic

gro

up fo

r diff

eren

t qua

ntile

s of

the

refe

renc

e gr

oup

Cou

ntry

FG

T(0

)H

ead

coun

t rat

ioF

GT

(1)

Pove

rty

gap

ratio

FG

T(2

)Se

verit

y of

pov

erty

10th

25th

50th

10th

25th

50th

10th

25th

50th

Aze

rbai

jan

2006

Ref

eren

ce: A

zerb

aija

ni10

.025

.050

.01.

02.

66.

40.

160.

461.

28(0

.5)

(0.8

)(1

.0)

(0.1

)(0

.1)

(0.2

)(0

.02)

(0.0

3)(0

.05)

Com

paris

on:

M

inor

ities

16.4

38.7

69.9

1.6

4.3

9.9

0.3

0.8

2.1

(2.0

)(3

.0)

(2.9

)(0

.3)

(0.4

)(0

.6)

(0.1

)(0

.1)

(0.2

)

Indi

a 20

05–0

6R

efer

ence

: Non

e

of th

em10

.025

.050

.03.

29.

120

.91.

44.

611

.7(0

.3)

(0.4

)(0

.5)

(0.1

)(0

.2)

(0.3

)(0

.1)

(0.1

)(0

.2)

Com

paris

on23

.651

.277

.47.

520

.639

.43.

310

.924

.3(0

.3)

(0.3

)(0

.3)

(0.1

)(0

.2)

(0.2

)(0

.1)

(0.1

)(0

.2)

SC25

.853

.480

.48.

822

.341

.54.

012

.126

.0(0

.6)

(0.6

)(0

.4)

(0.2

)(0

.3)

(0.3

)(0

.1)

(0.2

)(0

.3)

ST39

.374

.990

.611

.332

.153

.94.

716

.935

.5(0

.8)

(0.6

)(0

.4)

(0.3

)(0

.4)

(0.4

)(0

.2)

(0.3

)(0

.3)

OB

C19

.345

.273

.16.

017

.435

.42.

69.

021

.1(0

.3)

(0.4

)(0

.3)

(0.1

)(0

.2)

(0.3

)(0

.1)

(0.1

)(0

.2)

Nep

al 2

011

Ref

eren

ce: H

ill

B

rahm

in10

.025

.050

.02.

15.

012

.00.

71.

74.

2(1

.0)

(1.6

)(1

.7)

(0.3

)(0

.4)

(0.6

)(0

.1)

(0.2

)(0

.3)

Com

paris

on48

.663

.977

.518

.726

.235

.69.

514

.220

.6(0

.7)

(0.7

)(0

.6)

(0.4

)(0

.4)

(0.4

)(0

.3)

(0.3

)(0

.3)


274

Tabl

e 8.

4 (c

ontin

ued)

Cou

ntry

FG

T(0

)H

ead

coun

t rat

ioF

GT

(1)

Pove

rty

gap

ratio

FG

T(2

)Se

verit

y of

pov

erty

10th

25th

50th

10th

25th

50th

10th

25th

50th

Nep

al 2

011

Hill

Chh

etri

37.4

55.1

71.8

12.6

19.3

28.7

5.8

9.4

15.0

(1.3

)(1

.4)

(1.4

)(0

.5)

(0.6

)(0

.7)

(0.3

)(0

.4)

(0.5

)H

ill D

alit

55.5

74.7

87.3

20.0

29.0

40.1

9.8

15.1

22.6

(1.9

)(1

.8)

(1.3

)(0

.9)

(1.0

)(1

.0)

(0.6

)(0

.7)

(0.8

)H

ill Ja

naja

ti42

.659

.474

.815

.122

.231

.87.

311

.417

.4(1

.4)

(1.4

)(1

.3)

(0.6

)(0

.7)

(0.8

)(0

.4)

(0.5

)(0

.6)

Oth

er57

.769

.379

.924

.632

.441

.213

.218

.825

.7(1

.4)

(1.2

)(1

.0)

(0.8

)(0

.8)

(0.8

)(0

.6)

(0.6

)(0

.7)

Paki

stan

200

6R

efer

ence

: Urd

u10

.025

.050

.01.

94.

48.

40.

61.

52.

6(1

.3)

(2.1

)(2

.7)

(0.3

)(0

.5)

(0.6

)(0

.1)

(0.2

)(0

.3)

Com

paris

on58

.575

.788

.924

.832

.739

.114

.019

.123

.2(0

.7)

(0.6

)(0

.5)

(0.4

)(0

.4)

(0.4

)(0

.3)

(0.3

)(0

.3)

Punj

abi

47.6

69.2

85.9

16.0

23.9

30.8

7.9

12.0

15.7

(1.1

)(1

.1)

(1.0

)(0

.5)

(0.5

)(0

.5)

(0.3

)(0

.4)

(0.4

)Si

ndhi

68.8

83.1

92.7

34.4

42.3

48.2

21.3

27.1

31.5

(1.8

)(1

.5)

(1.2

)(1

.2)

(1.2

)(1

.1)

(0.9

)(1

.0)

(1.0

)Pu

shto

59.9

75.5

88.8

24.7

32.6

39.1

13.0

18.3

22.6

(1.8

)(1

.5)

(1.2

)(1

.0)

(1.0

)(1

.0)

(0.6

)(0

.7)

(0.8

)


275

Sira

iki

75.4

87.5

95.3

35.7

44.4

50.5

21.2

27.7

32.5

(1.7

)(1

.3)

(0.8

)(1

.0)

(1.0

)(1

.0)

(0.8

)(0

.9)

(0.9

)O

ther

63.6

76.2

87.8

33.5

40.3

45.7

21.5

26.8

30.7

(1.8

)(1

.7)

(1.4

)(1

.2)

(1.2

)(1

.2)

(0.9

)(1

.0)

(1.0

)

The

Phi

lippi

nes 2

008

Ref

eren

ce: T

agal

og10

.025

.050

.02.

45.

29.

80.

91.

93.

3(0

.7)

(1.0

)(1

.2)

(0.2

)(0

.3)

(0.4

)(0

.1)

(0.1

)(0

.2)

Com

paris

on34

.455

.075

.110

.216

.823

.94.

47.

611

.1(0

.6)

(0.6

)(0

.6)

(0.2

)(0

.2)

(0.3

)(0

.1)

(0.2

)(0

.2)

Ceb

uano

34.1

54.3

75.5

9.6

16.2

23.4

3.9

7.1

10.5

(1.0

)(1

.1)

(1.0

)(0

.4)

(0.4

)(0

.5)

(0.2

)(0

.3)

(0.3

)Ilo

cano

23.0

49.4

70.1

5.7

11.4

18.5

2.2

4.3

7.2

(1.3

)(1

.9)

(1.7

)(0

.4)

(0.6

)(0

.7)

(0.2

)(0

.3)

(0.4

)Ilo

nggo

32.2

53.3

73.7

9.0

15.6

22.6

3.8

6.8

10.1

(1.7

)(1

.9)

(1.7

)(0

.6)

(0.7

)(0

.8)

(0.3

)(0

.4)

(0.5

)O

ther

38.5

57.6

76.5

12.3

19.2

26.2

5.6

9.1

12.8

(0.9

)(1

.0)

(0.9

)(0

.4)

(0.4

)(0

.5)

(0.2

)(0

.3)

(0.3

)

Vie

t Nam

200

5R

efer

ence

: Vie

tnam

ese

10.0

24.9

50.0

1.9

5.1

9.9

0.6

1.7

3.2

(0.6

)(0

.9)

(1.0

)(0

.2)

(0.3

)(0

.3)

(0.1

)(0

.1)

(0.2

)C

ompa

rison

:

Min

oriti

es53

.977

.492

.115

.826

.134

.96.

712

.217

.2(2

.3)

(2.0

)(1

.1)

(1.0

)(1

.1)

(1.0

)(0

.6)

(0.7

)(0

.8)

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Fos

ter–

Gre

er–T

horb

ecke

val

ues m

ultip

lied

by 1

00. E

thni

c gr

oups

as l

isted

in

Tabl

e 8.

1. B

oots

trap

s sta

ndar

d er

rors

in p

aren

thes

es (3

00 re

plic

atio

ns).


276

CDF

Wea

lth in

dex

Indi

a, 2

005/

06

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Sch

edul

ed c

aste

Oth

er b

ackw

ard

clas

s

Sch

edul

ed tr

ibe

Non

e of

them

Ethnic poverty gap

Wea

lth in

dex

Indi

a, 2

005/

06

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Sch

edul

ed c

aste

Oth

er b

ackw

ard

clas

s

Sch

edul

ed tr

ibe

Ethnic poverty gap

CD

F r

efer

ence

gro

up

Indi

a, 2

005/

06

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Sch

edul

ed c

aste

Oth

er b

ackw

ard

clas

s

Sch

edul

ed tr

ibe

Wea

lth in

dex

CDF

Nep

al, 2

011

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Hill

Chh

etri

Oth

er

Hill

Dal

it

Bra

hmin

Hill

Jan

anja

ti

Wea

lth in

dex

Ethnic poverty gap

Nep

al, 2

011

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Hill

Chh

etri

Oth

er

Hill

Dal

itH

ill J

anan

jati

CD

F r

efer

ence

gro

up

Ethnic poverty gap

Nep

al, 2

011

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Hill

Chh

etri

Oth

er

Hill

Dal

itH

ill J

anan

jati


277

Wea

lth in

dex

CDF

Phi

lippi

nes,

200

8

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Ceb

uano

Oth

er

Iloca

no

Tag

alog

Ilong

go

Wea

lth in

dex

Ethnic poverty gap

Phi

lippi

nes,

200

8

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Ceb

uano

Oth

er

Iloca

noIlo

nggo

Wea

lth in

dex

CDF

Pak

ista

n, 2

006/

07

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Pun

jabi

Sira

iki

Sin

dhi

Oth

erU

rdu

Pus

hto

Wea

lth in

dex

Ethnic poverty gap

Pak

ista

n, 2

006/

07

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Pun

jabi

Sira

iki

Sin

dhi

Oth

er

Pus

hto

CD

F r

efer

ence

gro

up

Ethnic poverty gap

Pak

ista

n, 2

006/

07

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Pun

jabi

Sira

iki

Sin

dhi

Oth

erU

rdu

Pus

hto

CD

F r

efer

ence

gro

up

Ethnic poverty gap

Phi

lippi

nes,

200

8

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Ceb

uano

Oth

er

Iloca

noIlo

nggo

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1. N

onpa

ram

etric

den

sitie

s with

ada

ptiv

e op

timal

ban

dwid

th a

nd G

auss

ian

kern

els.

Figu

re 8

.5

Cum

ulat

ive

dist

ribu

tion

func

tions

and

eth

nic

pove

rty

gap

curv

es, g

(y)

and

�(t

), fo

r gro

ups i

n fo

ur c

ount

ries



probably the country with the largest diversity among ethnic groups; poverty tends to be substantially larger for Scheduled Tribes, achieving a differential with the reference group of 50 percentage points, followed by Scheduled Castes (30 percentage- point differential at its maximum), with Other Backward Class showing the smallest gap with respect to the population classifying as not being in any of these groups. In Nepal, the gap tends to be largest for most poverty lines for Hill Dalit and Other groups (also reaching 50 percentage points) than for Hilt Chhetri or Hill Janajati. In Pakistan, Punjabi generally show smaller poverty rates than Sindhi, Siraiki, and other groups (whose maximum ethnic poverty gap is about 60 percentage points), with Pushto having intermediate gaps. In the Philippines, the heterogeneity in ethnic poverty gaps is the smallest among the six countries; the gap for Ilocano tends to be the smallest, whereas the gap for those in the ‘Other’ category tends to be the largest.

The situation of selected ethnic groups (that tend to have largest ethnic poverty gaps) across countries is summarized in Figure 8.6. It reveals that Indian Scheduled Tribes face the largest absolute poverty gap among all considered ethnic groups in this study for a large range of low poverty thresholds, although its relative gap is smaller for lower percentiles as a result of the large proportion of poor people in the reference group. Pakistani Siraiki report the largest ethnic gap at higher levels of wealth (also at extremely low levels), and for the relative ethnic poverty gap. Vietnamese minorities and Nepalese Hill Dalit also show ethnic poverty gaps substantially larger than most disadvantaged ethnicities in the Philippines and Azerbaijan.

4.6 Trends in the Ethnic Poverty Gaps for India and the Philippines

To grasp the evolution of the ethnic poverty gap over time, Figure 8.7 reproduces the previous analysis for India and the Philippines in two different years (respectively, 1998–99 to 2005–06 and 2003–08). In these periods of strong growth in average wealth levels, both countries followed divergent trends. Both the absolute and the relative ethnic poverty gaps were generally lower in the Philippines. Especially relevant is the reduc-tion of the ethnic gap in severe poverty. However, although there was also a reduction in the ethnic gap in severe poverty in India, this was much smaller and contrasts with an increase when we use higher poverty lines (above 0.2) and a relative approach, indicating that the improvement in wealth was larger for the reference group than for the comparison group along the entire distribution of wealth. Figure 8.8 shows the change in the ethnic poverty gap in both countries for the most outstanding groups, and



reveals that the reduction in the ethnic poverty gap benefited all Filipino disadvantaged ethnicities but especially the Ilonggo. In India, the increase in the ethnic poverty gap was largest for the Scheduled Tribes, thus aggra-vating the relative situation of the most disadvantaged group. Similarly, the reduction in the ethnic gap in extreme poverty was largest for the Scheduled Castes.

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Eth

nic

pove

rty

gap

Wealth index

0

0

0.2

0.4

0.6

0.1

0.3

0.5

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Eth

nic

pove

rty

gap

CDF reference group

0

0

0.2

0.4

0.6

0.1

0.3

0.5

AzerbaijanPakistan (Siraiki)

India (ST)Philippines (Other)

Nepal (Hill Dalit)Viet Nam


Figure 8.6 Ethnic poverty gap curves, g(y) and �(t), for specific groups in the six countries


280

1998

/99

2005

/06

CDF

Wea

lth in

dex

Indi

a

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

ST

/SC

/OB

C 1

998/

99

Non

e of

them

199

8/99

ST

/SC

/OB

C 2

005/

06

Non

e of

them

200

5/06

Min

oriti

es 1

998/

99

Tag

alog

199

8/99

Min

oriti

es 2

005/

06

Tag

alog

200

5/06

Ethnic poverty gap

Wea

lth in

dex

Indi

a

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

2003

2008

2003

2008

1998

/99

2005

/06

Ethnic poverty gap

CD

F r

efer

ence

gro

up

Indi

a

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Wea

lth in

dex

Ethnic poverty gap

Phi

lippi

nes

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

CD

F r

efer

ence

gro

up

Ethnic poverty gap

Phi

lippi

nes

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

Wea

lth in

dex

CDF

Phi

lippi

nes

0

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1. N

onpa

ram

etric

den

sitie

s with

ada

ptiv

e op

timal

ban

dwid

th a

nd G

auss

ian

kern

els.

Figu

re 8

.7

Eth

nic

pove

rty

gap

tren

ds fo

r Ind

ia a

nd th

e P

hilip

pine

s



5 EXPLAINING THE ETHNIC POVERTY GAP

5.1 Competing Explanations

The previous section shows that there are substantial poverty gaps by ethnicity in Asian countries. We look at what might be

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Cha

nge

in e

thni

c po

vert

y ga

p

Wealth index

Scheduled caste Scheduled tribeOther backward class

India, 1998/99–2005/06

0

–1

1.0

–0.05

0

0.05

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Cha

nge

in e

thni

c po

vert

y ga

p

Wealth index

Cebuano Ilocano Ilonggo

Other

Philippines, 2003–08

0

–1

1.0

–0.05

0

0.05


Figure 8.8 Change in the ethnic poverty gap curve g(y) by groups in India and the Philippines



the determinants of those gaps in four countries. There are a few possible explanations.

First, there is one possible geographical explanation. Some ethnic groups live in the least- developed provinces of their countries or in rural or moun-tainous areas, in which people’s wealth is generally lower regardless of their ethnicity. A second possible explanation comes from disadvantaged ethnic groups having different demographic structures, for example, with more children or elderly people in their households as the consequence of higher fertility rates or migration flows. By increasing their needs, this reduces their ability to accumulate wealth. A third possible explanation is socioeconomic; it comes from the different levels of education and performance in the labor market. Disadvantaged groups might have lower attained education or a weaker attachment to the labor market, significantly reducing their ability to earn income. All these explanations have in common that disadvantaged groups have ‘worst’ characteristics, that is, a higher prevalence of those attributes typically associated with higher poverty, either because they imply lower income or higher needs. Note that in some cases the causality might go in both directions, giving that higher poverty of one group, for example, might also help to explain its higher fertility rates or its lower school enrollment.

Alternatively, higher poverty of some ethnic groups might be the direct consequence of unobserved factors such as earnings discrimination in the labor market, or the lower quality of some attributes, such as education or location (for example, living in more inaccessible rural areas), producing lower returns in the labor market. In the conventional analysis of wage differentials, wage discrimination is usually identified as being part of the unobserved gap (or coefficients effect), once wage gaps coming from inter- group differences in productivity have been already considered. However, it is important to note that discrimination might affect higher poverty either directly by reducing the returns to their characteristics (captured by the unexplained or coefficients effect) or, indirectly, through the accumulation of lower education, exclusion from the labor market, lack of geographi-cal mobility, and so on (the characteristics effect). That is, discrimination might be at the root of the lower endowments that ultimately explain the ethnic poverty gap. Disadvantaged groups might live in remote areas as the consequence of their traditional communities being historically denied basic infrastructure by the government, or them being excluded from the most profitable lands. Disadvantaged groups might have higher fertility rates, not as the consequence of having different cultural views about family, but different access to family planning. Also, they might have lower education and labor- force participation as the result of their lower oppor-tunities for schooling or their anticipation of segregation and lower returns in the labor market owing to discrimination.



In this section, we aim at disentangling what explanations (geographi-cal, demographic, socioeconomic, or unobserved factors) are relevant and significant in explaining the poverty gaps for the selected Asian countries. More detailed research on which of these are the result of discrimination or how, and by how much, they are producing this, is beyond the scope of this chapter and needs a much more thorough country- specific analysis.

For our purposes, we include several variables that might determine household wealth as explanatory variables in the logit regression. We measure location by a dummy variable indicating whether the area is urban or rural, and by the region of residence. In Nepal, region refers to each of the 13 subregions. In India and in the Philippines, states and provinces, respectively, were grouped by deciles according to their average wealth. Similarly, districts in Pakistan were grouped into wealth quartiles.

We also consider some demographic factors such as marital status (currently versus formerly married), teenage marriage (if age of first mar-riage was below 18), household type (two related adults, three or more related adults, rest of the households), the number of household members, the number of children below 5 years old in the household, and the total number of living children. Age is collected for each individual and the householder (also its squared value). The sex of the latter is also included. Immigration status reflects whether the individual was immigrant or not, and, in the affirmative case, whether they arrived less than five years ago, and were from rural or urban areas. Education is captured by the completed level of education (incomplete primary, primary, incomplete secondary, secondary, or higher) for the householder, for each eligible individual, and for their partner, in the case of married women. Individual literacy is also considered. Regarding labor- related variables, we use infor-mation about occupation (1- digit level) for each eligible individual (and their partner), whether they worked during the last 12 months, or had a non- paid job.19 All other factors, including direct wage discrimination or differences in the quality of education, are captured by the unobserved component that remains unexplained. Summary statistics of the explana-tory variables, and regression coefficients and standard errors of the logit probability, estimated for reweighting the comparison’s distribution, are reported in Tables 8A.4–8A.7 in the Appendix.

5.2 Decomposition of the Ethnic Poverty Gap

We now present the results of the decomposition of the relative ethnic poverty gap in the four countries, India, Nepal, the Philippines and Pakistan, at three different percentiles (tenth, twenty- fifth, and fiftieth) of the corresponding reference group, applying the methodology described in



section 3.3. These results are reported in Tables 8.5 and 8.6 and the share explained by each factor is summarized in Figures 8.9 to 8.11. Tables 8.7 to 8.9 report the distribution of some relevant characteristics.

In all countries, a large part of the observed ethnic poverty gap is associ-ated with the divergence of the distribution of observable characteristics by ethnic group (characteristics effect). Regarding the underlying factors, we distinguish three different patterns. India and Nepal outstand for socio-economic factors being at the root of the higher poverty of disadvantaged ethnic groups. The Philippines stand out for the higher poverty of disadvan-taged ethnic groups being associated with their location. Both location and socioeconomic factors play a substantial role in shaping ethnic inequalities in poverty levels in Pakistan. We now discuss these matters in more detail.

IndiaThe characteristics effect in India is able to account for about 80 percent of the ethnic poverty gap (the remaining 20 percent remains unexplained).20 The extent of the gap varies with the percentiles of the reference group used as the poverty line, as seen before, but the determinant factors are rather stable. Socioeconomic factors jointly account for 56–57 percent of the ethnic gap in poverty rates. The lower education of ethnic disadvan-taged groups alone accounts for more than 40 percent of the gap. This means about 11 percentage points of higher poverty (at the twenty- fifth and fiftieth percentiles) among disadvantaged groups and does not come as a surprise. For example, about two- thirds (65 percent) of the eligible population in the disadvantaged groups (SC/ST/OBC) are illiterate, and only the households heads of 46 percent of the eligible women completed primary education (see Table 8.7). These figures sharply contrast with 39 percent and 65 percent, respectively, for the reference group.

The majority of the population from disadvantaged groups living in rural areas (74 percent compared with 49 percent of the reference group, see Table 8.8) and in the poorest states, respectively, explain about 7 percent and between 2 and 7 percent (depending on the threshold) of the gap. Demographic factors are at least as important as geographical variables to explain the ethnic gap in poverty rates, about 12 percent (for example, there is a higher prevalence of immigration from rural areas, 9 percent higher, and for teen marriage, 16 percent higher).

These features of the ethnic (caste) gaps in poverty levels in India apply to all three disadvantaged groups: SC, ST and OBC (see Table 8.5 and Figure 8.10). We now consider the case when the poverty line is fixed at the twenty- fifth percentile of the reference group. The ethnic gap in poverty rates, as mentioned in the previous section, is larger for ST (51 percent-age points), and much smaller for SC (28 percentage points) and OBC


285

Tabl

e 8.

5 D

ecom

posi

tion

of th

e et

hnic

pov

erty

gap

for d

iffer

ent p

erce

ntile

s of

the

refe

renc

e gr

oup

Cou

ntry

Eth

nic

pove

rty

gap

Exp

lain

ed g

apU

nexp

lain

ed

gap

Tota

lR

egio

nA

rea

Dem

ogra

phic

Edu

catio

nL

abor

Indi

a 20

05/0

610

th13

.610

.80.

40.

81.

65.

92.

12.

8(0

.4)

(0.3

)(0

.2)

(0.1

)(0

.2)

(0.2

)(0

.2)

(0.3

)25

th26

.221

.21.

81.

52.

810

.84.

35.

0(0

.5)

(0.4

)(0

.3)

(0.3

)(0

.3)

(0.4

)(0

.3)

(0.4

)Sc

hedu

led

Cas

te28

.422

.02.

90.

54.

012

.52.

16.

4(0

.7)

(0.6

)(0

.4)

(0.4

)(0

.4)

(0.5

)(0

.4)

(0.6

)Sc

hedu

led

Trib

e50

.042

.73.

79.

13.

716

.610

.57.

3(0

.8)

(1.3

)(0

.9)

(0.7

)(0

.7)

(0.9

)(0

.8)

(1.3

)O

ther

Bac

kwar

d

Cla

ss20

.218

.53.

91.

22.

27.

43.

63.

7(0

.6)

(0.8

)(0

.7)

(0.3

)(0

.3)

(0.4

)(0

.3)

(0.5

)50

th27

.422

.41.

51.

83.

211

.64.

44.

9(0

.5)

(0.5

)(0

.3)

(0.3

)(0

.3)

(0.4

)(0

.3)

(0.5

)

Nep

al 2

011

10th

38.6

26.5

1.3

−1.

23.

223

.00.

112

.2(1

.3)

(1.1

)(0

.6)

(0.5

)(0

.7)

(1.3

)(1

.0)

(1.3

)25

th39

.028

.00.

3−

1.7

3.6

26.3

−0.

610

.9(1

.7)

(1.4

)(0

.8)

(0.6

)(0

.8)

(1.5

)(1

.3)

(1.6

)50

th27

.524

.3−

0.8

−2.

03.

424

.6−

1.0

3.2

(1.8

)(1

.7)

(0.9

)(0

.8)

(0.8

)(1

.7)

(1.2

)(1

.9)


286

Tabl

e 8.

5 (c

ontin

ued)

Cou

ntry

Eth

nic

pove

rty

gap

Exp

lain

ed g

apU

nexp

lain

ed

gap

Tota

lR

egio

nA

rea

Dem

ogra

phic

Edu

catio

nL

abor

Paki

stan

200

610

th48

.643

.06.

218

.90.

015

.32.

65.

7(1

.5)

(1.3

)(0

.9)

(1.5

)(0

.6)

(1.7

)(1

.1)

(1.2

)25

th50

.745

.87.

319

.3−

0.2

17.3

2.2

4.9

(2.1

)(1

.6)

(1.0

)(1

.6)

(0.7

)(1

.8)

(1.1

)(1

.79

50th

39.0

34.2

6.4

13.5

−0.

613

.61.

44.

8(2

.6)

(1.9

)(0

.9)

(1.2

)(0

.7)

(1.6

)(0

.9)

(2.7

)

The

Phi

lippi

nes 2

008

10th

24.4

23.3

20.0

−2.

0−

1.6

5.1

1.8

1.2

(0.9

)(0

.8)

(0.9

)(0

.6)

(0.6

)(0

.8)

(0.7

)(0

.8)

25th

30.1

27.9

25.4

−2.

6−

2.5

5.6

1.9

2.1

(1.2

)(1

.1)

(1.2

)(0

.7)

(0.8

)(1

.0)

(0.8

)(1

.4)

50th

25.1

17.7

16.3

−2.

1−

2.4

4.5

1.3

7.4

(1.4

)(1

.3)

(1.3

)(0

.6)

(0.7

)(1

.0)

(0.8

)(1

.7)

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1. R

ewei

ghtin

g de

com

posit

ion

desc

ribed

in

sect

ion

3.3.

Boo

tstr

aps s

tand

ard

erro

rs in

par

enth

eses

(300

repl

icat

ions

).


287

Tabl

e 8.

6 D

ecom

posi

tion

of th

e ch

ange

in th

e et

hnic

pov

erty

gap

for d

iffer

ent p

erce

ntile

s of

the

refe

renc

e gr

oup

Cou

ntry

Cha

nge

in

EPG

Exp

lain

ed g

apU

nexp

lain

ed

gap

Tota

lR

egio

nA

rea

Dem

ogra

phic

Edu

catio

nL

abor

Indi

a 19

98/9

9–20

05/0

6(I

ndia

n- sp

ecifi

c w

ealth

inde

x)10

th4.

54.

31.

10.

30.

71.

31.

00.

1(0

.6)

(0.4

)(0

.3)

(0.2

)(0

.2)

(0.3

)(0

.2)

(0.5

)25

th4.

55.

71.

80.

41.

30.

81.

3−

1.1

(0.7

)(0

.5)

(0.4

)(0

.4)

(0.3

)(0

.6)

(0.4

)(0

.6)

50th

3.0

3.3

0.7

0.2

1.5

0.1

0.8

−0.

3(0

.6)

(0.6

)(0

.4)

(0.4

)(0

.3)

(0.6

)(0

.4)

(0.7

)T

he P

hilip

pine

s 200

3–08

(Phi

lippi

ne- s

peci

fic w

ealth

inde

x)10

th−

7.1

−4.

4−

3.8

−2.

70.

10.

41.

6−

2.7

(1.3

)(1

.2)

(1.3

)(0

.8)

(0.9

)(1

.1)

(1.0

9(1

.3)

25th

−5.

0−

5.1

−3.

5−

3.0

−0.

40.

31.

30.

1(1

.8)

(1.6

)(1

.6)

(0.8

)(1

.1)

(1.5

)(1

.1)

(2.0

)50

th−

0.4

−4.

0−

2.1

−2.

40.

0−

0.1

0.7

3.6

(1.9

)(1

.8)

(1.6

)(0

.7)

(1.1

)(1

.4)

(1.1

)(2

.3)

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1. R

ewei

ghtin

g de

com

posit

ion

desc

ribed

in

sect

ion

3.3.

Boo

tstr

aps s

tand

ard

erro

rs in

par

enth

eses

(300

repl

icat

ions

).


288

–2020 0406080100

120

Reg

ion

Rur

alD

emog

raph

icE

duca

tion

Lab

orU

nexp

lain

ed

–2020 0406080100

120

–2020 0406080100

120

10th361212434316162121

0102030405060708090100

3612431621

25th761111414116161919 7611411619

50th561212434316161818 5612431618

Indi

a 20

05/0

6

Prop

ortio

n (p

erce

ntag

e) o

f the

rela

tive

ethn

ic p

over

ty g

ap e

xpla

ined

by

each

cha

ract

erist

ic

Nep

al 2

011

Paki

stan

200

6/07

Phili

ppin

es 2

008

10th

25th

–3 –38 3606003232

10th013133939323251212

10th–7–7

25th–8–8

50th

–10

–10

–8–86565181852929

–8–88484191967

–8–88282212175

28281212 9090 1212 –3–3 –4–4–7–7

6868 –1–1–4–419

50th

25th014143838343441212

50th–2 –216163535353531212

–38 360032

0133932512

–7–8

–10

–86518529

–8841967

–8822175

2812 90 12 –3 –4–7

68 –1–419

0143834410

–2163535312

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.

Figu

re 8

.9

Dec

ompo

sing

the

ethn

ic p

over

ty g

ap



(20 percentage points). The proportion of the poverty gap explained by characteristics is also largest among ST (86 percent) and smallest among SC (77 percent), and, after controlling for characteristics, the remaining gap is similar for SC and ST (6–7 percentage points) and still smaller for OBC (4 percentage points).

In all three groups, the socioeconomic explanation accounts for more than half the observed gap. In absolute terms, the gaps explained by education, labor and location are larger for ST (associated with 17, 11 and 13 percentage points of higher poverty, respectively). Only the demo-graphic gap is a bit higher for SC.

The distribution of the importance by factor shown in the overall results basically reflects what happens with OBC. The largest contribution to the characteristics effect, 39 percent of the gap, comes from education,

0

10

20

30

40

50

All

0102030405060708090

100

All76

1111

4141

1616

1919

SC101021414

4444

7

2323

ST7

1818

7

3333

2121

1414

OBC

1919

61111

3737

1818

9

SC ST OBC

Region Rural Demographic Education Labor Unexplained

GapLevel and proportion of the relative ethnic poverty gap explained by each characteristic

% Gap

7611

41

16

19

10214

44

7

23

7

18

7

33

21

14

19

611

37

18

9

Notes: Eligible population: ever- married women, 15–49 years old. Ethnic groups as listed in Table 8.1.

Figure 8.10 Decomposing the ethnic poverty gap in India (twenty- fifth percentile)



followed by 15 percent from labor variables, 13 percent from demograph-ics, and 9 percent from living in rural areas. In the case of the most disad-vantaged group, ST, education is relatively less relevant (33 percent) but labor variables (21 percent) and their overrepresentation in rural areas (18 percent) are much more important than in any other group. For SC, education (44 percent) and the region where they live (10 percent) are more relevant than in the other two groups.

Regarding the change over time in the contribution of each factor in India (see Table 8.6 and Figure 8.11), we observe that the increase in the gap between 1998/99 and 2005/06, discussed above (4 percentage points at the tenth and twenty- fifth percentiles), was driven by an increasing contribu-tion from all factors.

Region Rural Demographic Education Labor Unexplained

–2

–1

1

0

2

3

4

5

6

7

10th 25th 50th

India 1998/99–2005/06

Change in absolute values between both surveys

–10

–8

–4

–6

–2

0

2

4

6

10th

Philippines 2003–08

25th 50th


Figure 8.11 Trends in ethnic poverty gap


291

Tabl

e 8.

7 E

duca

tion

and

ethn

icity

Cou

ntry

Indi

vidu

alH

ouse

hold

he

adC

ount

ryIn

divi

dual

Hou

seho

ld

Hea

d

Illite

rate

Prim

ary

com

plet

edPr

imar

y co

mpl

eted

Illite

rate

Prim

ary

com

plet

edPr

imar

y co

mpl

eted

Indi

a 20

05/0

6In

dia

1998

/99

Non

e of

them

38.6

62.1

65.1

Non

e of

them

41.7

50.6

59.6

SC69

.531

.941

.7SC

70.9

23.0

37.7

ST78

.422

.132

.7ST

76.9

16.5

30.6

OB

C60

.941

.051

.4O

BC

58.8

35.4

49.5

SC/S

T/O

BC

65.5

36.1

46.4

SC/S

T/O

BC

65.1

28.9

43.1

Nep

al 2

011

Paki

stan

200

6H

ill B

rahm

in22

.267

.064

.4U

rdu

31.5

44.9

Res

t of

grou

ps54

.034

.936

.8R

est o

f gr

oups

74.3

21.8

The

Phi

lippi

nes 2

008

The

Phi

lippi

nes 2

003

Taga

log

2.8

95.6

88.7

Taga

log

2.8

93.9

87.8

Res

t of

grou

ps11

.185

.476

.2R

est o

f gr

oups

11.2

82.5

74.3

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.



NepalThe proportion of the ethnic poverty gap that is explained by charac-teristics in Nepal is smaller at the bottom, about 68 percent, but sharply increases for higher poverty lines. The proportion explained by education is even larger than in India, 60 percent at the tenth quantile, and goes up to 90 percent at the median. This implies that education is associated with between 23 and 26 percentage points of higher poverty among ethnic dis-advantaged groups in this country. This, again, does not come as a surprise considering that inequality in education turns out to be even stronger in Nepal than in India because of the higher education, in average, of the reference group. In the disadvantaged groups, 54 percent of eligible women are illiterate, whereas, for only 37 percent of them, the household head had completed primary studies, compared with 22 percent and 64 percent in the case of the reference group. Demographic factors are also of some rel-evance (3 percentage- point differential) especially for explaining moderate poverty (about 12 percent of the gap). Location and labor variables here are of little relevance, in general.

PakistanThe characteristics effect also explains the largest part (near 90 percent) of the observed gap in poverty rates by ethnicity in Pakistan. A large part of this gap is associated with location. In this case, it is the overrepresenta-tion in rural areas (71 percent of the eligible population of disadvantaged

Table 8.8 Area of residence and ethnicity

Country Rural Country Rural

India 2005/06 India 1998/99None of them 57.9 None of them 65.8SC 72.6 SC 78.9ST 89.9 ST 89.3OBC 71.1 OBC 76.0SC/ST/OBC 73.9 SC/ST/OBC 78.8

Nepal 2011 Pakistan 2006Hill Brahmin 82.7 Urdu 15.4Other 87.5 Other 71.1

The Philippines 2008 The Philippines 2003Tagalog 27.9 Tagalog 22.9Other 55.9 Other 55.3




ethnic groups compared with only 15 percent of Urdu), the main factor behind the ethnic poverty gap. This explains 35–39 percent of the differ-ential, that is, about 19 percentage points of higher poverty at the tenth and twenty- fifth percentiles. The district of residence also matters. The fact that about 60 percent of Urdu reside in the richest quartile of dis-tricts, compared with only 19 percent of disadvantaged groups, explains about 13–16 percent of the poverty differential (about six–seven additional percentage points). However, the educational gap is also responsible for about one third of the overall gap in poverty (14–17 percentage points). Again, this is owing to a huge gap in attained education. Similar to what was shown for Nepal, 74 percent of eligible women in disadvantaged eth-nicities are illiterate, whereas, for only 45 percent of them, the household head had completed primary studies, compared with 31 percent and 73 percent in the case of Urdu.

Table 8.9 Location and ethnicity: the Philippines

Region 2008 2003

Wealth Tagalog Other Wealth Tagalog Other

I – Ilocos Region 0.543 1.3 6.2 0.472 1.3 6.3II – Cagayan Valley 0.487 1.0 4.1 0.418 1.3 4.7III – Central Luzon 0.568 20.9 6.8 0.527 18.7 7.6IV- a – Calabarzon 0.594 36.3 3.6 0.567 36.4 3.7IV- b – Mimaropa 0.369 5.2 1.9 0.291 4.1 2.4V – Bicol Region 0.424 1.3 7.3 0.366 1.8 6.7VI – Western Visayas 0.423 0.3 10.1 0.320 0.5 9.2VII – Central Visayas 0.467 0.2 9.9 0.406 0.3 10.7VIII – Eastern Visayas 0.388 0.1 5.5 0.303 0.2 5.8IX – Zamboanga Peninsula 0.393 0.2 5.3 0.279 0.2 5.4X – Northern Mindanao 0.410 0.1 6.1 0.388 0.1 5.9XI – Davao Peninsula 0.414 0.1 6.7 0.417 0.3 6.8XII – Soccsksargen 0.377 0.3 5.4 0.323 0.4 5.7XIII – Caraga 0.406 0.2 3.4 0.346 0.1 3.5National Capital Region 0.643 32.2 10.1 0.599 34.2 8.4Cordillera Administrative Region

0.493 0.4 2.2 0.461 0.3 2.1

Arm 0.269 0.1 5.5 0.199 0.0 5.3Urban 0.592 72.0 44.1 0.550 77.1 44.7Rural 0.391 27.9 55.9 0.316 22.9 55.3




The PhilippinesIn contrast with the socioeconomic explanation dominant in India and Nepal, the Philippines is a remarkable case in which location turns out to be of extraordinary importance. Owing to historical reasons, minori-ties are strongly linked to specific regions and there is large inequality in wealth across regions that are along ethnic lines (see Table 8.9). The wealth decile of the region of residence explains more than 80 percent of the gap at the tenth and twenty- fifth percentiles (having 20 and 25 percentage points of differential, respectively). For these two percentiles, the characteristics effect globally accounts for more than 90 percent of the gap. The proportion explained by region goes down to 65 percent at the median, where a larger proportion of the gap (29 percent) remains unexplained. Education is also important, about 20 percent of the dif-ferential (five percentage points) but much less than in India and Nepal because the educational gap is also smaller (see Table 8.7). Looking at the trend over time reveals that the reduction of the ethnic poverty gap in the Philippines between 2003 and 2008 was driven by a lower contri-bution from location (region and area) owing to the larger increase in wealth in rural areas and in regions with proportionally more popula-tion from ethnic minorities (for example, regions IX, VI, or VIII, see Table 8.9) and lower in urban areas and in regions where Tagalog are disproportionally represented (for example, IV- a, III, and the National Capital Region).

6 SUMMARY AND CONCLUSIONS

Ethnicity is definitely a matter of concern in Asian countries. The results of this study showed that in the six selected countries there are some ethnic groups facing higher poverty risk than others when an index of wealth is used to measure economic status. There is, however, an important level of cross- country heterogeneity in both the extent of the ethnic poverty gap and the main explanatory factors, as well as in the evolution over time.

The poverty gap between some ethnic groups and their country’s refer-ence is astonishingly large. In some cases, the differential in poverty rates is above 50 or even 60 percentage points for some wealth cutoffs. This is especially the case of Siraiki and other linguistic groups in Pakistan, Scheduled Tribes in India, Hill Dalit in Nepal, or ethnic minorities in Viet Nam. Clearly, ethnic minorities surveyed in Azerbaijan enjoy, not only higher levels of wealth, but also a smaller poverty gap with respect to Azerbaijani (about 20 percentage points at its maximum). To a lesser extent, the ethnic poverty gap in the Philippines also tends to be smaller,



around 30 percentage points in its peak, similar to Scheduled Cates in India or Hill Chhetri and Hill Janajati in Nepal.

With regard to the reasons for this ethnic inequality in poverty rates, we know that some ethnic groups usually accumulate a number of disadvan-tages across different dimensions such as having lower education, higher unemployment, larger families, or lower development of their communi-ties that help to explain their higher poverty. Among the studied cases, this is probably a good description of the higher poverty gap of Scheduled Tribes in India, the group showing the largest absolute poverty rates among all those included in our analysis.

We have, however, found significant cross- country differences in what factors are more strongly associated with the ethnic poverty gap in the four countries we have analyzed in more detail. We show that the higher poverty rates of disadvantaged groups in India and Nepal are mostly driven by the extraordinarily high inequality in attained education by ethnicity prevail-ing in these two countries. As mentioned before, in the specific case of the Scheduled Tribes in India, their higher concentration in rural areas and their poorer performance in the labor market are also remarkable determinant factors. On the contrary, the Philippines stands out for having regional wealth inequalities as the main factor associated with most of the ethnic poverty gap of their disadvantaged ethnicities. Pakistan resembles India and Nepal in the remarkable importance of the poorer education of the disadvantaged groups, but it also stands out for their concentration in rural areas being associated with their higher poverty.

We also showed that, in a period of generally strong economic growth in the region, the wealth of all ethnic groups in India and the Philippines has increased. This implies a reduction of the ethnic poverty gap only in the Philippines (driven by diminishing interregional inequality), whereas some ethnicities in India, especially the Scheduled Tribes, took less advantage of growth than the reference group and the relative ethnic poverty gap increased (driven by the contribution of all factors).

The nature of this study does not allow us to make very specific policy recommendations because there is no causal analysis and because that would need a more in- depth analysis of the mechanisms that work to keep each particular ethnicity with higher poverty in each country. However, we can draw a few lessons that might be of help in orienting policy when it comes to reduce the ethnic poverty gap.

The significant extent of the ethnic poverty gap for many groups in Asian countries, as described in this chapter, suggests that the situation of their ethnic disadvantaged groups should be taken very seriously. Ethnicity should definitely be a matter of concern and be part of any agenda of poverty reduction in the region for the next years. This calls for a higher



visibility of ethnicity in statistics to be able to monitor the progress made during these years of intense economic growth in the region, establishing specific goals of poverty reduction, and designing appropriate strategies to achieve them. The fact that most of the ethnic poverty gap seems to be associated with a set of basic observed characteristics suggests that it should not be difficult to identify what policies are generally expected to have a larger impact on reducing the poverty gap in each case. The indi-cated factors associated with the ethnic poverty gap point in the direction of policies aimed at closing the gap, which should be addressed for improv-ing the basic endowments of the poorest ethnic groups.

In Nepal and India, where education is identified as the main factor associated with the ethnic poverty gap, we expect little improvement in the relative situation of ethnic disadvantaged groups (castes and tribes) without addressing this extraordinarily high inequality in the attained levels of education. We note that the inter- ethnic difference in education starts at the elementary level, with a large gap in literacy rates and in the proportion of the population that has completed primary- level educa-tion. Thus, it is at these basic levels that most efforts should be addressed improving and enhancing the existing infrastructure as well as promoting the enrollment among the poorest ethnic groups. For example, there exists wide empirical evidence of the success of conditional cash transfers in promoting schooling jointly with improvements in incomes among the poor in many countries (for example, the meta- analysis in Saavedra and García 2012) that suggests enhancing this type of transfers might have a formidable impact on reducing the ethnic gap too. The fact that India has a long tradition of affirmative- action quotas in politics, public employ-ment and education has probably prevented the gap from being even higher. However, India has been unable so far to substantially close the poverty gap for ethnic groups or, as shown here, to prevent an increase over time.

A similar conclusion applies to Pakistan, a country that, together with India and Nepal, has high ethnic inequality in access to basic education, which is a determinant factor of the large poverty rates of linguistic disad-vantaged groups. In Pakistan, we might also expect a significant reduction in the ethnic gap by reducing the urban–rural gap through development of rural communities where disadvantaged groups overwhelmingly live, some-thing that is also extremely important in India, especially for Scheduled Tribes. In the Philippines, any policy that aims to reduce the large geo-graphical inequality, thereby, increasing the economic opportunities in the least- developed provinces, is also expected to have an extraordinary impact on closing the ethnic gap. In fact, we have shown that a reduction in geo-graphical inequalities between 2003 and 2008 account for the reduction in



the ethnic poverty gap, both absolute and relative, that occurred during that period in this country.

NOTES

1. For example, in the PRC (Hannum and Wang 2012; Gradín 2015), India (Borooah 2005; Gang et al. 2008; Das et al. 2012), the Lao People’s Democratic Republic (King and van de Walle 2012) and, especially, Viet Nam (van de Walle and Gunewardena 2001; Baulch et al. 2004, 2007, 2008, 2010; Swinkels and Turk 2006; Hoang et al. 2007; Baulch 2008; Pham et al. 2010; Imai et al. 2011; Dang 2012).

2. For a comparison of ethnic inequalities among non- Asian countries (blacks and whites in the US, Brazil and South Africa), see Gradín (2014).

3. Multiple correspondence analysis is an extension of correspondence analysis, which allows us to analyze the pattern of relationships of several categorical dependent vari-ables and can be seen as a generalization of principal component analysis when the vari-ables to be analyzed are categorical instead of quantitative (Abdi and Valentin 2007).

4. Recently these approaches have been followed to analyze differences in well- being between blacks and whites in Brazil and South Africa, or among Chinese Han and minorities, among many other examples (for example, Gradín 2009, 2013, 2015).

5. See the information provided in its web page (http://www.measuredhs.com, accessed 20 December 2015) for details about available datasets, design, questionnaires and variables.

6. In all cases, we use the standard DHS, except in the case of Viet Nam, for which we use the standard AIDS Indicator Survey (AIS) because it is the only one with data on ethnicity. Previous releases of the DHS for Nepal were not used given the difficulty of matching ethnic groups in different years.

7. The main exception is Azerbaijan, which excludes two regions in the border with Armenia (Kalbajar- Lachin and Nakhchivan). The survey for India 1998/99 indicates that its coverage is more than 99 percent.

8. The use of common weights for all countries might be criticized on the basis that the implication of a household falling in a given category might differ across countries. For that reason, we also computed a country- specific index of wealth estimating separately the MCA scores for each country. The linear correlation between the indices constructed using common weights and country- specific weights is above 0.94 in Azerbaijan, and above 0.97 in the other countries. So we would not expect this choice to have a significant impact on the results.

9. We do not aim here at producing results representative of Asia as a whole or of a spe-cific region. We want a comparable wealth index to be meaningful in each country. In the case that each country were weighted according to its population, the index would be strongly influenced by the Indian survey.

10. This index is just a linear transformation of the predicted value, usually standardized to have zero mean and standard deviation equal to one.

11. The index, estimated using the first dimension, explains 58 percent of total variability (inertia). Given the high correlation of this index with a similar country- specific index (which explains a much higher proportion of each country’s inertia), we expect most of the unexplained inertia being variability between countries. As expected, the index assigns a zero weight (poorest profile) to households using an unprotected well as their main source of drinking water, using natural materials for their floor and roof and having no walls in the dwelling. They are overcrowded (more than ten people per sleep-ing room), use animal dung for cooking fuel and lack any facility for a toilet as well as most assets (all but a bicycle and a cart).

12. The variability (inertia) explained by the first dimension used to construct the wealth



index was 86 percent in India and 90 percent in the Philippines. The within- country cor-relation with the main index (with common weights across all six countries) was 0.93 in India (2005/06) and 0.94 in the Philippines (2008).

13. The corresponding densities are estimated using Gaussian kernels with adaptive optimal bandwidth, computed with the akdensity STATA routine, written by P. Van Kerm.

14. This relative threshold deviates from that most commonly used in the literature (for example, 60 percent of the country’s median income is used in the European Union) in that it is indexed to the entire distribution (not only one specific quantile). Furthermore, the reference here is a specific ethnic group, the most advantaged one. We could alterna-tively define the reference to be the rest of the groups or the society as a whole, having different implications.

15. As Butler and McDonald (1987) pointed out, this approach was implicit in the notion of economic advantage of one group over another in Vinod (1985).

16. Note that, by construction, f(t) = FGT1(t; a = 0) − FGT0(t; a = 0), with FGT0(t; a = 0) = t. Similarly, we could construct ethnic poverty curves of higher order that would be related to other members of the FGT class.

17. See Sastre and Trannoy (2002) for a general formalization of the procedure to get the Shapley decomposition.

18. For very low levels of wealth, the estimated proportion of poor is slightly higher for the reference group in India (wealth below 0.036, where we find about 0.5 percent of the population of each group) and Nepal (below 0.025, a level only about 0.2 percent of the population of each group does not reach).

19. Although questionnaires are very similar across countries, they still are country specific and come from different phases and survey types and, thus, some variables were not available in specific samples.

20. This result implies a smaller unexplained proportion of the ethnic poverty gap than that reported in Borooah (2005) for income poverty: 27 percent and 46 percent when the coefficients effects use the average characteristics of Hindu (Hindu treated as SC or ST). In that chapter, the poverty rates are 29 percent for Hindu, and 46 percent and 47 percent for SC and ST. Differences in the decomposition technique, in the set of charac-teristics used, or in the well- being variable might account for the divergence. Gang et al. (2008) also provide a similar decomposition (with 40 percent and 51 percent of the gap unexplained) but focused on rural poverty and with the coefficients effects obtained with the average characteristics of SC/ST.

REFERENCES77*

Abdi, H. and D. Valentin (2007), ‘Multiple correspondence analysis,’ in N. Salkind (ed.), Encyclopedia of Measurement and Statistics, Thousand Oaks, CA: Sage.

Asian Development Bank (ADB) (2002), Indigenous Peoples/Ethnic Minorities and Poverty Reduction, regional reports for Viet Nam, the Philippines, Pacific Region, Indonesia, and Cambodia, Environment and Social Safeguard Division, Regional and Sustainable Development Department, Asian Development Bank, Manila: ADB.

Bhalla, A.S. and D. Luo (2013), Poverty and Exclusion of Minorities in China and India, New York: Palgrave Macmillan.

* The Asian Development Bank recognizes China as the People’s Republic of China, Laos as the Lao People’s Democratic Republic, and Vietnam as Viet Nam.



Baulch, B. (2008), Ethnic Minority Underdevelopment in Vietnam: Full Research Report, Economic and Social Research Council (ESRC) End of Award Report, RES- 167- 25- 0157, Swindon: ESRC.

Baulch, B., H.T.M. Nguyen, P.T.T. Phuong and H.T. Pham (2010), ‘Ethnic minor-ity poverty in Vietnam’, Working Paper 169. Manchester: Chronic Poverty Research Centre.

Baulch, B., H.T Pham and B. Reilly (2008), ‘Decomposing the Ethnic Gap in Living Standards in Rural Vietnam: 1993 to 2004’, mimeo, Institute of Development Studies, University of Sussex, Brighton.

Baulch, B., C. Truong, D. Haughton and J. Haughton (2004), ‘Ethnic minor-ity development in Vietnam: a socioeconomic perspective’, in P. Glewwe, N. Agrawal and D. Dollar (eds), Economic Growth, Poverty and Household Welfare in Vietnam, World Bank Regional and Sectoral Studies, Washington, DC: World Bank.

Baulch, B., C. Truong, D. Haughton and J. Haughton (2007), ‘Ethnic minority development in Vietnam’, Journal of Development Studies, 43 (7), 1151–76.

Blinder, A.S. (1973), ‘Wage discrimination: reduced form and structural estimates’, Journal of Human Resources, 8 (4), 436–55.

Borooah, V.K. (2005), ‘Caste, inequality, and poverty in India’, Review of Development Economics, 9 (3), 399–414.

Butler, R.J. and J.B. McDonald (1987), ‘Interdistributional income inequality’, Journal of Business and Economic Statistics, 5 (1), 13–18.

Chantreuil, F. and A. Trannoy (2013), ‘Inequality decomposition values: the tradeoff between marginality and consistency’, Journal of Economic Inequality, 11 (1), 83–98.

Dang, H.- A. (2012), ‘Vietnam: a widening poverty gap for ethnic minorities’, in G. Hall and H.A. Patrinos (eds), Indigenous Peoples, Poverty and Development, New York: Cambridge University Press.

Das, M.B., G.H. Hall, S. Kapoor and D. Nikitin (2012), ‘India: the scheduled tribes’, in G. Hall and H.A. Patrinos (eds), Indigenous Peoples, Poverty and Development, New York: Cambridge University Press.

DiNardo, J., N. Fortin and T. Lemieux (1996), ‘Labor market institutions and the distribution of wages, 1973–1992: a semiparametric approach’, Econometrica, 64 (5), 1001–44.

Filmer, D. and L. Pritchett (2001), ‘Estimating wealth effects without expenditure data – or tears: an application to educational enrollments in states of India’, Demography, 38 (1), 115–32.

Foster, J. and A.F. Shorrocks (1988a), ‘Poverty orderings’, Econometrica, 56 (1), 173–7.

Foster, J. and A.F. Shorrocks (1988b), ‘Poverty ordering and welfare dominance’, Social Choice and Welfare, 5 (2–3), 179–98.

Gang, I.N., K. Sen and M- S. Yun (2008), ‘Poverty in rural India: caste and tribe’, Review of Income and Wealth, 54 (1), 50–70.

Gradín, C. (2009), ‘Why is poverty so high among Afro- Brazilians? A decomposi-tion analysis of the racial poverty gap’, Journal of Development Studies, 45 (9), 1–38.

Gradín, C. (2013), ‘Race, poverty, and deprivation in South Africa’, Journal of African Economies, 22 (2), 187–238.

Gradín, C. (2014), ‘Race and income distribution: evidence from the US, Brazil and South Africa’, Review of Development Economics, 18 (1), 73–92.



Gradín, C. (2015), ‘Rural poverty and ethnicity in China’, Research on Economic Inequality, 23, 221–47.

Hall, G. and H.A. Patrinos (eds) (2012), Indigenous Peoples, Poverty and Development, New York: Cambridge University Press.

Hannum, E.C. and M. Wang. (2012), ‘China: a case study in rapid poverty reduc-tion’, in G. Hall and H.A. Patrinos (eds), Indigenous Peoples, Poverty and Development, New York: Cambridge University Press.

Hoang, H., G. Pham, M. Tran and H. Hansen (2007), ‘Ethnicity and poverty reduction’, in T. Nguyen and H. Hansen (eds), Markets, Policy and Poverty Reduction in Vietnam, Ha Noi: Viet Nam Academy of Social Sciences/National Political Publishing House.

Imai, K.S., R. Gaiha and W. Kanga (2011), ‘Poverty, inequality and ethnic minori-ties in Vietnam’, International Review of Applied Economics, 2 (3), 249–82.

King, E.M. and D. van de Walle (2012), ‘Laos: ethno- linguistic diversity and disad-vantage’, in in G. Hall and H.A. Patrinos (eds), Indigenous Peoples, Poverty and Development, New York: Cambridge University Press.

Le Breton, M., A. Michelangeli and E. Peluso (2012), ‘A stochastic dominance approach to the measurement of discrimination’, Journal of Economic Theory, 147 (4), 1342–50.

Oaxaca, R.L. (1973), ‘Male- female wage differentials in urban labor markets’, International Economic Review, 14 (3), 693–709.

Pham, T.H., T. Le Dang and V.C. Nguyen (2010), ‘Poverty of the ethnic minorities in Vietnam: situation and challenges from the poorest communes’, MPRA Paper 50372, Munich: Munich Personal RePEc Archive.

Rutstein, S.O. and K. Johnson (2004), ‘The DHS Wealth Index’, DHS Comparative Reports No. 6, Calverton, MD: US ORC Macro.

Saavedra, J.E. and S. García (2012), ‘Impacts of conditional cash transfer programs on educational outcomes in developing countries: a meta- analysis’, RAND Labor and Population Working Paper Series, WR- 921- 1, RAND Corporation, Santa Monica, CA.

Sastre, M. and A. Trannoy (2002), ‘Shapley inequality decomposition by factor components: some methodological issues’, Journal of Economics, 9 (supplement 1), 51–89.

Shorrocks, A. (2013), ‘Decomposition procedures for distributional analysis: a unified framework based on the Shapley value’, Journal of Economic Inequality, 11 (1), 99–126.

Swinkels, R. and C. Turk (2006), ‘Explaining ethnic minority poverty in Vietnam: a summary of recent trends and current challenges’, background paper for Committee for Ethnic Minorities Affairs/Ministry of Planning and Investment of Vietnam (CEM/MPI) meeting on Ethnic Minority Report 28, Ha Noi: World Bank.

Van de Walle, D. and D. Gunewardena (2001), ‘Sources of ethnic inequality in Viet Nam’, Journal of Development Economics, 65 (1), 177–207.

Vinod, H. (1985), ‘Measurement of Economic Distance between Blacks and Whites’, Journal of Business and Economic Statistics, 3 (1), 78–88.


301

APP

EN

DIX

8A

.1

Tabl

e 8A

.1

Com

pone

nts o

f th

e we

alth

inde

x an

d di

stri

butio

n by

cou

ntry

and

gro

up

Varia

bles

and

cat

egor

ies

MC

A

Scor

eN

orm

aliz

ed

Wei

ght

Elig

ible

pop

ulat

ion:

dist

ribut

ion

by c

ount

ry a

nd g

roup

(%)

(C =

com

paris

on R

= R

efer

ence

)

sq jw

q jA

zerb

aija

nIn

dia

Nep

alPa

kist

anT

he

Phili

ppin

esV

iet N

am

CR

CR

CR

CR

CR

CR

Sour

ce o

f dr

inki

ng w

ater

Pipe

d in

to d

wel

ling

1.51

40.

0817

.532

.28.

421

.74.

19.

526

.463

.519

.533

.44.

121

.8Pi

ped

to y

ard/

plot

0.51

30.

0516

.819

.810

.313

.414

.724

.76.

15.

36.

45.

10.

00.

0Pu

blic

tap/

stan

dpip

e−

0.94

20.

025.

93.

818

.112

.123

.624

.62.

23.

36.

12.

92.

01.

1Tu

be w

ell/b

oreh

ole

−0.

932

0.02

9.5

9.9

46.1

40.9

43.8

29.0

15.0

11.0

21.8

19.2

0.0

0.0

Prot

ecte

d w

ell

−0.

317

0.03

22.5

5.4

2.0

3.0

1.7

2.8

41.9

8.7

6.1

1.4

34.2

54.0

Unp

rote

cted

wel

l−

1.77

00.

000.

00.

211

.26.

62.

40.

71.

30.

05.

11.

15.

21.

9Su

rfac

e w

ater

−0.

623

0.03

26.1

15.9

2.7

1.2

8.6

6.9

4.2

0.7

11.6

3.4

53.9

7.5

Rai

nwat

er0.

215

0.05

0.0

0.0

0.2

0.1

0.1

0.0

0.3

0.0

0.8

0.0

0.5

12.8

Tank

er tr

uck

1.68

70.

081.

710

.30.

40.

40.

70.

70.

93.

01.

00.

60.

00.

8C

art w

ith sm

all t

ank

0.54

10.

050.

01.

00.

10.

20.

00.

00.

50.

20.

60.

50.

00.

0B

ottle

d w

ater

1.55

80.

080.

00.

10.

30.

30.

51.

10.

23.

719

.431

.50.

00.

1O

ther

−0.

003

0.04

0.0

1.4

0.2

0.1

0.0

0.0

1.1

0.7

1.6

0.7

0.2

0.1

Type

of

toile

t fac

ility

Flu

sh to

pip

ed

se

wer

syst

em1.

760

0.09

16.6

42.7

5.6

17.3

5.2

9.3

27.7

80.6

1.3

8.4

4.2

35.3

Flu

sh to

sept

ic ta

nk0.

571

0.06

0.0

1.1

17.1

32.2

32.3

59.1

18.9

8.4

62.4

85.0

0.0

0.0

Flu

sh to

pit

latr

ine

−0.

543

0.04

0.0

0.0

5.0

8.3

4.0

6.6

0.0

0.0

14.1

1.7

0.0

0.0

Flu

sh to

som

ewhe

re e

lse0.

145

0.05

0.5

0.7

1.5

1.1

0.3

0.1

9.3

5.4

0.8

1.0

0.0

0.0


302

Tabl

e 8A

.1

(con

tinue

d)

Varia

bles

and

cat

egor

ies

MC

A

Scor

eN

orm

aliz

ed

Wei

ght

Elig

ible

pop

ulat

ion:

dist

ribut

ion

by c

ount

ry a

nd g

roup

(%)

(C =

com

paris

on R

= R

efer

ence

)

sq jw

q jA

zerb

aija

nIn

dia

Nep

alPa

kist

anT

he

Phili

ppin

esV

iet N

am

CR

CR

CR

CR

CR

CR

Type

of

toile

t fac

ility

Vent

ilate

d im

prov

ed

pi

t lat

rine

0.14

60.

050.

00.

00.

30.

41.

01.

10.

90.

11.

40.

23.

59.

8

Pit l

atrin

e w

ith sl

ab0.

437

0.06

59.4

40.2

2.0

3.8

8.4

11.7

1.1

1.6

2.6

0.3

47.5

42.2

Pit l

atrin

e w

ithou

t

slab/

open

pit

0.07

30.

0522

.914

.61.

01.

97.

15.

43.

10.

33.

10.

20.

00.

0

No

faci

lity/

bush

/fiel

d−

2.16

50.

000.

60.

366

.733

.941

.66.

230

.51.

012

.13.

144

.912

.7C

ompo

stin

g to

ilet

−1.

285

0.02

0.0

0.0

0.1

0.1

0.2

0.4

0.0

0.0

0.7

0.0

0.0

0.0

Buc

ket/d

ry to

ilet

−1.

328

0.02

0.0

0.0

0.4

0.9

0.0

0.0

1.7

0.1

0.2

0.0

0.0

0.0

Han

ging

toile

t/lat

rine

−0.

480

0.04

0.0

0.0

0.0

0.0

0.0

0.0

6.7

2.6

1.4

0.2

0.0

0.0

Oth

er0.

017

0.05

0.0

0.4

0.3

0.2

0.0

0.0

0.2

0.0

0.0

0.0

0.0

0.0

Shar

e to

ilet w

ith o

ther

hou

seho

lds

No

0.84

90.

0796

.491

.223

.849

.940

.774

.959

.488

.565

.178

.449

.679

.2Ye

s0.

094

0.05

2.8

8.3

9.5

15.8

17.6

18.8

9.6

10.1

21.3

17.2

5.1

7.7

No

faci

lity/

unkn

own

−2.

129

0.00

0.8

0.5

66.7

34.3

41.8

6.3

31.1

1.5

13.7

4.5

45.3

13.2

Has

ele

ctri

city

No

−2.

413

0.00

0.4

0.6

36.4

21.0

26.2

4.8

11.0

0.5

19.9

5.5

15.4

2.5

Yes

0.42

10.

0799

.699

.563

.679

.073

.895

.389

.099

.680

.194

.584

.797

.5


303

Has

tele

phon

eN

o−

0.65

90.

0043

.742

.190

.274

.292

.478

.352

.926

.691

.480

.694

.763

.9Ye

s1.

593

0.05

56.3

57.9

9.8

25.8

7.6

21.7

47.1

73.4

8.6

19.4

5.3

36.1

Has

radi

oN

o−

0.37

40.

0054

.153

.971

.258

.751

.330

.765

.873

.536

.427

.466

.457

.3Ye

s0.

449

0.02

45.9

46.1

28.8

41.3

48.7

69.3

34.2

26.5

63.6

72.6

33.7

42.7

Has

tele

visi

onN

o−

1.75

00.

004.

24.

159

.138

.053

.327

.044

.610

.332

.411

.442

.39.

8Ye

s0.

845

0.06

95.8

95.9

40.9

62.0

46.7

73.0

55.4

89.7

67.6

88.6

57.7

90.3

Has

refr

iger

ator

No

−0.

866

0.00

30.1

23.8

90.2

69.9

90.7

79.8

64.4

26.4

65.3

47.2

95.2

75.9

Yes

1.71

10.

0669

.976

.29.

830

.19.

320

.235

.673

.634

.752

.84.

824

.1

Has

bic

ycle

No

0.09

70.

0189

.891

.443

.642

.355

.155

.756

.964

.175

.474

.541

.916

.4Ye

s−

0.12

90.

0010

.28.

656

.457

.744

.944

.443

.135

.924

.625

.558

.183

.6

Has

mot

orcy

cle/

scoo

ter

No

−0.

270

0.00

94.2

99.1

85.0

70.2

89.8

77.5

80.2

61.0

76.1

75.3

56.3

35.6

Yes

0.92

90.

035.

90.

915

.029

.810

.222

.619

.839

.023

.924

.743

.764

.4

Has

car

/truc

kN

o−

0.16

80.

0082

.077

.098

.594

.598

.497

.092

.987

.792

.084

.099

.798

.8Ye

s2.

013

0.05

18.0

23.0

1.5

5.5

1.6

3.0

7.1

12.3

8.0

16.0

0.3

1.2

Has

an

anim

al- d

raw

n ca

rtN

o0.

049

0.02

96.3

96.6

93.2

93.3

95.3

99.1

88.3

97.0

95.4

99.1

95.8

95.7

Yes

−0.

802

0.00

3.7

3.4

6.8

6.7

4.7

0.9

11.7

3.0

4.6

0.9

4.2

4.3


304

Tabl

e 8A

.1

(con

tinue

d)

Varia

bles

and

cat

egor

ies

MC

A

Scor

eN

orm

aliz

ed

Wei

ght

Elig

ible

pop

ulat

ion:

dist

ribut

ion

by c

ount

ry a

nd g

roup

(%)

(C =

com

paris

on R

= R

efer

ence

)

sq jw

q jA

zerb

aija

nIn

dia

Nep

alPa

kist

anT

he

Phili

ppin

esV

iet N

am

CR

CR

CR

CR

CR

CR

Mai

n flo

or m

ater

ial

Nat

ural

−1.

812

0.00

1.1

2.8

53.2

33.0

71.4

44.2

52.6

6.3

9.2

6.9

35.3

11.9

Rud

imen

tary

(woo

d

plan

ks, p

alm

. . .

)0.

906

0.06

87.0

78.6

6.5

7.1

1.5

1.3

0.0

0.0

29.3

6.5

29.5

3.4

Parq

uet,

polis

hed

woo

d2.

433

0.10

6.5

10.0

0.1

0.1

0.3

1.1

0.0

0.0

0.5

0.3

0.3

0.5

Vin

yl, a

spha

lt st

rips

1.09

50.

070.

00.

00.

21.

11.

32.

30.

00.

03.

16.

30.

00.

0C

eram

ic ti

les

1.20

30.

070.

00.

03.

57.

60.

21.

30.

91.

78.

116

.29.

654

.9C

emen

t0.

454

0.05

0.7

1.5

31.1

39.0

19.5

36.0

28.0

54.3

48.6

60.9

25.3

29.3

Car

pet/m

ats

1.32

50.

070.

00.

40.

20.

45.

713

.91.

03.

50.

20.

20.

00.

1O

ther

fini

shed

(pol

ished

ston

e, m

arbl

e)1.

391

0.07

2.7

3.9

5.2

11.7

0.0

0.0

17.6

34.0

0.9

2.7

0.0

0.0

Oth

er1.

446

0.08

2.1

2.8

0.0

0.1

0.1

0.0

0.0

0.2

0.0

0.1

0.0

0.0

Mai

n wa

ll m

ater

ial

No

wal

ls−

2.13

40.

000.

00.

10.

10.

10.

00.

08.

50.

80.

00.

00.

20.

0C

ane/

palm

/trun

ks/g

rass

−1.

570

0.01

0.2

0.1

3.8

2.5

2.5

0.8

0.0

0.0

2.0

1.3

17.9

10.0

Dirt

/mud

/san

d−

2.07

70.

000.

92.

527

.214

.36.

62.

223

.13.

60.

00.

03.

10.

4B

ambo

o w

ith m

ud−

1.80

00.

010.

00.

02.

62.

923

.85.

70.

00.

020

.34.

513

.21.

3St

one

with

mud

−1.

297

0.02

10.7

5.5

3.2

2.1

26.9

27.8

0.0

0.0

0.0

0.1

0.9

0.1

Plyw

ood/

reus

ed w

ood

−0.

433

0.04

0.5

0.3

0.3

0.2

1.0

2.0

0.1

0.0

12.9

9.6

19.6

3.5

Car

dboa

rd−

1.38

00.

020.

00.

00.

00.

00.

00.

06.

10.

51.

10.

70.

20.

1


305

Unc

over

ed a

dobe

/unb

urnt

−0.

529

0.04

5.4

1.0

1.7

1.1

0.0

0.0

0.0

0.0

0.0

0.0

0.4

0.1

Cem

ent

0.67

60.

070.

70.

632

.850

.424

.548

.20.

00.

021

.238

.50.

10.

2St

one

with

lim

e/ce

men

t0.

317

0.06

11.7

4.3

5.8

5.9

0.9

1.6

0.6

0.9

0.4

0.6

1.4

1.2

Bak

ed b

ricks

0.20

00.

055.

33.

719

.216

.17.

86.

121

.614

.40.

10.

115

.776

.9C

emen

t blo

cks

0.98

00.

071.

70.

32.

73.

80.

81.

939

.579

.129

.039

.80.

10.

4C

over

ed a

dobe

1.00

80.

078.

53.

70.

00.

00.

00.

00.

00.

00.

10.

22.

60.

7W

ood

plan

ks/s

hing

les

−0.

725

0.03

1.7

0.5

0.1

0.1

4.6

3.3

0.0

0.0

11.6

3.6

20.9

3.2

Oth

er fi

nish

ed1.

955

0.09

52.6

76.8

0.3

0.4

0.0

0.0

0.0

0.3

1.0

0.9

0.0

0.0

Oth

er−

0.68

00.

030.

00.

60.

20.

20.

50.

30.

60.

30.

10.

33.

91.

9

Mai

n ro

of m

ater

ial

No

roof

−0.

881

0.03

0.0

0.1

0.1

0.1

0.1

0.0

0.0

0.0

0.0

0.2

0.0

0.0

Nat

ural

−2.

015

0.00

0.0

0.1

17.7

8.9

19.2

4.7

35.5

6.1

16.6

2.8

15.3

6.2

Rud

imen

tary

−1.

872

0.00

0.4

0.3

7.2

4.2

1.2

0.2

0.2

0.2

1.0

0.4

3.1

0.3

Met

al0.

290

0.05

3.0

4.3

9.2

11.5

28.0

38.3

2.8

6.4

80.4

94.4

13.1

29.2

Woo

d−

0.15

00.

040.

00.

10.

81.

10.

20.

239

.324

.90.

10.

20.

00.

0C

alam

ine/

cem

ent

−0.

122

0.04

0.0

0.0

0.8

1.1

2.2

1.1

0.0

0.0

0.1

0.1

15.8

10.8

Cem

ent

0.91

30.

072.

56.

329

.745

.418

.933

.822

.162

.20.

91.

54.

016

.9C

eram

ic ti

les

−0.

766

0.03

4.7

3.4

13.8

13.1

29.4

21.7

0.0

0.0

0.2

0.2

48.8

36.2

Oth

er fi

nish

ed1.

224

0.08

87.7

84.0

20.5

14.5

0.0

0.0

0.0

0.0

0.8

0.2

0.0

0.4

Oth

er−

0.10

40.

041.

71.

40.

30.

20.

90.

10.

10.

30.

00.

10.

00.

0

Hou

seho

ld m

embe

rs/ro

oms u

sed

for s

leep

ing

<1

0.72

00.

052.

51.

30.

70.

92.

56.

60.

92.

31.

10.

90.

41.

61–

20.

667

0.05

20.9

15.6

9.0

14.6

24.6

42.6

5.6

8.9

14.6

14.7

14.9

23.2

2–2.

50.

515

0.05

28.3

27.0

14.3

18.5

21.6

22.5

9.1

11.4

16.9

19.0

16.1

26.1

2.5–

30.

370

0.04

16.7

14.8

9.1

11.2

10.6

8.7

8.3

8.0

11.1

12.5

8.7

13.1

3–4

−0.

052

0.03

15.9

20.3

21.0

20.0

18.7

11.6

23.2

23.1

22.2

22.2

18.1

17.5

4–5

−0.

331

0.03

9.3

13.3

18.0

15.8

10.3

5.1

18.1

16.1

13.9

11.6

14.5

11.3

5–10

−0.

998

0.01

6.5

7.5

27.0

18.2

11.4

2.8

32.0

28.1

19.3

18.9

25.2

7.0

>=

10−

1.43

20.

000.

00.

21.

00.

80.

40.

22.

82.

31.

00.

32.

30.

2


306

Tabl

e 8A

.1

(con

tinue

d)

Varia

bles

and

cat

egor

ies

MC

A

Scor

eN

orm

aliz

ed

Wei

ght

Elig

ible

pop

ulat

ion:

dist

ribut

ion

by c

ount

ry a

nd g

roup

(%)

(C =

com

paris

on R

= R

efer

ence

)

sq jw

q jA

zerb

aija

nIn

dia

Nep

alPa

kist

anT

he

Phili

ppin

esV

iet N

am

CR

CR

CR

CR

CR

CR

Type

of

cook

ing

fuel

Ele

ctric

ity1.

437

0.07

20.0

22.6

0.3

0.5

0.1

0.0

0.2

0.2

1.0

1.1

0.2

0.5

Gas

1.66

50.

0854

.968

.417

.438

.115

.532

.125

.982

.824

.353

.73.

335

.4B

ioga

s−

0.07

70.

040.

00.

00.

40.

72.

313

.02.

00.

40.

00.

00.

01.

7K

eros

ene

0.30

20.

050.

00.

42.

12.

90.

30.

00.

00.

01.

13.

90.

00.

0C

oal/l

igni

te0.

508

0.05

0.3

0.1

1.7

2.8

0.0

0.0

0.0

0.0

0.1

0.2

0.9

11.1

Cha

rcoa

l0.

278

0.05

2.0

0.5

0.3

0.5

0.1

0.7

0.5

0.0

16.1

16.4

0.0

0.0

Woo

d−

1.03

90.

0222

.87.

555

.734

.270

.753

.555

.314

.056

.423

.894

.738

.7St

raw

/shr

ubs/

gras

s−

1.21

70.

010.

00.

04.

76.

33.

50.

14.

40.

70.

00.

01.

012

.5A

gric

ultu

ral c

rop

−1.

505

0.01

0.0

0.0

4.0

4.2

1.1

0.2

4.1

0.4

0.9

0.8

0.0

0.0

Ani

mal

dun

g−

1.78

90.

000.

20.

313

.59.

76.

20.

27.

51.

50.

00.

00.

00.

0N

o fo

od c

ooke

d in

hou

se−

0.06

90.

040.

00.

00.

00.

00.

10.

20.

00.

00.

10.

00.

00.

0O

ther

−0.

571

0.03

0.0

0.1

0.0

0.0

0.0

0.0

0.1

0.0

0.0

0.1

0.0

0.1

Not

e:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

.



Table 8A.2 Variables used for the Indian- and Philippine- specific wealth indices

India 1998/99–2005/06


House Motorcycle Water pump Cot or bed

Type of toilet facility

Acres of land under cultivation

Car Thresher Chair

People/sleeping rooms

Electricity Telephone Tractor Mattress

Main cooking fuel

Radio Clock or watch

Fan Pressure cooker

Purify water Refrigerator Bullock cart Television (b/w)

Table

Separate room used as a kitchen

Bicycle Household owns livestock

Television (color)

Sewing machine

The Philippines 2003–08


Main wall material

Refrigerator Has landline telephone

CD/VCD/DVD player

Time to get to water source

Electricity Bicycle/trisikad

Cellphone Component/karaoke

Type of toilet facility

Radio Motorcycle/scooter/tricycle

Personal computer

Owns a tractor

Main floor material

Television Car/truck Washing machine

Tenure status of lot


308

Tabl

e 8A

.3

Sam

ple

size

s

Sam

ple

Ref

eren

ceC

ompa

rison

Com

paris

on (a

ll)To

tal

Aze

rbai

jani

Oth

erA

zerb

aija

n 20

0653

1845

845

857

76N

one

of th

emSC

STO

BC

Indi

a 20

05/0

631

763

15 81

411

789

30 31

857

921

89 68

4In

dia

1998

/99

37 46

715

256

10 90

626

028

52 19

089

657

HB

HC

HD

HJ

Oth

erN

epal

201

113

5925

2811

5422

4125

5584

7898

37U

rdu

Punj

abi

Sind

hPu

shto

Sira

iki

Oth

erPa

kist

an 2

006/

0773

530

9813

1820

5713

7214

3792

8210

017

Taga

log

Ceb

uano

Iloca

noIlo

nggo

Oth

erT

he P

hilip

pine

s 200

821

3121

8689

784

131

3970

6391

94T

he P

hilip

pine

s 200

323

4224

1010

1483

427

2469

8293

24V

ietn

ames

eO

ther

Vie

t Nam

200

542

2078

578

550

05A

ll22

8 494

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.


309

Tabl

e 8A

.4

Sum

mar

y va

riab

les a

nd lo

git r

egre

ssio

n: p

roba

bilit

y of

bel

ongi

ng to

the

refe

renc

e gr

oup,

Indi

a 20

05/0

6

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Stat

e D

ecile

19.

529

.313

.334

.0R

ef.

Stat

e D

ecile

26.

224

.213

.534

.1−

0.33

60.

055

Stat

e D

ecile

312

.733

.316

.637

.20.

006

0.05

3St

ate

Dec

ile 4

15.3

36.0

4.6

21.0

1.49

10.

059

Stat

e D

ecile

56.

624

.98.

227

.40.

189

0.05

6St

ate

Dec

ile 6

7.5

26.4

9.4

29.1

0.20

30.

059

Stat

e D

ecile

73.

819

.115

.536

.2−

1.41

20.

058

Stat

e D

ecile

819

.939

.99.

929

.80.

918

0.05

2St

ate

Dec

ile 9

7.2

25.8

5.0

21.8

0.56

00.

058

Stat

e D

ecile

10

11.3

31.7

4.1

19.8

0.82

30.

051

Rur

al57

.949

.473

.943

.9−

0.14

40.

028

2 re

late

d ad

ults

28.9

45.3

32.9

47.0

Ref

.3+

rela

ted

adul

ts67

.446

.962

.848

.30.

015

0.03

1O

ther

hou

seho

ld ty

pe3.

618

.74.

320

.40.

177

0.07

1Fe

mal

e he

ad12

.633

.212

.633

.20.

190

0.04

7H

H a

ge46

.513

.544

.112

.9−

0.01

90.

006

HH

age

squa

re ×

100

0.02

60.

006

Age

32.4

8.3

31.3

8.4

0.08

50.

012

Age

squa

red

x 10

0−

0.09

50.

017

N M

embe

rs5.

62.

95.

83.

00.

004

0.00

7N

chi

ldre

n (5

yo)

0.7

1.0

0.9

1.1

−0.

063

0.01

7N

livi

ng c

hild

ren

2.3

1.6

2.6

1.7

−0.

019

0.00

9


310

Tabl

e 8A

.4

(con

tinue

d)

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Prim

ary

educ

atio

n (in

com

plet

e)37

.948

.563

.948

.0Pr

imar

y ed

ucat

ion

(com

plet

e)7.

927

.06.

925

.30.

346

0.04

6Se

cond

ary

educ

atio

n (in

com

plet

e)36

.648

.223

.342

.30.

225

0.05

0Se

cond

ary

educ

atio

n (c

ompl

ete)

6.1

24.0

2.7

16.2

0.38

10.

070

Hig

her e

duca

tion

11.5

31.9

3.2

17.5

0.72

10.

071

Illite

rate

38.6

48.7

65.5

47.5

−0.

338

0.04

5H

ead’

s Prim

ary

(inco

mpl

ete)

34.9

47.7

53.6

49.9

Ref

.H

ead’

s Prim

ary

(com

plet

e)6.

825

.28.

027

.10.

099

0.04

9H

ead’

s Sec

onda

ry (i

ncom

plet

e)37

.348

.429

.545

.60.

273

0.04

0H

ead’

s Sec

onda

ry (c

ompl

ete)

6.7

25.0

3.6

18.6

0.50

10.

062

Hea

d’s H

ighe

r edu

catio

n14

.335

.05.

422

.50.

439

0.06

8N

o im

mig

rant

15.7

36.4

17.8

38.2

Ref

.Im

mig

rant

(<=

5 ye

ars/

visit

or)

27.1

44.5

23.9

42.7

0.15

30.

039

Imm

igra

nt (6

+ y

ears

)57

.249

.558

.349

.30.

198

0.03

7Im

mig

rant

from

cou

ntry

side

53.5

49.9

62.6

48.4

−0.

073

0.02

8C

urre

ntly

mar

ried

94.7

22.4

93.9

23.9

Ref

.Fo

rmer

ly m

arrie

d5.

322

.46.

123

.9−

0.04

10.

051

Teen

mar

riage

49.2

50.0

65.1

47.7

−0.

189

0.02

4N

o w

orki

ng/o

ther

74.1

43.8

57.1

49.5

Ref

.Pr

ofes

siona

l, cl

eric

al4.

320

.22.

114

.50.

347

0.07

8Se

rvic

es, S

kille

d10

.831

.113

.333

.90.

386

0.05

8A

gric

ultu

re10

.831

.027

.444

.6−

0.24

80.

064


311

Wor

ked

last

12

mon

ths

70.9

45.4

48.5

50.0

0.89

20.

052

Not

pai

d w

ork

6.9

25.3

13.8

34.5

0.21

70.

051

Part

ner:

did

not w

ork/

no p

artn

er2.

314

.81.

713

.0R

ef.

Part

ner:

prof

essio

nal/t

echn

ical

/man

ager

ial

11.1

31.4

5.0

21.9

−0.

111

0.08

3Pa

rtne

r: cl

eric

al6.

023

.73.

518

.3−

0.18

80.

087

Part

ner:

sale

s17

.137

.710

.330

.40.

027

0.07

8Pa

rtne

r: ag

ricul

tura

l25

.743

.735

.247

.80.

059

0.07

6Pa

rtne

r: se

rvic

es5.

923

.55.

422

.6−

0.37

20.

084

Part

ner:

skill

ed a

nd u

nski

lled

man

ual

31.7

46.5

38.7

48.7

−0.

255

0.07

5Pa

rtne

r: un

know

n oc

cupa

tion

0.2

4.9

0.2

4.8

0.17

70.

222

Part

ner:

no e

duca

tion/

don’

t kno

w17

.438

.032

.746

.9R

ef.

Part

ner:

prim

ary

educ

atio

n13

.233

.917

.137

.70.

090

0.03

8Pa

rtne

r: se

cond

ary

educ

atio

n49

.950

.042

.449

.40.

015

0.03

9Pa

rtne

r: hi

gher

edu

catio

n19

.439

.57.

826

.80.

158

0.06

2In

terc

ept

−2.

986

0.24

7N

obs

erva

tions

89 68

4W

ald

chi2

8010

Pseu

do R

20.

177


312

Tabl

e 8A

.5

Sum

mar

y va

riab

les a

nd lo

git r

egre

ssio

n: p

roba

bilit

y of

bel

ongi

ng to

the

refe

renc

e gr

oup,

Nep

al 2

011

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Eas

tern

Mou

ntai

n1.

110

.31.

813

.5R

ef.

Cen

tral

Mou

ntai

n2.

114

.41.

913

.81.

075

0.23

7W

este

rn M

ount

ain

0.5

7.1

3.2

17.7

−0.

762

0.35

0E

aste

rn H

ill2.

515

.68.

127

.3−

0.54

80.

262

Cen

tral

Hill

11.6

32.0

11.4

31.8

0.33

50.

225

Wes

tern

Hill

25.9

43.8

10.0

30.0

1.52

80.

207

Mid

- Wes

tern

Hill

4.0

19.7

5.5

22.8

0.57

50.

241

Far-

Wes

tern

Hill

2.9

16.8

3.3

17.9

1.07

10.

242

Eas

tern

Ter

ai14

.935

.714

.635

.40.

550

0.21

9C

entr

al T

erai

13.0

33.6

21.0

40.7

0.51

70.

229

Wes

tern

Ter

ai13

.233

.88.

327

.61.

192

0.21

6M

id- W

este

rn T

erai

3.4

18.1

5.5

22.9

0.50

10.

239

Far-

Wes

tern

Ter

ai4.

921

.65.

222

.20.

947

0.24

6R

ural

82.7

37.9

87.5

33.1

0.28

60.

106

2 re

late

d ad

ults

29.1

45.4

29.4

45.5

Ref

.3+

rela

ted

adul

ts58

.949

.257

.449

.40.

170

0.11

9O

ther

12.1

32.6

13.2

33.9

−0.

203

0.17

2Fe

mal

e he

ad28

.845

.328

.145

.00.

167

0.13

5H

ead’

s age

−0.

001

0.02

0H

ead’

s age

squa

red

× 1

000.

020

0.01

9A

ge33

.28.

631

.18.

70.

052

0.04

1A

ge sq

uare

d ×

100

−0.

010

0.06

0N

mem

bers

4.5

1.9

5.3

2.6

−0.

151

0.03

3


313

N c

hild

ren

(5yo

)0.

50.

70.

80.

9−

0.10

40.

069

N li

ving

chi

ldre

n2.

21.

42.

51.

70.

067

0.04

0Pr

imar

y ed

ucat

ion

(inco

mpl

ete)

33.0

47.0

65.1

47.7

Ref

.Pr

imar

y ed

ucat

ion

(com

plet

e)5.

723

.26.

524

.7−

0.04

20.

206

Seco

ndar

y ed

ucat

ion

(inco

mpl

ete)

23.7

42.5

18.1

38.5

0.29

90.

147

Seco

ndar

y ed

ucat

ion

(com

plet

e)18

.538

.86.

524

.70.

906

0.17

5H

ighe

r edu

catio

n19

.139

.33.

819

.11.

336

0.20

5Ill

itera

te22

.241

.654

.049

.8−

0.69

10.

137

Hea

d’s P

rimar

y (in

com

plet

e)35

.647

.963

.248

.2R

ef.

Hea

d’s P

rimar

y (c

ompl

ete)

5.6

23.0

7.4

26.1

0.37

10.

199

Hea

d’s S

econ

dary

(inc

ompl

ete)

23.3

42.3

18.2

38.6

0.52

50.

149

Hea

d’s S

econ

dary

(com

plet

e)16

.637

.26.

524

.60.

760

0.18

1H

ead’

s Hig

her e

duca

tion

18.9

39.1

4.8

21.4

0.66

80.

212

Cur

rent

ly m

arrie

d96

.019

.696

.518

.4R

ef.

Form

erly

mar

ried

4.0

19.6

3.5

18.4

−0.

033

0.20

1Te

en m

arria

ge41

.149

.261

.448

.7−

0.17

30.

090

No

wor

king

/oth

er35

.848

.038

.748

.7R

ef.

Prof

essio

nal,

cler

ical

8.8

28.3

2.2

14.5

−0.

184

0.21

1Se

rvic

es, s

kille

d13

.434

.111

.231

.5−

0.38

90.

163

Agr

icul

ture

41.8

49.3

46.0

49.8

0.02

40.

188

Uns

kille

d m

anua

l0.

35.

11.

913

.8−

1.32

20.

521

Wor

ked

last

12

mon

ths

23.2

42.2

22.6

41.9

−0.

274

0.14

5N

ot p

aid

wor

k41

.449

.342

.549

.40.

065

0.15

6


314

Tabl

e 8A

.5

(con

tinue

d)

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Part

ner:

prof

essio

nal/t

echn

ical

/man

ager

ial

18.2

38.6

5.0

21.8

Ref

.Pa

rtne

r: cl

eric

al9.

729

.611

.531

.9−

0.45

70.

169

Part

ner:

agric

ultu

ral

22.0

41.4

28.1

44.9

−0.

303

0.15

8Pa

rtne

r: se

rvic

es36

.148

.123

.742

.5−

0.12

70.

136

Part

ner:

skill

ed m

anua

l6.

524

.616

.236

.8−

0.83

60.

191

Part

ner:

unsk

illed

man

ual

7.4

26.2

15.5

36.2

−0.

637

0.19

2Pa

rtne

r: no

edu

catio

n/do

n’t k

now

7.2

25.9

24.0

42.7

Ref

.Pa

rtne

r: pr

imar

y ed

ucat

ion

11.5

31.9

25.4

43.5

0.17

90.

171

Part

ner:

seco

ndar

y ed

ucat

ion

49.1

50.0

42.0

49.4

0.56

20.

172

Part

ner:

high

er e

duca

tion

32.2

46.8

8.7

28.2

0.89

90.

214

Inte

rcep

t−

4.52

40.

871

N o

bser

vatio

ns98

37W

ald

chi2

1000

Pseu

do R

20.

228

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.


315

Tabl

e 8A

.6

Sum

mar

y va

riab

les a

nd lo

git r

egre

ssio

n: p

roba

bilit

y of

bel

ongi

ng to

the

refe

renc

e gr

oup,

Pak

ista

n 20

06/0

7

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Dist

rict Q

uart

ile 1

8.2

27.4

27.1

44.4

Ref

.D

istric

t Qua

rtile

215

.336

.027

.144

.40.

488

0.19

1D

istric

t Qua

rtile

316

.637

.226

.444

.10.

209

0.19

3D

istric

t Qua

rtile

459

.949

.019

.539

.61.

116

0.17

0R

ural

15.4

36.1

71.1

45.3

−1.

621

0.14

12

rela

ted

adul

ts25

.143

.421

.541

.1R

ef.

3+ re

late

d ad

ults

72.4

44.7

75.5

43.0

−0.

237

0.16

3O

ther

2.5

15.7

3.0

17.2

−0.

026

0.39

1Fe

mal

e he

ad8.

327

.69.

729

.5−

0.27

90.

262

Hea

d’s a

ge46

.012

.347

.414

.70.

036

0.02

9H

ead’

s age

squa

red

× 1

00−

0.04

50.

029

Age

33.3

7.9

32.2

8.7

0.05

80.

054

Age

squa

red

× 1

00−

0.07

20.

078

N m

embe

rs7.

94.

28.

64.

80.

026

0.02

7N

chi

ldre

n (5

yo)

1.2

1.3

1.5

1.5

−0.

098

0.06

7N

livi

ng c

hild

ren

3.3

2.4

3.5

2.5

0.05

00.

036

Prim

ary

educ

atio

n (in

com

plet

e)28

.745

.373

.544

.1R

ef.

Prim

ary

educ

atio

n (c

ompl

ete)

10.7

30.9

9.2

28.9

0.70

00.

216

Seco

ndar

y ed

ucat

ion

(inco

mpl

ete)

12.6

33.2

6.6

24.8

0.71

60.

245

Seco

ndar

y ed

ucat

ion

(com

plet

e)21

.641

.26.

123

.81.

254

0.25

9H

ighe

r edu

catio

n26

.444

.14.

721

.21.

768

0.28

5Ill

itera

te31

.546

.574

.343

.7−

0.18

80.

209

Hea

d’s P

rimar

y (in

com

plet

e)26

.844

.354

.849

.8R

ef.


316

Tabl

e 8A

.6

(con

tinue

d)

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nsd

Hea

d’s P

rimar

y (c

ompl

ete)

10.2

30.3

11.5

31.8

0.17

60.

218

Hea

d’s S

econ

dary

(inc

ompl

ete)

16.3

37.0

11.1

31.4

0.32

40.

245

Hea

d’s S

econ

dary

(com

plet

e)18

.839

.113

.033

.60.

022

0.22

9H

ead’

s Hig

her e

duca

tion

27.8

44.8

9.7

29.6

0.10

50.

265

Cur

rent

ly m

arrie

d95

.620

.595

.321

.1R

ef.

Form

erly

mar

ried

4.4

20.5

4.7

21.1

0.24

30.

259

Teen

mar

riage

32.8

47.0

47.9

50.0

−0.

078

0.12

5N

o w

orki

ng/o

ther

83.9

36.8

74.3

43.7

Ref

.Pr

ofes

siona

l, cl

eric

al4.

821

.42.

014

.0−

0.40

80.

337

Serv

ices

, ski

lled

9.0

28.6

9.8

29.8

0.15

60.

272

Agr

icul

ture

0.6

7.6

12.3

32.8

−1.

367

0.50

9U

nski

lled

man

ual

1.8

13.3

1.5

12.3

0.63

80.

456

Wor

ked

last

12

mon

ths

79.8

40.2

68.8

46.3

0.19

90.

222

Not

pai

d w

ork

1.1

10.7

3.6

18.6

0.36

80.

411

Part

ner:

did

not w

ork

2.9

16.8

4.0

19.5

Ref

.Pa

rtne

r: pr

ofes

siona

l/tec

hnic

al/m

anag

eria

l16

.336

.98.

427

.70.

118

0.31

4


317

Part

ner:

cler

ical

3.7

18.8

3.8

19.0

−0.

586

0.36

3Pa

rtne

r: sa

les

23.4

42.4

12.4

33.0

0.27

10.

301

Part

ner:

agric

ultu

ral

4.5

20.7

22.9

42.0

−0.

223

0.34

3Pa

rtne

r: se

rvic

es11

.732

.110

.230

.20.

038

0.31

4Pa

rtne

r: sk

illed

man

ual

20.1

40.1

15.4

36.1

0.11

90.

296

Part

ner:

unsk

illed

man

ual

17.5

38.0

23.0

42.1

0.22

20.

294

Part

ner:

no e

duca

tion/

don’

t kno

w18

.238

.636

.948

.3R

ef.

Part

ner:

prim

ary

educ

atio

n11

.131

.416

.537

.20.

043

0.20

2Pa

rtne

r: se

cond

ary

educ

atio

n38

.348

.733

.347

.10.

012

0.20

1Pa

rtne

r: hi

gher

edu

catio

n32

.446

.813

.233

.9−

0.09

80.

263

Inte

rcep

t−

4.83

41.

171

N o

bser

vatio

ns10

017

Wal

d ch

i266

6Ps

eudo

R2

0.27

6

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.


318

Tabl

e 8A

.7

Sum

mar

y va

riab

les a

nd lo

git r

egre

ssio

n: p

roba

bilit

y of

bel

ongi

ng to

the

refe

renc

e gr

oup,

the

Phi

lippi

nes 2

008

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nSd

Prov

ince

Dec

ile 1

2.7

16.4

14.8

35.6

Ref

.Pr

ovin

ce D

ecile

22.

616

.014

.034

.7−

0.02

10.

164

Prov

ince

Dec

ile 3

1.8

13.2

14.2

34.9

−0.

512

0.17

9Pr

ovin

ce D

ecile

46.

624

.813

.734

.40.

781

0.15

7Pr

ovin

ce D

ecile

53.

017

.213

.534

.2−

0.02

00.

176

Prov

ince

Dec

ile 6

6.6

24.9

9.4

29.2

1.23

00.

166

Prov

ince

Dec

ile 7

21.5

41.1

6.1

23.9

2.87

70.

154

Prov

ince

Dec

ile 8

23.3

42.3

4.3

20.2

3.47

90.

167

Prov

ince

Dec

ile 9

16.6

37.2

4.7

21.2

2.99

70.

169

Prov

ince

Dec

ile 1

015

.236

.05.

322

.32.

802

0.17

4R

ural

27.9

44.9

55.9

49.7

0.32

60.

090

2 re

late

d ad

ults

33.9

47.3

40.4

49.1

Ref

.3+

rela

ted

adul

ts55

.449

.750

.550

.00.

096

0.09

4O

ther

10.7

30.9

9.1

28.7

−0.

187

0.13

1Fe

mal

e he

ad15

.035

.711

.031

.30.

114

0.11

0H

ead’

s age

43.5

12.5

42.8

12.1

0.01

30.

018

Hea

d’s a

ge sq

uare

d ×

100

−0.

011

0.01

8A

ge33

.78.

534

.38.

5−

0.05

60.

034

Age

squa

red

× 1

000.

057

0.04

8N

mem

bers

5.7

2.3

5.7

2.4

0.00

20.

022

N c

hild

ren

(5yo

)0.

90.

91.

01.

0−

0.02

20.

047

N li

ving

chi

ldre

n2.

51.

83.

02.

20.

013

0.02

5


319

Prim

ary

educ

atio

n (in

com

plet

e)4.

420

.614

.635

.3R

ef.

Prim

ary

educ

atio

n (c

ompl

ete)

10.8

31.1

15.6

36.3

0.22

70.

167

Seco

ndar

y ed

ucat

ion

(inco

mpl

ete)

12.3

32.8

16.4

37.0

0.11

60.

178

Seco

ndar

y ed

ucat

ion

(com

plet

e)35

.347

.826

.644

.20.

376

0.17

1H

ighe

r edu

catio

n37

.248

.326

.844

.30.

432

0.18

1Ill

itera

te2.

816

.411

.131

.4−

0.59

40.

190

Hea

d’s P

rimar

y (in

com

plet

e)11

.331

.723

.842

.6R

ef.

Hea

d’s P

rimar

y (c

ompl

ete)

15.4

36.1

16.8

37.4

0.30

60.

127

Hea

d’s S

econ

dary

(inc

ompl

ete)

11.4

31.8

13.0

33.6

0.05

60.

158

Hea

d’s S

econ

dary

(com

plet

e)26

.444

.122

.041

.4−

0.01

40.

148

Hea

d’s H

ighe

r edu

catio

n35

.447

.824

.442

.90.

027

0.15

0N

o im

mig

rant

29.1

45.4

31.3

46.4

Ref

.Im

mig

rant

(<=

5 ye

ars/

visit

or)

32.4

46.8

29.6

45.6

−0.

456

0.09

3Im

mig

rant

(6+

yea

rs)

38.5

48.7

39.2

48.8

−0.

279

0.08

8Im

mig

rant

from

rura

l are

a25

.843

.833

.847

.3−

0.09

00.

087

Cur

rent

ly m

arrie

d91

.228

.393

.524

.6R

ef.

Form

erly

mar

ried

8.8

28.3

6.5

24.6

0.03

80.

122

Teen

mar

riage

18.8

39.0

22.6

41.8

0.28

20.

094

No

wor

king

/oth

er53

.549

.952

.749

.9R

ef.

Prof

essio

nal,

cler

ical

20.4

40.3

17.7

38.1

0.01

70.

118

Sale

s, se

rvic

es, s

kille

d21

.641

.216

.437

.00.

318

0.11

5A

gric

ultu

re2.

314

.911

.632

.0−

0.19

60.

192

Uns

kille

d m

anua

l2.

314

.81.

712

.90.

193

0.24

8W

orke

d la

st 1

2 m

onth

s38

.948

.838

.448

.60.

061

0.10

3N

ot p

aid

wor

k0.

68.

05.

622

.9−

1.85

70.

302


320

Tabl

e 8A

.7

(con

tinue

d)

Ref

eren

ceC

ompa

rison

Coe

f.St

d. e

rr.

Mea

nsd

Mea

nSd

Part

ner:

did

not w

ork/

no p

artn

er4.

721

.12.

916

.8R

ef.

Part

ner:

prof

essio

nal/t

echn

ical

/man

ager

ial

16.9

37.5

10.6

30.8

0.18

90.

175

Part

ner:

cler

ical

2.6

15.9

1.4

11.7

0.29

60.

256

Part

ner:

sale

s7.

626

.65.

222

.20.

314

0.19

7Pa

rtne

r: ag

ricul

tura

l10

.530

.629

.945

.8−

0.03

00.

182

Part

ner:

dom

estic

2.1

14.3

1.0

10.0

0.38

00.

269

Part

ner:

serv

ices

8.0

27.2

6.6

24.8

−0.

142

0.18

9Pa

rtne

r: sk

illed

man

ual

34.4

47.5

27.9

44.8

−0.

008

0.16

2Pa

rtne

r: un

skill

ed m

anua

l13

.133

.814

.635

.30.

125

0.17

4Pa

rtne

r: no

edu

catio

n/do

n’t k

now

0.3

5.6

2.6

16.0

Ref

.Pa

rtne

r: pr

imar

y ed

ucat

ion

17.8

38.2

32.3

46.8

0.89

70.

433

Part

ner:

seco

ndar

y ed

ucat

ion

43.2

49.5

38.3

48.6

1.07

90.

435

Part

ner:

high

er e

duca

tion

38.8

48.7

26.8

44.3

1.21

70.

437

Inte

rcep

t−

3.18

90.

853

N o

bser

vatio

ns91

94W

ald

chi2

1823

Pseu

do R

20.

287

Not

es:

Elig

ible

pop

ulat

ion:

eve

r- m

arrie

d w

omen

, 15–

49 y

ears

old

. Eth

nic

grou

ps a

s list

ed in

Tab

le 8

.1.


321

Index

Aaberge, R. 6, 150, 157, 180, 184–5social poverty function of 162–5

Adger, N. 73, 142Alkire, S. 150, 156–7, 167, 177, 185,

200concept of

‘association decreasing rearrangement among the poor’ 160

‘weak transfers’ 160counting approach of 149, 153–5,

158, 185, 190–91, 199dimension-adjusted

multidimensional poverty measure 160–61

use of dual cutoff method 24Almond, D. 218Amin, S. 71

study of household-level vulnerability 69–70

Aristotle 37Arrow–Pratt absolute risk-aversion

measure 86–7Asian Development Bank (ADB) 13,

19, 253estimation of national poverty lines

19Asian financial crisis (1997–98) 201Atkinson, A.B.

measurement of poverty lines 31Azerbaijan 255–6, 259

ethnic groups in 294comparison group 256mean wealth differences 266, 268reference group 256

vulnerability-adjusted poverty line of 98

Balicasan, A. 151Bangladesh 69

national poverty line in 20

poverty rate in 98Beetsma, R.M.

estimation of CRRA 97Binswanger, H.P.

observation of vulnerability to poverty in India 54

Bossert, W.concept of ‘S-convexity’ 160counting measures of 160

Bourguignon, F. 153measurement of poverty lines 31

British Household Panel Survey (BHPS) 35

Bulgaria 55Butler, R.J.

interdistributional Lorenz curve 263

Calandrino, M.study of vulnerability at household

level in PRC 68Calvo, C. 55, 66, 85–6, 126–7, 141, 143

role in development of axiomatic measure of vulnerability to poverty 60–65, 67, 76

Cambodia 6, 150–51, 166–8, 191, 199, 201, 253

economic growth in 151, 165, 200poverty reduction 166, 176, 184–5,

201Mountains region of 178, 184–5, 200Phnom Penh 177, 184–5, 200Plains and Coastal regions of 177–8

Cameroon 75Casimiro, G.G. 151Chakravarty, S. 6, 38, 48, 86, 133, 150,

153, 157, 180, 184–5, 199concept of amalgam poverty line

31–3concept of non-decreasingness

of marginal social exclusion (NMS) 159, 161



counting measures of 160, 165multidimensional class of poverty

measures 161–2study of partial ordering of

vulnerability 65Chaudhuri, S. 85–6, 129, 143

role in development of expected poverty measure of vulnerability to poverty 56–7

Chen, S. 30–31estimation of national poverty lines

22proposal for ‘weakly relative’

international poverty line 22–3China, People’s Republic of (PRC) 5,

13, 22, 59extreme poverty rate in 1national poverty line of 2, 13

headcount ratio in 99, 103poverty rate in 98Shanghai 68subjective well-being in 35–6vulnerability-adjusted poverty line

of 98vulnerability to poverty in 68–9

Chinese Household Income Project 68

Chiwaula, L.S.role in development of expected

poverty measure of vulnerability to poverty 75

Christiaensen, L. 86use of FGT measure in definition of

vulnerability 59Clark, A.E. 32, 35–6

use of European Social Survey 36climate change 118–20

El Niño–Southern Oscillation (ENSO) 121, 129–30, 136–7, 140–42

physical impact of 119droughts 120–22, 125, 134, 136,

139, 141flooding risk 119–20, 122, 124–5,

131, 134, 136, 139, 141–3relationship with vulnerability to

poverty 73–4, 129, 140–41counting approaches 151–3, 160, 185,

190individual poverty function 154–6

social poverty function 157–8, 162–4Currie, J. 218

D’Ambrosio, C. 6, 150, 157, 180, 184–5, 199

concept of non-decreasingness of marginal social exclusion (NMS) 159, 161

counting measures of 160, 165multidimensional class of poverty

measures 161–2Dang, H.H. 87

estimation of poverty and vulnerability in Papua New Guinea 71

Datt, G. 155, 159Deaton, A.

observation of development of global poverty 20

Dercon, S. 55, 66, 70, 78, 85–6, 126–7, 141, 143

role in development of axiomatic measure of vulnerability to poverty 60–65, 67, 76

Devarajan, S. 78Dhanani, S. 78Duesenberry, J. 2, 34Dutta, I. 86

role in development of axiomatic measure of vulnerability to poverty 64–5

EconLit 53Elbers, C. 75

role in development of welfarist measure of vulnerability to poverty 55

use of Ramsey model 55Eswaran, M. 53Ethiopia 67, 70–71ethnic groups 69, 253–6, 281–2, 290,

292–3, 295–7absolute poverty gap curves 269–72advantaged 7differences in mean wealth 266–7disadvantaged 253, 262, 268, 278,

282–3, 295outstanding groups 272–3poverty gap curves 262–4, 278

cumulative distribution functions


Index 323

(CDFs) 262–3, 265, 268, 272–3

relative poverty gap curve 272European Social Survey 97

third wave of 36European Union (EU)

measurement of poverty lines 31

Fafchamps, M. observations of subjective welfare in

Nepal 35Ferrer-i-Carbonell, A. 2Foster, J. 150, 156–7, 167, 177, 185, 200

concept of ‘association decreasing

rearrangement among the poor’ 160

‘weak transfers’ 160counting approach of 149, 153–5,

185, 190–91, 199dimension-adjusted

multidimensional poverty measure 160–61

Foster–Greer–Thorbecke (FGT) poverty measure 59–60, 63, 67, 86, 131–2, 160, 263, 268, 272

use of dual cutoff method 24Fuente, A. de la 71

Gaiha, R. 85Gallup World Poll 97Gandelman, N.

estimation of CRRA 97Georgia


Gerry, C.use of quantile regression 71

Glewwe, P. 71Global Financial Crisis (2007–09) 215Gradin, C. 264Greb, F.

observation of development of global poverty 20

Gunning, J.W. 75, 85role in development of welfarist

measure of vulnerability to poverty 55

use of Ramsey model 55Günther, I. 70, 77, 86

Hadi, P.U. 122Hall, G. 71Hardeweg, B. 65, 69Hartley, R. 97

estimation of CRRA 97Harttgen, K. 70, 77, 86Health and Demographic Surveys 6Hernández-Murillo, R.

estimation of CRRA 97Hersch, P.L.

estimation of CRRA 97Hinduism

caste system of 256Hirschman, A.O. 35Hoddinott, J. 85–6Holzmann, R.

role in development of expected poverty measure of vulnerability to poverty 85

human capability 7–8Human Development Index (HDI)

149, 167

Imai, K. 85study of vulnerability at household

level in PRC 68income gap 35, 38India 7, 22, 69, 255, 259, 294

Constitution of 256ethnic groups in 256, 272, 296

absolute poverty gap curve 272mean wealth differences 266Other Backward Class (OBC) 256,

268, 278, 284, 289poverty gap trends 278–9, 283,

290 Scheduled Castes (SC) 256, 268,

278–9, 284, 289–90Scheduled Tribes (ST) 256, 268,

278, 284, 289–90, 294–6national poverty line of 2, 13,

20–21 headcount ratio in 99, 103

poverty rate in 98vulnerability to poverty in

uninsured weather risk 54Indonesia 6, 13, 58, 78, 130, 141,

150–51, 165–8, 199, 201, 253Borneo

Kalimantan 179, 190



climate change impact in droughts 120–22, 125flooding risk 119–20, 122, 124–5

economic growth inpoverty reduction 166, 178, 201

ethnic groups in 256poverty amongst 261

Jakarta 120Jakarta Province Regional

Disaster Mitigation Agency 122

Java 178–9, 185Bali 179, 185

Manokwari 120Ministry of Agriculture

Directorate General of Crop Protection 122

National Disaster Management Agency (Badan Nasional Penanggulangan Bencana) 118–19

national poverty line of 120headcount ratio in 99, 103

poverty rate in 98Ruteng Park 121Sulawesi 179, 185, 190, 199

Indonesia Family Life Survey (IFLS) 122–5, 130–31, 133

first round of (IFLS 1) 122–3, 133 fourth round of (IFLS 4) 131, 133–4,

137, 139–40second round of (IFLS 2) 122–3,

133–4, 137, 139third round of (IFLS 3) 122–3, 133

Indonesian Central Bureau of Statistics (BPS)

Consumer Price Index (CPI) 126National Socio-Economic Survey

(SUSENAS) (1993) 122information theory 152Intergovernmental Panel on Climate

Change (IPCC)Working Group II

Fifth Assessment Report 119Islam, I. 78

Jack, W. 78Jalan, J.

definition of total poverty 60study of rural poverty in PRC 68–9

JapanMeiji Restoration (1868) 8

Jha, R.analysis of poverty and vulnerability

in Tajikistan 69estimation of poverty and

vulnerability in Papua New Guinea 71

Jorgensen, S.role in development of expected

poverty measure of vulnerability to poverty 85

Kabubo-Mariara, J. 218Kamanou, G. 86

use of expected poverty measure of vulnerability 58

Kenya 59Klasen, S. 19, 70, 85

proposed use of standardized methodology in setting national poverty lines 23

Knight, J. 35Kolm, S.C.

study of multidimensional inequality 159–60

Koren, M.analysis of volatility of GDP growth

76Korkeala, O. 121Kotwal, A. 53Kramer, R.A. 121–2Krishnan, P. 70Kurosaki, T. 71

study of vulnerability 70

labor 283force status 36market 71, 282

Lanjouw, P.F. 87Le Breton, M.

first-order discrimination curve of 263

Li, C.use of quantile regression 71

Ligon, E. 63, 69, 85, 89definition of vulnerability

103 role in development of welfarist


Index 325


static analytic framework 55Lorenz curve 108

macroeconomics 77Madagascar 70Malaysia


McCulloch, N.study of vulnerability at household

level in PRC 68McDonald, J.B.

interdistributional Lorenz curve 263McDougall, G.S.

estimation of CRRA 97Mexico

remittance flows in 71microcredit 78Mongolia

national poverty line in 20Montalbano, P. 76Monte Carlo method 58Morduch, J. 78, 86

use of expected poverty measure of vulnerability 58

Moser, C. 74

Naudé, W. 76Naylor, R. 121Nepal 7, 255, 294

ethnic groups in 256, 272, 294–6absolute poverty gap curve 271mean wealth differences 266outstanding groups 278poverty gap 283–4, 292

national poverty line in 20poverty rate in 98subjective welfare in 35

Nigeria 75nutrition 215–17

children 218–19, 234, 237–8, 241–3childcare 219, 238

malnutrition 217–18measurements of 217

Body Mass Index (BMI) 217–18, 227

height-for-age (HFA) 217, 227, 233, 242

weight-for-age (WFA) 217–18, 220, 226–7, 233–4, 241–2

weight-for-height (WFH) 217, 227relationship with poverty 6, 216, 219,

227, 230, 234, 237, 244undernourishment/undernutrition

215–18, 242–3

Oaxaca–Blinder procedure 265ordinary least squares (OLS) regression

128–9, 220, 234, 238, 241models

Durban–Wu–Hausman test 220Oswald, A.J. 32, 35

Pakistan 7, 70, 255ethnic groups in 272, 283, 296

absolute poverty gap curve 271poverty gap 283, 292–3

national poverty line of headcount ratio in 99, 103

poverty rate in 98Papua New Guinea 71Pasaribu, S.M. 122Pasha, A. 25Pattanayak, S.K. 121–2Peluso, E. 6, 150, 157, 180, 184–5

social poverty function of 162–5Penn World Tables 21Peru 71Philippines 6–7, 150–51, 166–8, 199,

201, 253, 255economic growth in 151

poverty reduction 166, 190Tagalog 256

ethnic groups in 272, 294–6absolute poverty gap curve 272comparison groups 256mean wealth differences 266, 268outstanding groups 278poverty amongst 261poverty gap trends 278–9, 283, 294

Mindanao 180national poverty line of

headcount ratio in 103poverty rate in 98, 165Visayas 180

Povel, F. 70, 85poverty 8, 37, 68, 72, 87, 152, 176–7,

179–80, 190, 201–2, 233–4



chronic 60consumption-based definition of

125–6extreme 1

decline in 1–2global

observed 20international 20–21line 2–6, 14–15, 18–19, 23–4, 53, 55,

86–7, 89–90, 97, 120, 125–6, 140, 153, 227, 233

absolute 5, 30, 32–3, 37–8, 48adjusted 5, 91, 93amalgam 33, 38Asian 14–15, 17, 19–21, 241country-level income 21estimation of 13–14, 17–20impact of vulnerability on 89international income 13issues of 20–22modification of 2multidimensional 17PPP-adjusted 14vulnerability-adjusted 97–8

multidimensional 3–4, 6, 13, 16, 149–50, 158, 165, 167–8, 176, 200

Aaberge and Pelusio extension (Ext. AP) 190, 200

Chakravarty and D’Ambrosio (CDA) measure 190, 200

counting approaches to 151–65multidimensional transfer

principle (MTP) 160non-monetary 3rates of 1reduction efforts 13, 166, 176,

184–5, 201, 215–16, 223–4, 233, 253

relationship with nutrition 6, 216, 219, 227, 230, 234, 237, 244

social 150urban 178, 185, 190–91, 201

poverty gap 31poverty trap 53Powdthavee, N. 34Praag, B. van 34Pritchett, L. 57purchasing power parity (PPP) 4,

20–21, 26, 97

-adjusted poverty 4line 14

exchange rates 23

Quisumbing, A.R. 71, 85–6

RAND Corporationconducting of IFLS (1993–94) 122–3

Ravallion, M. 30–31, 37definition of total poverty 60estimation of national poverty lines

19, 22use of linear model 19

proposal for ‘weakly relative’ international poverty line 22–3

role in development of expected poverty measure of vulnerability to poverty 55, 85

study of rural poverty in PRC 68–9

Reddy, S.proposed use of standardized

methodology in setting national poverty lines 23

Rippin, N. 6, 150, 153, 156, 184counting measures of 160, 191

multidimensional class of ordinal poverty measures 161

Rippin decomposition 191non-decreasingness under inequality-

increasing switch (NDS) 159, 161

risk aversion 87–9coefficient of relative risk aversion

(CRRA) 97–8, 103constant absolute 89–90constant relative 91–2noise 91

additive noise model 89, 93multiplicative noise model 93

risk aversion indicators 87absolute

Arrow–Pratt measure 87–8Roma

vulnerability to poverty 70Rosenzweig, M.

observation of vulnerability to poverty in India 54

Rothschild, M. 35, 89Runciman, W.G. 34


Index 327

Russian Federation 71nutritional well-being in 75

Schechter, L. 63, 69, 85, 89definition of vulnerability 103role in development of welfarist


static analytic framework 55Schotman, P.C.

estimation of CRRA 97Second World War (1939–45) 53Sen, Amartya 7

role in development of poverty definitions 152, 159

rank-dependent framework 162Senik, C. 35

use of European Social Survey 36Seth, S. 153

counting approach of 158Shilpi, F.

observations of subjective welfare in Nepal 35

Shorrocks, A. 38Silber, J. 150, 153, 155–6, 184, 199

concept of ‘social poverty function’ 157, 159, 163–4

Skoufias, E. 70Smith, Adam

Wealth of Nations 30South Africa 76Soviet Union (USSR) 21Stiglitz, J.E. 89Stillman, S.

study of nutritional well-being in Russian Federation 75

Subbarao, K. 86use of FGT measure in definition of

vulnerability 59subjective well-being 33

reference groups 33–5, 48Sumarto, S. 86


observations of vulnerability in Indonesia 58

Survey METRErole in conducting of IFLS 4 123

Suryahadi, A. 86


observations of vulnerability in Indonesia 58

Szpiro, G.G.estimates of CRRA 97

Tajikistan 69national poverty line in 15, 20–21vulnerability to poverty in 54vulnerability-adjusted poverty line

of 98Tanzania 54taxation 37Tenreyro, S.

analysis of volatility of GDP growth 76

Thailand 6–7, 215Buri Ram 221Nakhon Phanom 221poverty reduction efforts in 220, 233

poverty headcounts 221, 223–4rural population of

nutritional status of 220, 226, 230, 232–3, 238, 241–3

Ubon Ratchathani 221vulnerability-adjusted poverty line

of 98Thomas, D.

study of nutritional well-being in Russian Federation 75

Thorbecke, E. 152Tobit models

Smith–Blundell test for 220Tobit regression 237Townsend, P.

role in development of poverty definitions 152

Turkmenistanvulnerability-adjusted poverty line

of 98

unemployment 71United Kingdom (UK) 97United Nations (UN)

Children’s Fund (UNICEF) 151, 218nutritional framework of 218–19

Development Programme (UNDP) 149, 167



Multidimensional Poverty Index (MPI) 3–4, 6, 14, 16, 24–5, 149–53, 156–7, 161, 166, 199

Millennium Development Goals (MDGs) 1, 13, 20, 150–51, 216–17

United States (US) 123Agency for International

Development (USAID)Demographic and Health Surveys

(DHS) 151, 166, 199, 219–20, 255–6, 259

Monitoring and Evaluation to Assess and Use Results Demographic and Health Surveys (MEASURE DHS) 255

University of California, Los Angelesrole in conducting of IFLS 2 122

University of Gadjah Mada Center for Population and Policy

Studiesrole in conducting of IFLS 4 123

Population Research Center role in conducting of IFLS 3 123

University of IndonesiaLembaga Demografi

role in conducting of IFLS 2 122US Health and Retirement Survey 97utility function 32, 87, 91–2

constant relative risk aversion 56household 218instantaneous 55–6Taylor’s expansion 90–91von Neumann-Morgenstern 85

Viet Nam 6–7, 13, 69, 215, 253, 255Central Highlands

Dak Lak 221, 241Hat Thinh 221, 241

ethnic groups in absolute poverty gap curve 271comparison group 256mean wealth differences 266reference group 256

poverty reduction efforts in 220, 233poverty headcounts 221, 223–4

rural population of nutritional status of 220, 226, 230,

232–3, 237, 238, 241–3


vulnerability 2–3, 5–6, 53–4, 68, 72–3, 77, 84–5, 89, 91, 93–4, 103, 215

additive model of 5climate change 73

impacts of 73relationship with vulnerability to

poverty 73–4, 129, 140, 143definition of measures of 59–60,

70–71, 84axiomatic 60–65, 67, 76expected poverty 56–8, 75, 85welfarist 55–6

impact of trade openness 75–6impact on poverty line 89long panel data issues 77as low expected utility 85measures of 126–9, 134, 141, 143multiplicative model 94–7non-monetary outcomes 74–5nutritional well-being 75potential policies addressing 77–8role of personal networks 71

Wan, G. 38definition of vulnerability 59

study of vulnerability at household level in PRC 68

Watts, H.W. poverty measure of 86, 133

wealth index 258–9construction of 260–62

welfare function 56World Bank

measurement of international poverty line 13–14, 17–18, 20–23, 25, 97

World Development Indicators 126World Health Organization (WHO)

217, 227, 242, 244Global Targets 2025 216threshold for child malnutrition

242–3World Values Survey 97

Yaari, M.E.rank-dependent framework of 162

Yalonetzky, G. 6, 150, 153, 155–6, 184, 199


Index 329

concept of ‘social poverty function’ 157, 159, 163–4

Yemennational poverty line in 20–21

Zhang, Y.definition of vulnerability 59

study of vulnerability at household level in PRC 68

Zimbabwe 56


adbi series on asian economic integration and · pdf filemonetary and currency policy ......

Documents