Mapping Hunger: A Report on Mapping Malnutrition Prevalence in the

Dominican Republic, Ecuador, and Panama

Beatrice Lorge Rogers, James Wirth, Kathy Macías, Parke Wilde

Gerald J and Dorothy R Friedman School of Nutrition Science and Policy, Tufts University

Boston, Massachusetts USA

March 22, 2007

Acknowledgments

This report is the result of a collaboration between the World Food Programme, Office for Latin America and the Caribbean, and the Friedman School of Nutrition Science and Policy, Tufts University, Boston. The authors deeply appreciate the advice and support of our project officers, Judith Thimke, Carlos Acosta, and Mahadevan Ramachandran. We received helpful advice on the PovMap method from Qinghua Zhao and Peter Lanjouw of the World Bank. We appreciate the responsiveness of individuals responsible for management of the data sets in the three countries with which we worked. In Panama, Roberto González Batista and Edith de Kowalczyk Arosemena of the Ministerio de Economía y Finanzas (MEF); in Ecuador, Reinaldo Cervantes, César Amores, and Adriano Rodríguez of the Frente Social, INEC, and Paul Stupp, CDC/ENDEMAIN. We received many responses from other individuals in these offices; their help is acknowledged even though we cannot identify all of them by name. Patrick Florance at Tufts University provided help in implementing GIS analysis. We also received support from the offices of the World Food Program: Maria Paz Salas and Pavel Isa in the Dominican Republic, Nelson Herrera in Ecuador and Xinia Soto and Lilibeth Herrera at WFP/LAC in Panama, for which we are grateful.

ii

Table of Contents

1. Introduction................................................................................................................. 1 1.2 Goals of the Project............................................................................................. 3

2. Data............................................................................................................................. 3 2.1 General considerations........................................................................................ 3 2.2 Data Used in the Present Analysis ...................................................................... 4

2.2.1 Dominican Republic ................................................................................... 5 2.2.2 Ecuador ....................................................................................................... 5 2.2.3 Panama........................................................................................................ 6

3. Results......................................................................................................................... 8 3.1 Synthesis ............................................................................................................. 8 3.2 Country level results ......................................................................................... 16

3.2.1 Dominican Republic ................................................................................. 16 3.2.2 Ecuador ..................................................................................................... 19 3.2.3 Panama...................................................................................................... 23

4. Discussion................................................................................................................. 27 5. Next steps.................................................................................................................. 28 6. References................................................................................................................. 30 Appendix...........................................................................................................................A1

iii

List of Tables

Table 2.1. Characteristics of Data Sets Used in the Analysis 4 Table 3.1: Comparison of Malnutrition Prevalence and Mean HAZ

of Small Area Estimation and Survey Estimates: Dominican Republic 9

Table 3.2: Comparison of Malnutrition Prevalence and Mean HAZ of Small Area Estimation and Survey Estimates for Ecuador 10

Table 3.3: Comparison of Malnutrition Prevalence and Mean HAZ of Small Area Estimation and Survey Estimates: Panama 11

Table 3.4: Country Descriptive Statistics on Chronic Malnutrition 12

List of Figures

Figure 1: Range and Number of Malnourished Children In the Dominican Republic (based on HAZ) by Province and Zone 18

Figure 2: Range and Number of Malnourished Children In Ecuador

(based on HAZ) by Province and Zone 22 Figure 3: Range and Number of Malnourished Children In Panama

(based on HAZ) by Province 26

List of Maps Map 1. Prevalence of Chronic Malnutrition, Standardized Categories, Dominican Republic 13 Map 2 Prevalence of Chronic Malnutrition, Standardized Categories, Ecuador 14 Map 3. Prevalence of Chronic Malnutrition, Non-Indigenous,

Standardized Categories, Panama 15 Map 4. Prevalence of Chronic Malnutrition, Indigenous, Standardized Categories, Panama 15 Map 5 Prevalence of Chronic Malnutrition: Estimates Based on Small Area Estimation, by Municipio, Dominican Republic 16 Map 6 Number of Chronically Malnourished Children Based on Small Area Estimation by Municipio, Dominican Republic 17 Map 7 Prevalence of Chronic Malnutrition: Estimates Based on Small Area Estimation, by Canton, Ecuador 19 Map 8 Number of Chronically Malnourished Children Based on Small Area Estimation, by Canton, Ecuador 21 Map 9 Prevalence of Chronic Malnutrition, Non-Indigenous, Based on Small Area Estimation, By District, Panama 23

iv

Map 10 Prevalence of Chronic Malnutrition, Indigenous, Based on Small Area Estimation, by District, Panama 24

Map 11 Number of Chronically Malnourished Children, Non- Indigenous, Based on Small Area Estimation, by District, Panama 25 Map 12. Number of Chronically Malnourished Children, Indigenous, Based on Small Area Estimation, by District, Panama 25

Appendix Tables Table A1 Data Requirements for the Prediction of Malnutrition Table A2 Variables Used in the Analysis of Malnutrition,

Dominican Republic Table A3 Variables Used in the Analysis of Malnutrition, Ecuador Table A4 Variables Used in the Analysis of Malnutrition, Panama

v

Mapping Hunger: A Report on Mapping Malnutrition Prevalence in the

Dominican Republic, Ecuador, and Panama

Beatrice Lorge Rogers, James Wirth, Kathy Macías, Parke Wilde

Gerald J and Dorothy R Friedman School of Nutrition Science and Policy, Tufts University

Boston, Massachusetts USA

March, 2007

1. Introduction Reducing poverty and hunger is the first Millennium Development Goal (UN 2001). “Hunger” is commonly defined in terms of nutritional status, which in turn is measured as children’s anthropometric status. A first step toward realizing the goal of reducing malnutrition (here interpreted to refer to undernutrition, not nutritional excess) is to identify the places where the problem is the most severe. Localizing malnutrition makes it possible to understand better the underlying causes of the problem in different places, and to target resources appropriately to the areas in which they will make the most difference. In addition, mapping provides a powerful tool for visualization of the nature of the nutrition problem in a country. Maps, because they are intuitively interpretable, can be useful for evidence based advocacy purposes. In this study, the outcome of interest is childhood malnutrition, measured in terms of anthropometric status (height-for-age) of children under five years of age. A child falling below negative two standard deviations of the mean HAZ is considered malnourished.1 Nutrition surveys typically collect information representative at the Province level, but previous studies have found significant variation in nutritional outcomes among smaller administrative units within provinces (Fujii, 2003; Larrea et al. 2005; Benson 2006). The present study used the technique of Small Area Estimation (SAE) to analyze data from 1 Each anthropometric indicator has a different interpretation. Low height-for-age (HAZ), or stunting, is an indicator of chronic malnutrition, the result of long-term food insufficiency, often combined with other conditions (low birth weight, frequent illness). Low weight-for-height is a measure of thinness; if severe, it indicates wasting or acute malnutrition: a lack of food in the short term (also often combined with current illnesses like diarrhea and infection). Global malnutrition is measured by weight-for-age (WAZ). A child can be malnourished by this criterion if s/he is stunted or wasted; either condition would result in a child having low weight-for-age. These indicators are calculated with reference to standardized growth curves for children under age five. It is widely recognized that the growth trajectory of healthy, well-nourished children is similar among populations, irrespective of nationality or ethnicity, so that international standards are appropriate to assess nutritional status at the population level across countries (WHO 1995; 2006). The surveys used in this study made use of the NCHS/CDC/WHO anthropometric standards for growth (Waterlow et al 1977). New standards were published in 2006 (WHO 2006) that might slightly alter the distribution of malnutrition prevalence reported here (deOnis et al 2006).

1

three countries: Ecuador, Panama, and the Dominican Republic, in order to produce sub-provincial estimates of child malnutrition. 1.1 Analytic Approach: Small Area Estimation The technique of Small Area Estimation (SAE) makes it possible to use sample survey data, combined with a national census, to develop malnutrition prevalence estimates at highly disaggregated levels. The approach of Small Area Estimation is to identify a sample survey, representative of the national population, that contains information on the outcome of interest – in this case, malnutrition. A predictive model is developed based on the information contained in the survey, using variables that are also present in the national census. The parameters derived from that predictive model are then applied to the census data, producing estimates of malnutrition for every geographic unit in the census. The level of disaggregation is constrained only by the desired level of precision: the smaller the unit, the less precise the estimate (Hentschel et al 2000). We used a program developed by the World Bank for estimating poverty prevalence, and adapted it for use with nutrition indicators.2 The program performs both steps of the process: it first performs the regression analysis using a statistically representative household sample survey, and then applies the results to census data, producing percent prevalence estimates at any level of geographic disaggregation. The PovMap program uses a statistical technique called bootstrapping to estimate a standard error around each estimate from the census data; this makes it possible to assess the precision of the estimate produced, construct confidence intervals, and determine whether two geographic units are statistically significantly different from each other.3 The present report provides estimates of malnutrition prevalence for children under age five in the Dominican Republic, Panama, and Ecuador. The key indicator of malnutrition in these countries, as in most of Latin America, is low height-for-age (HAZ), or stunting. Wasting is quite uncommon in most Latin American settings.

2 At the writing of this report, PovMap v. 2.0 (beta version) can be downloaded free at the World Bank Website: http://iresearch.worldbank.org/PovMap/index.htm. See Zhao 2005 3 SAE estimates the outcome variable for each child in a community according to the following equation: 1) Y = β0 +β1W + β2 X + β3 Z + u, where X, W, and Z are vectors of individual, household, and community characteristics respectively. The term u represents the disturbance or error term, which may be decomposed into two parts as follows: 2) u ci = η c + εci , where η c is the variance accounted for by the community, and εci is the variance accounted for at the individual levelTo produce an estimate of the variance around the estimate of the individual’s nutritional outcome, the computer repeats the regression a given number of times, sampling randomly from the variances around the parameter estimates (βs) and the error terms (η and ε). The variances are derived from the regression model estimated using the survey data. The approach is described in a number of papers that explain the underlying statistics and give examples of applications to estimation of the prevalence of poverty (Elbers, Lanjouw, and Lanjouw, 2003, 2001; Zhao, 2005; Hentschel et al. 2000; Demombynes et al. 2002) and malnutrition (e.g., Fujii 2003; 2005; Larrea 2005; Gilligan et al. 2003; Haslett and Jones 2005).

2

The prevalence of stunting varies among the three countries: at the national level, average prevalence based on our SAE results is 10.1% for the Dominican Republic, 27% for Ecuador, and 24.3% for Panama. The precision of the malnutrition estimates falls as the prevalence falls, and this is reflected in the model fit, or R2 of the regressions used to develop the predictions in PovMap, and in the standard errors of the estimates (Demombynes et al 2002). It stands to reason that in countries and regions with a high prevalence of malnutrition, environmental and socioeconomic causes predominate; as conditions improve in a country, the remaining malnutrition may be due in large part to idiosyncratic characteristics of a child’s household and caretaker – factors that cannot be included in the present model because such information is not included in the census.

1.2 Goals of the Project The goal of the present project was to apply the PovMap method to the estimation of malnutrition prevalence in three countries: The Dominican Republic, Ecuador, and Panama. The purpose was to see whether the method, developed for poverty estimation, would produce estimates of malnutrition prevalence that were consistent with previous survey results, and then to use the technique to estimate the prevalence of malnutrition at a more disaggregated level. A second purpose of this project was to develop clear guidance to others wishing to adapt the PovMap method to the estimation of malnutrition prevalence. A companion to this report provides detailed documentation of the methods used (Rogers et al. 2007). A longer and more detailed report of this study is available from the authors.

2. Data

2.1 General considerations Causal models of malnutrition depend on having information on the child, his household, and the community in which he lives. A wide literature on the causes and the factors associated with malnutrition provides guidance on the types of information that could be used to predict the nutritional status of a child (UNICEF 1990, 1991; Smith and Haddad 2000). Appendix Table A1 shows examples of the key variables and the level (individual, household, community) at which they are measured. The SAE process depends on having enough variables that are comparable in both the survey and the census. One challenge of using the SAE method is to find suitable proxies for variables that would ideally be included in a predictive model. The survey and the census must contain similar information on the individual child and on the household. Additional variables can be added to the predictive model, which describe the location in which the child lives. Every segment represented in the survey will also be represented in the census4, so that information on the segment or the

4 Not every sample survey uses the census sampling frame. In such cases, it may not be possible to match census segments with survey segments (see, for example, Simler 2006). In such cases, information can be calculated at the smallest unit for which the survey and sample segments can be matched – the community, the district. In all the countries in the present study, survey and census segments could be matched. Since

3

community in which the segment is located can be derived from the census and added to the survey data. In addition to information on individual children and their households and communities, secondary sources can provide institutional and geographic information that contributes to a more accurate prediction of malnutrition prevalence, such as access to health and schooling services, and coverage by social programs; land use (percentage of land in agriculture, forests, swamp, and other uses), climate (rainfall, history of flooding and drought), elevation and slope. (Appendix Tables A2– A4 show the variables used in each separate analysis, with their sources.) Merging data from diverse sources poses challenges, as geographic units of institutional and geographic data must correspond to the units of the survey and the census. Data sets, both administrative and geographic, cannot be merged until the definitions of administrative levels are made consistent. Further, the survey and the census must have been implemented fairly close to each other in time, and there should not have been any major social or economic disruptions between the survey and the census that would be likely to change the situation with regard to nutrition, food security, or poverty.

2.2 Data Used in the Present Analysis Table 2.1 summarizes the characteristics of the core data sets used in the three countries.

Table 2.1. Characteristics of Data Sets Used in the Analysis Dominican Republic Ecuador Panama Name of Survey ENDESA

Encuesta Demográfica y de Salud

ENDEMAIN Encuesta Demográfica y de Salud Materna Infantil

ENV Encuesta de Niveles de Vida

Year of Survey 2002 2004 2003 Survey representative at what level

Province, Development Zone, Urban/Rural

Province, Stratum Urban/Rural

Province, Sampling Domain, Urban/Rural

Number of households in survey

27,135 16,530 (MEF survey)b 6,363

Number of children in survey (of ages included in analysis)a

10,291

4,021

1,955

Number of children in Census

851,174 1,079,772

248,731

Year of Census 2002 2001 2000 a. Children ages 6 to 59 months in the Dominican Republic; children from 1 to 5 completed years in Ecuador and Panama. b. Survey of women of childbearing age (households in which child anthropometry was measured). See discussion of data sources.

official boundaries of administrative units may change, a key step in preparation for analysis is to ensure that the same boundaries are used for these units in both the census and the survey.

4

2.2.1 Dominican Republic Analysis for the Dominican Republic was based on the 2002 Demographic and Health Survey (Achécar et al 2003), and the National Census (VIII Censo Nacional de Población y Vivienda), also for the year 2002. We were fortunate that a donor-funded project known as PROSISA had recently been undertaken, whose goal was to bring together a wide range of health related information that could be linked to geographic indicators.5 Other geographic data came from publicly available sources. There is no information on the race/ethnicity of individual children or members of their households in the ENDESA data set, even though this information, specifically, whether a child is of Haitian origin, is of great policy relevance in the country (as is indigenous status in both Panama and Ecuador).

Appendix Table 2 shows the variables included in the analysis and the sources of the data.

The Domincan Republic is divided into ten Development Zones that represent distinct ecological/economic strata. The first administrative level below the national level is that of the Province; there are 31 provinces in the country, plus the Distrito Nacional, which has the status of a province. The ENDESA was designed to produce statistically valid information at the level of the provinces. The second administrative level is the municipio. There are 225 municipios in the country. The population of the Dominican Republic is 8,562,541; the number of households is 654,541, and the number of children 6 to 59 months included in the analysis is 851,174 (based on the 2000 Census). Our estimates were performed for the municipio level.

2.2.2 Ecuador Data for the analysis of Ecuador came from the National Demographic and Maternal-Child Health Survey (Encuesta Demográfica y de Salud Materna Infantil – ENDEMAIN), conducted in 2004 (CEPAR 2005), and the VI National Census of Population and Housing, conducted in 2001. For survey sampling, the country was divided into seven strata: Sierra Rural and Urban; Costa Rural and Urban, Guayaquil, Quito, and Amazonia. Province is the first administrative unit of Ecuador below the national level; the second level is cantón, and the third administrative level is parroquia. There are 24 provinces, 220 cantones, and 995 parroquias.6 We initially performed our

5 This data set (called SIGpaS: Sistema de Información Geográfica para la Salud ) is available on CD from the Health Secretariat (SESPAS). The data were collected using the same administrative boundaries that were used for the 2000 ENDESA survey and the Census. 6 The number of parroquias varies over time, and thus by data set. The ENDEMAIN was drawn from a sampling frame of 995 parroquias; the Census also provides data on 995 parroquias. However, older data sources are based on a different sampling frame, containing, for example, only 960 parroquias (Ecuamapalimentaria 2004). The different number is not explained by a simple division or merging of contiguous parroquias. Using data sources, particularly geographic data, with the same number of geographic units as the census is imperative as this will affect the validity of the model and presentation of

5

estimates at the level of the parroquia, but found that due to small population sizes and quite imprecise estimates at the parroquia level, the canton-level estimates were more reliable. Two issues arise in connection with the Ecuador data sets. The first is that of child age. Age of the child is a key variable for predicting nutritional status. Typically, an infant grows close to his recommended growth trajectory for the first six months of life (that period for which breastfeeding is sufficient for growth), and then begins to drop below it. Malnutrition rates increase for children up to the age of about 24 months, and then stabilize or even drop slightly up to age five. For this reason, malnutrition rates for children under age five should be measured starting at six months; including children younger than this will underestimate malnutrition, since younger infants have not yet fallen off their growth path. In the Ecuador census (as in Panama), only age in completed years is available in the census, even for infants. The predictive model was thus estimated using age in completed years. We did not include children below 12 months in the model. We had the choice of including children below six months, or excluding children 6 to 11 months. We concluded that including infants 0 to 6 months would produce misleadingly low estimates of malnutrition prevalence. A second issue is that in ENDEMAIN, anthropometric information was collected only on a subset of households included in the survey. In order to reduce the respondent burden on any individual household, detailed demographic information such as education level of members was collected on a different subset of households. As a result, some key variables relating to characteristics of adults in the child’s household could not be included in the predictive model. Appendix Table A3 shows the variables that were included in the analysis.

2.2.3 Panama The data sets used for the Panama analysis were the Encuesta de Niveles de Vida (ENV), implemented in 2003, and the National Census, implemented in 2000. The ENV was a Living Standards Measurement Survey supported by the World Bank, with a sample of 8,000 households. Note this is a much smaller survey than the surveys used for the Dominican Republic and Ecuador (see Table 2.1). Geographic information was obtained from a variety of publicly available sources. Appendix Table A4 shows the variables included in the Panama analysis along with their source.

The first administrative level below national is the Province; there are 9 provinces, 3 indigenous areas, called comarcas, with the status of provinces, and two comarcas with the status of corregimiento (Atlas 2006). The second administrative unit is the District, of which there are 75, and the third administrative unit is the corregimiento. There were 593 corregimientos in Panama at the time of the Census; these are the corregimientos and their boundaries included in the present analysis.7 The ENV collected information to be

results in map form. The Tufts team closely accounted for disparities in data, as merging data from numerous sources requires consistency in hierarchical identification codes. 7 New corregimientos have been formed since then; as of 2006 there are 621 corregimientos in Panama.

6

representative of fourteen domains, or “dominios”. These domains correspond approximately to the country’s nine provinces, except for the province of Panama, which was divided into five sampling domains: Panama City, the rest of the District of Panama, the District of San Miguelito, and Panama Oeste and Panama Este. The survey was representative of these domains.

Indigenous areas constituted a single domain. In addition to the three comarcas, whose population is largely or entirely indigenous, there are five provinces (Bocas del Toro, Chiriqui, Darien, Panama, Veraguas) that have significant indigenous populations. In these provinces, if a segment had a population that was greater than 50% indigenous by self-report, that segment was included in the domain Indigenous Areas; otherwise, it was included in the province in which it was located geographically. Thus, all the domains for which ENV results are reported correspond to a Province or a defined contiguous geographic area within a province, except for the domain “Indigenous Areas”, which is geographically dispersed. In order to apply the SAE method and compare our resulting estimates to the results from the ENV, we classified every census segment according to its percentage indigenous population, and place all the segments with more than 50% indigenous individuals into the “Indigenous Areas” domain.8 This division affects the presentation of our results in the form of maps. Our estimates were at the district level, but some districts have two separate estimates, one for the indigenous portion, and one for the non-indigenous portion. Thus, estimates are presented separately for the indigenous and non-indigenous domains. This separation allows us to see clearly the dramatic differences in malnutrition prevalence between the indigenous and non-indigenous areas of the country. As is the case for Ecuador, the Panama census contained information on age only in completed years. We therefore excluded children under 12 months of age because including children ages 0 to 5 months) would have produced misleadingly low estimated prevalence of malnutrition. 2.4 Level of Disaggregation of the Estimates Our original intent was to produce prevalence estimates for the three countries at the third administrative level.9 After running the estimates at this level in all three countries, we assessed the precision of the estimates by examining the standard errors around the point estimates, and concluded that we could produce estimates with satisfactory precision at the secondary level. These levels were municipio for the Dominican Republic, cantón for Ecuador, and distrito for Panama.

8 In ENV, indigenous segments are those with greater than 50% indigenous population (see http://www.mef.gob.pa/ ). To assure comparability between our estimates and the survey results, we recalculated the percentage of indigenous population in the survey segments. 9 These levels are: sección for the Dominican Republic, parroquia for Ecuador, and corregimiento for Panama.

7

In the case of Ecuador and Panama, it might be possible to improve the precision of the estimates, and thus permit estimation at the level of the parroquia and, corregimiento by modifying the model. Some specific steps that might be taken are: remove from the model any variables whose coefficients are non-significant and have high standard errors; introduce interaction terms that would modify the effects of certain key variables according to child’s age or selected geographic characteristics.10 Another possibility is to perform the original survey regressions in a statistical program that allows the use of a robust regression technique to identify multivariate outliers; these cases could be removed prior to conducting the estimation using PovMap.

3. Results

3.1 Synthesis Tables 3.1 – 3.3 show malnutrition prevalence (based on HAZ), and average HAZ, comparing the estimates derived from SAE with the same measures derived from the survey data alone. These figures provide an initial sense of the plausibility of the SAE estimates. These results are reassuring: in general, the SAE procedure, estimated at the first sub-provincial level and aggregated to the level of province, produces estimates that are within 2 SE’s of the survey-derived figures.

10 PovMap offers some options for testing variations on the predictive (Beta) model. It also offers a test to determine whether the model is overfitted to the specific data set being used.

8

Table 3.1: Height-for-Age Malnutrition Prevalence Rates and Small Area Estimation Results, Dominican Republic

Development Zone

Province n

Prevalences Based on

SAE

Standard Errors on

SAE estimates

Malnutrition Prevalence Rates (DHS

Survey)

Mean HAZ Scores

Based on SAE

Mean HAZ Scores (DHS

Survey) Cibao Norte Espaillat 252 8.78 0.0136 8.33 -0.30 -0.33 N = 1014 Puerto Plata 281 9.85 0.0181 9.61 -0.29 -0.25 K = 98 Santiago 481 8.46 0.0139 5.41 -0.25 -0.24 R-sq = 0.2175 Cibao Sur La Vega 368 11.95 0.0142 9.78 -0.41 -0.55 N = 991 Monsenor Nouel 349 7.98 0.0115 9.46 -0.20 -0.40 K = 97 Sanchez Ramirez 274 8.11 0.0137 7.66 -0.36 -0.41 R-sq = 0.2630 Cibao Nordeste Duarte 334 9.57 0.0097 5.99 -0.38 -0.34 N = 1170 Ma. T. Sanchez 261 10.31 0.0114 9.16 -0.30 -0.24 K = 97 Salcedo 285 8.58 0.0119 7.37 -0.31 -0.28 R-sq = 0.1534 Samana 290 11.05 0.0127 6.90 -0.36 -0.32 Cibao Noroeste Dajabon 280 8.75 0.0124 8.90 -0.22 -0.24 N = 938 Monte Cristi 215 8.45 0.013 6.05 -0.25 -0.31 K = 101 Stgo Rodriguez 176 10.54 0.0168 7.95 -0.42 -0.49 R-sq = 0.2463 Valverde 267 6.70 0.0112 6.74 -0.22 -0.38 Valdesia Azua 377 9.09 0.0121 8.99 -0.45 -0.55 N = 1419 Peravia 322 9.18 0.0114 9.01 -0.41 -0.40 K = 102 San Cristobal 483 9.79 0.0127 13.25 -0.48 -0.47 R-sq = 0.1728 San Jose de Ocoa 237 10.13 0.0167 8.02 -0.43 -0.49 El Valle Elias Pina 323 19.35 0.0275 17.34 -0.73 -0.79 N = 634; K = 97; R-sq = 0.2960 San Juan 311 12.54 0.0179 9.97 -0.44 -0.45 Enriquillo Bahoruco 396 15.39 0.0148 16.83 -0.64 -0.74 N = 1380 Barahona 363 11.03 0.0139 9.92 -0.44 -0.42 K = 103 Independencia 297 13.70 0.0164 13.47 -0.38 -0.52 R-sq = 0.2350 Pedernales 324 15.32 0.017 14.51 -0.63 -0.72 Yuma El Seybo 300 8.38 0.0192 6.67 -0.29 -0.40 N = 966 La Altagracia 287 11.11 0.0198 6.97 -0.27 -0.23 K = 96 La Romana 379 12.67 0.0263 11.35 -0.47 -0.45 R-sq = 0.1443 Higuamo Hato Mayor 241 13.27 0.0246 9.13 -0.51 -0.44 N = 889 Monte Plata 321 14.13 0.0222 14.95 -0.52 -0.53 K = 96 San Pedro de Macoris 327 6.75 0.0121 8.56 -0.13 -0.28 R-sq = 0.2277 Ozama Distrito Nacional 268 7.00 0.0116 8.21 0.03 -0.11 N = 730; K = 81 Santo Domingo 462 11.49 0.0157 10.15 -0.12 -0.18 R-sq = 0.2014

9

Table 3.2: Comparison of Malnutrition Prevalence and Mean HAZ of Small Area Estimation and Survey Estimates for Ecuador

Malnutrition Prevalence Mean HAZ

Zone Models Dominio Hunger Mapping Results

*Survey Estimates

Hunger Mapping Results

*Survey Estimates

Quito † (N=396) (K=66) (R2=0.2643)

1 Quito 0.331 (0.031) 0.345 -1.406 (0.122) -1.525

Guayaquil † (N=361) (K=63) (R2=0.3173)

2 Guayaquil 0.169 (0.022) 0.146 -0.841(0.107) -0.824

3 El Oro 0.164 (0.018) 0.156

-0.817 (0.103) -0.998

4 Esmeraldas 0.154 (0.018) 0.170

-0.732 (0.112) -0.958

5 Guayas 0.226 (0.017) 0.221

-1.131 (0.065) -1.267

6 Los Rios 0.194 (0.016) 0.240

-0.950 (0.086) -1.191

Costa † (N=1278) (K=86) (R2=0.2041)

7 Manabi 0.196 (0.017) 0.185 -0.996 (0.078) -1.080

8 Azuay 0.289 (0.027) 0.367

-1.408 (0.080) -1.638

9 Bolivar 0.426 (0.030) 0.459

-1.801 (0.090) -1.976

10 Canar 0.448 (0.033) 0.422

-1.857(0.108) -1.622

11 El Carchi

0.284 (0.028) 0.344

-1.434 (0.078) -1.586

12 Cotopaxi 0.428 (0.026) 0.409

-1.816 (0.083) -1.744

13 Chimborazo 0.424 (0.035) 0.446

-1.805 (0.102) -1.905

14 Imbabura 0.408 (0.022) 0.373 -1.731 (0.068) -1.575

15 Loja 0.352 (0.027) 0.342

-1.576 (0.082) -1.622

16 Pichincha 0.272 (0.027) 0.248

-1.325 (0.087) -1.335

Sierra (N=1716) (K=87) (R2=0.2561)

17 Tungurahua 0.393 (0.027) 0.504

-1.690 (0.092) -1.817

Amazon Region** (N=270) (K=42) (R2=0.3190)

18 Amazon 0.325 (0.032) 0.277 -1.202 (0.192) -1.339

Note: Hunger mapping results are based up predictions of a national-level model calculated in PovMap 2.0 * Survey estimates of malnutrition prevalence and mean HAZ scores are based on children 1 - 5 years of age with HAZ = ± 5. Prevalences weighted using the inverse of the household sampling fraction (varname=factor) † No locational effect used in Beta Model ( ) Standard Errors Displayed in Parentheses ** Amazon Region model run with individual, household, and dwelling-level variables due to small N

10

Table 3.3: Comparison of Malnutrition Prevalence and Mean HAZ of Small Area Estimation and Survey

Estimates: Panama Malnutrition Prevalence Mean HAZ National Model** Dominio N Hunger Mapping

Results *Survey

Estimates Hunger Mapping

Results *Survey

Estimates 0.128 0.113 -0.515 -0.380 1 Panama City 123

(0.023) (0.132) 0.179 0.195 -0.851 -0.763 2 District of Panama 106

(0.035) (0.168) 0.108 0.039 -0.493 -0.283 3 San Miguelito 61

(0.024) (0.150) 0.181 0.169 -0.845 -0.721 4 Panama Oeste 189

(0.028) (0.125) 0.223 0.210 -0.985 -1.040 5 Panama Este 143

(0.039) (0.160) 0.168 0.279 -0.757 -1.212 6 Bocas del Toro 296

(0.037) (0.197) 0.251 0.238 -1.196 -1.006 7 Cocle 96

(0.041) (0.139) 0.161 0.137 -0.779 -0.702 8 Colon 119

(0.031) (0.138) 0.193 0.115 -0.868 -0.592 9 Chiriqui 156

(0.032) (0.145) 0.203 0.151 -1.003 -0.820 10 Darien 112

(0.053) (0.206) 0.174 0.134 -0.882 -0.674 11 Herrera 66

(0.034) (0.145) 0.175 0.094 -0.900 -0.275 12 Los Santos 47

(0.035) (0.145) 0.175 0.253 -1.007 -1.032 13 Veraguas 105

(0.037) (0.146) 0.670 0.629 -2.466 -2.297

(N=1955) (K=82) (R2=0.2940)

14 Indigenous Areas 336 (0.042) (0.132)

Note: Hunger mapping results are based up predictions of a national-level model calculated in PovMap 2.0 * Survey estimates of malnutrition prevalence and mean HAZ scores are based on children 1 - 5 years of age with HAZ = ± 5. Prevalences weighted using the inverse of the household sampling fraction (varname=factor) ** Hunger mapping results calculated using "cluster locational effect" ( ) Standard Errors Displayed in Parentheses

11

Table 3.4 compares the mean, median and range of prevalence of malnutrition according to HAZ as estimated by SAE, to show how the three countries differ from each other. Many of the maps shown below describe malnutrition prevalence in terms of the distribution within the country, that is, by national-level quartile. By contrast, this table, and the three maps that follow, demonstrate that the three countries differ dramatically in the nutrition situation.

Table 3.4: Country Descriptive Statistics on Chronic Malnutrition Malnutrition Prevalence – Height-for-Age z-scores Dominican Republic Ecuador Panama

Mean a 12.04 30.86 37.92 Median 10.75 30.41 24.73 IQR 7.78 – 14.45 22.20 – 38.07 19.00 – 58.71 Minimum 3.53 9.33 10.56 Maximum 45.78 61.80 92.48 National Prevalence (mean prev, weighted) 10.10 26.96 24.32 a Simple mean of the small areas, based on small area estimates at the municipio, cantón and distrito levels, respectively. The following four maps show prevalence of stunting in all three countries. (As discussed above, there are two maps for Panama because of the separate estimates for indigenous areas.). These maps show malnutrition prevalence by HAZ, using common categories, rather than a relative measure of quartiles.

12

Map 1: Dominican Republic, Standardized Categories

13

Map 2: Ecuador, Standardized Categories

14

Map 3: Panama, Non-Indigenous, Standardized Categories

Map 4: Panama, Indigenous Areas, Standardized Categories

15

These maps demonstrate the sharp differences among the three countries. In the Dominican Republic, only one municipio has a malnutrition prevalence above 40 percent, and no municipio has a prevalence above 60 percent. In contrast, a number of cantones in Ecuador show prevalences between 41 and 60 percent; these are all located in the Sierra region. Only one cantón shows a prevalence rate above 60 percent. Using these standardized categories to compare countries and regions is even more instructive for Panama. Two separate maps for Panama show estimated prevalence for the non-indigenous and indigenous areas. Among the non-indigenous districts, not one has a prevalence rate above 40 percent. Among the Indigenous areas, only three have prevalence rates below 40 percent, and none is below 20 percent. In the discussion that follows, individual country results are presented, largely in terms of quartiles of malnutrition prevalence and numbers of children affected. The quartiles emphasize how areas within the countries vary with respect to each other, not with respect to an absolute level. The quartile boundaries vary widely from one country to another, and in some cases, the quartile spans a very wide range. It is important to pay attention to these details in interpreting the results that follow.

3.2 Country level results

3.2.1 Dominican Republic Map 5 shows how stunting prevalence varies within provinces at the municipio level.

Map 5

16

The municipio level map demonstrates the value of providing disaggregated estimates of malnutrition prevalence. For example, in the eastern province of La Altagracia there are a few municipios where the malnutrition problem appears to be severe. In La Romana, clearly there is a more severe problem in the northern than in the southern municipio. And on the frontera, all of Elías Piña is in the worst quartile, but there are notable differences among the municipios within the provinces of Bahoruco, San Juan, and Independencia, suggesting priorities for the targeting of any nutrition program. Monte Plata also shows wide variation among the municipios, with only a few municipios in the highest quartile of prevalence. The story told by these prevalence estimates is incomplete, however. The following map (Map 6) shows how different the picture is if the total number of children affected, rather than percent prevalence, is used as a basis for conclusions about the severity of the malnutrition problem. Of course, this is because many of the most severely affected municipios are in less densely populated rural areas, while concentrations of population are in the urban areas in and around Santo Domingo and in the north.

Map 6

Targeting programs according to prevalence would risk missing some of the largest concentrations of malnourished children, simply because they are in areas where better off populations also live. This point is demonstrated in the graph below, which compares absolute numbers and percentages of children by province, according to our SAE estimates.

17

Number of Malnourished Children

0 1000 2000 3000 4000 5000 6000

Distrito Nacional

Santo Domingo

San Pedro De Macoris

Monte Plata

Hato Mayor

La Romana

La Altagracia

El Seibo

Pedernales

Independencia

Barahona

Bahoruco

San Juan

Elias Piña

San Cristobal

San Jose De Ocoa

Peravia

Azua

Valverde

Santiago Rodriguez

Monte Cristi

Dajabon

Samana

Salcedo

Maria Trinidad Sanchez

Duarte

Sanchez Ramirez

Monseñor Nouel

La Vega

Santiago

Puerto Plata

Espaillat

Range of prevalence estimates

0.0010.0020.0030.0040.0050.00

7,118

21,634

Number of children Percent of children

Figure 1. Range and Number of Malnourished Children (based on Height-for-Age Z-scores) by Province and Development Zone, Dominican Republic

18

If the municipios in the highest prevalence quartile (prevalence range 14.5 to 45.8 percent) were targeted with a nutrition intervention, 21,376 children would be reached, or about 25 percent of the 85,970 chronically malnourished children in the country. If the ten municipios with the highest number of malnourished children were to be targeted, 34,918 children, or 41 percent would be reached. This is not in any way to suggest that the more remote and less densely populated regions of high malnutrition prevalence should be ignored, but it does reinforce the importance of taking on the difficult challenge of reaching the malnourished in areas where they are the minority. Different targeting strategies will be appropriate for reaching the malnourished in rural and remote areas and for reaching those in the cities.

3.2.2 Ecuador The results for Ecuador are presented in the following maps. Map 7 shows the prevalence of stunting at the cantón level, according to national quartiles.

Map 7

19

Once again, we see the value of disaggregating to the canton level, because of the variability among cantones within a province. The coastal provinces show significant variation, and there are important variations in the Amazon region and in the southern province of Loja. Note also that the lowest quartile of prevalence in Ecuador ranges from 9.33 to 22.2 percent, a stark contrast to the lowest quartile in the Dominican Republic, where the lowest quartile boundaries are 3.6 to 7 percent, and a prevalence rate of 22 percent would fall in the highest (worst) quartile. In Ecuador, no cantón has a malnutrition prevalence below 9 percent. Once again, though, prevalence rates tell only part of the story. The following map (Map 8) shows how the distribution of malnutrition appears based on absolute numbers of affected children rather than on percent prevalence. Again, note the very wide range represented by the highest quartile. Because there are only a few very densely populated cantones, including Quito and Guayaquil, the highest quartile of cantones based on number of malnourished children extends into some much less densely populated regions of the country. These numbers reflect the combination of population density and prevalence rate. The Sierra region is worst off in terms of both percent prevalence and absolute numbers, as is the southern cantón in the southwest province of Guayas. The differences between numbers and percent prevalence are less dramatic for Ecuador than for the Dominican Republic. Still, a program targeted to the quartile of cantons with the highest prevalence would reach an estimated 77,029 chronically malnourished children, or 26 percent of the total, while targeting the ten cantones with the highest numbers of malnourished children would reach 122,799, or 42 percent of the estimated total of 291,421 malnourished children in the country. Clearly, a dual strategy is required to reach both areas of high prevalence, and areas where high numbers of children are affected with malnutrition.

20

Map 8

The figure below (Figure 2) shows in graphic form at the province level how the prevalence and absolute numbers of malnourished children vary.

21

Number of Malnourished Children

0 10000 20000 30000

Quito

Guayaquil

El Oro

Esmeraldas

Guayas

Los Rios

Manabi

Azuay

Bolivar

Canar

Carachi

Cotopaxi

Chimborazo

Imbabura

Loja

Pichincha

Tungurahua

Morona Santiago

Napo

Pastaza

Zamora Chinchipe

Sucumbios

Orellana

47,756

Range of prevelance estimates

0204060

Figure 2 Range and Number of Malnourished Children In Ecuador (based on HAZ) by Province and Zone

Number of children Percent of children

22

3.2.3 Panama As with the Dominican Republic and Ecuador, we show malnutrition prevalence estimates at the district level. Two maps are shown for the district level estimates: one for the non-indigenous areas, and one for the indigenous areas.

Map 9

23

Map 10

Note the pronounced difference in the quartile boundaries for the indigenous and non-indigenous districts. Among the indigenous, the very lowest quartile starts at 35 percent prevalence, and reaches to 56 percent. Among the non-indigenous areas, a prevalence of 35 percent puts the district squarely in the worst, highest prevalence quartile. Thus if national quartile boundaries were used rather than quartiles calculated separately for the indigenous and non-indigenous districts, the indigenous area map would be entirely in the highest quartile, while all the non-indigenous districts would be in the lower three. The next set of maps show that while prevalence is clearly far worse among the indigenous areas, the actual numbers of children affected by malnutrition are higher in the non-indigenous areas, reflecting the fact that the indigenous population as a whole represents only 11.3 percent of the population of Panama (Bermudez 2006). Given the small percentage of the population that is indigenous, it is striking that fully 39 percent of the 60,522 chronically malnourished children in Panama are living in the indigenous areas of the country. Taken together, these maps show that the indigenous are a relatively small but highly vulnerable population, concentrated in rural areas; among the indigenous districts, those closer to the cities (Panama and Colón) have lower rates of malnutrition than those areas that are further away from these population centers.

24

Map 11

Map 12:

25

Range of prevelance estimates

0255075100

Percent of children

Number of Malnourished Children

0 2500 5000 7500 10000

Panama City

District ofPanama

District of SanMiguelito

PanamaOeste

Panama Este

Bocas delToro

Coclé

Colón

Chiriqui

Darien

Herrera

Los Santos

Veraguas

ComarcaKuna Yala

ComarcaEmbére

ComarcaNgöbe Bungle

IndigenousSegments

12440

Number of children

Figure 3 Range and Number of Malnourished Children In Panama (based on HAZ) by Province

26

The maps and graph above show once again how different conclusions about targeting would be drawn based on percent prevalence and based on absolute numbers. The five domains that represent the province of Panama show widely varying prevalence rates, but all are relatively low; yet together, these areas account for over 16,000 malnourished children or about 25 percent of the national total of 60,522. The indigenous segments and the indigenous comarca of Ngöbe Buglé show the highest prevalence and also account for the largest number of malnourished children: close to 20,000, or about a third of the total. As we did for the Dominican Republic and for Ecuador, we can calculate the consequences of targeting interventions based on prevalence alone, or based on numbers of children affected. The ten districts with the highest number of malnourished children account for 27,958 children, or 46 percent of all malnourished children. If the districts with the highest prevalence were targeted (with prevalence calculated nationally, without separating Indigenous from non-Indigenous), 18,030 children or about 30% of the total would be reached. Once again, we do not wish to suggest that areas of high prevalence and small numbers be ignored; rather, we point out the importance of considering both numbers and prevalence in the design of nutrition interventions.

4. Discussion Small Area Estimation has enormous potential to guide policy decisions to address malnutrition. Nutrition surveys cannot generally be disaggregated below the level of province, and we have seen that there is wide variation in malnutrition prevalence within provinces. With the information provided by these estimates, much better targeting is possible. A key result of the analysis is the demonstration that areas of high malnutrition prevalence are not always those where the greatest numbers of malnourished children are living. These results suggest establishing targeting mechanisms and priorities based on a dual strategy: one based on locating the large number of malnourished children living in relatively more affluent urban areas, and another based on reaching those areas that have very high prevalence, many of them in more remote and rural locations. Malnutrition is in many ways a more complex and less predictable phenomenon than poverty, subject to the influence of more unobservable (in a survey setting) factors relating to child and caretaker characteristics and family environment. For this reason, it would build confidence in the application of SAE techniques to estimating malnutrition prevalence if a body of empirical work were produced that confirmed their accuracy with on-the-ground verification. The criterion for accuracy is correct classification: if areas are divided into highest prevalence, high, medium, and low, does the measurement of malnutrition on the ground preserve these same classifications? We hoped to provide even more disaggregated estimates of malnutrition prevalence, but these estimates were generally too imprecise for us to have confidence in these results. It might well be possible to improve precision by modifying the model, eliminating imprecisely estimated parameters and judiciously adding interaction terms, and by

27

eliminating multivariate outliers through robust regression implemented prior to running the data in PovMap. It would be worth making this effort particularly for those cantones and distritos that comprise large populations: Quito and Guayaquil in Ecuador, and the urban districts of Panama. At the same time, it may well be that for most policy purposes, the first sub-provincial level is sufficient for most purposes given the administrative structure in place in these countries. Once a geographic area is identified as high-risk, further targeting at the household or individual level is needed in any case. The power of the SAE technique is that it allows conclusions to be drawn about conditions on the ground at a level of detail that would be impossible if these conclusions had to be based on primary data collection. Once a model is developed, SAE could presumably be used also to track changes in the local situation, and thus probable changes in the prevalence of malnutrition. Recall, though, that the disaggregated prevalence estimates are based on having census data. Typically, censuses are repeated at best at ten-year intervals, while surveys may be repeated every five or six years. A new survey would make it possible to explore whether the underlying factors associated with malnutrition in the country had changed, so that the direction or strength of the association of a particular condition or characteristic with nutritional status was altered. But it is the census that provides the information about the distribution of those conditions, for example, notable changes in housing quality, sanitation, access to public services or to roads and markets, that would permit re-estimation of malnutrition prevalence.

5. Next steps A number of institutions have committed themselves to promoting the use of SAE for hunger mapping in Latin America, including the World Bank, the World Food Programme, and national governments. Our experience suggests that this is both feasible and potentially valuable as a means of understanding the nature of the nutrition problem in different areas of a country, improving the targeting of resources, and advocating for attention to be given to problems of hunger and malnutrition. If this is to be a region-wide effort, then institutionalizing the capacity to implement the analysis and apply it to mapping the results should be a high priority. Plans are under way to develop training manuals and intensive course modules for the purpose. Key decisions include not only the content and structure of the training, but also the best place within the government or the research and academic communities to institutionalize the capacity for the analysis.

A second effort that is already in process is the modification of PovMap to adapt it to the specialized needs of malnutrition estimation: specifically, to adjust for the multiple layers of clustering, and also to allow for negative values of the outcome variable. This will produce estimates that are more statistically defensible, with standard errors that are a more accurate reflection of the precision of the estimate. Whether this improves the results in the sense that is changes the classification of the areas is an empirical question that would be worthwhile to investigate.

28

Field-testing of the prevalence estimates produced by the SAE technique should be a high priority. Promotion of the use of SAE and hunger mapping is based on the confidence that the estimates produced are valid and accurate. Up to now, there have been few field studies that would verify the SAE approach, and more of these have related to poverty than to malnutrition. We recommend harmonizing data collection efforts within countries. Often, different government agencies are responsible for different data collection efforts, but there is no reason why they could not consider the possibilities for improving the consistency of these efforts. A series of workshops in various countries that would bring together the responsible people in a country to consider the possibilities might start a productive process that, in the longer term, would improve the usefulness of all the data collection efforts.

29

6. References Achécar, M.M.; N.Ramirez; J.J.Polanco; L.H.Ochoa; G.Lerebours; B.Garcia, 2003. Encuesta Demográfica y de Salud (ENDESA) República Dominicana. Santo Domingo: Centro de Estudios Sociales y Demográficos (CESDEM) and Calverton, MD: ORC Macro/MEASURE DHS. Atlas Universal y de Panamá 2006. Panamá:.Promotora Educativa S.A. Benson, T. 2006. Insights from poverty maps for development and food relief program targeting. International Food Policy Research Institute. Food Consumption and Nutrition Division Discussion Paper 205. Benson, T.; J.Chamberlin; I.Rhinehart, 2005. A Investigation of the Spatial Determinants of the Local Prevalence of Poverty in Rural Malawi. Food Policy 30:5-6 Oct/Dec 2005, 532-50. Bermudez, Odilia 2006. Situación Nutricional, Patrón de Consumo y Acceso a Alimentos: Informe Final de Consultoría. Panama: Min. de Economia y Finanzas, Dirección de Políticas Sociales, April. CEPAR 2005. Endemain 2004 Informe Final. Quito: Centro de Estudios de Población y Desarrollo Social (CEPAR) Demombynes, G.; C.Elbers; J. Lanjouw; P.Lanjouw; J.Mistiaen; B.Ozler, 2002. Producing an Improved Geographic Profile of Poverty: Methodology and Evidence from Three Developing Countries. WIDER Discussion Paper 2002-39. Elbers, C.; J.O.Lanjouw; P.Lanjouw, 2002; Micro Level Estimation of Welfare. Washington DC World Bank Policy Research Paper WPS2911, October.

Elbers C., Lanjouw J. and Lanjouw P. 2003. “Micro-level estimation of poverty and inequality”, Econometrica, 71, 355-364.

Elbers, C., J.O.Lanjouw, P. Lanjouw, 2004. Imputed Welfare Estimates in Regression Analysis. Washington DC: World Bank Policy Research Paper WPS 3294, April. Engle, P.; P.Menon; L.Haddad, 2001. Care and Nutrition: Concepts and Measurement. Washington DC: International Food Policy Research Institute. Fujii, T. 2003. Micro-Level Estimation of the Prevalence of Stunting and Underweight Among Children in Cambodia. Report to Ministry of Health, Royal Government of Cambodia (preliminary report). UN World Food Programme, March (mimeo).

30

Fujii, T., 2005. Micro-level Estimation of Child Malnutrition Indicators and Its Application in Cambodia. Washington, DC: World Bank, Policy Research Working Paper WPS3662, July. Gilligan, D.; A. Veiga; M.H.D.Benicio; C.A.Monteiro, 2003. An Evaluation of Goegraphic Targeting in Bolsa Alimentação in Brazil: Report Submitted to the Government of Brazil. Washington DC: International Food Policy Research Institute, April. Haslett, S., Jones, G. 2005. Small area estimation using surveys and censuses: some practical and statistical issues. Statistics in Transition. 7(3), 541 – 555. Haslett, S.; G.Jones; with D. Parajuli, forthcoming. Small Area Estimation of Poverty, Caloric Intake, and Malnutrition in Nepal. Kathmandu: Government of Nepal, Central Bureau of Statistics. World Food Programme and World Bank. Hentschel, J.; J.O.Lanjouw; P.Lanjouw; J. Poggi, 2000. Combining Census and Survey Data to Trace the Spatial Dimensions of Poverty. World Bank Economic Review 14:1, (January) 147-165.

Larrea, C., 2005. Poverty, Food Poverty, and Malnutrition Regression Models for Ecuador. Taken from the EcuaMapAlimentaria website on August, 18, 2006. http://www.ecuamapalimentaria.info/ Morillo P., Antonio, 2003. Focalización de la Pobreza en la República Dominicana (Edición corregida y ampliada), Segunda Edición. Distrito Nacional, R.D.: Secretariado Técnico de la Presidencia, Oficina Nacional de Planificación (ONAPLAN). Morillo, A.; A. Guerrero; Y. Alcántara, 2005. Focalización de la Pobreza en la República Dominicana 2005. Santo Domingo: Secretariado Técnico de la Presidencia, Oficina Nacional de Planificación (ONAPLAN). deOnis, M.; A.W.Onyango; E.Borghi; C.Garze; H.Yang, 2006. Comparison of the World Health Organization (WHO) Child Growth Standards and the National center for Health Statistics/WHO international growth reference: implications for child health programmes. Public Health Nutrition 9:7, 942-947. Rogers, BL.; J.Wirth; P.Wilde; K.Macías, 2007 Introduction to the Estimation of Malnutrition Prevalence by Small Area Estimation using the PovMap Program. Boston, MA: Tufts University Friedman Nutrition School; Report submitted to World Food Programme/LAC, Panama, February 2007. Simler, K. 2006. Nutrition Mapping in Tanzania: An Exploratory Analysis. Washington DC: International Food Policy Research Institute, Food Consumption and Nutrition Division Discussion Paper #204, March.

31

Smith, L.; L.Haddad, 2000. Overcoming Child Malnutrition in Developing Countries: Past Achievements and Future Choices. Washington DC: International Food Policy Research Institute. Agriculture, Food and Environment Discussion Paper #30. UNICEF (United Nations Children’s Fund), 1991. Strategy for improved nutrition for children and women in developing countries. UNICEF Policy Review. New York: UNICEF. United Nations 2001. Road Map Towards the Implementation of the United Nations Millennium Declaration: Report of the Secreary General. A/56/326, 6 September 2001 Waterlow, J.C.; R.Buzina; W.Keller; JM Lane; MZ Nichaman; JM Tanner. 1977. The presentation and use of height and weight data for comparing the nutritional status of groups of children under the age of 10 years. Bulletin of the World Health Organization 55: 489-498 WHO (World Health Organization) WHO Child Growth Standards: Length/Height-for-age, Weight-for-age, Weight-for-length, Weight-for-height and Body Mass Index-for age Methods and Development. Geneva: World Health Organization 2006 WHO (World Health Organization) 1995. Physical Status: The Use and Interpretation of Anthropometry. Geneva: WHO. Zhao, Q., 2005. User Manual for PovMap 1.1a. Development Research Group. From the World Bank website, August 12, 2006. http://iresearch.worldbank.org/PovMap/index.htm

32

Appendix Table A1 Data Requirements for the Prediction of Malnutrition Table A2 Variables Used in the Analysis of Malnutrition,

Dominican Republic Table A3 Variables Used in the Analysis of Malnutrition, Ecuador Table A4 Variables Used in the Analysis of Malnutrition, Panama

A1

Table A1: Data Requirements for the Prediction of Malnutrition

Level

Variable

Possible Source

Individual Age in months Census, Survey Gender Census, Survey Birth Order Survey; rare in census Food consumption Rare in survey; never in census Illness Survey, not census

Household Household size Census, survey Number of children under 5 yrs of age Census, survey Number of adult females Census, survey Number of persons per room - crowding Census, survey

Education of child’s mother Survey; usually no link to mother in

census Education levels of adult household members Census, survey

Economic status, wealth – ownership of key consumption goods

Census, survey

Food consumption: Adequacy Diversity Sources – purchase, home production, etc.

Rarely if ever available in survey;

never in census

Quality of housing Census, Survey Household water source Census, Survey Household sanitation: latrine, garbage disposal Census, Survey Electricity, fuel, telephone Census, survey Income, total, by source, earner Rarely collected in survey or census Livelihoods: income sources, earners Limited information Female/Male household head Census, Survey Ethnicity of members Usually available census and survey Location: urban/rural Census, survey Household food insecurity Rarely collected in survey or census

Community/Cluster Economic Inequality Can be computed from hh assets

Marketing infrastructure: Access to roads Transportation infrastructure Volatility of prices

GIS GIS

Secondary sources; rarely avail.

Services: Access to health services Access to/enrollment in school

Government sources

Local livelihoods: Dependence on agriculture Unemployment Remittances

Variable data, often not consistent

between survey and census

Distance in kilometers to urban centers, markets GIS Ethnic diversity Census

Province/Region Land type, quality, land uses GIS Climate:Rainfall; Droughts; Floods GIS

Topography - Elevation, slope

GIS

A2

a. Disenventar is a project of the Latin American Network of Social Studies on Disaster Prevention (“La Red”): http://www.desinventar.org/

Table A2: Variables Used in the Analysis of Malnutrition, Dominican Republic

Level

Variable

Source

Individual Child Age Census, ENDESA Gender Census, ENDESA Relationship to household head Census, ENDESA

Household Household size Census, ENDESA Number of children under five yrs of age Census, ENDESA Number of adult females Census, ENDESA Number of persons per room - crowding Census, ENDESA Education levels of adult household members Census, ENDESA

Economic status, wealth – ownership of key consumption goods

Census, ENDESA

Quality of housing Census, ENDESA Household water source Census, ENDESA

Household sanitation: methods of garbage disposal, access to toilet facilities

Census, ENDESA

Electricity, fuel, telephone Census, ENDESA

Environmental contaminants affecting household and surrounding areas

Census, ENDESA

Female/Male household head Census, ENDESA Marital status of household head Census, ENDESA Location: urban/rural Census, ENDESA

Community/Sección

Local livelihoods: Dependence on agriculture (self-reported) Remittances

Census Census

Ethnic diversity Census Access to/enrollment in school Secretariat of Education Female headship at the community level Census Education level of heads of households Census Quality of housing at the community level Census Sanitation at the community level Census Household services – electricity, fuel, telephone Census

Region/Municipio Access to roads PROSISA Access to health services PROSISA Distance in kilometers to health care facility PROSISA Distance in kilometers to major urban center GIS Topography EROS System, USGS Land type DIARENA Population density Census

Region/Province Province on the frontera with Haiti GIS

Climate: Rainfall Droughts Floods

PROSISA

Desinventara Desinventara

A3

a The Earth Resources Observation and Science system, part of the US Geological Survey

Table A3: Variables Used in the Analysis of Malnutrition, Ecuador

Level

Variable

Source

Individual Child Nutrition Status ENDEMAIN Age Census, ENDEMAIN Gender Census, ENDEMAIN

Household Household size Census, ENDEMAIN Number of children under five yrs of age Census, ENDEMAIN Number of adult females Census, ENDEMAIN Number of persons per room - crowding Census, ENDEMAIN Primary language of members Census, ENDEMAIN Primary ethnicity of members Census, ENDEMAIN

Number of children (6-11 yrs of age) in/out of primary school

Census, ENDEMAIN

Number of children (12-16 yrs of age) in/out of secondary school

Census, ENDEMAIN

Tenancy status Census, ENDEMAIN

Dwelling type and material of roof, walls, and floor

Census, ENDEMAIN

Household water source and waste water disposal Census, ENDEMAIN

Household sanitation: methods of garbage disposal, access to toilet facilities

Census, ENDEMAIN

Electricity, cooking fuel, telephone Census, ENDEMAIN Location: urban/rural Census, ENDEMAIN

Community/Sector Female headship (percent) Census Percent of Spanish-speaking and mestizo households Census Education level and literacy of household heads Census Housing tenancy, quality and building materials Census Sanitation services: water, septic, garbage Census Household services – electricity, fuel, telephone Census

Region/Parroquia Access to roads WFP Access to markets CIAT Topography EROS System, USGSa Land type, erosion SIISEb Earthquake and volcano risk SIISE Landslides and flooding Desinventarc Infant mortality rate SIISE # of Primary/secondary students per classroom SIISE/SINECd # of out/ inpatient health facilities per 1000 SIISE/ERASe Malaria and Dengue incidence per 100,000 SIISE/MSPf Ethnicity (Indigenous, Black, Mestizo, White) Census Population density Census/WFP

b Sistema Integrado de Indicadores Sociales del Ecuador c Sistema de Inventario de Desastres d Sistema Nacional de Estadísticas Educativas e Encuesta de Recursos y Actividades de Salud f Ministerio de Salud Pública

A4

a The Earth Resources Observation and Science system, part of the US Geological Survey

Table A4: Variables Used in the Analysis of Malnutrition, Panama

Level

Variable

Source

Individual Child Nutrition status ENV Age Census/ENV Gender Census/ENV

Household Household size Census/ENV Number of children under five yrs of age Census/ENV Number of adult females Census/ENV Number of persons per room - crowding Census/ENV % of employed household members Census/ENV Female/Male household head Census/ENV Marital status of household head Census/ENV Education status of household head Census/ENV Education level of adult household members Census/ENV Per capita household salary

Household sanitation: methods of garbage disposal, access to toilet facilities

Census/ENV

Electricity, cooking fuel, telephone Census/ENV Housing type and building materials Census/ENV Household water source and access Census/ENV

Community/Segmento Education level of household heads Census Sanitation services: water, septic, garbage Census Household services – electricity, fuel, telephone Census Ethnicity of household heads Census Segment is urban or rural Census

Region/Corregimiento Access to all-year roads CCAD Distance to urban center GFK Macon Topography EROS System, USGSb Land type CCADc Floods since 1998 Desinventard Population density Census/WFP

b The Earth Resources Observation and Science system, part of the US Geological Survey c Comisión Centroamericana de Ambiente y Desarrollo, http://www.ccad.ws d Sistema de Inventario de Desastres

A5