a manual: measuring and interpreting malnutrition · measuring and interpreting malnutrition and...

A Manual:Measuringand InterpretingMalnutrition and Mortality

ASMR Age-specific mortality rateASMR-U5 Age-specific mortality rate for children under 5BMI Body Mass IndexCDC Centers for Disease Control and PreventionCMR Crude Mortality Rate/Crude Death RateDHS Demographic and Health SurveysENA Emergency Needs AssessmentEPI WHO Expanded Programme on ImmunizationHIV/AIDS Human immunodeficiency virus/acquired immune deficiency syndromeHPLC High-performance liquid chromatographyICCIDD International Council for Control of Iodine Deficiency DisordersIDD Iodine deficiency disordersIMR Infant mortality rateINACG International Nutritional Anaemia Consultative GroupLBW Low birth weightMICS Multi-indicator cluster surveysMMR Maternal mortality rateMUAC Mid-upper arm circumferenceNCHS National Center for Health StatisticsNGO Non-governmental organizationPPS Probability proportional to sizePSU Primary sampling unitRBM Results-based managementSCN United Nations Standing Committee on NutritionSRS Systematic (or simple) random samplingU5MR Under-5 mortality rateUI Urinary iodineUNICEF United Nations Children’s FundVAD Vitamin A deficiencyVAM Vulnerability Analysis and MappingWFP World Food ProgrammeWHO World Health Organization

Acronyms

Acknowledgements 3

Purpose of the Manual 5

Foreword 7

Introduction 9

Chapter 1 Defining and measuring malnutrition 15

Chapter 2 Defining and measuring mortality 33

Chapter 3 Designing a survey 53

Chapter 4 Using and interpreting survey results 107for decision making

Chapter 5 Ethical issues 127

Chapter 6 The end point: example of a good survey report 131

Annexes

1.1 Policy Paper Brief: Food for Nutrition 1751.2 Policy Paper Brief: Micronutrient Fortification 1771.3 Policy Paper Brief: Nutrition in Emergencies 1792 Anthropometry 1813.1 Weight for height reference in Z-scores 1913.2 Weight for height reference in percentage of the median 195

both boys and girls3.3 Height for age index in Z-scores 1973.4 Weight for age reference values for boys 1993.5 Weight for age reference values for girls 2014 Sample data collection forms for a nutrition 203

and mortality survey5 Random number table 2076 Example of ethical clearance form 2097 Resources 2178 Glossary of terms 2199 Q&A sheet 227

Table of Contents

MEASURING AND INTERPRETING MALNUTRITION AND MORTALITY 3

As a joint collaboration between the World Food Programme (WFP) and the Centers forDisease Control and Prevention (CDC), we would like to acknowledge the hard work ofthe persons involved in the development of this manual.

The key technical advisors were Bradley Woodruff and Reinhard Kaiser from the CDC.Many thanks for their technical wisdom, patience and persistence. In addition, warmthanks are given to Roodly Archer and Fiona Galloway of the CDC for their consistentsupport and input into the manual development process.

The core WFP team consisted of Rita Bhatia and Leah Richardson. Patrick Webb,Andrew Thorne-Lyman and Soha Moussa were also intimately involved in the develop-ment process - warm thanks for their technical input and illuminating visions.

Additional thanks to Kevin Sullivan for his technical review and comments.

There are many people who have contributed by providing comments on this manu-al and who deserve special mention, for without their help, advice and support, thismanual would not have been possible. In particular, thanks are given to all the staffof Nutrition Service and to the staff involved in the accompanying Pilot AdvancedNutrition Training.

Acknowledgements


The Purpose of this ManualWHY THIS MANUAL?• To provide guidance on issues relating to nutrition and mortality surveys.• To standardize survey methodology used by WFP staff, consultants and implementing

partners as a means of ensuring quality.• To standardize survey data interpretation and reporting.

WHO IS IT FOR?• WFP staff involved in nutrition-related data collection and/or nutrition interventions.• WFP staff involved in nutrition intervention decision-making. • WFP consultants and partners involved in nutrition surveys and interventions.

WHAT DOES IT CONTAIN?• A chapter on anthropometry and micronutrient deficiencies.• A chapter on mortality.• Guidance on deciding when secondary data are of sufficient quality for WFP purposes,

thereby rendering a new survey unnecessary.• A chapter on survey methodology, including an essential guide on selecting appropriate

survey methods and calculating sample size.• Guidance on interpreting and using survey results along with available complementary

materials to make informed decisions.• Guidance on reviewing survey reports for quality.• Example of a good survey in comparison to a less valid one.

HOW SHOULD IT BE USED?• As a reference guide.• As complementary material for the WFP Advanced Nutrition Training.• As a basis for framing questions/discussions with nutrition experts.

WHAT IT DOES NOT DO!• Does not absolve you of carefully thinking about decisions.• Does not provide programmatic guidance on what to do with survey results (guidance

on such issues can be found in the WFP Food & Nutrition Handbook).• Does not provide guidance on identifying malnutrition on an individual basis, but

focuses on population-based malnutrition.• Does not provide guidance on screening for either malnutrition or nutrition surveillance.


While nutrition has been central to WFP's mandate for the past 40 years, its importanceto WFP is growing. Already recognized as a key player in nutrition, WFP plays an increa-singly visible role in micronutrient fortification, HIV/AIDS-nutrition programming, scho-ol nutrition and enhanced maternal and child interventions. Following the ExecutiveBoard's endorsement of three nutrition policies in 2004, WFP's Strategic Plan for 2006-2009 reconfirms the organization's commitment to support improved nutrition amongchildren, mothers and other vulnerable people in both crisis and non-crisis settings.

However, achieving greater impact, and more clearly documenting results, will be a chal-lenge. The adoption of nutrition indicators in the context of Results-Based Managementrepresents a significant shift in WFP's use of nutrition data - going beyond monitoringand evaluation to the corporate reporting of nutrition and mortality outcomes. Enhancingthe capacity of WFP staff, partners and national counterparts to collect and interpret sur-vey-derived data will be a key to future successes. It is of utmost importance that suchdata are collected using appropriate methods, interpreted meaningfully (in light of com-plementary information on food security, health, markets, etc.), and presented and usedin transparent, appropriate ways.

This manual was prepared to support such capacity-enhancement, providing rigorousyet accessible advice on survey “dos and don'ts”. It provides step-by-step guidanceintended not only for nutritionists and nutrition focal points, but for all staff involvedwith data management, programme design and reporting. This is not to suggest thatWFP staff collect most nutrition information; such data are collected frequently bypartners. However, even when they do not themselves design and conduct surveys,WFP staff must be conversant with data quality issues to be able to interpret and useinformation appropriately. In other cases, WFP country offices, often with headquar-ters and regional bureaux support, will indeed directly oversee data collection andanalysis. This manual tells you how.

The manual (and its associated training modules) is a team product, combining inputfrom many individuals whose contributions are much appreciated. Future editions areenvisaged, with updates occurring according to developments in nutrition science, sur-vey methods and WFP needs. Feedback is therefore always welcome. We hope that theinformation provided will be widely used and lead to further improvements both in pro-gramming and in our ability to report on successful interventions.

Patrick Webb and Rita BhatiaNutrition Service

WFP, RomeJuly 2005

Foreword


Introduction

THE IMPORTANCE OF NUTRITION ANDMORTALITY INFORMATION FOR WFPSince its first operations in the early 1960s,the World Food Programme (WFP) hasprovided food to beneficiaries with thedual objectives of saving lives in crises andseeking to enhance lives and livelihoods indevelopment settings. Very often this hasrequired explicit attention to nutrition. Forexample, in emergencies malnutrition isan important determinant of mortality;food interventions therefore play a key rolein saving lives through their impact on thenutrition and health of affected popula-tions. That is, WFP emergency interven-tions seek to prevent the deterioration, orpromote recovery, of nutritional status ofcrisis-affected people. Outside of emergen-cies, WFP also seeks to improve the nutri-tion of vulnerable subpopulations at criti-cal times in their lives (particularly pre-gnant and lactating women, preschoolchildren, and certain adolescents).

In both contexts, operational success(i.e., positive impacts on nutrition)depends on effective use of nutrition data- information that helps define the pro-

blem, design appropriate responses, docu-ment change and allow for reporting oneffectiveness. Since 2004, the use of dataon nutrition (and now mortality) has takenon a much higher significance. This can betraced to five important developments:

First, in 2004, WFP's Executive Boardendorsed three new nutrition policy papers(see Annexes 1.1-1.3). These policies requi-re WFP to more systematically analysenutrition problems and define the mostappropriate responses based on up-to-dateknowledge and best practice. To determinethe nature and scale of problems to beaddressed by WFP requires greater inve-stment in the collection of nutrition informa-tion. As noted in the Executive Board deci-sion above, this means that WFP has toenhance staff capacity in nutritional asses-sment and data collection and management.Much needs to be done in terms of staff trai-ning, technical guidance, analytical supportand interaction with field partners engagedin nutrition and other humanitarian work.

Second, in 2004, the Executive Boardendorsed WFP's adoption of Results-Based

“WFP will mainstream nutrition in its programmes, advocacy and partnerships in orderto (i) tackle malnutrition directly, responding to and/or preventing malnutrition whenfood can make a difference, and (ii) enhance national and household capacities torecognize and respond to nutritional challenges. WFP will expand its efforts to achie-ve and document positive nutritional outcomes. This will include putting in place appro-priate staff capacity at country, regional and Headquarters levels in nutritional asses-sment, programme design, project implementation and data collection and manage-ment. WFP will engage more fully in global and national policy dialogues on malnutri-tion problems and solutions in collaboration with appropriate partners.”

World Food Programme Executive Board Decision, 2004 Annual Session:WFP/EB.A/2004/5-A/1

Management (RBM). Evidence-based pro-gramming is essential to achieve all ofWFP's goals, but it is particularly critical inthe domain of saving lives and protectingthe world's most nutritionally vulnerablepeople. The adoption of nutrition and mor-tality indicators in the context of RBM repre-sented a significant shift in WFP's approach.

Third, there has been increased attention tonutrition on the international developmentagenda in recent years. For example, theWorld Bank (2003) has stated that reducingmalnutrition is central to reducing poverty.As long as malnutrition persists, develop-ment goals for the coming decade will not bereached. Similarly, the Hunger Task Force ofthe Millennium Project (2005) has arguedthat since “adequate nutrition lies at theheart of the battle to fight hunger […] stan-dardized methods and approaches are nee-ded for measuring and documenting theeffectiveness of humanitarian responses bymeasuring child malnutrition and mortality.”

Fourth, WFP has developed an updated andfundamentally revised Memorandum ofUnderstanding with UNICEF (2005) whichstates that while “UNICEF will generallytake the lead in undertaking nutrition sur-veys representative at the national level, ingeographic regions or among certain benefi-ciary groups where WFP intervenes, WFPcan request UNICEF to collect data or WFPwill organize collection itself. WFP will alsotake lead responsibility for baselines, M&Esample surveys, impact assessments and inthe context of specific operations-researchor pilot activities where nutrition is a keyelement of the activity.” In other words, asof today WFP has a clearly recognized man-date - and hence responsibility - for collec-ting and using nutrition and mortality datawhere necessary (i.e., where UNICEF orother partners are not doing it). Fifth, WFP has become increasingly invol-ved in supporting local micronutrient fortifica-

tion of food aid commodities. While experien-ces in this area go back at least 10 years, thescale of local fortification activities has grownrapidly, as has WFP's role in the promotion ofnational fortification policies and its asses-sments of the prevalence of micronutrientdeficiencies while seeking to affect changes inmicronutrient status of nutritionally vulnera-ble populations. This requires WFP to be fullyproficient in the collection and use of data onmicronutrient malnutrition.

Given these important developments, it isimperative that WFP work more effectivelywith suitable nutrition and mortality infor-mation, particularly as they relate to the per-formance of various nutrition interventions.The WFP Food and Nutrition Handbookand related Food and Nutrition Trainingcourses already provide extensive guidanceto WFP staff and partners on these issues.However, understanding information, orensuring that information about the preva-lence of malnutrition simply exists, is nolonger enough. It is equally essential thatdata are collected using reliable and soundmethods, and that staff can appropriatelyinterpret the information for programmaticpurposes. This means that WFP staff haveto take on greater responsibility to:

1) Engage in dialogue with partners on data collection issues before surveysare undertaken (to ensure that surveys meet relevant standards and fulfil WFP information needs);

2) Build WFP's capacity, and that of key partners, in the skills related to the review and analysis of nutrition information;

3) Collect nutrition information to help quantify WFP's contribution to achievement of the Millennium Development Goals (MDGs); and

4) Competently discuss and disseminate nutrition-related information to the outside world.

10

NUTRITION INFORMATION FORTARGETING NUTRITIONALLYVULNERABLE PEOPLEThe effectiveness of WFP's programmeshinges on the organization's ability toidentify people in need and to make deci-sions about where food, along with neces-sary non-food resources, can be deliveredmost effectively. While malnutrition hasmany causes, an extended shortage offood inevitably leads to potentially life-threatening malnutrition. In combinationwith food security indicators, indicators ofmalnutrition can therefore be a powerfulinstrument for identifying geographicareas where people are either already mal-nourished or are nutritionally vulnerable,including specific sub-groups of vulnera-ble populations who have priority needs.This holds true for WFP country program-mes and for emergency needs assessments(ENA). Information about the prevalenceof malnutrition during normal times, or inthe early stages of an emergency, helps

WFP to judge the pre-crisis level of foodinsecurity. It also provides insights on thecurrent status of entire populations (notjust at-risk individuals) and suggests likelytrends. The prevalence of malnutrition istherefore important as a “baseline” com-parison and as a predictor of the expectedeffects of disruptive events.

Nutritional surveillance or repeated sur-veys can also reveal trends in the preva-lence of malnutrition, as seen on a natio-nal level in Bangladesh over time (Figure1.1). Such information can help interpretwhether changes in malnutrition duringa crisis are actually due to the crisis orwhether they are associated with expec-ted (normal) seasonal variations.

Indicators of malnutrition generallyreflect what has happened to an indivi-dual or population in the past, while foodsecurity indicators tend to be more focu-sed on the present and even the future.


INTRODUCTION

25

20

15

10

5

0

June ‘90

December’90

Febrary

‘91

August ‘91

December’91

Febrary

‘92

August ‘92

December’92

Febrary

‘93

August ‘93

December’93

Febrary

‘94

August ‘94

December’94

Febrary

‘95

August ‘95

December’95

Febrary

‘96

August ‘96

December’96

Febrary

‘97

June ‘97

Source: based on data published by Helen Keller International, 1999

Figure 1.1. Seasonality of acute malnutrition in Bangladesh (for children 0-59 months)

Some indicators, such as child stunting,reflect the totality of conditions that haveinfluenced children's growth over a num-ber of years. Others, particularly wasting,are more likely to reflect recent processesrelated to weight loss that may be causedby short-term hunger or illness. As such,it is important to remember that nutritioninformation is most useful for purposesof assessment when it is used in conjun-ction with food security data.

NUTRITION INFORMATION FOR PROJECT DESIGN, IMPLEMENTATION AND DOCUMENTATIONBeyond assessments, nutrition surveyinformation also has an important role inproject design, monitoring and manage-ment. Collecting baseline informationrelated to nutrition in a project area (asthrough a baseline survey, if needed) is akey part of designing the most appropria-te nutrition activities. Prevalence infor-mation is useful to set objectives andindicators as part of preparing a logicalframework and activity summary. Anexample is the goal of “reducing the pre-valence of underweight by 15%” outli-ned in WFP's Mother and Child HealthCare activity for Nepal (2002-2006).

Baseline surveys ideally should try toexplore the potential causes of malnutri-tion and document its prevalence before(or at the outset of) an intervention. Inaddition to a lack of food, those causesmay include inadequate access to healthservices, a lack of clean water or sanita-tion, illness or sub-optimal caring practi-ces. While Vulnerability Analysis andMapping (VAM) and ENA identify popu-lations in need of food aid, the samepopulations are often in need of manyother interventions. Such complementa-ry interventions and resources may needto be in place for food assistance to signi-

ficantly reduce the prevalence of malnu-trition. Where malnutrition is linked topoor caring practices, baseline surveyscan reveal more about the prevalence ofsuch practices. The information can thenbe used to tailor communications andnutrition education interventions.Information from nutritional surveysabout causes of malnutrition also can beinvaluable in persuading partners tofocus their activities in the same areas asWFP. Leveraging joint activities withsuch partners (governmental agencies,non-governmental organizations, UNICEF,etc.) can greatly enhance the effective-ness of those efforts.

Appropriately collected nutrition indica-tors are not only important for assessingcorporate progress towards meetingstrategic priorities (RBM), they are alsocritical to effective programming - formonitoring trends in nutritional statusand effecting project design changesthat may be needed to ensure measura-ble impact.

NUTRITION INFORMATION FOR ADVOCACY AND RESOURCE MOBILIZATIONThe basic information that comes out ofa nutrition survey can be a powerfuladvocacy tool for convincing expertsand the general public of the need foraction. Statements about the prevalenceof malnutrition or the number of mal-nourished children are easy to under-stand and to put into a headline. Forexample, following the nutritional sur-vey undertaken by WFP and CDC inSudan, the following headline was pic-ked up by the media: “More than one-fifth of Darfur children malnourished.”Such a statement quickly conveys to thereader the extent of the malnutritionproblem in a given setting.

12

The results of nutrition surveys can alsobe used for advocacy with donors to fund“forgotten” emergencies. Those includethe situations where “donor fatigue”stems from the protracted duration of agiven situation, particularly those notbeing covered by the media. For exam-ple, funding for WFP's 17-year opera-tion supporting the Saharawi refugeesin Algeria ran so low that the supple-mentary feeding program was phasedout in 2002. To determine the extent ofmalnutrition in the resource-reducedsetting, WFP and the United NationsHigh Commissioner for Refugees con-tracted the Institute of Child Health,London, to undertake a nutrition sur-vey. The survey provided data on theprevalence of different indicators ofmalnutrition. Largely as a result of suc-cessful advocacy with the surveyresults, WFP was able to persuadedonors to once again fund the supple-mentary feeding program and toaddress micronutrient deficiencies.

THE STRUCTURE OF THIS MANUALWFP staff will always collect nutritionand mortality data. While such data arenormally and frequently collected byimplementing partners, it is importantthat all such nutrition information pro-vided by partners or other secondarysources in the future better conform toWFP reporting needs, be reported moresystematically and be collected andanalysed in statistically appropriatefashion. Additionally, there will beinstances where WFP staff will need towork with counterparts to oversee datacollection or conduct and analyse sur-vey data themselves. Over time this willcontribute to greater understanding andownership of nutrition informationwithin WFP.

The present manual, and its associatedtraining modules (for which the manualserves as the main reference text), willenable WFP staff and its partners togain the skills necessary in all of theabove domains. The first chapter brieflyreviews the many types of malnutrition(child and adult, macronutrient andmicronutrient deficiencies) that matterto WFP, and it introduces the measuresused to assess such problems. Thesecond chapter focuses on the measure-ment of mortality (in the context ofemergency settings). The third chapterelaborates on deciding whether a newsurvey is needed, and if so, the appro-priate design and implementation of it.Chapter 4 discusses the analysis andinterpretation of survey results, particu-larly in relation to programmatic deci-sion-making needs. The fifth chapterfocuses on ethical issues relating to sur-veys of distressed populations, be theyquestionnaire surveys or more invasivemethods (involving, for example, a fin-ger prick for blood to assess anaemia).The sixth and final chapter considerswhat “ideal” survey reports may looklike (i.e., details of a well conductedsurvey and a survey report containingessential information).

The manual is a joint product of WFP'snutrition service and the Centers forDisease Control and Prevention (CDC),intended for use by WFP staff and par-tners. Comments on how to improve itare always welcome.


INTRODUCTION

Malnutrition literally means “bad nutrition”and technically includes both over- andunder- nutrition. In the context of develo-ping countries, under-nutrition is generallythe main issue of concern, though indu-strialization and changes in eating habitshave increased the prevalence of over-nutri-tion. Nonetheless, within the context ofWorld Food Programme (WFP) programsand assessments, malnutrition refers tounder-nutrition unless otherwise specified.

WFP defines malnutrition as “a state inwhich the physical function of an indi-vidual is impaired to the point wherehe or she can no longer maintain ade-quate bodily performance process suchas growth, pregnancy, lactation, physi-cal work and resisting and recoveringfrom disease.”

Malnutrition can result from a lack ofmacronutrients (carbohydrates, pro-tein and fat), micronutrients (vitaminsand minerals), or both. Macronutrient

deficiencies occur when the bodyadapts to a reduction in macronutrientintake by a corresponding decrease inactivity and an increased use of reser-ves of energy (muscle and fat), ordecreased growth. Consequently, mal-nourished individuals can beshorter(reduced growth over a prolon-ged period of time) and/or thinnerthan their well-nourished counter-parts. 'Hidden Hunger', or micronu-trient malnutrition, is widespread indeveloping countries. It occurs whenessential vitamins and/or minerals arenot present in adequate amounts in thediet. The most common micronutrientdeficiencies are iron (anaemia), vita-min A (xerophthalmia, blindness), andiodine (goiter and cretinism). Others,such as vitamin C (scurvy), niacin(pellagra), and thiamin or vitamin B1(beriberi), also can occur during acuteor prolonged emergencies when popu-lations are dependent on a limited,unvaried food source.


Defining and MeasuringMalnutrition

1CHAPTER

• Measurement of malnutrition in children and adults

• Anthropometric indices � · What they are

· How to calculate Z-scores and percentage of median · How to calculate body mass index (BMI)

• Micronutrient deficiencies: definitions, clinical and biochemical assessments, andcommodity testing

· Iron· Vitamin A

� · Iodine

MEASURING MALNUTRITIONAnthropometric1 information can be usedto determine an individual's nutritionalstatus compared with a reference mean. Italso can be used to determine the preva-lence of malnutrition in a surveyed popu-lation. Acute and chronic malnutrition ismeasured and quantified through anthro-pometric tools. Within both emergencyand development contexts, population-based nutrition indicators can be a usefultool for assessment, prioritization and tar-geting. The basic information and measu-rements that constitute anthropometricmeasurements in children are:

These measurements are the key buildingblocks of anthropometrics and are essen-tial for measuring and classifying nutritio-nal status in children under 5 years. Moredetailed information on anthropometricmeasurement techniques and recommen-ded equipment can be found in Annex 2.

Measurement of malnutrition in children under 5 yearsPhysical growth of children (under 5 years)is an accepted indicator of the nutritionalwell-being of the population they repre-sent. Adults and older children can accessproportionally larger reserves of energythan young children during periods ofreduced macronutrient intake. Therefore,the youngest individuals are most at riskfor malnutrition. For assessment of acutemalnutrition, children are more vulnerableto adverse environments and respond rapi-

dly to dietary changes, they are also moreat risk of becoming ill, which will result inweight loss. Consequently, their nutritionalstatus is considered a good gauge for popu-lation-based malnutrition. For assessmentof chronic malnutrition, children during thedevelopmental years are susceptible to ske-letal growth failure in ways that adults arenot and are a good reflection of long-termnutritional issues. Therefore, the surveyresults of the under-5-years population areused to draw conclusions about the situa-tion of the whole population, not just ofthat age group.

Reference populationTo determine a child's nutritional status,you need to compare that child's statuswith a reference for healthy children.References are used to compare a child'smeasurement(s) with the median for chil-dren of the same sex and age for height-for-age and weight-for-age, or to childrenof the same sex and height for weight-for-height. The internationally accepted refe-rence was developed by the CDC and itsNational Center for Health Statistics(NCHS) using data collected from a popu-lation of healthy children2.The World Health Organization (WHO)adopted the NCHS reference curves forinternational use. Evidence has shownthat the growth patterns of well-fed, heal-thy preschool children from diverse ethnicbackgrounds are similar and consequentlyare applicable for children from all racesand ethnicities. These references are usedby agencies involved with nutritionalassessments and analysis, including WFP.

16

1 Anthropometry is the measurement of the proportions of the human body.2 The NCHS reference was established in 1977 using two different child populations: 0-36 months, lying recumbent,

and 2-18 years, measured standing. Length measurement is always greater than height measurement for children.When interpreting data for children around 24 months it should be noted that wasting and stunting rates may peakas a result of the overlapping data sets.

Age Sex Length Height Weight Oedema

MEASURING AND INTERPRETING MALNUTRITION AND MORTALITY

DEFINING AND MEASURING MALNUTRITION

17

1CHAPTER

EXPRESSIONS OF NUTRITION INDICES Anthropometric indices can be expressedin relationship to the reference popula-tion in two different statistical terms:standard deviations from the median orpercentage of the median.

1.1 Standard deviations, or Z-scoresThis is the preferred expression foranthropometric indicators in surveys. It isthe difference between the value for anindividual and the median value of thereference population for the same age orheight, divided by the standard deviationof the reference population. In other

words, by using the Z-score, you will beable to describe how far a child's weightis from the median weight of a child atthe same height in the reference value.

1.2 Percentage of median The percentage of median is common-ly used and recommended for admis-sion/discharge criteria for selectivefeeding programs.

Percentage of median is the ratio of thechild's weight to the median weight of achild of the same height in the referencedata, expressed as a percentage.

Z-SCORE =measured value - median of reference population

standard deviat ion of the reference populat ion

Example 1.1 Calculation of Z-score for weight-for-length

A little boy measures 84 cm in length and weighs 9.9 kg.

By referring to the reference population data in Annex 3.1, you find that the referen-ce median weight for boys of 84 cm is 11.7 kg and that the standard deviation for thereference distribution for boys of 84 cm is 0.908. Using these values, and the formu-la provided, you can calculate a weight-for-length Z-score for this child.

Weight-for-length Z-score = 9.9 kg - 11.7 kg

0.908

= -1.98 Z-scores

Therefore, this child is 1.98 standard deviations (or Z-scores) below the mean weight-for-length. Similar calculations could be performed for height-for-age andweight-for-age.

PERCENTAGE OF THE MEDIAN = x 100measured weight of the child

median weight of the reference population

18

1.3 Oedema as a confounding factor Oedema as a confounding factor: Childrenwith edema should always be classified ashaving severe acute malnutrition regardlessof their weight-for-height or weight-for-agez-score or percent of median. Theseanthropometric indices can be calculatedfor children with edema, and computer pro-grams, such as the EpiNut module of EpiInfo, will automatically calculate them forchildren with edema along with all otherchildren in the survey. Regardless, even ifthe indices are calculated and included inthe final survey dataset, the weight-for-height and weight-for-age indices should beignored when deciding which children haveacute malnutrition. Of course, edema hasno effect on height-for-age indices.Oedema increases weight due to the accu-

mulation of fluids; therefore, indices such asweight-for-height and weight-for age will notbe representative of the true anthropometricstatus. Z-score should not be calculated forthese children, as the weight measurementwill not be valid. When oedema is present inboth feet (bilateral pitting oedema) a child isconsidered severely malnourished, regar-dless of his weight-for-height Z-score

Example 1.2 Calculation of percentage of the median weight-for-length

A little boy is 84 cm tall and his weight is 9.9 kg. Using the weight-for-height table inAnnex 3.2, you can see that the median height for a boy measuring 84 cm is 11.5 kg.Using the percent median formula given, you can calculate the weight-for-height percentage of the median for this boy:

Percentage of the median weight-for-length = 9.9 kg x 100

11.5kg

= 86.1% of the median weight-for-length

Therefore, this child is 86.1% of the median weight-for-length. A similar approachcan be used for height-for-age and weight-for-age.

3 Nutritional oedema is always bilateral. If the accumulation of fluid is only in one foot, it might be the symptomof another medical condition that will require further investigation from the medical team.

Oedemous pitting occurring as a result of malnutrition



19

1CHAPTER

Reporting of survey results in Z-score or percentage of the median?

The preferred method of expressing prevalence of malnutrition obtained through sur-vey results is in Z-scores, primarily because the percentage of the median does nottake into account the standard deviation associated with the reference distribution ofweight for each height category. As the child grows, the standard deviation associa-ted with the reference median increases more slowly than the median weight. Theweight of a child who is 100 cm tall is further from the reference distribution than thatof the child who is 80 cm tall, and his weight deficiency, compared with the referencestandard deviation, is greater than that of the child of 80 cm.

The Z-score expression takes into account the standard deviation of the distributionand thus standardizes weight deficiencies, regardless of the height of the child. Use ofthe median height-for-age and weight-for-age is also calculated without taking intoaccount the distribution around the median in the reference population.

Moreover, the Z-score is a more statistically valid comparison to the reference popu-lation than the percentage of the median. When using Z-scores, all malnourished chil-dren, regardless of age and/or height, are likely to be actually classified as malnouri-shed. Since the percentage of the median only uses two factors to calculate malnu-trition, as opposed to the three factors used in Z-score calculations, percentage of the median has less likelihood of capturing all the malnourished children. Therefore, whenZ-scores are used to define malnutrition, the number of children classified as malnou-rished is higher than if the percentage of the median is used, and it is a more statisti-cally uniform approach to defining malnutrition.

Percentage of the median is primarily used as a programmatic tool for selective fee-ding programs (because of ease of calculation and understanding); therefore, programreports will often express malnutrition in percentage of the median. Bearing thesepoints in mind, WFP recommends that anthropometric survey results are expressedforemost in Z-scores. If circumstances call for, results can be presented secondarilyin percentage of the median along with Z-scores.

EXPRESSION OF NUTRITION INDICATORSTo define nutritional status based onanthropometric indices, cutoff values areused. Nutrition indicators are a tool tomeasure and quantify the severity of mal-nutrition and provide a summary of thenutritional status of all children in themeasured group. It provides a method bywhich the nutritional status of a group canbe compared easily over time or with othergroups of interest.

Prevalence of malnutrition in childrenOnce the comparisons are made betweenindividual nutritional status and the referen-ce population, you can calculate the preva-lence of malnutrition among the populationthe individuals represent. The prevalence ofmalnutrition is equal to the number of mal-nourished children divided by all childrenassessed in the population. To help you inyour calculations, you can use statisticalsoftware, such as Epi Info™ or the newlydeveloped Nutrisurvey, which will automa-tically calculate the nutritional indices.

20

Table 1.1 Classification of malnutrition for weight-for-height, height-for-age,and weight-for-age based on Z-scores

Classification

Adequate

Moderately malnourished

Severely malnourished

Z-score values

-2 < Z-score < + 2

-3 < Z-score < - 2

Z-score < - 3

Table 1.2 Classification of malnutrition for weight-for-height, height-for-age,and weight-for-age based on percentage of the median

Classification

Adequate

Mildly malnourished

Moderately malnourished

Severely malnourished

Weight-for-height (%)

90-120

80-89

70-79

<70

Height-for-age (%)

95-110

90-94

85-89

<85

Weight-for-age (%)

60-80

<60

Example 1.3 Calculation of prevalence of wastingTo calculate the prevalence of wasting, count all the children in the sample with a weight-for-height less than -2 Z-scores. Report the result as a percentage of the total sample. You measured 900 children.101 had a weight-for-height less than -2 Z-scores.The prevalence of wasting would be 101/900 or 11.2%.



21

1CHAPTER

Determining the prevalence of acute malnu-trition in a population can be useful inmany different ways. Malnutrition prevalen-ces are used to define emergency levels, tojustify initiation or suspension of nutritionprograms and to verify needs assessment.However, the decision to implement nutri-tion programs should be based on thoroughanalysis of factors such as the environment,food security and public health issues. Evenif the overall food needs of a population aremet, there may be inequities in the distribu-tion system, disease outbreaks and othersocial factors that can cause an increase inthe prevalence of malnutrition among cer-tain vulnerable groups.

Measuring malnutrition among adultsThe anthropometric indices used with chil-dren (weight-for-height, height-for-age andweight-for-age) cannot be applied to adults.There is no internationally accepted anthro-pometry reference for adults, and the prin-ciples of a standardized growth curve arenot applicable to adults. Consequently, analternative measure is used for adults.

Body Mass Index (BMI)The most useful measure of malnutritionin adults is the body mass index (BMI)4.BMI is calculated by dividing the weight(in kilograms) by the height (in meterssquared). Pregnant women or adults withoedema are excluded from surveys toassess BMI because of the bias introdu-ced by weight gain not related to nutritio-nal status. BMI is calculated as:

An example of a BMI calculation is providedin Example 1.4. The BMI cutoff values areapplied equally to both sexes (Table 1.3)and the same cutoffs are applicable to alladults except pregnant women and indivi-duals with oedema. BMI is not used forpregnant women due to the weight gainassociated with the pregnancy.

Example 1.4 Calculation of BMIA young, non-pregnant woman's height is 1.60 m and her weight is 50 kg. Using these values in the BMI formula, you would calculate her BMI as follows:

BMI = 50kg1.6m2

= 19.5Therefore, the woman's BMI is 19.5

4 Some populations, such as the Kenyan Samburu and the Sudanese Dinke, are genetically very tall and shouldnot be assessed using BMI cutoffs.

Table 1.3 Classification of adult malnutrition (also called Chronic Energy Deficiency)using Body Mass Index (BMI)

Malnutrition classification

Mild

Moderate

Severe

Cutoff point using BMI

17 ≤ BMI <18.5

16 ≤ BMI <17

BMI < 16

BODY MASSINDEX (BMI) =

weight of the adult (kg)

height of the adult2 (m2)

22

Low birth weight as a measurement of mother and infantnutritional statusSmall babies - especially low-birth-weight (LBW) babies - are effectivelyborn malnourished and are at higher riskof dying in early life. LBW is defined as abirth weight of less than 2,500 g. Thisindicator is widely used because itreflects not only the status (and likelynutritional health risks) of the newborn,but also the nutritional well-being of themother. That is, while a low birth weightresults from many other factors (includ-ing smoking, alcohol consumption dur-ing pregnancy, genetic background andother environmental factors), it remains agood marker for a mother's weight gainand the fetus' development during preg-nancy. The growth and development ofbabies are affected by their mother's pastnutritional history. Malnutrition is anintergenerational phenomenon.

A low-birth-weight infant is more likelyto be stunted (low height-for-age) by theage of 5 years. Such a child, without ade-quate food, health and care, will becomea stunted adolescent and later, a stuntedadult. Stunted women are more likely togive birth to low-birth-weight babies, per-petuating the cycle of malnutrition fromgeneration to generation. In addition, thelow-birth-weight infant remains at muchhigher risk of dying than the infant withnormal weight at birth. The proportion oflow-birth-weight infants in a populationis the major determinant of the magni-tude of the mortality rates and a proxyindicator for maternal malnutrition.

Low birth weight as an indicator is usuallycollected through monitoring data such asbirth records and clinic registrations. Assuch, there is usually uncertainty and biasassociated with such records because it is

a self-selective sample. In some instances,where reliable birth weight data is avail-able at the household level, birth weightcan be collected through a survey. Moreinformation on equipment used to meas-ure birth weight and infant weight can befound in Annex 2.

MICRONUTRIENT DEFICIENCIESMicronutrient deficiencies represent aless visible, but often devastating, formof malnutrition that can be particularlyprevalent among WFP's beneficiarypopulations already lacking sufficientquantity and/or quality of food. Thereis a close relationship between malnu-trition, which is often linked to lack offood, and specific micronutrient defi-ciency diseases that are associatedwith consumption of foods poor inmicronutrients. Since WFP's benefici-aries frequently have limited access toa varied diet, a large proportion ofthem are also likely to suffer multiplemicronutrient deficiencies. WHOprevalence data for micronutrient prob-lems suggest that 4 million women andyoung children are vitamin-A deficient,almost 7 million school children areiodine-deficient and 7 million womenof childbearing age are anaemic.Deficiences of one or more of thesemicronutrients usually means there arealso deficiencies of other micronutri-ents, because the origin of these defi-ciencies, a deficient diet, means thatother micronutrients are also presentin insufficient amounts.

Currently, most international effortsare directed toward reducing theprevalence of deficiencies of iron, vita-min A, iodine, zinc and folic acid.According to WHO, deficiencies iniron, vitamin A and zinc each rankamong the top 10 leading causes of

death in developing countries. Mostpeople affected by micronutrient defi-ciencies do not show overt clinicalsymptoms, nor are they necessarilyaware of the deficiency. Micronutrientdeficiencies represent a particularthreat to the health of children under5 years and pregnant women.

The following section will focus on thethree most common micronutrient defi-ciencies (iron, vitamin A and iodine).Known effects of these micronutrientdeficiencies include impaired physicaland mental growth among children, iron-deficiency anaemia, maternal mortality,low adult labor productivity and blind-ness. Although micronutrients arerequired in tiny amounts, the conse-quences of severe deficiencies can becrippling or fatal. However, deficienciesin other micronutrients can occur in apopulation where the food supply is inad-equate or not diversified. Severe niacindeficiency causes pellagra, a diseaseaffecting the skin, gastrointestinal tractand the nervous system. Pellagra is oftencalled the “4 Ds”: dermatitis, diarrhea,dementia and death. Severe thiaminedeficiency can cause beriberi, whilesevere vitamin C deficiency will causescurvy. Scurvy is recognized by painfuljoints, swollen and bleeding gums, andslow healing or re-opening of wounds.

Recently zinc deficiency has been gar-nering more attention. Although severezinc deficiency is rare, mild-to-moderatezinc deficiency is quite commonthroughout the world. It is estimatedthat some form of zinc deficiency affectsabout one-third of the world's popula-tion, with estimates ranging from 4% to73% across subregions. Worldwide,zinc deficiency is responsible for approx-imately 16% of lower respiratory tract

infections, 18% of malaria and 10% ofdiarrhoeal disease. In total, 1.4% ofdeaths worldwide (2002) were attributa-ble to zinc deficiency. Serum and plasmazinc concentrations are the most widelyused biochemical markers of zinc status .Circulating zinc concentrations is a usefulindex in assessing zinc status at the pop-ulation level. The collection and prepara-tion of intravenous blood samples forzinc analysis should be performed in acontrolled environment to ensure accu-rate assessment. Contaminant sources ofzinc can also be introduced by the techni-cian handling the blood, through sweat,fingernails or saliva (via sneezing orcoughing), zinc being present on theequipment used (needles, tubes, etc), andtransportation of dust particles; therefore,it requires an extremely controlled envi-ronment and special equipment to ensurethat the results are accurate. In most set-tings be introduced by the technicianhandling the blood, through sweat, fin-gernails or saliva (via sneezing or cough-ing), and transportation of dust particles;therefore, it requires an extremely con-trolled environment to ensure that theresults are accurate. In most settings inwhich WFP would be involved in a nutri-tion survey these conditions would bevery hard to achieve. Expert adviceshould be sought before attempting toassess zinc status of the population.

To assess some micronutrient deficien-cies, blood or urine needs to be collect-ed. Trained phlebotomists or lab techni-cians should be hired to collect bloodsamples when necessary. Where onlyfingerprick samples are needed, surveystaff can be trained. Because of the inva-siveness of such procedures, care mustbe taken to assure and respect therights of individuals by followingeach country's guidelines in this area.



23

1CHAPTER

It may be necessary in some regions toobtain written consent from parents toallow their children to participate in thesurvey. In addition, even given parentalapproval, consent from the child is nec-essary. Confidentiality of results alsoneeds to be considered. Feedback to theindividuals, families and communitiesregarding significant health problemsshould also be considered. For moreinformation about ethical issues, pleaserefer to Chapter 5.

If blood is going to be collected, alwaysfollow these universal precautions foryour own safety and the safety of theothers. These steps will prevent thetransmission of diseases such as hepati-tis B, HIV and other infections carriedin the blood.

1. Always explain the procedure to the individual (child and adult). Some micronutrient deficiency testing is more invasive than measuringweight and height. The sight ofblood or a needle prick might fright-en some individuals. Use reassuringterms and be empathic.

2. Always obtain informed consent. Ifthey do not agree, do not take asample.

3. Always be careful around biohaz-ardous materials. Never allow achild or any individual to play witha piece of equipment.

4. Always wear sterile latex gloves. 5. Only use one needle or lancet per

person. 6. After pricking the skin, place the

needle in a puncture-resistant con-tainer such as the commerciallyavailable red biohazardous contain-ers with the logo for biohazardouscontent. Do not leave it on the tableor the floor.

7. Always dispose of all biohazardousmaterial properly. The biohazardouscontainers should be disposed at thelocal health facility that uses standardprocedures for biohazardous contents.

Iron deficiencyBecause anaemia is the most commonindicator used to screen for iron deficien-cy, the terms anaemia, iron deficiency,and iron deficiency anaemia are oftenused interchangeably. There are differ-ences between these conditions whichare explained later. Prior to the develop-ment of iron deficiency anaemia, thereare mild-to-moderate forms of iron defi-ciency, in which various cellular func-tions are impaired.

Iron deficiencyAccording to WHO, iron deficiency is themost common nutritional disorder in theworld . It affects at least half of all preg-nant women and young children indeveloping countries. Iron deficiencyoften results from a lack of bioavailableiron in the diet, but also can occur dur-ing a period of rapid growth (pregnancyand infancy), when the body needs moreiron. Another common cause is increasedblood loss, such as gastrointestinalbleeding due to hookworm or urinaryblood loss due to schistosomiasis.

Anaemia Anaemia is defined by low hemoglobinlevels and can be caused by nutritionaldeficiencies of iron, vitamin B12, vitamin

24

WFP recommended tests for micronutrientdeficiencies• Anemia: Hemoglobin• Vitamin A Deficiency: Night Blindness

and/or Serum Retinol• Iodine Deficiency: Urinary Iodine

A and folic acid. It also can result fromchronic infections (malaria, worm infesta-tion, etc.), severe blood loss or inheritedabnormalities such as thalassaemia.Multiple causes of anaemia can coexist inan individual or populations and con-tribute to its severity; however, the mostcommon cause of anaemia is iron deficien-cy. Children younger than 24 months areespecially at risk for anaemia, which slowstheir mental and psycomotor develop-ment, only part of which may be reversiblelater in life. In older children, the ability toconcentrate and perform well in school ishindered. Among adults, anaemia is a seri-ous risk to mothers in childbirth: everyday some 140 women die in childbirthbecause of severe anaemia.

Iron deficiency anaemiaA sufficiently large lack of iron can causeanaemia. Although some functional con-sequences may be observed in individu-als who have iron deficiency withoutanaemia, cognitive impairment,decreased physical capacity and reducedimmunity are commonly associated withiron deficiency anaemia. In severe irondeficiency anaemia, capacity to maintainbody temperature may also be reduced.Severe anaemia is also life threatening.

Because anaemia can contribute to mater-nal mortality, infant morbidity, infant mor-tality, intrauterine growth retardation andlow birth weight, WHO recommendsscreening of all pregnant women foranaemia.

Clinical signs and biochemical test for anaemiaUsing clinical pallor of the nails or eyes(inferior conjunctiva) to diagnose anaemiaon a population basis should be avoidedbecause these clinical signs are very sub-jective and not precise. A more reliable

and easy method is to test the hemoglobinconcentration in the blood. Specific equip-ment is needed for the testingo.• latex gloves for you and your assistant• alcohol pads• sterile, dry gauze pads• disposable needles (lancets such

as Tenderlett) • microphotometer (such as the

HemoCue)• microcuvettes for the photometer• adhesive bandages

Hemoglobin testing using the HemoCue methodIn the field, hemoglobin levels are deter-mined by using a photometer, such asthat manufactured by HemoCue. Thiscompany also offers essential training forthe proper use and care of the testingequipment. A video is also available onHemoCue's Web site at URL:http://www.hemocue.com/hemocueus/sida_3.asp. The basic procedure is as follows:

1. Ensure ethical clearance from thehost government and obtain personalconsent from each individual.

2. Have the analyser turned on and thecuvette holder in the outer position;the screen should say “READY.”

3. Take a microcuvette out of the vialand reseal the vial.

4. After cleaning the finger of the childor adult with alcohol pads, hold thefinger firmly and prick with a dispos-able lancet (small disposable needle).

5. After the puncture has been made,apply gentle pressure as needed toextrude a large drop of blood.



25

1CHAPTER

Portable HemoCue machine

6. Release the pressure on the finger andwipe off the drop with a dry, lint-free wipe.

7. Wipe away an additional one or twolarge drops, alternately applyingand releasing pressure on the fingeras needed.

8. Apply the microcuvette of the HemoCueto a drop of blood from the same finger-prick. Blood is drawn into the cuvetteby capillary action. Hold the cuvette inplace until the entire teardrop-shapedcavity is filled with blood.

9. After wiping off any excess bloodfrom the sides of the cuvette, place itin the cuvette holder and insert it intothe HemoCue.

10. Read the hemoglobin concentration[Hb] and record the hemoglobin con-centration to one decimal point.

11. Apply an adhesive bandage on the fin-ger of the individual.

Errors that can occur due to incorrecthandling of microcuvette are:• microcuvette not completely filled;• contamination of the optical eye with-

in the hemoglobin instrument;• introduction of air bubbles (i.e., when

filled from the edge instead of the tip); and

• cuvette exposed to heat and humiditybecause of incorrect storage (i.e., whenthe lid is not closed properly). Notethat once the container has beenopened they may not stay active untilthe indicated expiry date.

Other methods, such as WHO's hemoglo-bin color scale and the Sahli method, havebeen used to determine hemoglobin con-centration; however, these methods areboth highly subjective and therefore lessaccurate than the more objectiveHemoCue method.are not recommendedby WFP. Compared to the HemoCuemethod, an objective method, the hemo-globin color scale and the Sahli methodhave low accuracy. They are not recom-mended by WFP. A program officer pre-sented with a report containing hemoglo-bin concentrations measured with thehemoglobin color scale or the Sahlimethod should be aware that such meth-ods lack precision.

Use of hemoglobin concentration to determine statusInternational cutoffs have been created toclassify the status of individuals based onthe amount of hemoglobin in the blood.

26

Table 1.4 Hemoglobin cutoffs to define anaemia in individuals living at analtitude <1000 m and non smokers

Age in years and sex

Children (both sexes)

0.5 < age in years < 5.0

5.0 < age in years < 12.0

12.0 < age in years < 15.0

Non-pregnant females > 15.0 years

Men > 15.0 years

Hemoglobin cutoff (g/dL)

11.0

11.5

12.0

12.0

13.0

UNICEF/UNU/WHO (2001) and INACG (2002)

Adjustments should be made on thebasis of pregnancy status, altitude andfrequency of cigarette smoking. Theconcentration of hemoglobin in bloodnormally increases as children get older.During adolescence, hemoglobin pro-duction increases even more as a resultof accelerated growth. For these rea-sons, age-specific values must be usedto define anaemia in children. Also,men have higher hemoglobin concentra-tions than women.

In women with adequate iron nutrition,hemoglobin concentration starts to fallduring the early part of the first trimester,reaches its lowest point near the end ofthe second trimester and then graduallyrises during the third trimester; trimester-specific adjustments hence have beendeveloped. At elevations above 1000 m,hemoglobin concentrations increase as anadaptive response to the lower partialpressure of oxygen and reduced oxygensaturation of blood.



27

1CHAPTER

Table 1.5 Adjustments to hemoglobin cutoffs for pregnancy, altitude andcigarette smoking (INACG 2002)

Stage of pregnancy (trimester)FirstSecondThirdTrimester unknown

Altitude (m) rangem < 1000

1000< m <12501250< m <17501750< m <22502250< m <27502750< m <32503250< m <37503750< m <42504250< m <47504750< m <52505250< m

Cigarettes smoked per dayFewer than 10 cigarettes/day10 < cigarettes/day < 20 20 < cigarettes/day < 4040 < cigarettes/daySmoker, amount unknown

Hemoglobin adjustment (g/dL)-1.0-1.5-1.0-1.0

No adjustment+0.2+0.5+0.8+1.3+1.9+2.7+3.5+4.5+5.5+6.7

No adjustment+0.3+0.5+0.7+0.3

Note: the adjustment is subtracted from or added to the hemoglobin cutoff values presented in Table 1.4.

Vitamin-A Deficiency (VAD)Vitamin A is a fat-soluble vitamin requiredfor normal growth and development. It isinvolved in the functioning of the eyes aswell as the immune and reproductive sys-tems, while also helping to keep skinhealthy. For children, lack of vitamin A maycause severe visual impairment and blind-ness. Note that clinical signs (nightblind-ness and other xeropthalmia) present thetip of the iceberg of VAD. Many more chil-dren, not suffering from clinical signs ofVAD, have low circulating levels of vitaminA (biochemical indicator of VAD) andhence suffer consequences of higher risk ofmorbidity and mortality. VAD significantlyincreases the risk of severe illness, and evendeath, from such common childhood infec-tions as diarrheal disease and measles. Notonly is VAD is the leading cause of child-hood blindness across developing coun-tries, it also affects children's immune sys-tems and is directly responsible for around10.8 million deaths each year. Eliminatingvitamin A deficiency would cut child deathsdue to measles alone by 50 percent.

Women and vitamin-A deficiencyWomen, whether pregnant or not, shouldbe asked about nightblindness during theirprevious pregnancy in the last 3 years, andthat should have been a pregnancy carried

to full term. However, pregnant women areparticularly vulnerable to VAD, particularlyduring the last trimester of pregnancywhen demand by both the fetus and themother is highest. Among pregnantwomen in high-risk areas (where food con-taining vitamin A is rare), the prevalence ofnight blindness often increases during thelast trimester. Night blindness during preg-nancy is highly associated with malnutri-tion, anaemia and increased morbidity inwomen and their infants. To assess theprevalence of night blindness among preg-nant women, you ask them about theirnight blindness history for their most pre-vious pregnancy.

Clinical assessment of VAD: night blindnessNight blindness is the inability to see afterdusk or at night and is the most commonvision problem resulting from severe vita-min-A deficiency. In many regions, a localterm is used to define night blindness. Toassess night blindness, ask the individual ifhe or she has any problem seeing in thedark, at night or in a darkened room com-pared to their eyesight during the day or in alighted room. For children, you may need toobtain the information from the child'smother or caregiver. Whenever possible, usethe local term for night blindness.

28

Table 1.5 Cutoffs for vitamin A deficiency (VAD) using retinol concentration

Serum retinol(micrograms/dL)

< 10

10 - 19.9

20 or above

Serum retinol(micromols/L)

< 0.35

0.35 - 0.69

0.7 or above

WHO definition ofdeficiency*

Severe

Moderate

None

IVACG definitionof deficiency**

Deficient

None

* WHO. Indicators of vitamin A deficiency and their application in monitoring and evaluating interventionprogram. WHO/NUT/96.10. World Health Organization, 1996. Geneva, Switzerland.

** Sommer A, Davidson FR. “Assessment of vitamin A deficiency: the Annecy Accords”. Journal of Nutrition2002;132:2845S-2850S.

WHO has created a scheme for classifying night blindness by interview, using four questions:1. Does your child have any problem see-

ing in the daytime?2. Does your child have any problem see-

ing at nighttime?3. If (2) is yes, is this problem different

from other children in your communi-ty? (this question is particularly appro-priate where VAD is not very prevalent)

4. Does your child have night blindness(use the local term that describes thesymptom)?

Biochemical assessment of VAD:serum retinol concentrationsAt the individual level, retinol does notreflect liver stores of vitamin A and may beaffected by other factors, such as infectionand protein-energy malnutrition. However,it does allow for the detection of subclini-cal vitamin-A deficiency at a level that doesnot lead to vision problems, but does lowerimmune response and hence increases therisk of morbidity and mortality.

At the population level, you measure theserum or plasma retinol concentration todetermine vitamin A status. The proportionof individuals with low retinol levels reflectsthe prevalence of VAD in children and adults.The prevalence in pregnant women may bea bit higher than the overall adult prevalence.

For assessment of vitamin A status, thehigh-performance liquid chromatography(HPLC) method currently is used, but it isexpensive and time-consuming. Thismethod requires the handling and trans-portation of blood specimens, with skilledtechnicians needed to operate the equip-ment. Further, it requires freezing of sam-ples, and transportation can be difficult.

A surrogate for plasma retinol is plasmaretinol-binding protein (RBP). It can be

measured by radical immunodiffusion, atechnique that is much simpler and lessexpensive than HPLC. RBP can also bemeasured in a rapid field test using driedblood spots. One alternative is to measureretinol levels by using filter paper bloodspot samples. Retinol can be measured ina small sample of serum obtained from afinger pricked by sterile lancet.

At this time there is no field-based methodfor testing for Vitamin A content in oil.

Iodine Deficiency Disorders (IDD) Iodine is a mineral that is part of the hor-mones produced by the thyroid glandlocated in the front of the neck. Wheniodine intake falls below recommendedlevels, the thyroid may no longer be able tosynthesize sufficient amounts of thyroidhormone. The resulting low level of thyroidhormones in the blood is responsible forthe damage done to the developing brainand the other harmful effects known col-lectively as the iodine deficiency disorders.

Iodine deficiency can cause a goiter - aswelling of the thyroid gland in the neck.Iodine deficiency is also associated withsevere mental disabilities due to perma-nent brain damage in the fetus and infantand retarded psychomotor developmentin the child. Such disorders can be pre-vented by iodising all edible salt.

Clinical signs of IDD: goiterPalpation of the thyroid is performed as anindicator of iodine deficiency. However, thistechnique is less reliable when there are fewgoiters and/or when the goiters are relativelysmall. The thyroid size is slow to respond tochanges in iodine nutrition. Therefore assess-ment of thyroid size through palpation maynot be representative of the current iodinenutrition status. Consequently, palpation isnot preferred by WFP; nonetheless, it often isstill used in the absence of other tests.



29

1CHAPTER

Biochemical assessment of IDD: urinary iodineThe measurement of the iodine concen-tration in urine is the recommended wayto assess the current iodine status of apopulation5. Urinary iodine (UI) concen-tration is a good indicator of iodineintake because most of the ingestediodine is excreted in the urine.

At the individual level, iodine excretionvaries throughout the day due to hydra-tion and iodine intake. At the populationlevel, the median UI concentration ofcasual specimens will be representativeof the population's recent iodine intake.For assessing the iodine status of a pop-ulation, urine specimens from individu-

als do not need to be collected over a 24-hour period. The goal is to have a medi-an urinary iodine concentrationbetween 100-300 µg/L(WHO/UNICEF/International Councilfor Control of Iodine DeficiencyDisorders [ICCIDD]). The urinary iodinesurvey should be used to help determinethe level of iodine needed in the salt toachieve median urinary iodine valuesbetween 100-300 µg/L.

Blood samples or blood filter paperspecimens for assessing thyroid func-tion (such as thyroid stimulating hor-mone [TSH], thyroglobulin, or T4) inchildren or adults are not recommendedfor survey settings.

30

Table 1.6 Epidemiologic criteria for assessing iodine nutrition based onmedian urinary iodine concentrations in school-age children

Median urinaryiodine (ºg/L)

< 2020-4950-99100-199200-299

>300

Iodine intake

InsufficientInsufficientInsufficientAdequateMore than adequateiodine intake

Excessive iodine intake

Iodine nutrition

Severe iodine deficiencyModerate iodine deficiencyMild iodine deficiencyOptimalRisk of iodine-induced hyperthyroi-dism within 5 or 10 years followingintroduction of iodised salt insusceptible groupsRisk of adverse health consequen-ces (Iodine-induced hyperthyroi-dism, autoimmune thyroid disease)

5 The current recommendations by WHO/UNICEF/ICCIDD on urinary iodine and goiter are specific to school-agechildren, those within the range of 6-12 years, although a narrower age range is acceptable, e.g., 8-10 years.

6 Testing kits are available to test for either iodite or iodate. Attention should be paid to this detail whendetermining which one to order and use in the field.

Source: WHO/NHD/01.1, 2001

Materials and procedures for urine collection You will need the following materials forcollecting urine samples:• Disposable cups for collecting urine specimens• Screw-capped tubes for urine storage

and transportation• Disposable pipette for transferring urine

from cup to tube• Tube labels• Tube racks• Cardboard with styrofoam-insert boxes• Mailing/shipping labels• Coolant• Disposable gloves (for handling of urine

which may pose an infectious diseaserisk to handlers)

• Permanent ink pens for labels• Sealable plastic bags• Waste disposal bags

Follow these guidelines when collectingurine samples:• Always wear gloves while handling

urine specimens to reduce the risk ofinfections from the urine.

• Provide each participant with a disposa-ble paper cup for urine collection.

• Ask the individual to urinate directlyinto the cup. It should be filled approxi-mately halfway with urine.

• Transfer approximately 3-5 mL of the urinespecimen to the screw-capped tube using adisposable pipette. (Note that urinary iodi-ne analysis generally requires 1 mL or less;however, the provision of extra urineallows for repeat analysis if necessary.)Dispose of used cups and pipettes proper-ly. The urine specimen should be labeledappropriately and placed into a tube rack.

At the end of the collection of survey informa-tion, urine specimens should be packed in batch-es into sealed plastic bags and then into a ship-ping box or padded bag. Refrigeration during theshipping process is preferable, but not required.Various techniques are used to measure uri-nary iodine. A detailed description of the

various methods can be found in a docu-ment produced by WHO entitledAssessment of Iodine Deficiency Disordersand Monitoring their Elimination. A guidecan also be found online at URL:http://www.who.int/nut/documents/assessment_idd_monitoring_elimination.pdf

Field methods to test for iodine in saltIodised salt is often a first-line defenceagainst iodine deficiencies, and there are cir-cumstances in which it is useful to test saltin order to determine coverage of iodisedsalt. When iodine deficiency prevalence isestimated to be high, or when the region ofinterest is landlocked, coverage of iodisedsalt can act as a proxy indicator for iodinedeficiency. An easy-to-use field-basedmethod for testing salt has been developed.

Method:- Place a small amount of the salt to be testedon a saucer and moisten with two drops oftest reagent (a dilute acid, potassium iodideand starch solution). - If iodate (or iodite)6 is present, the saltshould immediately turn blue-purple andremain blue for several minutes before fading.Disadvantages:- If the result is not interpreted immediate-ly, colour fading may occur over time andlead to incorrect results.- These kits are specific to the form ofiodine, either potassium iodate (KIO3) orpotassium iodide (KI) salt.

Testing kits:ICCIDDc/o Centre for Community MedicineAll India Institute of Medical SciencesNew Dehli-29, IndiaEmail: [email protected]

WYD Iodine Checker Salt Research InstituteChina National Salt Industry Corporationhttp://www.chinasalt.com.cn/SALT-3/Product1%20-%20ChinaSalt_com.htm



31

1CHAPTER

32

REFERENCES

I World Food Programme. Food and Nutrition Handbook. Rome: World FoodProgramme; 2000.

II Prudhon C. Assessment and treatment of malnutrition in emergency situations. Paris:Action Contre la Faim; 2002.

III Habicht JP, Martorell R, Yarbrough C, Malina RM, Klein RE. “Height and weight stan-dards for preschool children: How relevant are ethnic differences in growth potential?”Lancet 1974;i:611-5.

IV Centers for Disease Control and Prevention. What is Epi Info™? Atlanta: US Departmentof Health and Human Services. Available at URL: http://www.cdc.gov/epiinfo/.

V Erhardt J, Gross R. Nutrition survey software. Available at URL:http://www.nutrisurvey.de/.

VI McIntire D et al. “Birth weight in relation to morbidity and mortality among newborninfants”. N Engl J Med 1999340:1234-1238.

VII World Health Organization. The World Health Report 2002. Geneva: World HealthOrganization; 2002.

VII International Zinc Nutrition Consultative Group (IZiNCG). “Assessment of the Risk ofZinc Deficiency in Populations and Options for Its Control, 2004”. Available at URL:http://www.izincg.ucdavis.edu/publications/FNBv25n1supp2zinc.pdf

IX World Health Organization. Micronutrient deficiencies: battling iron deficiency anaemia.Geneva: World Health Organization; 2003. Available at URL:http://www.who.int/nut/ida.htm.

X Sanchez-Carrillo CI. “Bias due to conjunctiva hue and the clinical assessment of anaemia”.J Clin Epidemiol 1989;42:751-4.

XI Sharmanov A. Anaemia testing manual for population-based surveys. Calverton,Maryland: Macro International Inc.; 2000.

XII HemoCue AB. Corporate Web site. URL: http://www.hemocue.com/.

XIII International Nutritional Anaemia Consultative Group (INACG), 2002. “AdjustingHemoglobin Values in Program Surveys”.http://inacg.ilsi.org/file/HemoglobinValues2004.pdf

XIV Tanumihardjo SA, Blaner WS, Jiang T. MOST Technical Report. Dried blood spotretinol and retinol-binding protein concentrations using enzyme immunoassay assurrogates of serum retinol concentrations. Arlington: The MOST Project; 2002 Jun.Available at URL: http://www.mostproject.org/Assesmt%20Methods.PDF.

XV World Health Organization. “Vitamin A”. Available at URL: http://www.who.int/vaccines-diseases/en/vitamina/science/sci05.shtml.

XVI Sullivan KM, May S, Maberly G. Urinary iodine assessment: a manual on surveyand laboratory methods. 2nd ed. Atlanta: Emory University and UNICEF; 2000.Available at URL: http://www.sph.emory.edu/PAMM/lab/UIAssessment2.pdf.

Deaths, and counting deaths, is a crucialpublic health indicator for many reasons.Death is the final and most definitive healthoutcome of many important public healthproblems. And most important causes ofpoor health in a population, when commonand severe enough, produce an elevation inthe mortality rate. Death is easily defined,making it a health outcome for which astandardized case definition is easilyapplied. Because death results from somany health problems - from chronic dis-eases, infectious diseases and injuries - themortality rate can provide an overall indica-tor of general health status of a population.Mortality rates can also provide informationon nutritional status because widespreadmalnutrition among children or adultsalmost always results in an elevation of themortality rate, especially if the level of com-municable diseases is high. Nonetheless,the mortality rate is a relatively insensitivemeasure of population health statusbecause conditions often must be quitepoor before it is markedly elevated. Themortality rate is also relatively non-specific;there are many causes of elevated mortali-ty, any one of which might lead to an

increase in the mortality rate. As a result, anelevated mortality rate can indicate thatthere is indeed a health problem in a popu-lation, but it cannot indicate the cause.

Mortality rates have been measured in manycountries for hundreds of years. Deaths areoften counted by various authorities, includ-ing religious leaders, civil authorities andpublic health professionals. In stable popula-tions, mortality rates are usually monitoredby registering deaths using a vital statisticssystem, and reporting deaths is mandatoryin most countries. However, in many coun-tries in the developing world, vital statisticssystems are far from complete. Even if func-tioning, such systems usually are disruptedearly in situations of civil conflict or displace-ment. As a result, cross-sectional surveys areoften necessary to determine accurate mor-tality rates.

Often, especially in emergency situations,initial surveys will combine the assessmentof many different health and nutrition out-comes. Recently, nutrition assessmentsurveys have begun routinely includingmortality measurement. When combin-


Defining and Measuring Mortality

2CHAPTER

Key messages• Description of the different types of mortality rates, with formulas used� · Crude mortality rate/crude death rate (CMR)� · Age-specific mortality rate (ASMR)� · Age-specific mortality rate for children under 5 years (ASMR-U5)� · Under-5 mortality rate (U5MR)� · Cause-specific mortality rate� · Infant mortality rate (IMR)� · Maternal mortality rate (MMR)• Retrospective cross-sectional mortality surveys• The use of a “recall period” in mortality surveys• Determining the causes of mortality

ing nutrition and mortality assessment inthe same survey, it is important to con-sider the sampling strategy so as not tochoose a sample which is biased foreither outcome (See Chapter 3 for more details). Forexample, you cannot include only house-holds with children 6-59 months in thesurvey sample; the mortality rate in suchhouseholds may not reflect the mortalityrate in all households in the population.

COUNTING DEATHSJust counting the number of deaths in apopulation is not sufficient. For example,if you hear that there have been 139deaths in a certain population, this tellsyou nothing about the rate of death ifneither the size of the population fromwhich these deaths were reported nor thetime period during which these deathsoccurred were known. If you subse-quently learn that the deaths came froma population of 5,000 persons, you stillhave insufficient information: If thesedeaths occurred over 10 years, the rate ofdeath in this population is quite low; onthe other hand, if these deaths occurredin the same month, the rate of death isextremely high. This example illustratesthat every mortality rate must have:

• The number of deaths (the numeratorof the mortality rate);

• The size of the population from whichthe deaths came (the populationdenominator of the mortality rate);and

• The time period during which thedeaths occurred.

If 127 deaths occurred in a population of19,546 over 9 months time, the deathrate would be 127 deaths per 19,546 peo-ple per 9 months. However, this rate can-not be compared to the mortality rate in

other populations of different sizes or tonumbers of deaths counted over differenttime periods. To determine if the deathrate is high, low or normal as comparedwith rates in the same population in pre-vious time periods or compared to ratesin other populations, the mortality rate ina given population must be converted toa rate using a standard populationdenominator and time period.

The first step in such a conversion is todecide how the final rate should beexpressed. There are three differentways to express the same mortality rate. a) # deaths/1,000/year: For many vital

statistics systems, which recorddeaths for longer time periods for anentire nation or province, mortalityrates are often expressed as the num-ber of deaths per 1,000 populationper year.

b) # deaths/1,000/month: In some dis-placed populations, when the acuteemergency is over and the health sit-uation stabilized somewhat, mortalityrates are sometimes expressed as thenumber of deaths per 1,000 popula-tion per month.

c) # deaths/10,000/day: During acutehumanitarian emergencies, when thenumber of deaths is totalled each dayor every few days, the mortality rateis often expressed as the number ofdeaths per 10,000 population per day.

These are three different ways to expressthe same mortality rate; the conversionamong them is merely an exercise inmathematics. For example, to convertfrom 34 deaths per 1,000 per year todeaths per 1,000 per month, just dividethe numerator by 12 (1/12th the numberof deaths will occur in one month as inone year): 34 / 12 = 2.8 deaths per 1,000per month.

34


DEFINING AND MEASURING MORTALITY

35

2CHAPTER

Because the actual number calculatedvaries depending on the populationconstant and the time period, a mortal-ity rate should never be stated withoutthese parameters. For example, you can-not say that the mortality rate is 1.86;you must say that the mortality rate is1.86 per 10,000 per day. The number1.86 would mean very different things ifthe population constant and time periodwere 10,000 per day, 1,000 per year or1,000 per month.

SOURCES OF DATA TO CALCULATE MORTALITY RATESIn stable populations, it is better to col-lect information of death prospectivelywhere, as each death occurs, it is report-ed to public health or governmentauthorities. Such systems allow the cal-culation of recent death rates as fre-quently as is required, and this data canbe collected easily. In vital statisticssystems, death reporting is oftenmandatory for persons charged withburying bodies. If deaths occur pre-dominantly in clinics or hospitals, med-ical personnel may be responsible fordeath reporting. In many societies, reli-

gious leaders record deaths. In humani-tarian emergencies, someone may behired to count deaths and monitor thearea designated as the graveyard or cre-mation site.

All of these systems of counting deathsrequire separate information on the sizeof the population from which the deathsoccurred in order to calculate mortalityrates. Such information for the popula-tion denominator may come from censuscounts; census projections; populationregistration; or the monitoring of births,deaths, immigration and emigration.. Inhumanitarian emergencies and other sit-uations in which data on population sizeis poor, techniques have been developedto estimate population size.

But which population size is used as thedenominator if deaths are counted for atime period during which the popula-tion size fluctuates? One approachwould be to calculate the average popu-lation during the time period (add thepopulation at the beginning of the timeperiod to the population at the end ofthe time period and divide by 2, produc-ing the arithmetic mean population).

Example 2.1 Expression of mortality ratesEven though they may appear to be very different, mortality rates expressed using different population constants or time periods indicate the same rate of death in apopulation. For example:

• Take a mortality rate of 9.6/1,000 population/year. • This could be expressed as 0.8/1,000 population/month.• This could also be expressed as 0.26/10,000 population/day.

The equivalency of these rates can be seen if we calculate from one rate to the other. For example, if the rate of 0.26 deaths per 10,000 per day remained constant for an enti-re year, we would expect that about 95 people would die (0.26 x 365 days in a year) outof each 10,000 people in the population.

A second approach would be to deter-mine the population at the mid-point ofthe time period and use that (this iscalled the “mid-interval population”and is the method used most commonlyin vital statistics systems). However, in many situations, the popu-lation denominator for a mortality rateitself is only a rough population esti-mate and no accounting can be madefor changes in the population during thetime period of interest. This is often truein humanitarian emergencies whichhave the additional complication of highrates of in-migration and out-migration.

TYPES OF MORTALITY RATESMany different rates are used to measuremortality:• Crude mortality rate/crude death rate

(CMR)• Age-specific mortality rate (ASMR)• Age-specific mortality rate for children

under 5 years (ASMR-U5)• Under-5 mortality rate (U5MR)• Cause-specific mortality rate• Infant mortality rate (IMR)• Maternal mortality rate (MMR)

Crude Mortality Rate (CMR)The crude mortality rate (CMR), alsocalled the crude death rate or CDR, is

defined as the number of people of allages and both sexes who die in a giventime interval divided by the total pop-ulation at the mid-point of that timeinterval. The CMR always includesthe length of the time interval and astandard population size, called thepopulation constant. For example, aCMR may be 8.5 deaths per 1,000 per-sons per year.

It is calculated by the following formulawhere:• the numerator of the fraction in

parentheses is the number of deathswhich occurred in a specific popula-tion during a certain time period.Only deaths which occurred duringthis time period should be included inthe numerator of the mortality rate.

• the denominator of the fraction is thenumber of people in the population inwhich these deaths occurred. Thispopulation should be well defined,and only persons fitting this definitionshould be included in the denomina-tor. For example, if you are calculat-ing the mortality rate for a certainprovince, only people who lived inthat province during the time periodof interest should be included in thepopulation denominator.

36



37

2CHAPTER

This means that each person has an average likelihood of 0.006393 (or 0.6393% chance)of dying during the 8-month period. Stating that the CMR is 377 deaths per 58,975 per 8 months has little meaning; it can-not be directly and meaningfully compared to other CMRs from previous periods orfrom other populations. However, if this rate is converted to deaths per 1,000 per year,as follows, it can be compared to other rates expressed in the same way:

1. To convert this likelihood to a likelihood for a standard population size, we multiplythis rate by 1,000 to obtain 6.4. This means that during the period of 8 months, 6.4 peo-ple died for each 1,000 people in the population.

2. To convert this likelihood to a likelihood for 1 year: 6.4 is divided by 8/12ths (8 monthsdivided by the 12 months in 1 year) = 9.6 deaths per 1,000 population per year. This ratemeans that, if the death rate for the 8-month period continued for an entire year, forevery 1,000 people in the population, there would be 9.6 deaths.

Using the formula above, the rate of deaths per 1,000 per year would be calculated asfollows:

Example 2.2 Calculation of CMR with standard time unitIn a specific population, 377 deaths occurred during a period of 8 months in a popula-tion with a mid-interval size of 58,975.

Using the information in the formula you get:377 58,975 = 0.006393 or 0.6393%

Vital statistics systems which calculate mor-tality on an annual basis use the size of thepopulation on July 1 to indicate the averagepopulation between January 1 and December31. Such systems count the number of deathsduring a year's time, divide it by the mid-interval population, then multiply by 1,000(the population constant) to get the numberof deaths per 1,000 population per year.

Since the CMR reflects the overall risk ofdeath in the population among all ages andboth sexes, it is the least specific indicator ofmortality. Mortality reflected in the CMRmay result from causes as varied as thosefrom violent deaths from massacres to thosefrom neonatal tetanus. If only one indicator

of mortality can be calculated, CMR is usu-ally the one chosen. Ideally, a newly calcu-lated CMR should be compared with a pre-vious CMR from the same population todetermine whether the mortality rate is ris-ing or falling. Such trend information can beused as an overall evaluation of health,nutrition and other interventions. However,when prior mortality data are unavailable, arough rule-of-thumb can be used: • a CMR of less than 1 death per

10,000/day indicates a reasonable healthsituation;

• a CMR of more than 1 death per10,000/day reflects elevated mortality; and

• a CMR of more than 2 deaths per10,000/day indicates a health emergency.

Calculating CMR using person-time unitsThe denominator of a mortality rate is thenumber of people in the population; howev-er, the denominator can also be seen asbeing based on person-time instead of thenumber of persons. That is, a rate of 10deaths per 10,000 per day may be seen asthe risk of death in 10,000 people during aperiod of one day, or the risk in 5,000 peo-ple during 2 days, or the risk in 1,000 peo-ple during 10 days. In all three examples, thedenominator is 10,000 person-days (thenumber of people in the population multi-plied by the number of days in the time peri-od). This will be important later when dis-cussing the measurement of mortality incross-sectional surveys. The sample sizerequired to achieve a certain precisionaround the estimate of the CMR is the num-ber of person-time units required in thedenominator of the rate.

In general, if births and deaths are distribu-ted evenly throughout the time period, theneach person who was born or who diedduring the time period contributes, on ave-rage, _ a person-time period to the denomi-nator. For example, vital statistics systems

assume that births and deaths occur evenlythroughout the year. Therefore, each personwho was born or who died during the yearcontributed about _ a year to the denomina-tor. Use of the mid-interval population, asdescribed above, captures half of deaths andhalf of births and is one way of adjusting thepopulation denominator for the incompletecontribution of births and deaths.

Age-Specific Mortality Rates (ASMR)Age-specific mortality rates (ASMR) restrictboth the numerator and denominator to per-sons of a certain age. For example, a morta-lity rate for persons 15-49 years of age is thenumber of deaths of persons 15-49 years ofage divided by the mid-interval population ofpersons 15-49 years of age (adjusted for thelength of the time period).

ASMR is often used to determine if the rate ofdeath is substantially different from thatexpected in any specific age group. Forexample, in a survey in Kosovo, the deathrate among young men was much higherthan expected, indicating that some factordisproportionately increased the risk of deathin this age group.

38

Figure 2.1 Age-specific mortality rates, Badghis Province, March 2001 - April 2002.

3.5

3

2.5

2

1.5

1

0.5

0

<5 5 - 14 15 - 49 50+

Mor

talit

y ra

te (d

eath

s/10

,000/

day)

Age group (year)



39

2CHAPTER

Age-Specific Mortality Rate for children under 5 (ASMR-U5)The age-specific mortality rate for childrenunder five (ASMR-U5) is calculated bydividing the number of deaths of childrenunder age 5 during a specific time periodby the number of children under age 5.This is often called the “under-5 mortalityrate”; however, as described below, thisterm is also used for a completely differentmeasure of mortality in children under 5years of age. For this reason, the term“ASMR-U5” is used here to maintain adistinction between these two differentmeasures.

The ASMR-U5 often is used as a more sen-sitive indicator of the affect of emergencyconditions on mortality. When health andnutritional status in a population deteriora-tes, the ASMR-U5 often changes earlierand to a greater extent in a crisis situationthan the crude mortality rate becauseyoung children are more susceptible tohealth and nutrition insults than olderchildren and adults.

Cause-specific mortality rateThe cause-specific mortality rate mea-sures the rate of death due to a specificcause (i.e., malnutrition-related deaths,violence related deaths, etc.), and itincludes only those deaths attributed tothe cause of interest in the numerator.The population denominator may inclu-de the entire population, or if the causeof death occurs predominantly in a sub-group of the population, it may includeonly that subgroup. For example, if 28people in a population of 78,904 diefrom tuberculosis during the course of ayear, the cause-specific death rate fortuberculosis during this time period

would be 35.5 deaths from tuberculosisper 100,000 population per year.If data on the causes of deaths in apopulation are available, cause-specificmortality rates provide informationabout the most important causes ofdeath. Such information can be used todesign intervention programs addres-sing these causes. However, in manypopulations, such information is lackingand the causes of death must be obtai-ned from other sources (see below).

Because the number of deaths from a sin-gle cause is usually far lower than thenumber of deaths from all causes (mea-sured by the crude mortality rate), thedenominator of cause-specific death ratesis often expressed as per 100,000 popula-tion. This allows the actual rate to be anumber greater than one. In the exampleabove, the cause-specific death rate fortuberculosis could also be expressed as0.355 deaths from tuberculosis per 1,000per year, but that rate is less easily under-stood than the larger number of 35.5deaths per 100,000 per year.

As mentioned above, the denominator ofmortality rates for causes of death whichoccur predominantly or only in a certainpopulation group may include only thosemost susceptible to death from thatcause. For example, death rates for uteri-ne or ovarian cancer are almost alwaysexpressed as the number of deaths per100,000 adult women because deathsfrom these causes are confined towomen. Similarly, death rates for prosta-te cancer are calculated using a denomi-nator consisting of the number of adultmen in the population.

40

MORTALITY “RATIOS”Some mortality rates are not truly rates,even though they may be called a morta-lity "rate." These include rates whichuse live births as a denominator, such asthe UNICEF under-five mortality rate, theinfant mortality rate and the maternalmortality rate.

The under-five mortality rate, or U5MRas the term is most commonly used, isthe probability of dying before the age offive, usually expressed per 1,000 livebirths. The U5MR cannot be calculated

directly from data on births and deathsby age in a single year because thedeaths, for example, of four-year-oldsoccur to children born four to five yearsbefore the occurrence. U5MR can be cal-culated using life table methods, frombirth history data (the recording of thedate of each birth and the age of deathfor those that have died), or using indi-rect methods that are beyond the scope ofthis manual. This measure of childhoodmortality is used most frequently by UNICEFand others who conduct large healthassessment surveys.

Comparison of the two measures of under-5 mortality rates

Both U5MR and ASMR-U5 measure the mortality risk for children under 5 years of age,but the two indicators express the risk differently.• In the case of U5MR, the risk is expressed as the cumulative probability of dying

before age 5 years in a hypothetical group of 1,000 births.• In the case of ASMR-U5, the risk is expressed relative to the mid-interval

population as with the crude and other age-specific mortality rates.

Because U5MR expresses risk over 5 years, whereas ASMR-U5 expresses risk peryear, U5MR is often almost five times as large as ASMR-U5.

As each measure (U5MR and ASMR-U5) has its uses and its advocates, there is nooverwhelming reason to recommend one over the other; they are two different waysof expressing much the same data. Since relief agencies tend to be more familiar withage-specific mortality rates, which are derived in the same way as the CMR (countingdeaths during a period of time and dividing by a population denominator), that methodwill usually be the one used in emergency assessments. Moreover, the usual nutritionand mortality assessment survey done in emergency situations does not gather thedata necessary to calculate the U5MR; only the ASMR-U5 can be calculated.

On the other hand, U5MR is used by UNICEF and others when measuring child morta-lity in stable populations. It is presented as the measure of child mortality in manysummary publications, such as the State of the World's Children. To avoid confusion,any reporting on under-five mortality should specifically note whether it is calculatedas an age-specific mortality rate or as the probability of dying by the age of 5 years.For WFP purposes, the ASMR-U5 should be used.

Infant Mortality Rate (IMR)*The infant mortality rate (IMR) is the num-ber of deaths in children under 1 year ofage in a given period divided by the num-ber of live births in the same time period.The infant mortality rate allows assess-ment of the rate of death in the most vul-nerable age group - children less than 1year of age. It often rises earlier and fasterin the face of poor health and nutritionthan other mortality rates.

The infant mortality rate also is describedas a ratio since many of the children whodie and are recorded in the numeratorwere born before the beginning of the timeperiod, and thus their birth is not recordedin the denominator. If the time period is1996, for example, then a six-month-oldchild dying in March 1996 would havebeen born in 1995; her death would add tothe numerator but her birth would not beadded to the denominator. Likewise, achild born late in the year would still be atrisk of dying under the age of one formuch of the following year.

The difference between a ratio such asthe infant mortality rate and the usualmortality rate can also be seen anotherway. The infant mortality rate calculatesthe chances that a live-born infant willdie before his first birthday - it is a cumu-lative incidence. The denominator is thepopulation at the beginning of the timeinterval of interest. A normal mortality

rate gives the average risk of dying dur-ing the time interval for a person in apopulation and represents a true rate. Ituses as the denominator the averagepopulation during the time interval, ormid-interval population.

Maternal Mortality Rate (MMR) **The maternal mortality rate (MMR) usesas the numerator all deaths of pregnantwomen or pregnancy-related deaths with-in 42 days of the end of pregnancy. Thedenominator is live births. Maternaldeaths are often a relatively rare cause ofdeath and, as such, the rates should becalculated only for relatively large popula-tions (more than 1,000,000). Randomvariation in the maternal death rate calcu-lated in small populations with few birthsmay be misinterpreted as significanttrends, when they are not actually so.

The maternal mortality rate is critical todetermine the need for antenatal andobstetric services. Although the actualproportion of all deaths in a populationresulting from pregnancy-related causesis often small, the effects of a woman ofchildbearing age dying are often muchgreater for her family and the society thanthe deaths of others in the population.Therefore, in a certain country, if therewere 34,459 births in 2004 and 78 mater-nal deaths, the maternal mortality ratewould be 226.4 deaths per 100,000 livebirths.



41

2CHAPTER

*

**

CROSS-SECTIONAL SURVEYS FORRETROSPECTIVE MORTALITY RATESAs mentioned above, it is better tocount deaths prospectively so that themortality rates calculated representrecent events. However, sometimes thesystem to count deaths does not exist ordid not exist during an earlier timeperiod in which you want to measuremortality. In such cases, mortality canbe measured using surveys. Just as withnutritional status or other health outco-mes, mortality information can be col-lected from the randomly selected hou-seholds. The people living in these hou-seholds report the number of deathswhich have occurred in that householdduring a specified time period. Duringdata analysis, the information ondeaths for all the households includedin the survey sample is put together.The total number of deaths in all thehouseholds is counted and becomes thenumerator of the mortality rate. Thedenominator of the mortality rate is thetotal number of people in the all thehouseholds included in the survey sam-ple. The time period is the period in therecent past during which deaths areasked about. Hence, you do not need toknow the size of the entire populationsurveyed or a count of the total numberof deaths in that population in order tomeasure the mortality rate.

Determining the recall periodIn order to calculate a mortality ratefrom data obtained by a survey, onlydeaths which occurred in a definedperiod in the past, called the recallperiod, should be included. To improvethe accuracy of mortality estimates incross-sectional surveys, the beginningof the recall period should be a memo-rable date known to everyone in thepopulation. For example, the start of the

recall period may be a major holiday orfestival (Christmas, beginning ofRamadan, etc.), an election, an episodeof catastrophic weather or other remar-kable event. The end of the recallperiod is usually the date the interviewtakes place. You can then calculate thelength of the recall period by countingthe days between the holiday or otherevent marking the start of the recallperiod and the date of data collection.Of course, this is rarely a nice roundnumber of days, like 90 or 180 days.

To detect such deaths, survey workersask respondents living in randomlyselected households to tell them aboutdeaths which occurred during thisrecall period. The denominator of themortality rate, being the total numberof people living in selected households,can also be seen as the number of per-son-time units (i.e., the number of per-sons in the selected households timesthe number of time units in the recallperiod). For example, if a survey selec-ted 569 households in which 3,243 peo-ple lived and the recall period was 7.3months, the denominator would be23,674 person-months. The same num-ber of person-time units could be obtai-ned from a recall period of 3.65 monthsand a survey sample of 6,846 people.The denominator of the mortality ratecan therefore be increased either byincreasing the number of persons in thesurvey sample or by increasing thenumber of time units (that is, the lengthof the recall period).

42

Lengthening the recall period is one wayto minimize the sample size. However,as you lengthen the recall period, you areasking survey respondents to reportdeaths which occurred in the moredistant past; naturally,an individual'smemory may become less reliable overtime. This is especially true if, along withrecalling the death, you also ask respon-dents about the circumstances or causesof the death. Moreover, by lengtheningthe recall period, you will produce anestimate of the mortality rate for a longerperiod in the past, which may be lessrelevant to current needs than a morerecent mortality rate.

Thus the sample size for a mortality rateis the number of person-time units nee-ded to obtain the desired precisionaround the estimate of the mortality rate.In acute emergencies, the person-timeunit is usually person-days to express themortality rate in terms of the number ofdeaths per 10,000 population per day. Instable populations, the mortality ratemay be expressed as the number ofdeaths per 1,000 population per monthor per year. The procedure for calculatingthe minimum sample size to achieve a

certain precision for the mortality rate isdescribed in the chapter on surveys.

If the mortality rate calculated from sur-vey data includes all reported deathsduring the recall period, that rate is anaverage for the entire recall period. Italso may also be possible to recorddeaths as having occurred in specificparts of the recall period. For example, ifthe recall period is 9.5 months, one couldrecord the death as occurring in one ofthree intervals: 1-3 months, 4-6 months,or 7-9.5 months prior to the interviewdate. Because this increases the comple-xity of the mortality survey, it shouldonly be included if the additional infor-mation is useful and if the persons inter-viewed can reliably place deaths intothese shorter intervals. Most people tendto recall important or traumatic events ashaving occurred more recently thanactual fact, so care must be taken to besure responses are accurate. In practice,in populations where calendar time isnot closely followed and dates are notwell remembered, recalling exactly whendeaths occurred can be very difficult.Nonetheless, if a traumatic or importantdate can be identified within the recall



43

2CHAPTER

Advantages of a longer recall period

• A smaller sample size (i.e., number ofhouseholds) is needed for the sameprecision, potentially saving resour-ces and time.

• If deaths are recorded for specificparts of the recall period, it is possibleto look at sub-intervals of time (e.g.,before and after displacement) or toexamine monthly trends (dependingon sample size).

Disadvantages of a longer recall period

• Mortality rate may be less relevant tocurrent needs than a mortality ratecalculated for a more recent timeperiod.

• Additional information, such as causeof death, becomes increasingly unreliable as the recall period lengthens.

period, separate mortality rates can becalculated for both the period before andthe period after such a date. For exam-ple, some surveys done in displacedpopulations have asked if a specific

death in the household occurred beforeor after the household left its home villa-ge. Separate mortality rates could thenbe calculated for the pre-displacementand post-displacement periods.

44

How long should the recall period be?

There is no absolutely “correct” length for a recall period for surveys measuringmortality rates. The recall period should be based on the objectives of the surveyand the following factors:

• Accuracy: the recall period should be short enough to allow accurate recall ofinformation about the death. For most purposes, a recall period greater than oneyear probably will result in less accuracy.

• Statistical precision: the recall period should be sufficiently long to provide enoughperson-time units to obtain the desired precision around the estimate of the morta-lity rate. For sample sizes used in many surveys, such as 1,000 households, a recallperiod substantially less than 90 days produces relatively poor precision.

• Recent changes in mortality rates: if mortality rates are changing rapidly, you maynot be interested in the average rate over the last year, but rather the average rateover the prior few months. The population should also have a relatively constantmortality rate during the recall period. This may have to be assumed if no informa-tion is available.

• Seasonality in mortality: if you are trying to measure the impacts on mortality of fac-tors not determined by season, the recall period should be chosen to cover severalseasons so these effects can be mitigated.

• Logistic considerations: longer recall periods reduce the number of householdswhich need to be included in the survey sample and therefore the time needed tocomplete the survey.

During the acute phase of an emergency, it may be advisable to use a short recallperiod, such as 1-3 months, because you may be most interested in the mortalityrate for a very recent time period. When measuring mortality in stable populationswith less fluctuation in the mortality rate, much longer recall periods (such as oneor more years) can be used.

Mortality interview To estimate a mortality rate from a sur-vey we need to know (a) the number ofpeople at risk, and (b) the length of timeover which they were at risk. However,the composition of some of the house-holds will have changed during the recallperiod (death, birth, migration into andout of the household). As a result, thenumber of people within each of thehouseholds may not have been constantduring the recall period.

Figure 2.2 demonstrates this concept gra-phically. Time moves from left to right,with the vertical line on the left being thebeginning of the recall period and thevertical line on the right being the end ofthe recall period - usually the date of datacollection at each household. The asses-sment of mortality will count only thosedeaths which occurred during the recallperiod (those deaths occurring betweenthe vertical lines). Each horizontal lineillustrates a household member. The topline shows a household member wholived in the household at the beginningof the recall period and still lived therewhen the survey team visited the house-hold. The other horizontal lines repre-sent other household members whoentered or left the household during the

recall period by various means, includingbirth and death. The dotted horizontallines represent household members whodied during the recall period; these per-sons would be counted in the numeratorof the mortality rate. The householdshown in Figure 2.2 had three membersat the beginning of the recall period andalso had three members at the end of therecall period; however, only one personwas in the household during the entireinterval. At one time in the middle of therecall period, the household had sixmembers.

Two main methods have been used tocount the number of people in a house-hold in order to calculate a denominatorfor mortality rates: the past householdcensus method and the current householdcensus method. For both methods, a hou-sehold census is taken, whereby a list ismade of all the people living in the house-hold. In the past household method, thecensus is done as of the beginning of therecall period. Interviewers might pose aquestion such as, “At the time of {name ofholiday or event}, who lived in this hou-sehold?” In the current household censusmethod, the census is done as of the timeof the interview; the question often posedis, “Who lives in this household now?”



45

2CHAPTER

Beginning ofrecall period

Time End of recall period(Usually when survey data collected)

HH member

Moved into HHduring recall period

Birth during recall period

Moved out of HHduring recall period

Death during recall period

Birth and death duringrecall period

Figure 2.2. Household members' experience during the recall period

In this manual we recommend a modifica-tion of the current household censusmethod. Essentially, a census is done atboth the end and the beginning of therecall period. In order to calculate thedenominator you need to: 1. Ask the household respondent to list all

the household members at the time ofthe survey (the end of the recall period).

2. Ask the household respondent if eachof these household members werepresent at the start of the recall period.

3. Add to the household list all the mem-bers of the household who were presentat the start of the recall period but arenot currently present in the household.

4. Ask the household respondent the cur-rent status of each of the members whowere in the household at the beginningof the recall period but are no longer inthe household. Status may include alivebut living elsewhere, dead or unknown.

5. Ask the household respondent if eachperson on the household list is youngeror older than 5 years of age. Thisallows calculation of an age-specificmortality rate for children under 5years of age.

6. Ask the household respondent if anybabies were born during the recall peri-od and where these newborns are now.

The interviewer also can ask for additionalinformation if other indices of mortalityare to be calculated:1. The age of each member. These data

confirm whether an individual is aboveor below 5 years of age and allow ademographic pyramid of the popula-tion to be constructed. In addition,other age-specific mortality rates couldbe calculated, such as those shownabove for Badghis Province.

2. The sex of each member. These dataallow calculation of separate mortalityrates for males and females.

3. The date of each death if mortalityrates are to be calculated for sub-periods within the recall period.

4. The cause of death if cause-specificmortality rates or proportional causesof death are to be calculated.

These data are collected on a form, usinga separate sheet for each household.Figure 2.3 shows an example of the form.

46



47

2CHAPTER

Figure 2.3 Example of household mortality data collection form

1ID

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

2Sex

F

M

M

F

F

F

M

M

F

M

M

F

M

M

M

3Current

age (in years)

23

26

54

48

18

12

8

2 mos

4

1

3

29

33

8

31

Survey district: Ambo Village: Limbo Cluster number: 4

HH number: 23 Date: 12- Aug - 04 Team number: 2

4Present

now

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

NO

NO

NO

NO

5Present at

beginning of recallperiod

Yes

Yes

Yes

Yes

NO

Yes

Yes

NO

Yes

Yes

NO

Yes

Yes

Yes

Yes

6Current status(1=Alive;2=Dead;

3=Unknown)

1

1

1

1

1

1

1

1

1

1

1

2

1

1

3

Tally (if data on household members will not entired into the computer)

a. Number of HH members at end of recall period

b. Number of children < 5 years at end of recall period

c. Number of HH members at beginning of recall period

d. Number of children < 5 years at beginning of recall period

e. Total number of deaths

Numberabove

11

4

12

2

1

Data comefrom:

Column 4

Columns 3 & 4

Column 5

Columns 3 & 5

Column 6

Sometimes survey managers may be tem-pted to ask each household respondentonly the number of people in the house-hold rather than listing each householdmember. Although this may be faster, it isfar less accurate than asking the respon-dent to list all the household members. Wetherefore strongly recommend that thehousehold members are listed on a formsuch as that above.

The numerator of the crude mortality ratein the survey sample is simply the totalnumber of all deaths counted in the surveysample during the recall period - that is,the sum of all the numbers in row “e” forall the households selected for the surveysample. The population denominator ofthe mortality rate is the average of the totalpopulation in selected households at thebeginning of the recall period (the sum ofrow “c” for all households) and the totalpopulation in selected households at theend of the recall period (the sum of row

“a” for all households). This averagepopulation is then multiplied by thenumber of time units in the recall periodto derive the person-time denominatorfor the mortality rate. The time andpopulation constants are then applied toconvert the mortality rate to a standardform which can be compared to othermortality rates.

For under-five mortality, a comparableprocess is followed. The numerator of theunder-five mortality rate is the number ofdeaths in children under 5 years of agewhich occurred in selected households, orthe total of row “f” for all households. Thedenominator for this mortality rate is theaverage of the number of children under 5years of age at the beginning and end ofthe recall period - that is, the average ofrows “d” and “b.” This mid-interval popu-lation of children under 5 years of age ismultiplied by the number of time units inthe recall period, as described above.

48

What to do with people whose status in the household is reported as unknown?

If such persons represent relatively few people who are no longer in the households,they can be excluded from both the numerator and denominator.

If they represent a large proportion of household members who are no longer in the hou-sehold, two mortality rates can be calculated: one assuming that the unknown membersare alive and one assuming that the unknown members are dead. Some probing with mem-bers of the community may also give you an idea whether those whose status is unknownare more likely to be dead or alive but living elsewhere. For example, in a population wheremany young men fled the household to join a fighting group which has not lost many mem-bers in battle, heads of households may report their status as unknown, but they may belikely to be alive. In other situations where attackers take household members away andthere have been mass executions, such household members may be more likely to be dead.

In any situation where the number of persons with unknown status is larger than 10 per-cent of the number of deaths, calculate two mortality rates (one rate excluding unknownpersons from the numerator, and therefore assuming that they are still alive, and thesecond rate including unknown persons in the numerator and therefore assuming thatthey are dead).

CAUSE OF MORTALITY One of the survey objectives may be todetermine the major causes of death dur-ing the recall period. During a cross-sec-tional survey, this is done by asking ques-tions of a surviving household memberwho was present just before and duringthe death of their relative. The questionsare meant to elicit a description of thesigns and symptoms experienced by theperson who died in order to determinewhat illness caused the death. Thisprocess is notoriously difficult even withan extended interview of a closely relatedperson very soon after death. Lists of vali-dated questions exist; however, the inter-view is much too long for most emergencyassessments and requires highly skilledinterviewers. More abbreviated lists ofquestions have been used in some sur-veys, but such lists have not been proper-ly validated.

Nonetheless, some causes of death may bedistinct enough to diagnose with fewerquestions. For example, malnutrition maybe suspected as a cause of death if thepopulation being surveyed has experi-enced famine and food insecurity, if thesurviving relative reports lack of food inthe household and if the deceased personhad rapid weight loss in the few weeksbefore death. In addition, some diseasesare well known by mothers and others inthe society. Many cultures and languageshave specific terms for measles, neonataltetanus and other common illnesses withrelatively unique appearances.

In many situations of conflict, it may be ofinterest to determine if deaths have beencaused by war-related violence. Becausesuch violence is easily identified by laypeople, a short series of questions, such asthose below, can often accurately identifydeaths due to this cause.

Questions to detect war-related violence:

1) “Did (the person) die from some sort ofinjury such as being assaulted, shot orstabbed, a car accident, a fall, drown-ing, poisoning, burn, bite or sting?”If YES, go to next question. If NO, record death as not related toinjury or violence.

2) “Was this injury caused by someonefighting the war such as from a bullet,bomb, mine, machete or assault?” If NO, record non-war-related injury orviolence as cause of death. If YES, record war-related injury or vio-lence as cause of death.

War-related violence normally affectsonly particular areas within the wholesurvey area. As a result, if war-relatedviolence was a major cause of death dur-ing the recall period, the sample size mayhave to be substantially larger to meas-ure mortality rates with any precision(see chapter on surveys for more detailedexplanation). Nevertheless, determiningthe contribution of war-related violenceto overall mortality may be important inmany situations.

In general, because of the complexity ofdetermining the causes of deaths in cross-sectional surveys through interviews,alternate sources of data should be usedto determine the causes of death. Suchsources could be disease surveillance,death registration, clinic or hospital logbooks, or others. If other sources of dataon causes of deaths do not exist, suchinformation could be collected duringsurveys if the following conditions exist:

• If local terms exist for causes of inter-est, and respondents can reliably iden-tify them.

• If the causes of interest consist only ofviolence.



49

2CHAPTER

50

Example 2.3 Calculation of mortality rates from a cross-sectional survey

The data for a cluster sample survey were gathered between June 11 and 17. The sam-ple contained 387 households. At the end of Ramadan, the November 14, before the sur-vey, 2,818 people lived in the households selected for the survey sample, of which 535were children under 5 years of age. At the end of the recall period (i.e., at the time of sur-vey data collection), the population of these selected households was 2,827, of whom 578were children under 5 years of age. During the recall period, 44 people died; 31 of thesedeaths were children under 5 years of age. Seventeen adults had left the householdduring the recall period and were living elsewhere; 4 adults were no longer in the house-hold, and their current status was unknown.

The first step in calculating any mortality rate is to determine the length of the recallperiod. The mid-point of data collection fell on June 14; this date can be used as the endof the recall period. There are 213 days between November 14 and June 14. These 213days are equivalent to 7 months (November 14 - June 14).

Crude mortality rate (in deaths per 10,000 population per day)

The denominator equals the average population size:

(2,818 + 2,827) = 2,822.52

multiplied by the length of the recall period:

2,822.5 persons x 213 days = 601,192.5 person-days

The numerator equals the 44 deaths reported in all ages during the recall period. Therefore,the mortality rate equals:

44 deaths x 10,000 = 0.73 deaths per 10,000 population per day601,192.5 person-days

Crude mortality rate (in deaths per 1,000 population per month)

If the CMR is to be expressed as deaths per 1,000 per month, the recall period would be 7months. Therefore, the denominator would be:

2,822.5 persons x 7 months = 19,757.5 person-months

and the mortality rate would be:

44 deaths x 1,000 = 2.22 deaths per 1,000 population per month19,757.5 person-months



51

2CHAPTER

Age-specific mortality rate for children under 5 years of age (in deaths per 10,000 population per day)

The denominator equals the average size of the population of children under 5 years of age:

(535 + 578) = 556.52

multiplied by the length of the recall period:

556.5 persons x 213 days = 118,534.5 person-days

The numerator equals the 31 deaths reported in children under 5 years of age during therecall period. Therefore, the mortality rate equals:

31 deaths x 10,000 = 2.62 deaths per 10,000 population per day118.534.5 person-days

Age-specific mortality rate for children under 5 years of age (in deaths per 1,000 population per month)

If the age-specific mortality rate for children under 5 years of age is to be expressed as deaths per 1,000 per month, the recall period would be 7 months. Therefore, the denominator would be:

556.5 persons x 7 months = 3895.5 person-months

and the mortality rate would be:

31 deaths x 1,000 = 7.96 deaths per 1,000 population per month3895.5 person-months

REFERENCES

Measuring Mortality, Nutritional Status and Food Security in Crisis Situations: SmartMethodology, Version 1, June 2005. UNICEF and USAID, New York and Washington, D.C.2005. Available at: http://www.smartindicators.org/.

Spiegel PB, Sheik M, Woodruff BA, Burnham G. “The accuracy of mortality reporting indisplaced persons camps during the post-emergency phase”. Disasters 2001;25:172-180.

Myatt M, Taylor A, Robinson WC. “A method for measuring mortality rates using pre-vious birth history”. Field Exchange 2002;14:13-15. Available at: http://www.ennonline.net/fex/17/index.html

Woodruff BA. “A method for measuring mortality rates using previous birth history:a commentary”. Field Exchange 2002;14:16. Available at: http://www.ennonline.net/fex/17/index.html

52

Carrying out a survey requires manyskills, much patience and a substantialinvestment of time but, when well con-ducted, the value to WFP will far out-weigh the costs. This section summarizessome of the many steps involved andprovides guidance on how each of thesesteps may be completed. It also high-lights, for many of the steps described,some of the more common mistakes andhow to avoid them. Additional guidance,and more thorough discussions of thetheoretical aspects of surveys, can beobtained from the references listed else-where in this document.

Key steps in carrying out a surveyinclude the following:1. Decide whether a survey is neces-

sary. If a survey is deemed necessary,then the following steps would apply.

2. Define survey objectives and definethe geographic target area to beincluded in the survey.

3. Determine what information to col-lect and from which population sub-group it will be collected.

4. Gather background information nec-essary for carrying out the survey andmeet with community leaders, localauthorities and organizations thatmay be concerned with nutrition.

5. Determine the appropriate samplingmethod.

6. Calculate sample size.7. Select the sample.8. Determine the schedule for the sur-

vey, and obtain and prepare suppliesand equipment.

9. Design the data collection form andsurveyor's manual.

10. Select and train survey workers.11. Field test data collection forms and

data collection procedures.12. Collect the data.13. Enter the data and assess its quality.14. Analyse the data and present the

results. 15. Disseminate findings in presenta-

tions and reports.

Although listed roughly in order of theirexecution, some of the tasks will have tobe done simultaneously or out of orderin some circumstances and may need fre-quent revisiting. No manual or set ofinstructions can possibly instruct anovice survey manager in all the poten-tial technical and logistic complications.For this reason, the advice of an experi-enced person should be sought at manypoints in the process of planning and car-rying out a survey.


Designing a Survey

3CHAPTER

Key messages• How to determine whether a survey is necessary• What information is needed to prepare for a survey• How to choose appropriate sampling methodology and calculate sample size• How to select the survey sample• How to train the survey team and develop survey materials• How to collect survey data and assess its quality• Presenting the results

1. DECIDE WHETHER A SURVEYIS NECESSARYThe decision to undertake a nutrition sur-vey usually will be made with local ornational government authorities and otherpartner agencies. The actual decision-mak-ing process will depend on local circum-stances and relationships. It is, however,always important to share informationabout when and where you plan to under-take a nutrition survey to prevent unneces-sary duplication and overlap of surveys.

Conducting a nutrition survey is expensiveand time-consuming. Before starting a sur-vey, you should answer the followingquestions:• Are the results crucial for decision-

making?If a population's needs are obvious,immediate programme implementationis the first priority. For example, if therehas been a natural disaster, such as anearthquake or landslide, and it is clearthat the population's main food sourcehas been destroyed, it may not be nec-essary to undertake a nutrition surveybefore distributing food. Similarly, ifanother agency has recently7 carriedout a nutrition survey in the same area,and the context has not changed con-siderably, then it should not be neces-sary to repeat the process. However, ifcurrent population needs must bedetermined or if WFP requires out-come or baseline data for a specificproject, then it may be necessary tocarry out a survey. Thus, the imple-mentation of WFP programmes may besufficient justification for carrying outa nutrition survey if data are unavail-able from other sources.

• Are the survey results going to be usedto take action?There is no point in undertaking anutrition survey if you know that aresponse will not be possible. If there isno capacity to implement an interven-tion to address the identified needs, orif the survey objectives do not addressWFP Strategic Priorities, furtherthought should be given to the useful-ness of initiating a survey.

• Is the affected population accessible?Insecurity or geographical constraintsmay limit access to the population ofinterest. If this is extreme, a surveycannot be conducted.

Unless these three prerequisites are ful-filled, there is probably insufficient rea-son to undertake a survey; other data col-lection techniques may be more appro-priate and efficient. Such data collectiontechniques may include nutrition anddisease surveillance, food market moni-toring, qualitative assessment of foodsupplies and household food economy,screening of all persons in the vulnerablepopulation and other techniques. In gen-eral, because surveys gather data at asingle point in time, a single survey has avery limited ability to explore causality.To more completely understand nutri-tional problems in a population, addi-tional information about the causes ofmalnutrition, which cannot be collectedin a survey, must be obtained by othermeans, such as qualitative assessmenttechniques.

Surveillance data are often especiallyuseful to answer public health or nutri-tion questions. In many populations,

54

7 The time involved in “recently” varies with the context and the use of the survey results. In general, survey resultsfrom the last year are useful; however, attention must be paid to changes in the target area.


DESIGNING A SURVEY

55

3CHAPTER

information on the number of cases ofcommunicable diseases is collected rou-tinely from health care providers, such asclinic and hospital personnel, both publicand private. Nutrition surveillance is alsooften carried out by reporting anthro-pometry from clinics where growth mon-itoring is done to central public healthauthorities. Of course, because most sur-veillance data are collected from healthfacilities, they are relevant to people whocome to clinics for curative or preventivecare. Often such people are quite differ-ent from those who do not come tohealth care workers for services. As aresult, surveillance data rarely are trulyrepresentative of the entire population.As will be described in this chapter, sur-veys can provide just such representativeinformation if carried out correctly.

Because surveillance systems gather datacontinuously over time, such data areespecially well suited for following trendsin the incidence or prevalence of the dis-ease reported in the system. For example,if the surveillance system has not under-gone major changes in a period of fiveyears, it can show whether the number ofcases of a specific disease has increased,decreased or remained the same duringthis time period. In general, because sur-veys gather data at a single point in time,data from a single survey cannot be usedto follow trends. Moreover, because sur-veys collect data on possible contributingfactors at the same time they collect dataon the disease of interest, you often cannottell which came first, the risk factor or thedisease. For this reason, surveys usuallyare not the best way to explore causality.To more completely understand nutritionalproblems in a population, additional infor-mation about the causes of malnutritionshould be obtained by other means, suchas qualitative assessment techniques.

2. DEFINE SURVEY OBJECTIVES ANDDEFINE THE GEOGRAPHIC TARGET AREATO BE INCLUDED IN THE SURVEYYou must be clear about your objectivesbefore starting a nutrition survey. Preciseand clear objectives will make it mucheasier for your team, collaborating organ-izations, the survey population anddonors to understand what you are tryingachieve. Clear objectives will also makethe analysis of survey data much easierby guiding which analyses will answerthe basic questions that prompted thesurvey. Often, a nutrition survey isundertaken to fulfil the following types ofobjectives:

• To measure the proportion of individu-als in specific groups, often children 6-59 months of age or women of child-bearing age, with malnutrition oranaemia. Such data can be used todetermine the need for specific nutri-tion interventions.

• To measure the coverage of feedingprograms - that is, the proportion ofindividuals in specific groups whoare beneficiaries of a nutrition pro-gram - for example, the proportion ofmoderately malnourished childrenwho receive food from a supplemen-tary feeding program or the propor-tion of families who have receivedfood from the general ration distribu-tion. These results can measurewhether a program is reaching theintended beneficiaries.

• To establish a baseline against whichchanges in nutritional status over timecan be compared by carrying out a fol-low-up survey. Although such data aresometimes used to measure the effec-tiveness of a program, there may bemany factors that can influence changeover time in addition to the implemen-tation of a specific nutrition program.

Serial surveys meant to demonstratechanges in nutritional status should beplanned carefully. For example, baselinesand follow-up surveys meant to detectchanges over time could be done in thesame population and at the same time ofyear in order to minimize the differencesbetween populations and the effect of sea-sonal changes in nutritional status. Inaddition, an apparent difference betweensurveys may be the result only of samplingerror, described later in this manual. Youshould not conclude that a statistically sig-nificant difference exists between theresults of two surveys without the use of astatistical test. In addition, surveys overtime in an intervention population (thatwhich receives a programme) and a non-intervention population (that which doesnot receive a programme) may provideevidence that a particular program hascaused a change in nutritional status. Theintervention population is the populationthat receives a program, and the noninter-vention group is similar population thatdoes not receive the program.

Undertaking a nutrition survey providesan ideal opportunity for agencies to learnmore about the population they are assist-ing or planning to assist. When undertak-ing a nutrition assessment, it may be use-ful to collect additional information on thepopulation, such as mortality, immuniza-tion and nutrition programme coveragedata. Objectives related to these healthoutcomes should be included in the list ofsurvey objectives. Such objectives couldinclude• Estimating the coverage of feeding pro-

grams;• Estimating mortality rates (crude and

under-5 years); and• Estimating the coverage of measles

vaccination and vitamin A supplemen-tation.

You must decide in which area the surveyshould be conducted, and you must definethis area carefully. Remember that the areafrom which the survey sample is selectedis the only area to which your surveyresults can be generalized. You will not beable to say anything about an area or pop-ulation that was not eligible to be selectedfor the survey sample.

In most cases, the area chosen will corre-spond to an administrative or governmentarea, such as a district or province. ForWFP purposes, sometimes the area chosenfor a survey will correspond to areaswhere WFP is implementing a programmeor is investigating the need to implement aprogramme. The survey ideally should beconducted in an area where the wholepopulation has a similar nutritional situa-tion because the final estimate of theprevalence of malnutrition can be appliedonly to the entire population from whichthe sample was selected. It cannot beapplied with any statistical precision toany subgroup or sub-area within the sur-vey population or area. If you conduct asurvey in an area in which different groupshave very different nutritional statuses, theresults will be an average of these differentnutritional statuses. For example, if someprovinces in a particular nation have suf-fered drought and crop failure while othershave not, it would not be very useful to doa nationwide survey to determine whichprovinces are in greatest need of assis-tance. The result of such a survey wouldproduce only an average measure for theentire nation.

56


DESIGNING A SURVEY

57

3CHAPTER

COMMON MISTAKE #1:Selecting the area to be assessed on the basis of accessibility or the presence of aid organizations, not on the need for a survey.

Explanation: Nutrition surveys should be done to answer specific programmaticquestions, such as “Does this population need feeding programmes?” or “Whattype of feeding programmes are needed by this population?” or “When we measu-re nutritional status in one year's time, will there be an improvement in compari-son to this baseline?” Therefore, the selection of the area to assess by carrying outa survey should contain a population about which someone needs answers to spe-cific questions. A population should not be chosen because they are easier to reachthan other populations (e.g., closer to the capital city or on a better road). Surveysdone in more easily accessible populations may be biased because populationswhich are easier to reach may be substantially better off than populations in moreremote or less secure locations. As a result of choosing the easier population,important resources of food, money and personnel may be directed to a populationin which the need for interventions is not as great as in other populations.

Solution: Be sure the area you select for a survey truly needs a survey; that is,be sure that the area has a potential need and that the capacity exists to takeaction on the basis of the survey results.

STOP

GO

COMMON MISTAKE #2:Always selecting the area to be assessed on the basis of administrative delineations, not on environmental or other characteristics more likely to determine risk of malnutrition.

Explanation: The decision regarding which geographic area to include in a surveyshould be based on (1) who needs the survey results, and (2) what decisions needto be made. For example, if knowledge of the prevalence of malnutrition in a pro-vince is required to decide whether to request food assistance, the selection of sur-vey sites should be made from all communities in the province, regardless of theprovince's geologic or ecologic boundaries. On the other hand, if there is strongsuspicion that, for example, the people living in the mountains have a greater shor-tage of food, then it may be best to target the mountainous districts for the surveyand exclude districts in the lowlands. Of course, if the survey is done to measurethe outcomes of an intervention, the population to be included in the surveyshould be potential recipients of that intervention. In general, decisions regardingall aspects of survey design should be determined by the questions the survey ismeant to answer and the use to which the results will be put.

Solution: Define the area to be included in the survey according to who needsthe information and what will be done with the information.

STOP

GO

3. DETERMINE WHAT INFORMATION TO COLLECT AND FROM WHICH POPULATION SUBGROUP IT WILL BECOLLECTEDIn large part, what information you willcollect in the survey will be decided whenformulating the survey objectives.However, additional information may beincluded in the list of major objectives.When making the list of data to collect,remember that data collection requirestime and money. Survey workers shouldnot be overburdened by collecting informa-tion which will not be analysed or used toimprove the health of the population. Also,remember to collect all the data necessaryto meet the survey objectives. Some typesof information may be necessary onlyduring data analysis. For example, if youare assessing the nutritional status of adultwomen by using body mass index (BMI),remember to determine pregnancy statusfor each woman included in the surveysample because pregnant women will haveto be excluded during the analysis of BMI.If cluster sampling is done, you will alsoneed to know how to identify which clu-ster each household or individual belongsto in order to calculate appropriate confi-dence intervals. Therefore, you mustrecord in each data set the cluster numberfor each household, child, woman and anyother unit of analysis.

At this point you will also need to decidewhat age group to measure. For manynutrition and health outcomes, the sub-groups selected to measure are the groupsat greatest risk of the nutritional deficiency.For example, anthropometric surveys ofacute and chronic malnutrition usually tar-get children aged 6-59 months. Younginfants may be included if you suspect thatthere is an acute nutritional problem in thisgroup. Young children and women of chil-dbearing age may be included because

these groups are at greatest risk for anae-mia, and adolescents, especially adolescentgirls, may also be included if there is suspi-cion that this age group is at particular riskfor anaemia or other nutrition conditions.If you are in doubt about which group toinclude in measurement of some health ornutrition outcome, consult an expert inthat disease or condition.

In order to organize the data, you couldmake a complete list of all the data youwish to collect during the survey. The par-tial list below details the data needed toassess some health and nutrition outcomesin children 6-59 months of age:

For children 6-59 months of age currentlyin household:A. Village name, cluster number, household

numberB. Date of birth, age in months, or bothC. SexD. Vaccination status

1. Tuberculosis (BCG) vaccine-by scar, mother's history or vaccination card

2. Measles- by mother's history or vaccination card

E. Recent morbidity1. Diarrheal disease2. Acute respiratory infection

F. Nutritional status1. Malnutrition

a. Weightb. Height or lengthc. Presence of bilateral oedema

2. Anaemiaa. Hemoglobin

3. Vitamin Aa. Presence of night blindness b. Serum vitamin A level c. Coverage of Vitamin A capsules

4. Enrolled inSupplementary/Therapeutic Feeding Program, currently or recently

58


DESIGNING A SURVEY

59

3CHAPTER

Such a list should be made for each tar-get subgroup of the population. Creatingsuch a list:• Helps to ensure that nothing essential is

forgotten; • Identifies the specific equipment and

supplies which will be needed for the survey;

• Allows others to review what data willbe collected in the survey; and

• Organizes the data in the order it may beincluded on the data collection form.

COMMON MISTAKE #3:Using mid-upper arm circumference (MUAC) as the only anthropometric indexof acute malnutrition.

Explanation: Some survey managers feel that measuring weight and height in thefield using scales and height boards is too complex, while measuring MUAC requi-res only a MUAC tape. However, MUAC is not the best index for use in nutritionassessment surveys for the following reasons:

• The size of a child's upper arm increases as the child gets older. Therefore, if thesame cut-off point (often <12.5 cm) is used to define malnutrition for all ages ofchildren 12-59 months of age, a younger child is more likely to be defined as mal-nourished than an older child, regardless of their actual nutritional status.

• Taking a MUAC measurement is difficult to do accurately. If the measurer pullsthe tape too tightly, the measurement will be spuriously low. If the measurer doesnot pull the tape tightly enough, the measurement is spuriously high. Several stu-dies demonstrate that inter-observer and intra-observer error are both quite highwith MUAC measurements. Moreover, if all measurers pull too tightly, then the sur-vey will overestimate the prevalence of acute malnutrition. Likewise, if the tape isconsistently not pulled tightly enough, the survey will underestimate the prevalen-ce of acute malnutrition.

• MUAC measures only acute malnutrition. As a result, MUAC cannot tell you ifchildren are chronically malnourished. In some populations, chronic malnutritionmay be more important than acute malnutrition.MUAC may be used to screen large numbers of children to determine who needsadmission to supplemental or therapeutic feeding. However, such data should notbe used to estimate the prevalence of malnutrition in the population.

Solution: Use weight-for-height instead of MUAC to measure acute malnutrition.

STOP

GO

60

COMMON MISTAKE #4:Always excluding infants under 6 months of age.

Explanation: Infants under 6 months of age often are not included in nutrition sur-veys for various reasons.

First, in most populations, infants under 6 months of age are considered less vul-nerable to acute malnutrition. Such children are relatively protected from foodshortages because typically most of their diet consists of breast milk. Unless themother is very ill or very malnourished, the supply of breast milk is usually suffi-cient for young infants; therefore, food shortage is less common in breastfeedinginfants. Nonetheless, there may be situations where maternal health or nutritionalstatus is so poor that some women may not be able to adequately breast feed theiryoung infants, or young infants may have substantial exposure to communicablediseases which lead to malnutrition. There may be other situations where the sur-vey is meant to evaluate specific programs targeting young infants.. In such cases,it may be desirable to include children under 6 months of age in the survey sam-ple. However, it should be kept in mind that the number of infants under 6 monthsof age included in most population-based household surveys will be relativelysmall. As a result, a separate estimate of the prevalence of malnutrition for just thisgroup will have very little precision.

Second, many survey workers are less comfortable handling such small children.Survey workers should be reassured that weighing and measuring young infants isnot difficult, especially if using the UniScale or another scale which allows wei-ghing of children in the mother's arms. The decision to include infants under 6months of age should be based upon the need to assess their nutritional status, noton perceived difficulties with their inclusion in the survey sample.

Solution: Include infants under 6 months of age in the survey target popu-lation if there is reason to believe that they are unusually vulnerable tomalnutrition.

STOP

GO

GATHER BACKGROUND INFORMATIONNECESSARY FOR CARRYING OUT THESURVEY AND MEET WITH COMMUNITYLEADERS, LOCAL AUTHORITIES ANDORGANIZATIONS THAT MAY BE CONCERNED WITH NUTRITIONMany types of information are necessaryto plan a nutrition survey, and a good vari-ety of it can be garnered from discussionswith community leaders and health careworkers, from reports of prior assess-ments, or from other sources. The logisticsof planning a survey often can be a moretime-consuming process than it wouldappear. It is very important that this plan-ning phase is comprehensive and welldeveloped: the positive effects will beapparent once the survey is underway. Tobegin designing the survey, it is advisableto gather the following information:

1) To calculate the sample size, it is nec-essary to obtain either data on themost current estimate of the preva-lence of malnutrition or informationthat can be used to estimate thisprevalence.

2) To determine which samplingmethod to use, it is necessary toobtain information about what dataare available at the local, regionaland national levels.

3) To carry out the first stage of multi-stage cluster sampling, it is oftennecessary to identify local geograph-ic units with corresponding popula-tion size.

4) To calculate the number of householdsto be selected in order to obtain dataon the required number of childrenand women, the average number ofwomen and children per household isnecessary.

5) To construct a local calendar used toestimate children's ages in populationswhere date of birth is frequently not

known, it may be necessary to obtaininformation on seasonal and historicalevents.

6) To establish referral systems for severemalnutrition, information on nutritionprograms and health infrastructure willbe needed.

7) And to aid in planning everything fromthe survey teams' work to the timeschedule to the geographic areasselected for the survey sample, mapsare of course necessary.

It is absolutely essential to meet with com-munity leaders and local authorities dur-ing the survey planning process and datacollection. During your visit you should:

1) Make sure the community fully under-stands the objectives of the survey. Ifthe population does not understandwhy you are doing a survey, they maynot cooperate with data collection.

2) Agree on the dates of the actual sur-vey with the community and localauthorities to avoid conflict withother community activities, such asmajor holidays, market days or fooddistribution days.

3) Obtain information on security andaccess in the survey area.

4) Obtain letters of permission from thelocal authorities, addressed to the dis-trict or village leaders, stating thatyou will be visiting. The lettersshould explain why you are conduct-ing a survey and ask for the popula-tion's cooperation.

Of course, national and local governmentofficials should be consulted regarding thenecessary permits and clearances. Ethicalreview by an institutional review boardmay be required for survey procedures.Some countries require permits forresearch and may define any survey data


DESIGNING A SURVEY

61

3CHAPTER

collection as research. In addition, specificexport permits may be required to removebiologic specimens from the country if lab-oratory testing is to be done elsewhere(such as in the case of testing for somemicronutrient deficiencies).

5. DETERMINE THE APPROPRIATE SAMPLING METHODAlthough there are many ways to choosesamples, the methods broadly fall intoone of two categories: non-random sam-pling and random sampling. Non-ran-dom sampling does not guarantee a sam-ple representative of the larger popula-tions and is rarely used in epidemiologicstudies or surveys. For these reasons, itwill not be further described in this man-ual. In contrast, random sampling veryfrequently is used to select samples forsurveys. Random sampling is defined assampling where (1) the selection of eachunit is done at random, and (2) the like-lihood that any one unit is selected isfinite and calculable. This means that, ifyou are selecting a random sample ofhouseholds from a population, you couldcalculate the probability that any specifichousehold in that population will be cho-sen for the sample.

DefinitionsTo discuss the various sampling methods,some terminology must be defined:

• Sampling universe - the entire group ofsampling units (commonly householdsor persons) who are eligible to beincluded in the survey sample. Thispopulation should match the popula-tion for which you are trying to esti-mate the outcomes measured in thesurvey. For example, if a survey is to bedone in a province in order to estimatewhat proportion of households have asafe water supply, the sampling uni-

verse would be every household resid-ing in that province on the day sam-pling is done.

• Sampling frame - the list of all the sam-pling units from which you will chooseyour sample. The sampling frameshould match the sampling universe asclosely as possible; however, in somesurveys, it does not match completely.For example, if a telephone surveywants to determine what proportion ofadults have arthritis, the sample maybe selected from persons listed in direc-tory assistance. As a result, an adultwho does not have a telephone cannotbe included in the sample, even thoughthe sampling universe includes alladults in the population.

• Sampling unit - the unit that is selectedduring the process of sampling. If youselect households from a list of allhouseholds in the population, the sam-pling unit is the household. If you areselecting districts in the first stage ofcluster sampling, the sampling unit atthe first sampling stage is district.

• Basic sampling unit or elementary unitthe sampling unit at the last stage ofsampling. In a multi-stage cluster sur-vey, if you select first villages, thenselect households within selected vil-lages, the basic sampling unit would behousehold.

• Sampling fraction - the proportion of allsampling units in the sampling framewhich were selected or will be selectedfor the sample. If you want a sample of100 households from a list of 10,000households, the sampling fraction is. 100÷10,000=0.01 Thus, one of every100 households in the sampling frameis selected for the survey sample.

62

• Sampling interval - the inverse of thesampling fraction. This is used whencarrying out systematic random sam-pling. For example, in the exampleabove, the sampling interval would be100. If you are selecting households bywalking down the street and selectingevery 100th household, you wouldcount households until 100 is reachedand include that household in the sam-ple. Then you would continue countinguntil the 200th household is reachedand include that household in the sam-ple, etc.

• Respondent - the person who answersthe questions during an interview.

• Survey subject - the person aboutwhom data are collected in the survey.

If the survey is collecting dietary infor-mation on young children, the childrenare the survey subjects, but the respon-dent is probably the child's mother orother adult caretaker who answers thequestions during the interview.

• Unit of analysis - that unit for which thedata are analysed. In the example abovewhere dietary data are obtained foryoung children, the unit of analysiswould be children. However, if you calculated the proportion of mothers providing adequate complementary feed-ing to their children, the unit of analysiswould be mothers. If you calculated the proportion of householdsin which children were provided withadequate complementary feeding, theunit of analysis would be the household.


DESIGNING A SURVEY

63

3CHAPTER

Example 3.1 Application of terminology to a hypothetical survey

As an example, a hypothetical two-stage survey is conducted in a province. The samplesize for the survey is 500 households out of 100,000 households in the province. Villagesare selected using the PPS methodology (described later) from a list of all possible villa-ges in the province. Households are then randomly selected from a list of all the house-holds in each selected village. In each selected household, the head of the householdreports on how frequently meat is served to the mother of any young children in the hou-sehold, and each child 6-59 months of age is weighed and measured. The elements of thissurvey are as follows:

• The sampling universe for the whole survey is all households in the province at the timeof the survey.

• The sampling frame for the first stage of sampling is the list of all villages in the provincefrom which the sample of villages is chosen. The sampling frame in the second stage ofsampling consists of the list of households in each selected village.

• The sampling unit is the village.• The basic sampling unit is the household.• The sampling fraction for the whole survey is 0.005 (500÷100,000 ). • The sampling interval, if systematic random sampling was used to choose households

directly from a list of all 10,000 households, would be 200.• The respondent for the question about meat is the head of the household.• The survey subject for the question about meat is the mother of a young child.• The unit of analysis when calculating the proportion of mothers who get meat at least

once a week is mothers.• The survey subject for the anthropometric measurements is the child 6-59 months of age.

SELECTING A SAMPLING METHODThe decision about which to make thebasic sampling unit (children or house-holds) depends on what information isavailable before beginning the sampling.For example, some villages or refugeecamps may have lists of children under 5years of age. In such cases, it may beeasier to directly select children toassess. However, in most places, thebasic sampling unit will be the house-hold because households may be listedin registration databases or because hou-seholds are easier to identify and selectafter arrival in a village or camp.

Systematic or Simple RandomSampling (SRS)Different sampling techniques are moreappropriate in different situations. If acomplete list of sampling units (house-holds or children) is available for the

entire population, you can choose a sam-ple by systematic or simple random sam-pling. In SRS, individual basic samplingunits are chosen directly from a list of alleligible basic sampling units. Thesetechniques will be described further.

Systematic random sampling can also bedone on the ground without having a listof all basic sampling units. If the actualshelters are arranged in some order andthe total number of households in thesampling universe is known, then a sam-pling interval (n) can be calculated toobtain a certain sample size and everynth household can be selected by wal-king or driving through the populationand counting houses. Figure 3.1 shows aphotograph of a refugee camp in whichthis technique was used to select a sam-ple, even though there was no paper orcomputer list of all households.

64

Figure 3.1 Cegrane Refugee Camp in Macedonia, May 1999.

Cluster samplingCluster sampling is a sampling methodin which the first sampling step invol-ves selecting collections of persons orhouseholds instead of sampling house-holds or persons directly. If there is nolist of all the sampling units in thepopulation and the houses are not pla-ced in a regular order, you may have todo cluster sampling.

For example, you may choose a sampleof villages in a province as the firststage of sampling because there is nosampling frame for the entire samplinguniverse; that is, there is no list of hou-seholds for the entire province. Onceyou arrive at each selected village, youthen select a sample of householdswithin that village.

Cluster sampling is done very frequently.It allows random sampling in situationswhere simple or systematic random sam-pling of households either is not possibleor is inefficient. However, as describedlater, cluster surveys often need largersample sizes because they are inherentlyless precise than surveys of the samesample size done with simple or syste-matic random sampling.

Another distinct advantage of clustersampling occurs when sampling a wide-ly dispersed population such as ruralhouseholds. In cluster sampling, all thebasic sampling units, such as house-holds, are grouped together into clu-sters, so that the distance between basicsampling units within a cluster can besmall. The only long-distance travel ofthe survey teams is between clusters. Ifsimple or systematic random samplingwere used, each basic sampling unitmight be quite far apart, resulting insubstantial travel to visit each one.

6. CALCULATE SAMPLE SIZESampling error, confidence intervals,precision and P valuesTo discuss sample size calculations, onemust understand some additional con-cepts. Most surveys which measure theprevalence of malnutrition or mortalityrates actually measure the nutritional sta-tus of a sample of children and mortalityin randomly selected households chosenfrom the population of concern. There willalways be some difference between thesurvey results and the population fromwhich the sample was selected. For exam-ple, if many samples are selected of 6 per-sons from a room which contains 10 menand 10 women, samples will sometimescontain 3 men and 3 women, but mightalso contain 4 men and 2 women, 2 menand 4 women and sometimes even 1 manand 5 women or 5 men and 1 woman.Samples which are very different from theentire population are much less likely to berandomly selected than samples which aresimilar to the entire population. The diffe-rence between the result calculated from asample and the true value for that outco-me in the entire population is called sam-pling error. Of course, we are usually doinga survey because we don't know theresults for the entire population, so weusually cannot compare the survey resultsto the population results.

The size of the sampling error can be mea-sured and expressed in many ways. Thestandard error is one such measure ofsampling error; however, the most com-mon method used when presenting surveyresults is confidence intervals (sometimescalled confidence limits). If a survey wererepeated many times with the same sam-pling methods and the same sample size,95 percent of the confidence intervalsaround mortality rates or malnutrition pre-valence calculated from these surveys


DESIGNING A SURVEY

65

3CHAPTER

would include the true population value.Thus, the 95 percent confidence interval isan expression of how certain we are thatthe actual result in the population is simi-lar to that obtained from a survey.

For example, a survey estimates a preva-lence of wasting of 8.7 percent amongyoung children, with a 95 percent confi-dence interval of 7.1 percent - 9.5 per-cent. We are then 95 percent sure that thetrue prevalence of wasting in the popula-tion in which the survey was done is bet-ween 7.1 percent and 9.5 percent. Thereis a small probability that the prevalenceis less than 7.1 percent or greater than 9.5percent. The confidence interval is oftenpresented as an interval, but it can alsobe written as ± X percent. This type ofexpression is often seen in newspaperaccounts of the results of a public opinionsurvey. For example, a report may saythat 68 percent ± 3 percent of peoplefavor a certain policy. This means that theconfidence interval for the estimate of 68percent is 65-71 (68 - 3 through 68 + 3).Note that one must be very careful todetermine if ± component is the confi-dence interval, a standard error or a stan-dard deviation.

Precision, that is, how similar the resultswould be if a survey were repeated overand over, is measured by the samplingerror. If the confidence interval is wide,sampling error may be responsible for asubstantial difference between the resultobtained by a survey and the actual valuein the population. In contrast, if the con-fidence interval is narrow, we know thatif the same methods were used in repea-ted surveys, the results would be quitesimilar. In general, in the absence of bias,the larger the sample, the closer the sur-vey estimate of malnutrition prevalenceor mortality will be to the actual value inthe population. Precision is increased, and

the confidence interval narrowed, with lar-ger sample sizes. This means that, allother factors being equal, the larger thesample size, the narrower the confidenceinterval and the more certain we are thatthe survey result is close to the actualpopulation value. Therefore, statisticallyspeaking, the larger the sample size, themore certain we are that the survey resultis close to the actual population value.However, it takes more time, money, per-sonnel and other resources to measuremore children or ask more householdsabout mortality. As with most things inlife, there are trade-offs. In this case, thetrade-off is between more certainty in theresult and spending more money and timedoing the survey. However, we can decidehow much certainty we want, or converse-ly, how much uncertainty we will be com-fortable with. On that basis, we can thencalculate how many children or house-holds will be included in the survey toobtain the desired level of precision, thussaving resources. Statistical equations andcomputer software can calculate the preci-se sample size needed to achieve a givendegree of sampling error in the results ofthe survey, therefore saving survey mana-gers from having to include too many peo-ple in the survey.

When one is interested in knowing whe-ther there is a statistically significant diffe-rence between two survey estimates, fre-quently a statistical test is applied and a Pvalue calculated. The P value is the proba-bility that the two estimates differ by chan-ce or sampling error. As a general conven-tion, if the P value is <0.05, we wouldstate that the two estimates are statistical-ly significantly different from one another.If the P value is > 0.05, then we wouldstate that there is no statistically significantdifference between the two estimates. TheP value is calculated assuming that there isno bias in the estimates.

66


DESIGNING A SURVEY

67

3CHAPTER

For example, suppose that a baseline sur-vey found a 50 percent prevalence ofanaemia among young children before anintervention, and a follow-up surveyfound 40 percent prevalence in this samegroup after the intervention. It may bepossible that this difference is due only tothe random selection of different childrenfor the two samples from a population inwhich there had been no change in theprevalence of anaemia. In this case theintervention probably had no effect.However, if we know that the P value forthis difference is <0.01, then we knowthat there was less than a 1-in-100 chan-ce that this difference was due only tosampling error. Therefore, we could con-clude with some certainty that, in theabsence of bias, the difference betweenthe two survey results is real and that theprevalence in young children in the enti-re population has declined between thetwo surveys. In general, if the precisionof the two surveys is high (that is, thesample size is sufficiently large), the like-lihood of the difference being due only tosampling error is decreased. The selec-tion below describes how to calculate thesample sizes for two surveys when theintention is to demonstrate some level ofdifference between them.

BiasSampling error is not the only source of dif-ference between a survey's result and theactual population value. As describedabove, sampling error is due to the randomselection of children or households fromthe population. Sampling error cannot beeliminated entirely, but it can be minimizedby selecting a larger sample and its extentcan be calculated based on statistical theo-ry. Unfortunately, bias cannot be calcula-ted. Bias is anything other than samplingerror that causes the survey result to differfrom the actual population prevalence orrate. Examples of bias might include:

1) The survey workers systematicallymeasured the height of each child as 2cm taller than they really are. As aresult, each child's weight-for-heightZ-score would be underestimated by asmall amount, thereby causing anoverestimate of the prevalence ofwasting.

2) Survey respondents misunderstandquestions about mortality in theirhouseholds. They tell survey intervie-wers, for example, that persons wholeft the household are dead, when infact the respondent does not reallyknow whether or not they are dead.This would result in an overestimateof the mortality rate, thus increasingthe difference between the mortalityrates calculated from the surveyresults and the actual population mor-tality rate.

Unlike sampling error, bias cannot be calculated by the computer after data col-lection is finished. Therefore, it is veryimportant to avoid or minimize biasduring sampling and data collection. Youdo this by standardizing measurementtechniques, training survey workers well,writing clear questions to be asked ofsurvey respondents and through othertechniques. Although a full description ofthe many sources of bias is beyond thismanual, survey workers must be verydiligent to avoid bias.

SAMPLE SIZE FORMULASThe sample size is the number of unitsof analysis (for example, children 6-59months of age or women of childbea-ring age) or the number of basic sam-pling units (for example, households)required for the survey. Calculating therequired sample size will allow you toinclude only the necessary number ofchildren, women or households in thesurvey.

Sample size for simple or systematic sampling If simple or systematic sampling is done,sample size calculation depends largely ontwo factors:1) The estimated prevalence of the outco-

me being measured (p in the formulabelow)

2) The precision you desire for your esti-mated prevalence (d in the formulabelow)

The formula used to calculate sample sizefor simple or systematic sampling is:

The values for p and d must be expressedas decimals. For example, if you expect tofind that 10 percent of children have mal-nutrition, then p = 0.10. If the precisiondesired equals ±5 percentage points, thend = 0.05. In this example, the sample sizewould be:

Sample size for mortality estimatesCalculating sample size for mortality mea-surement is essentially the same process.The actual calculation of sample size canbe done using computer programs specifi-cally written to calculate sample size formortality rates (SampleRate is one exam-ple of such a program; it is available athttp://www.brixtonhealth.com/index.html). Alternately, the sample size can be cal-culated as if the mortality rate were a pro-portion or percentage. That is, you makean assumption regarding the mortality ratein the population on interest, just as youwould make an assumption regarding the

prevalence of malnutrition. Then you calcu-late what proportion of the populationwould have died during the recall period toobtain this rate.

For example, you assume that the crude mor-tality is 2 deaths per 10,000 persons per day.This is equivalent to 2 persons dying eachday in a population of 10,000, or 0.02 percentof the population dying each day. If, forexample, you wish to start the recall periodwith a well-known holiday which produces arecall period of 107 days, then 2.14 percent ofthe population would die in 107 days (0.02percent x 107). This is the “prevalence” ofdeath in the population. Therefore, in the for-mula for sample size calculation, “p” equals0.0214 and “1-p” = 0.9786. Next, you mustconvert the desired precision, that is, the con-fidence intervals, into “prevalence.” In theexample above, if you want a 95 percent con-fidence interval of ± 0.5 death per 10,000 perday, you convert 0.5 death per 10,000 per dayinto a “prevalence.”

Using the method described above, 0.5death per 10,000 per day would mean that0.005 percent of the population would dieeach day. During the 107-day recall period,0.535 percent of the population would die(0.005 percent x 107). Therefore, the “d” inthe sample size equation equals 0.00535.The sample size for simple or systematicrandom sampling would then be:

If households will be sampled for the sur-vey, then this number is divided by the ave-rage number of persons per household. Forexample, if the average household size is5.5 persons, then the survey sample shouldcontain at least 511 households (2811 / 5.5).

68

Sample size for cluster samplingFor cluster sampling, another factor must betaken into account when calculating samplesize. A measure called the “design effect”must be included because, for a given samplesize, cluster sampling usually results in lessprecision than simple or systematic randomsampling. Essentially, the design effect tellsyou how much larger the sample size mustbe with cluster sampling to get the same pre-cision as simple or systematic random sam-pling. Therefore, once the sample size is cal-culated as above for simple or systematic ran-dom sampling, it must be multiplied by theassumed design effect to get the sample sizeappropriate for cluster sampling.

If we use the example above, with an assumed malnutrition prevalence of 10percent (p = 0.10), precision of 5 percent(d = 0.05) and a design effect of 1.5(DEFF = 1.5), the equation must be multiplied by 1.5:

HOW TO MAKE ASSUMPTIONS NEEDEDFOR SAMPLE SIZE CALCULATIONOne of the most difficult aspects ofdetermining a sample size is decidingwhat values to include in the formula forprecision, the prevalence of the outcomeand the design effect. Below is some gui-dance for making these assumptions.

Precision. If survey workers want greaterprecision, the sample size must be lar-ger, as described above. To determinejust how much precision you need, youmust consider what question the surveyis designed to answer. If a survey ismeant to determine whether there is a

large problem with malnutrition, thenyou may not need much precision. Onthe other hand, if you will compare thissurvey to a baseline or a follow-up sur-vey, you may want much more precisionto ensure that a difference detected bet-ween the two surveys has statisticalsignificance. Also, if precise subgroupestimates are needed, such as males vs.females, or by age, then a larger samplesize would be required. In fact, desiredprecision and expected prevalence areinterconnected. Therefore, if you expecta high level of malnutrition or mortalityin the population to be surveyed, youwill have to use a larger sample size toachieve a given precision.

However, you often will not need suchhigh precision for common outcomes orindicators. For example, if the prevalen-ce of stunting is 50 percent, there maybe no programmatic decision whichwould change if the prevalence were 45percent or 55 percent, so very narrowconfidence intervals may not be neces-sary; ±10 percentage points may besufficient. On the other hand, rarer out-comes may need greater precision. Ifthe prevalence of wasting is estimatedto be 10 percent, confidence intervals of±10 percentage points would not bevery useful. This would mean that thetrue prevalence of wasting is somewhe-re between 0 percent and 20 percent.The difference here is significant: 0 per-cent is excellent and no additional fee-ding programs are needed, while 20percent is a catastrophe, requiringwidespread food aid and supplementa-ry and therapeutic feeding programmes.To distinguish between these possibili-ties and make programmatic decisions,you will need much greater precision,such as ±3 or ±4 percentage points sothat the confidence interval is 7 percent- 13 percent or 6 percent - 14 percent.


DESIGNING A SURVEY

69

3CHAPTER

DEFF

70

The important point is that there is nostandard precision to be used, nor any wayto calculate the desired precision. Youmust consider why the survey is beingconducted and what questions need to beanswered. Then think about possibleresults and whether wide confidence inter-vals will be acceptable when making pro-gram decisions.

Expected malnutrition prevalence ormortality rateYou can use various techniques to obtaina gross estimate of the prevalence of mal-nutrition or the rate of mortality beforeyou conduct a survey. Surveillance data,for example, may include counts of thenumber of children presenting to clinicswith malnutrition or the number ofdeaths. A prior survey may have estimatedthese outcomes, and persons who haveworked in this population since that sur-vey may be able to give you a general ideaif malnutrition or mortality have changedsince that survey. An overall impression ofthe extent of malnutrition or mortality alsocan be obtained by more qualitativemeans. You can ask health workers if theysee many thin children, or you can askreligious leaders if they recently have beencalled upon to conduct more than theusual number of funerals.

In general, a prevalence of wasting of 20percent is very high. In such a situation,you will have already received manyreports of serious malnutrition whileorganizing the survey. The traditional 30x 30 survey assumes a prevalence of 50percent, a level of wasting seen only veryrarely in the worst emergency situations.However, if you have only the mostimprecise estimate of the prevalence ofmalnutrition, use an estimate closer to 50percent than you think is the true preva-lence. This will overestimate the samplesize required, ensuring that you have a

large enough sample size to get the nee-ded precision.

Design effect is a measure of how evenly orunevenly the outcome (for example,wasting, stunting, anaemia, or mortality) isdistributed in the population being sam-pled. For example, if you think that malnu-trition is about the same in all parts of thepopulation, then the design effect is proba-bly low. In many populations, the designeffect for malnutrition is usually in therange of 1.5-2.0. On the other hand, if youthink an outcome such as mortality is quitedifferent in different parts of the population,then the design effect may be quite high.For example, in emergencies where violen-ce causes a large proportion of deaths andthe violence is not evenly distributed, thedesign effect can be in the range of 4-10.

Probably the best source for estimates ofdesign effect are prior surveys done in thesame or similar populations. However,because the design effect changes withthe number of units of analysis in eachcluster, design effects from prior surveysshould not be used in the calculation ofsample size for a survey which will havea very different cluster size. In general,the greater the number of units of analy-sis in each cluster, the larger the DEFF.There are some published papers whichcan give some indication of the range ofpossible design effects for various outco-mes other than the prevalence of wastingand mortality rates.

Sample size calculation accounting fornon-responseSample size calculations should alsoaccount for non-response. If we assumethat, at most, 10 percent of householdswill be gone or will refuse participation,then we must select enough householdsfor the initial sample so that, if we lose 10percent of selected households to


DESIGNING A SURVEY

71

non-response, we will still have enough toobtain our desired precision. This can becalculated using the following formula:

Thus, if the non-response rate is 10 percent (or 0.10 in decimal format), thenthe number of households to be selectedin the example above is:

Assuming a 10 percent non-response, if 461 households are invited to participatein a survey, 415 would agree to participate.

Sample size for multiple outcomesMost surveys measure more than one outco-me. In such cases, the sample size for eachoutcome should be calculated. An example

is shown in the table below and the largestsample size is used in order to ensure that alloutcomes have an adequate sample size.

This table shows that in order to obtain atleast 305 children and 415 women for thesurvey, data must be collected from atleast 415 households.

Sample size calculation for detecting thedifference between two surveys (beforecommencing the first survey)

If a survey is designed with the specific inten-tion of comparing it to another survey whichwill follow it (i.e., baseline and follow-up sur-veys), the calculation to determine the neces-sary sample sizes in order to detect a statisti-cally significant difference between twosurveys is different than the regular sam-ple size calculation.

Table 3.1 Sample size calculations based on varying outcomes

Target group and type of malnutrition

Children 6-59 months

Wasting (< -2 SD)

Stunting (< -2 SD)

Anaemia (<11.0 g/dL)

Women of childbearing age

Malnutrition (BMI <18.5)

Anaemia (<11 g/dL)

Estimatefromprior surveys(%)

3.7

39.2

48.5

Not available

53.5

Assumed current value (%)

5

50

50

10

50

Precisiondesired(%)

± 3

± 10

± 10

± 5

± 10

Designeffectassumed

1.5

2

2

3

2

Samplesize needed

305

193

193

415

193

SD = standard deviation; BMI = body mass index

3CHAPTER

The two surveys may be done to comparethe current situation in two different popu-lations or they may be done to detectchange over time in the same population.Note that this process produces a samplesize sufficient only to make an overall esti-mate for the entire survey sample. If sepa-rate estimates are needed for subgroupswithin the sample, such as estimates forurban and rural dwellers or males andfemales, the calculated sample sizes applyto only one such group. The entire surveysample must be a combination of the sam-ple sizes calculated for each group.

This table shows that at least 305 children6-59 months are needed and at least 415adult women are needed. However, becau-se in most surveys you will be samplinghouseholds, you must now calculate howmany households must be included in thesurvey to obtain the required number ofchildren and women as shown in Table 3.2.

This table shows that in order to obtain atleast 305 children and 415 women for thesurvey, data must be collected from atleast 415 households.

Sample size calculation for detectingthe difference between two surveys(before commencing the first survey)If a survey is designed with the specificintention of comparing it to another survey which will follow it (i.e., baselineand follow-up surveys), the calculationto determine the necessary sample sizesin order to detect a statistically signifi-cant difference between two surveys isdifferent than the regular sample size calculation. The two surveys may bedone to compare the current situation intwo different populations or they may bedone to detect change over time in thesame population. Note that this processproduces a sample size sufficient only tomake an overall estimate for the entiresurvey sample.

If separate estimates are needed for subgroups within the sample, such asestimates for urban and rural dwellers ormales and females, the calculated samplesizes apply to only one such group. Theentire survey sample must be a combina-tion of the sample sizes calculated foreach group.

72

Table 3.2 Sample size household calculations

Target group and type of malnutrition

Children 6-59 months

Wasting (< -2 SD)

Stunting (< -2 SD)

Anaemia (<11.0 g/dL)

Women of childbearing age

Malnutrition (BMI <18.5)

Anaemia (<11 g/dL)

Sample size needed

305

193

193

415

193

Number of persons in target group perhousehold

1.1

1.0

Number ofhouseholdsneeded

277

175

175

415

193

SD = standard deviation; BMI = body mass index

The assumptions you need to make beforecalculating the sample sizes include:1) The estimated prevalence of the outco-

me being measured in one survey (p1in the formula below)

2) The estimated prevalence of the outco-me being measured in the other survey(p2 in the formula below)

3) The design effect for the two surveys(DEFF in the formula below)

The formula is:In this equation, 6.175 is the z valueassuming that the P value of significan-ce is 0.05 and the power required is 0.8.This equation also assumes that youwill have equal sample sizes for bothsurveys and that you are calculatingthese sample sizes before performingthe first survey. Remember that the dif-ference between the two surveys'results (P1-P2) is the minimum differen-ce which can be detected with statisti-cal significance. Any difference greaterthan this found after the two surveyshave been done will, of course, be sta-tistically significant.

For example, you assume that a baselinesurvey done before an iron supplemen-tation program is begun will find that 50percent of non-pregnant women of chil-dbearing age will be anaemic. You wishto detect a statistically significant diffe-rence of 7 percent or more in the preva-lence of anaemia in this group whenyou do a follow-up survey three yearslater. How big should the sample sizebe for the two surveys? The prevalencein the first survey (P1) equals 0.5, theprevalence in the second survey (P2)

equals 0.43, the design effect is assu-med to be 1.5, and the difference betwe-en the two prevalence rates is 0.07.

Therefore, each survey sample mustinclude at least 463 women of childbea-ring age to be able to detect a statistical-ly significant difference of 7 percent ormore in the prevalence of anaemia withstatistical significance.

Because many statistical analysis pro-grams for computers can calculate sam-ple sizes for both single surveys and thecomparison of two surveys, it is rarelynecessary to calculate sample sizes byhand. It is recommended that surveymanagers identify such a program whichthey can learn thoroughly and use foreach sample size calculation.

Sample size calculation for detectingthe difference between two surveys(after an initial survey has alreadybeen completed)If a survey is designed with the specificintention of comparing it to another surveywhich has already been completed (i.e.,a follow-up survey), the calculation todetermine the necessary sample sizes inorder to detect a statistically significantdifference between two surveys is differentthan the regular sample size calculation.This calculation is substantially morecomplex, and a survey statisticianshould be consulted for assistance. Acomputer program may be availablesoon to perform this type of calculation.When completed, it will be made availa-ble from the Nutrition Service at WFPheadquarters.


DESIGNING A SURVEY

73

3CHAPTER

74

COMMON MISTAKE #5:Always using cluster sampling, even if simple or systematic random samplingis possible.

Explanation: Some people tend to use cluster sampling in all situations becausethat is what they are accustomed to or that is what their organization recommends.However, cluster sampling usually requires a larger sample size to get the sameprecision, which in turn requires more survey workers, more time and moreresources to complete the survey. Therefore, it is much more efficient to use syste-matic random sampling when conditions permit, such as when there is a list of allthe sampling units or the arrangement of dwellings in neat rows.

Solution: Choose the most efficient sampling method on the basis of the dataavailable on the population of interest.

STOP

GO

7. SELECT THE SAMPLEOnce the type of sampling is selected andthe desired sample size is determined, thechildren or households to include in thesurvey can be selected. These samplingmethods are covered in more detail inmany books listed in the reference section.A summary description is presented below:

Simple random sampling Simple random sampling is the process inwhich each sampled unit (child or house-hold) is selected one at a time from a list ofall eligible sampling units. Simple randomsampling can be done many ways, but itrequires a complete list of all the basic sam-pling units in the population. a) The prototypical method for simple ran-

dom sampling is to write the name ofeach household or child on a piece ofpaper, then put all the pieces of paper ina bowl and randomly select as manypieces of paper as the number of house-holds or children needed for the survey.Of course, if there are 10,000 house-holds in the population, it would take avery long time to write the names of thehouseholds on 10,000 pieces of paper.

b) Another way would be to number thehouseholds. Then random numbersbetween 1 and the total number ofbasic sampling units could be selectedfrom a random number table or gene-rated on a calculator or computer. Thehousehold with the same number asthe random number would be includedin the sample. The quantity of randomnumbers selected equals the desirednumber of sampling units, as determi-ned by the sample size calculation.

Systematic random samplingSystematic random sampling is an alterna-te way to select a random sample. It maybe much easier when the population islarge. Systematic random sampling selectsone household at random, then selectsevery nth household thereafter, where “n”equals the sampling interval. The samplinginterval is the total number of samplingunits in the population divided by the desi-red sample size. The first household isselected by choosing a random numberbetween 1 and the sampling interval. Then the sampling interval is added to thatnumber to select the next sampling unit.

For all subsequent selections, the samplinginterval is added until the end of the list ofsampling units is reached. For example, ifthere are 10,000 households in the popula-tion and a sample of 1,000 households isneeded, the sampling interval is 10 (10,000divided by 1,000). A random number bet-ween 1 and 10, the sampling interval inthis example, is chosen to select the firsthousehold. Let us say that 3 was the ran-dom number. The second household isselected by adding the sampling interval of10 to the random start of 3 to choose hou-sehold number 13. Then 10 is added tothis number to select household number23, and so on.

Systematic random sampling can be donewith a list of all sampling units either onpaper or on a computer. You can also arran-ge the sampling units, such as households,in a pattern so that they can be easily coun-ted on the ground. For example, in CegraneCamp in Figure 3.1, the desired sample sizewas 350 households. The camp contained5,263 households; therefore, the surveyteam needed to include in the survey sam-ple every 15th household (5,263÷350=15). The team chose a random number bet-ween 1 and 15 inclusive to select the firsttent, then beginning in one corner of thecamp, counted tents along the first rowuntil that number was reached. This wasthe first tent selected for the survey. Thenproceeding down that row and subsequentrows of tents, the survey team counted eachtent and selected every 15th tent to includein the survey. The team did not have a listof all households in Cegrane Camp, butbecause the tents were in rows that wereeasily counted, a systematic random sam-ple could be selected easily.

However, as mentioned above, in manypopulations, no list of all children or allhouseholds exists, and the dwellings arenot arranged in neat rows.

In such populations, survey workers mayneed to do cluster sampling. Many organi-zations, especially those working in emer-gency situations, are very familiar withcluster sampling.

Cluster samplingCluster sampling is used in large geo-graphically disperse populations,where no accurate list of households isavailable for the entire population andhouseholds cannot be visited systema-tically. This is the most common situa-tion in most populations. Cluster sam-pling is often more convenient anduses fewer resources for transport thansimple random sampling because a clu-ster design reduces the distance thesurvey team has to travel between hou-seholds. However, the sample size isusually larger than simple randomsampling so that more households needto be visited.

With cluster sampling, the sampling issplit into multiple stages. In most clustersurveys, there are two stages of sampling: 1) The first stage of sampling selects

collections of persons or households,such as geographic areas within thepopulation to be surveyed. These geogra-phic areas may consist of political subdi-visions, such as districts, subdistricts,census blocks or other defined areas.Alternately, in rural regions, the areas tobe selected may be discrete concentra-tions of population, such as villages.

2) The second stage of sampling chooses,within each selected area, the house-holds to be included in the survey.

Stage one: selecting the primary sampling unitCluster sampling requires the division ofthe population into smaller geographicalunits, such as city blocks, subdistricts or villages.


DESIGNING A SURVEY

75

3CHAPTER

These geographic units may be referred toas enumeration areas, enumeration units,or other terms, and are usually availablefrom the organization within the countryresponsible for the census. A sample ofthese geographic units, called primary sam-pling units (PSUs), is then selected in thefirst sampling stage. Although many textsand people use the word “cluster” to meanthe same thing as PSU, they are not thesame. The PSU is the geographic unit selec-ted during the first stage of sampling. A clu-ster is the group of basic sampling units,usually households, chosen within a selec-ted PSU. Therefore, when carrying out thefirst stage of sampling, you are selectingthose PSUs in which a cluster will be selec-ted; you are not directly selecting clusters.

Probability Proportional to Size (PPS)When determining what to use as PSUs,the selection of smaller PSUs rather thanlarger PSUs will facilitate second stagesampling (see below). However, in orderto use a geographic unit as PSUs, somemeasure of the relative size of each unit,most commonly the population or thenumber of households, must be known.For example, if a survey is done in a city,it would be better to use city blocks ratherthan larger neighborhoods as PSUs. Inaddition, each geographic unit shouldhave at least the number of householdsrequired to form a complete cluster.Moreover, the geographic unit chosen tobe PSUs must be mutually exclusive; thatis, all households in the population beingsurveyed must be contained in one PSUand only one PSU.

The first stage selection of PSUs is done sothat the chance of any specific PSU beingselected is proportional to the size of thatPSU relative to the entire population. Thistype of selection is called “probability proportional to size” or PPS. Thus, if onePSU has a population of 5,000 and another1,000, then the former PSU has five timesthe chance of being chosen to contain acluster as the latter PSU. This is the mainreason why some measure of the size ofeach PSU is required.

PPS sampling is demonstrated schemati-cally in Figure 3.2. Each row of boxesrepresents 6 PSUs from which we mustselect one randomly. Each PSU, A throughF, is assigned to a box, and the populationof that PSU is written inside that box. Inthe first row, the size of the box allocatedto each PSU is equal. As a result, if wethrow a dart randomly at the first row ofboxes, the chance that PSU A will be hit isthe same as the chance that PSUs Bthrough F will be hit. As a result, eachPSU has the same chance of selection asany other PSU. This is called selectingPSUs with equal probability. On the otherhand, if we make the size of the box foreach PSU proportional to the population ofthat PSU, as is done in the second row ofboxes, the chances that a randomlythrown dart will hit PSU A is much smal-ler than the chance that it will hit PSU C,which has a much larger population. Thisresults in PPS sampling; that is, the proba-bility of selection for any single PSU isdirectly proportional to the size of thatPSU relative to the sizes of all other PSUs.

76

A B C D E F

Not PPS 231 912 3,099 376 484 763

A B C D E F

PPS 231 912 3,099 376 484 763

Figure 3.2 Schematic demonstration of (1) equal probability of selection, and (2) probability proportional to size.

In practice, we cannot draw boxes foreach of the PSUs from which the selec-tion will be made in a real survey.Moreover, throwing a dart is not really arandom way of making a selection.When carrying out sampling for a clustersurvey, the following method is usuallyemployed to select PSUs with probabilityproportional to size:

1) Determine the sample size usingmethods described above as if youwere using simple or systematic ran-dom sampling.

2) Obtain the best available data on sub-divisions of the populations whichmight be used as PSUs. You will needthe size of each unit as well as thelocation of each unit. Census datafrom the national or local gover-nment offices often provide suchinformation. In stable populationswith little in- or out-migration, a cen-sus that is several years old may stillbe acceptable. In emergency situa-tions, data may consist of populationestimates or registration data incamps. Alternatively, if no populationdata are available, you can estimatethe relative size of the populationliving in each PSU using data fromkey informants such as communityleaders or health workers.

3) Create a list of PSUs with a columncontaining the measure of the size ofeach PSU. The order in which PSUsare listed is not important. [Usually itis recommended to sort on a variableof interest for implicit sampling, suchas on rural/urban status, or by geo-graphic areas.] What is important isthat all PSUs be included on the list.Figure 3.3 gives a hypothetical exam-ple of a list of villages with the num-ber of households in each village.

4) Add columns to this list which con-tain the cumulative population.

Imagine that you are numbering allthe households in the entire popula-tion. In Figure 3.3, households num-ber 1-600 are located in the village ofUtural. Households number 601-1300are located in the village of Mina, etc.

5) Determine the sampling interval. Thesampling interval is the number ofPSUs you will select divided by thetotal number of households in thepopulation. For example, if you wan-ted to select 30 PSUs from the popu-lation listed in Figure 3.3, the sam-pling fraction would be 25,370 / 30= 845.7 H 845.

6) Select a random number between 1and the sampling fraction. The villa-ge where this number household islocated will be the first PSU contai-ning a cluster. For example, if 399 isthe random number, household num-ber 399 is located in Utural, and thefirst cluster will be located in that vil-lage.

7) Add the sampling fraction to the ran-dom number selected above. The vil-lage where this number household islocated will be the second PSU con-taining a cluster. For example, 399 +845 = 1,244. Household number1,244 is located in Mina; the secondcluster will be in this village.

8) Continue adding the sampling frac-tion to the previous number to deter-mine where the remaining clusterswill be located.

Some PSUs may be large enough to beselected more than once to contain a clu-ster. In such a case, when the survey teamarrives at this PSU, they should select asmany clusters from this PSU as indicatedduring the first stage sampling procedure.The selection of each cluster in a PSU withmore than one cluster should be complete-ly independent. The procedure for selec-ting households will be described below.


DESIGNING A SURVEY

77

3CHAPTER

78

Figure 3.3 A list of hypothetical villages to be included in a cross-sectional survey

No.

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950

Village name

UturalMinaBolamaTalumWar-YaliGaleyTarumHamtatoNayjaffNuviyaCatticalParalaiEgala-KuruUwanarpolHilandiaAssosaDimmaAishaNam YaoMae JarimPuaGambelaFugnidoDegeh BurMezanBan VinaiPuratnaKegalaniHamali-UraKameniKiroyaYamwelaBagviAtotaKogouvaAhekpaYondotMozopMapazkoLatohahVoattiganPlitokDopoltanCococopaFamegziJigpelayMewoahOdiglaSanbatiAndidwa

Number of households

60070035068043022043015090

300430150380310

2,00075025042018030010071019015045040022014080

41028033044032012060

3201780390

1,500960420270

3,50040021050

3501,440

260

Cumulative number of households-Lower

1601

1,3011,6512,3312,7612,9813,4113,5613,6513,9514,3814,5314,9115,2217,2217,9718,2218,6418,8219,1219,2219,931

10,12110,27110,72111,12111,34111,48111,56111,97112,25112,58113,02113,34113,46113,52113,84115,62116,01117,51118,47118,89119,16122,66123,06123,27123,32123,67125,111

Cumulative number of households-Upper

6001,3001,6502,3302,7602,9803,4103,5603,6503,9504,3804,5304,9105,2207,2207,9708,2208,6408,8209,1209,2209,930

10,12010,27010,72011,12011,34011,48011,56011,97012,25012,58013,02013,34013,46013,52013,84015,62016,01017,51018,47018,89019,16022,66023,06023,27023,32023,67025,11025,370

Cluster number

12

3

4

5

6

7, 8, 9

10

11

12

13

14

15

16

17,18,19

20,2122

2324,25,26,27

28

29,30


DESIGNING A SURVEY

79

3CHAPTER

How do you decide how many clustersshould be selected? For a given sample size,the more clusters in the survey sample, thelower the design effect and the greater theprecision obtained by the survey. However, alarger number of clusters often requires sub-stantial additional logistic and transport costsbecause travel between clusters may be dif-ficult or involve long distances. Determiningthe number of clusters to use in a survey the-refore requires weighing the advantage ofgreater precision with the disadvantage ofgreater cost. Studies have demonstrated that,for a given overall sample size, having fewerthan 25-30 clusters may lead to a high designeffect and an unacceptable loss of precision.Adding more than 30 clusters often does notincrease the precision enough to justify theadditional cost. Consequently, many surveysuse 30 clusters.

The size of clusters is determined from thenumber of clusters, as discussed above, andthe sample size calculation. The total samplesize is divided by the number of clusters todecide how many basic sampling unitsshould be included in each cluster. For exam-ple, if sample size calculations determinedthat 450 households should be selected for asurvey and 30 clusters were to be used, theneach cluster would contain 15 households.

Stage two: selection of households to form the clusters There are several methods of choosing thehouseholds within the selected PSUs.

1. Simple or systematic random samplingThe best way is to select households usingsimple or systematic random sampling, asdescribed previously. Simple random sam-pling can be done using a complete list ofall the households in the selected PSU.Village leaders sometimes have such a listavailable to keep track of tax obligations orfor some other purpose. If there is no writ-ten list, village leaders or elders can often

tell survey workers the names of the headsof all the households in the village whilesurvey workers write them down. Whencreating such a list, survey workers mustbe very careful to ensure that the infor-mants have not forgotten any household inthe PSU, such as households headed bywomen, households of poor people, house-holds of ethnic minorities or others. Once alist is located or created, households can benumbered and random numbers used toselect individual households for inclusionin the survey. Another method of randomsampling within PSUs is to select house-holds from the household list using syste-matic random sampling. Alternately, sur-vey workers can draw a rough map of allthe households within the PSU and thencarry out systematic random samplingusing the map to ensure that no house-holds are missed by the sampling.

2. SegmentationIf the PSU is large enough to make thetechniques above too time-consuming, thePSU can be divided into segments of roughlythe same size. One of these segments is thenchosen at random. In general, the segmentsshould contain fewer than 250 households.These households then are listed and therequired number of households is selected.

3. EPI methodIf it is absolutely not possible to select thehouseholds using random or systematicsampling, then the sampling method fre-quently used by WHO's ExpandedProgramme on Immunization (EPI) canbe used. While this method is simple,widely known and rapid, it may resultin a biased sample. See Common Mistake #7 for a more detailed explanation of thispotential bias. To employ this method, thefollowing procedures should be followedafter arrival at the selected PSU: 1) Go to the center of the selected PSU. Local

residents can help you locate the center.

2) Randomly choose a direction by spin-ning a bottle, pencil or pen on theground and noting the direction inwhich it points when it stops.

3) Walk in the direction indicated by the pen,from the center to the edge of the village,counting the number of houses on the way.

4) Draw a random number between 1 andthe number of houses counted on theline. Count the number of houses onthe line from the center of the village;the house matching the randomly selec-ted number is the first house to beincluded in the survey.

5) Include all children aged 6-59 months inthis household for the nutrition surveyand complete the mortality questionnaire.

6) Subsequent households are chosen byproximity. In a village where the housesare closely packed together, choose thenext house on the right. If the village isless densely inhabited, choose the housewith the door closest to the last house sur-veyed, whether it is on the right or left.

7) The original EPI method recommendsthat sampling continue until the requi-red number of children have beenenrolled in the survey.

In general, once a household is selected, alleligible persons in that household should beincluded in the survey. If there is more thanone eligible person and only one is selected,this makes the likelihood of selection diffe-rent for different persons and thus requires aweighted analysis. Such analysis is substan-tially more complex than an unweightedanalysis and is best avoided by including alleligible persons from each selected house-hold. For a more complete discussion of thisissue, see Common mistake # 10 below.

Regardless of the method of selecting hou-seholds, if there are two clusters located inthe same PSU, their selection should becompletely independent. For example, ifyou make a list of all households in a villa-ge, you should select one cluster by random

sampling, then select the households forthe second cluster. The lists of householdsfor each cluster should be kept separate.

Some potential operational problemsImplementation problems can arise in eventhe best-planned surveys. Typical amongthese are inaccessible clusters, non-responseand an insufficient number of households ina given PSU to complete an entire cluster.

Inaccessible clustersAt times, it may be impossible to reach asample cluster due to poor weather, impas-sable roads, insecurity or other reasons.Usually, the best recourse is to replace thecluster with another randomly chosen clu-ster with similar characteristics. For exam-ple, if the cluster in question is located inthe far northern part of the area included inthe survey, it should be replaced with ano-ther cluster in the same general area, butone that can be reached during the periodof survey fieldwork. To minimize the risk ofbias, replacement clusters should be chosenfrom among similar clusters; convenienceshould not be an issue. As far as possible,supervisory personnel should make deci-sions on replacement clusters.

Survey non-responseNon-response is an important issue in sur-veys. When households are selected, theremay be non-response at two levels: (1) enti-re households may be missing or refuse par-ticipation, and (2) individuals within con-senting households may refuse participationor be absent. The initial calculation of sam-ple size should compensate for the predic-ted level of both types of non-response. Thiswill help ensure that the final survey samplewill have the required precision in spite ofsome non-response.

When no one is at home in a selected house-hold, the survey team should inquire fromneighbors whether the dwelling unit is inha-

80


DESIGNING A SURVEY

81

3CHAPTER

bited and if so, where the residents are andwhen will they return. If they are to returnbefore the survey team must leave the PSU,a message can be left that the survey teamwill return at a prearranged time. If thehouse is not occupied, no further action isrequired. No pressure should ever beapplied to an individual within a selectedhousehold who refuses interview or biologicspecimen collection.

Non-response can bias the survey resultsbecause people who participate in a surveymay be systematically different than thosewho do not. These differences may be reflec-ted in the indicators that are being measured.As a result, non-response should be minimi-zed as much as possible by allowing adequa-te time to reattempt contact with absent hou-

sehold members. Moreover, the reasons forboth household and individual non-respon-se should be noted on data collection formsto allow the assessment of potential non-response bias during later data analysis.

Insufficient number of households If possible, before selecting the sample ofPSUs, survey managers should go throughthe list of PSUs to be sure each is largeenough to select a complete cluster. If certainPSUs are not large enough, they can be com-bined with adjacent PSUs. If this cannot bedone before sampling because the list is toolong, another procedure can be followed inthe field to complete a cluster. Once all avai-lable households are selected in a small PSU,selection of households can be continued ina neighboring PSU to complete the cluster.

COMMON MISTAKE #6:Using 30 x 30 cluster sample size regardless of required precision for the survey.

Explanation: Many survey managers select a sample consisting of 30 clusters with30 children or households in each cluster. Such a sample is often called “30 x 30”or “30 by 30.” Such sampling has also been recommended by many organizationsin order to ensure sufficient precision when the survey is completed. However, insome situations, the 30 x 30 cluster survey provides more precision than is reallyneeded. The 30 x 30 sample size assumes that: 1) the prevalence is 50 percent; 2) the desired precision is ±5 percentage points; 3) the design effect is 2; and 4) 15 percent of households or children will refuse or be unavailable. However, in most emergency-affected populations, the prevalence of acute malnu-trition is much lower than 50 percent, far less than 15percent of households refu-se or are unavailable and the design effect may be less than 2.0. Therefore, for mostnutrition surveys where wasting is the primary indicator of interest, the desiredprecision could be obtained with a sample size substantially less than the 900included in a 30 x 30 survey. Survey workers could design a more efficient surveyby making their own assumptions about the potential prevalence and design effectand by calculating a sample size necessary to achieve the precision they need, evenif different from ±5 percentage points.

Solution: Calculate the sample size which provides the desired statisticalprecision, but not more.

STOP

GO

82

COMMON MISTAKE #7:Selecting households using the EPI method for final-stage sampling instead oftrue probability sampling at this stage.

Explanation: Infants under 6 months of age often are not included in nutrition sur-veys for various reasons.

Other methods of selecting households may be less biased than the EPI methodand should be used if possible. These include the methods described in this chap-ter about selecting the sample.

Why should the EPI method be avoided? The EPI sampling method uses proximitysampling, which means houses are selected according to their proximity or distan-ce to the previous house. Proximity sampling may introduce bias into the sample.Figure 3.4 (below) represents a village with 30 households. The first house is oftenselected by spinning a bottle to choose random direction. Lines A and C representthe arc that makes house number 5 eligible to be the first house. Lines B and Crepresent the arc that would make house number 1 eligible to be the first householdin the sample. The likelihood that a randomly chosen line falls into arc AC is muchgreater than the likelihood that such a line would fall into arc BC. As a result, house5 is much more likely to be eligible to be the first house than house 1. In general,because houses in the center of the village are closer to the point of origin for therandomly chosen line, they are more likely to fall on this line.

Figure 3.4 Schematic of Village With 10 of 30 Households Selected by EPI Houses also tend to be closer together in the center of most villages than at theperiphery. Selecting the next closest house often moves the survey team toward thecenter of the village, thus making it more likely that houses in the center are selec-ted for the sample. In Figure 3.4, if house 1 is selected as the first house, choosingthe next closest house results in selection of houses 2-10. The circle includes the15 more central houses in this village. Note that 7 (houses 4-10) of the 10 selectedhouses fall within the circle. In many places in the world, families living in the center

STOP

Continued on the next page


DESIGNING A SURVEY

83

CHAPTER

of the village are different from families living at the edge of the village. Those inthe center may be richer or of a different ethnic group. Families in the center mayalso be village leaders or may be merchants, whereas families at the edge may befarmers. If the nutritional status of these groups differs, the nutritional status of thechildren included in the sample may not be representative of the nutritional statusof all children in the population.In addition, the choice of which house is actually the closest is often unclear, allo-wing the survey workers to decide which house will end up in the survey sample.Survey workers should never choose households on any basis other than randomselection. If two houses look equally distant from the last house in the sample, mostworkers will choose the house that looks nicer, doesn't have the angry dog in thefront garden, or has no children so that data collection can be completed more rapi-dly. This adds yet another potential source of bias to the overall sample selection. Finally, the EPI method of selection is not true probability sampling because you can-not calculate the probability that any one house in the village will be selected for thesample. True probability sampling requires that the total number of houses in the vil-lage be known so that this probability can be determined.

Solution: Use a method for the final stage of sampling which produces a trueprobability sample.GO

COMMON MISTAKE #8:Ending household selection after reaching target number of children, womenor households for that cluster.

Explanation: The sample size should be calculated and adjusted for possible non-response. This sample size is then used to determine how many households shouldbe sampled. The number of households then is divided by the number of clusters todetermine how many households will be included in each cluster. You next shouldselect this number of households in each cluster and attempt to gather survey dataon all eligible members of these households. Do not stop gathering data in the clu-ster if you reach the desired number of children or women. The EPI method recommends that survey teams select households until the desirednumber of children are recruited for the survey. Most of the time, for the sake of effi-ciency, survey teams will begin in one part of the village and move in a certain direc-tion until they have completed survey activities in that village. If survey workers stopwhen the desired number of children is found, they may stop before reaching theremaining eligible households in the last part of the village.

Solution: Determine how many households to select for each cluster by calculating the sample size, and then select and visit the predeterminednumber of households. Do not stop the survey before visiting all the selected households, even if the required number of eligible children hasalready been met.

STOP

GO

3Continued from the previous page

84

COMMON MISTAKE #9:Selecting only households with a child under 5 years of age if collecting infor-mation on other target groups or household-specific information.

Explanation: Since nutrition assessment surveys often include children 6-59months of age as the main target population, some survey teams may wish toexclude households that do not contain an eligible child. However, some surveysmay also measure nutritional status in other age groups, such as adult women, orthey may measure mortality rates among persons of all ages or gather other hou-sehold information, such as receipt of relief food or source of water.

Hence, households with children 6-59 months of age may not be representative ofall households. And excluding households without eligible children might therefo-re bias other indicators being collected, particularly if the survey is measuringcrude mortality. Households with young children tend to have young parents andsmall children; as a result, the mortality in such households may be less than ave-rage because the adults tend to be younger or, conversely, the mortality may behigher because young children are more likely to die than are older children andadults. Unfortunately, there is no way to determine which is true in any specificpopulation and there is therefore no way to know in which direction the bias isoperating. Thus, you cannot tell the true mortality rate in the population if morta-lity is measured only in households with young children.

Solution: Include all households selected for the survey sample, regardlessof whether there is a child under 5 years of age.

STOP

GO


DESIGNING A SURVEY

85

3CHAPTER

COMMON MISTAKE #10:Selecting only one child from surveyed households that have more than oneeligible child.

Explanation: If you include all eligible children from each selected household, eachchild in the population will have the same chance of being included as the house-hold in which the child lives. This is because once a child's household is selectedfor the survey, the child will automatically be included in the survey, regardless ofwhether there are other eligible children in that household. Therefore, because allhouseholds in the population have an equal chance of selection, all children in thepopulation have an equal chance of selection.

In contrast, if you include only one child in a household with more than one eligi-ble child, you have now changed the chances that children in that household willbe included in the survey. In households with only one child, the child's chance ofbeing included in the survey is the same as that household's chance, as explainedabove. If a selected household has two children, and you include only one childfrom that household, you have added another sampling step because you thenmust choose between the two children in that household. Therefore, each suchchild's chance of being included in the survey is equal to the chance that their hou-sehold is selected, divided by 2, since you have to choose between the two chil-dren in that household. As a result, the likelihood that any specific child in thepopulation is included in the survey sample is now different for different childrenin the population.

The ultimate goal of sampling is to obtain a sample of children that is representa-tive of all children in the population being surveyed. If you include only one childin each selected household, the sample no longer will be representative of thepopulation because it will contain a smaller proportion of children from multi-chil-dren households than does the population. A similar principle applies to womenof reproductive age and other target groups.

Solution: Always select all eligible children in every household chosen forthe survey.

STOP

GO

8. DETERMINE THE SCHEDULE FOR THE SURVEY, AND OBTAIN AND PREPARE SUPPLIES AND EQUIPMENTThe exact dates of the survey shouldbe chosen to determine when surveyworkers, supplies and equipment willbe needed. You should consult withcommunity leaders to help determinethe best times to conduct the survey.That way, you can avoid conflicts bet-ween the survey and market days,local celebrations, food distributiondays, vaccination campaigns, or othertimes when people may be absent.Take the agricultural calendar into con-sideration, because women may be inthe fields for most of the day duringcertain seasons. The survey scheduleshould allocate time for preparation,training, pilot testing, communitymobilization, data collection, data ana-lysis and report writing.

Moreover, you should take intoaccount the seasonality of acute mal-nutrition and interpret survey resultswith this seasonality in mind. If resultsof a survey will be compared withthose of a previous survey, try to con-duct both surveys at the same time ofthe year so as to eliminate potentialdiscrepancies associated with seasonalvariation.

Once the sample size has been calcula-ted, you can determine what type ofsupplies are needed and in what quan-tity. You should make a complete listof supplies and equipment requiredduring the survey. For some equip-ment, such as height boards and sca-les, usually only one item is neededper team. For other supplies, such aslancets, alcohol wipes, HemoCuecuvettes, etc., at least one item will be

needed per survey subject tested. Inaddition to the equipment needed toactually take the measurements ateach selected household, you willneed:1) stationery, such as data collection

forms, pencils, erasers, clipboardsand waterproof folders or pouchesto store completed forms;

2) material needed for transport, suchas fuel and spare parts; and

3) items needed for the survey teammembers, including food, blanketsor sleeping bags, cots, water bot-tles, electric torches, etc.

The process of ordering or obtainingequipment and supplies must be star-ted early; it always takes longer thanyou think. For material coming fromoutside the country where the surveywill take place, you must take intoaccount delays at customs and youoften must pay duty on such importeditems. Measuring instruments, such asscales, height boards and portable sta-diometers, should be in perfect condi-tion and tested regularly for accuracy.For example, during a survey, scalesshould be checked each day against aknown weight. If the measure does notmatch the weight, the scales should berecalibrated, discarded, or the springsreplaced.

9. DESIGN THE DATA COLLECTION FORMAND SURVEYOR'S MANUALThe list of data to be collected by sur-vey workers in the field must now beput into a workable format. This inclu-des writing out the specific questionsto ask survey respondents and creatingspecific places on the data collectionform to record the data. The data col-lection form should be organized clear-ly to facilitate ease of use by surveyworkers in the field. The questions or

86

description of the data to be collectedbegin on the left side of the form andare numbered so that survey workerscan easily find their place when inter-viewing a mother. All the data are writ-ten on the right side of the form so thatthe data entry person can scan downthe form quickly while entering theinformation. It is important to keepdata collection forms open and unclut-tered so that survey workers can findeach question easily during the inter-view and likewise can determine easi-ly where to record each item of infor-mation. The codes to be typed into thecomputer should also be apparent onthe form to ease data entry.

Items on the data collection formshould be grouped according to theperson from whom (or the source fromwhich) the data will be collected.Questions posed to the caretaker comefirst, then all the data collected by exa-mination of the child, then the anthro-pometric and laboratory measure-ments. For example, by listing all thequestions for the mother in one sectionbefore listing the anthropometric mea-surements, the survey workers cancomplete the mother's interview befo-re setting up the equipment to weighand measure the child. Although thepresence of oedema will be used whenanalyzing the anthropometry data, this

data need not be included with theanthropometric measurements on thedata collection form. In fact, it is mucheasier to include it in the section onexamination.

The data collection form also can bedesigned to make it easier to enter thedata into the computer once field datacollection is complete. The computerentry screen can be made to look asmuch as possible like the paper datacollection form. Recording data in acolumn on the right or left side of thepaper form allows the person enteringdata into the computer to scan down acolumn instead of searching on theform for each piece of data. An exam-ple of such organization is shown inAnnex 4.

If possible, create a “surveyor'smanual” as a guide for the surveyteams while working in the field.Include details on how each item ofdata, including questions for house-hold members, should be created.Such a manual could describe the sam-pling procedures to be followed at eachcluster to help standardize householdselection across clusters. It could alsodescribe the technique for collection ofbiologic specimens or field laboratorytesting methods.

DESIGNING A SURVEY

87

3CHAPTER


88

COMMON MISTAKE #11:Not writing questions in the language of respondents.

Explanation: Some information in nutrition assessment surveys is collected byasking questions of the mother or another household member. For example, dieta-ry information, such as how often a child eats meat, is collected by asking themother or other caretaker about a child's recent diet. To help ensure that all thesurvey workers ask the same question, these questions must be written down sothat the same words are used by all survey workers for all interviews. If the que-stions are written in a language the survey respondents do not understand, theinterviewer must translate each question into the language of the respondent. Thismental translation may vary from one interviewer to another, or may even changefor the same interviewer between households. As a result, a question may not beposed to all survey respondents in the same way, thus possibly causing differentinformation to be gathered from the respondents.

Survey managers or team supervisors often do not speak the language of the sur-vey respondents. In such cases, the questions can be written in more than one lan-guage so that interviewers may read the questions directly to respondents, andsupervisors may understand what questions are being asked at which point.

Solution: Write the questions in the language that the survey participantsunderstand.

STOP

GO

COMMON MISTAKE #12:Not back-translating questionnaire questions.

Explanation: Survey managers, especially expatriates or outside consultants, oftendo not speak the language understood by the survey respondents. They will pre-pare survey questions and data collection forms in their native language. This textthen must be translated into the language of the survey workers and respondents.No single translation is perfect. Because the managers may not speak this langua-ge, they cannot judge the accuracy of the translation. To make such a judgement,the translated questionnaire then must be translated back to the language of thesurvey managers to determine the accuracy of the original translation. A differenttranslator, who has never seen the original questions, should make this secondtranslation. The two translators and the survey managers then should discuss anydifferences between the original version and the back-translation to decide howbest to pose each question in the language of the survey respondents.

Solution: Always translate, then back-translate, the data collection form,especially those questions to be posed to survey respondents.

STOP

GO

10. SELECT AND TRAIN SURVEY WORKERSMake a list of all the tasks to be performedby survey teams when they go to the fieldto collect the survey data. This list mayinclude: 1) interviewing household members; 2) measuring the weight and height

of children; 3) measuring the weight and height

of women; 4) examining children and women for

signs of micronutrient deficiency; and 5) performing a fingerstick and measuring

hemoglobin.

Next, decide what type of worker you willneed to perform each of these tasks. Whenpossible, hire someone with experience intaking anthropometric measurements toweigh and measure children. Survey wor-kers with clinical training may be neededto examine children and women. You maywish to hire a laboratory technician to per-form the fingersticks and operate theHemoCue and other serologic tests. Onesurvey worker may be able to do morethan one task. However, in some cultures,it may be unacceptable for men to weigh,measure and examine adult women; theseworkers may need to be female. Other cul-tural factors may also play a role in selec-ting survey workers.

Team members do not have to be healthworkers. In fact, health professionals suchas physicians are sometimes more difficultto train because they have learned a diffe-rent method of measuring or examiningduring their medical education. Personsfrom many backgrounds can serve as sur-vey workers as long as they can endure therigors of field work and are literate andable to count.

You will need at least three people perteam, sometimes more. Each survey team

needs a supervisor who is responsible forthe quality and completeness of data col-lection. The supervisor should check eachcompleted data collection form at eachhousehold to be sure that all data havebeen collected and recorded correctlybecause it may be impossible to revisithouseholds to correct errors after the teamhas left the cluster. The team also needstwo individuals to perform anthropometricmeasurements and record the data on thedata collection form. If testing for micro-nutrient deficiencies, the team may needan additional person who is trained inphysical examination and the gathering oflaboratory specimens. It may also be use-ful to have a respected community mem-ber on the team. This person can introdu-ce the survey team to the population andhelp guide the team around the site.

The number of teams should be sufficientto complete data collection in the timeallotted in the survey schedule. Havingmore teams than absolutely needed can bea disadvantage because more teams pro-duce larger inter-observer variation. Inaddition, it is often difficult to superviseand organize a larger number of teams.

Survey team members should receivecomprehensive training in the tasks theywill perform as part of survey activities.All members performing a specific taskshould undergo the same training in thattask, regardless of their former experience;this will ensure standardization ofmethods. The training for a survey usuallytakes 2 to 5 days and should include:1) A clear explanation of the objectives

of the survey.2) An explanation of the sampling

method. This should stress theimportance of proper household andchild selection to ensure a representa-tive sample.

3) A demonstration and practice of


DESIGNING A SURVEY

89

3CHAPTER

weight and height measurements.Each measurer should practice heightand weight measurements and asses-sments of oedema twice on each of10 children. Standardization of mea-surements among survey team mem-bers during training will help detectand correct errors in measurementtechnique before the actual data col-lection. Figure 3.5 shows one exam-ple of a spreadsheet with which tocompare the measurements of thetrainees.

4) An explanation and discussion of thequestionnaire questions. This willensure that the questions are phrased

correctly and that each interviewerunderstands what data the surveymanagers wish to collect with eachquestion. Role play exercises are use-ful in practicing interviews.

5) A pilot test in the field. A pilot test isan essential component of the trai-ning and should test all the compo-nents of survey data collection underrealistic conditions. Be sure that youvisit a location that was not selectedfor the real survey but is similar tothose sites actually selected. Data col-lected during the pilot test should notbe used in the analysis of the surveyresults.

90

Figure 3.5 A spreadsheet for comparing height measurements among surveyworker trainees


DESIGNING A SURVEY

91

3CHAPTER

COMMON MISTAKE #13:Not having enough training or not training in enough detail for survey workers

Explanation: Many populations do not have enough people with experience con-ducting nutrition surveys or taking anthropometric measurements to make up theteams for a nutrition survey. As a result, some survey workers may have no expe-rience in weighing children, measuring height, conducting interviews, or otheractivities necessary for data collection. Even if your survey team members areexperienced, they may have different techniques than you require. For these rea-sons, the training given to survey workers must be detailed and extensive. Formany nutrition surveys, at least 3 days must be devoted to the initial training befo-re data collection begins.

Many of the steps in data collection, such as measuring height, appear to be sim-ple. However, many sources of error can make the measurements inaccurate andthus bias the final results of the survey. Therefore, each of the steps in data collec-tion should be described and practiced during the training. Children can bebrought to the training site or the teams can go to a clinic or feeding center to prac-tice weighing and measuring. In addition, survey workers who will be measuringchildren should participate in some type of formal standardization exercise. Thisexercise consists of the survey teams measuring at least height on the same chil-dren to compare inter- and intra-observer error. Many texts, such as the MSFNutrition Guidelines, have descriptions of standardization exercises.

Solution: Always do complete training for all survey workers, includingpracticing the sequence of activities and carrying out a standardizationexercise for anthropometric measurements

STOP

GO

11. FIELD TEST DATA COLLECTIONFORMS AND DATA COLLECTION PROCEDURESConduct pilot testing during the trainingto give survey team members practicalexperience in all aspects of data collec-tion and survey methodology beforethey embark on actual data collection.This pilot testing should include the fol-lowing:

1) Pre-testing the survey questions.Comments by both those interviewedand the survey workers who conduc-ted the interviews can greatly impro-ve the questions.

2) Sampling households or children. Theteam should practice selecting the hou-seholds.

3) Taking and recording measurements cor-rectly in the field. Survey workers shouldpractice weighing and measuring as ateam in order to determine which wor-ker should perform each specific task.

4) Introducing the survey and comple-ting the data collection form. All sur-vey workers should practice usingthe data collection form and surveyo-r's manual. It will often be necessaryto revise the form and the surveyor'smanual after the pilot testing.

5) Organizing transport and equipment.

Pilot testing should not take place in alocation where the proper survey will takeplace, but in a similar community. Forexample, you could take the survey teamsto a neighboring village that has not beenselected during the sampling. Do not for-get to get a permission letter from localauthorities, even for field testing.

At the end of the pilot test, the teammembers, team supervisors and surveymanagers should discuss together allaspects of the data collection. The pilottest also will provide an estimate of howlong the data collection will take for eachsurvey subject and for a household as awhole. This information will help youcalculate how many households or chil-dren you can expect to measure each dayduring the data collection. Keep in mind,however, that during the pilot test thesurvey team members are still practicing.After collecting information from the firstfew households in the actual survey,teams will become much more efficientand accurate.

12. COLLECT THE DATAData collection will differ for each sur-vey. In small, compact populations, allsurvey teams may be able to reconveneeach evening to discuss questions or dif-ficulties. Surveys over wider areas mayrequire teams to operate independently

for days or weeks. Ideally, there shouldbe some means of communicationamong teams and between teams and thesurvey manager so that questions anddifficulties can be answered and correc-ted in a standard way for all teams. Ifcommunications are unavailable, the sur-vey manager should provide answers toanticipated questions during the surveyworker training and in the surveymanual.

Additional ways to improve the quality ofthe data collected during a nutrition survey include the following:1) Provide rigorous training, even to sur-

vey workers who have prior experien-ce or who may think they already haveall the skills necessary for the survey.

2) Use good quality equipment that iscalibrated regularly.

3) Provide good supervision of surveyteams. For example, the team supervi-sor should double check each case ofoedema. Improperly trained surveyworkers may mistake a “fat” child forone with oedema (particularly in youn-ger children). In addition, team super-visors should check data collectionforms for blank entries or erroneousentries at the end of each householdbefore the team leaves that household.

4) Do not overwork your teams; peoplemake mistakes when they are tired.

92


DESIGNING A SURVEY

93

3CHAPTER

COMMON MISTAKE #14:Replacement of missing households during data collection.

Explanation: Survey teams will always encounter households where everyone inthe household is missing, or a member of a target population, such as a youngchild or woman of childbearing age, is missing. In order to include enough house-holds in the survey, some survey teams will be tempted to substitute missing hou-seholds or children with others chosen from the village. This should be avoided.Children may be unavailable for many reasons which may bias the result of a sur-vey. For example, an ill child may be unavailable for survey assessment becausehe was taken to the hospital. Of course, ill children tend to have more malnutritionthan well children. If the survey team fails to include an absent ill child, the esti-mate of the prevalence of malnutrition may be somewhat underestimated. If thatchild is then replaced by an available child, that available child may be less likelyto have malnutrition. As a result, the estimated prevalence of malnutrition will befurther underestimated because the sample now contains a child who is less like-ly than average to have malnutrition. Although having a sample size too small toachieve the desired precision is not a good thing, it is much worse to inject biasinto the estimated prevalence of malnutrition. Substitution therefore should beavoided.

One way to avoid the need to replace missing children is by attempting to predictwhat the non-response will be among children in the sample. The sample size isthen increased by this proportion. For example, if you need a sample size of 400children and you predict that you will not be able to collect data on 10 percent ofchildren, the sample size could be increased to 444 (90% of 444 = 400).

Solution: Do not substitute selected households which are absent or refuseparticipation with a neighbor or other household not selected in the origi-nal sample. Calculate the sample size to account for non-response andachieve adequate precision.

STOP

GO

94

COMMON MISTAKE #15:Misidentifying the presence or absence of oedema.

Explanation: Some persons who become acutely malnourished develop an accu-mulation of fluid in the tissues which is called oedema. This abnormal fluid addsto the child's weight. As a result, a child may have normal weight-for-height eventhough children with oedema have a substantially higher death rate than childrenwith malnutrition without oedema. For this reason, all children with malnutritionand oedema should be judged as severely malnourished.

Survey workers who examine children must be taught the correct way of detectingoedema, as follows: Moderate pressure is applied to the back of the foot or to theshin just above the ankle for a full 3 seconds (the survey workers should count slo-wly, 1… 2… 3, while applying pressure). Oedema is present if, when the pressureis removed, an indentation remains in the skin.

Frequent mistakes made by survey workers include: 1) applying momentary pressure, which is insufficient to make

an indentation if oedema is present; 2) pushing too hard and causing the child pain; 3) recording oedema because the examiner's finger which is applying

the pressure produces an indentation while the pressure is applied.

Many children have fat on the back of the foot or lower leg which can be inden-ted with pressure, but the indentation disappears immediately when the pressureis stopped. An indentation from oedema will last many seconds after cessation ofthe pressure.

When calculating the prevalence of severe and moderate acute malnutrition, persons with oedema should automatically be included in the “severe” categoryregardless of their weight-for-height Z-score or percent of median. In a populationin which a high proportion of malnourished children have oedema, failure to properly classify oedematous children will lead to a serious underestimate of theprevalence of severe acute malnutrition. Because weight-for-height and height-for-age Z-scores are invalid if oedema is present, children with oedema should not be included when plotting the distribution of these Z-scores or calculating the average Z-score.

Solution: Each child who is weighed and measured should be examined forthe presence of bilateral pitting oedema. All survey members should betrained to detect bilateral pitting oedema. If oedema is detected, the teamsupervisor should verify the accuracy of the diagnosis before it is recordedon the data collection form.

STOP

GO


DESIGNING A SURVEY

95

3CHAPTER

COMMON MISTAKE #16:Not providing enough supervision of survey teams.

Explanation: Each survey team must have a supervisor who is responsible for allthe activities of that team. Supervisors must observe closely all data collection pro-cedures, including interviews and weighing, measuring and examining selectedchildren. Although the survey workers should have learned correct proceduresduring the training period, the hard work and long hours that surveys entail maycause some survey workers to cut corners or use sloppy technique. Supervisorsshould continuously remind them of the proper method.

Data collection forms are often not completely filled out, especially early on whensurvey workers have not yet established a strong routine. Team supervisors shouldreview each data collection form before departure from the household because afuture revisit to a household is often impossible.

Solution: Close supervision of survey teams, especially early on during datacollection, is essential for the collection of good quality data. Supervisionshould include careful monitoring of data collection along with a review ofeach data collection form before leaving the household.

STOP

GO

13. ENTER THE DATA AND ASSESS ITS QUALITYIn order to analyse survey data on acomputer, the data from the paper datacollection forms must be typed into adatabase program. The software mostcommonly used in nutrition data analy-sis is EpiInfo, with the accompanyingEpiNut. Specifically designed to analysenutrition and public health information,this program can be downloaded freefrom the Centers for Disease Controlwebsite, www.cdc.gov/epiinfo. Othersoftware, such as NutriSurvey andSPSS, can also be used for analysis.

Computer software that allows range andlogic checking should be used for dataentry. That is, the computer program canbe set so that only entries within a speci-fied range can be entered into specificfields. For example, you might wish to

set the acceptable range of 6 - 59 for thefield recording the ages of young childrenwho were weighed and measured.

If the data entry person tries to enter anumber less than 6 or greater than 59,the computer makes a warning soundthat indicates an unacceptable value.Logic checking allows comparison of thevalues entered into two or more fields toensure that the data entered makes sense.

For example, suppose you are collectinginformation in a survey on infant feeding.You first type into the computer that theage at which a child stopped exclusivebreastfeeding was 5 months. Then youtype in the age at which the child firstreceived cow's milk was 3 months.

The computer will compare these valuesand show a message that indicates thatthese two values are contradictory.

Such checking may eliminate some dataentry errors, but additional proceduresare needed to be sure that the computerdata are correct. This data entry must be done carefullyin order to minimize mistakes. Dataentry is not an easy job; personnel ente-ring data need proper training andsupervision. If possible, employ peoplewho are familiar with computers, avariety of basic software programs, andeven data entry. Those who are trainingthe data entry personnel should haveexperience in the specific data entrysoftware program being used. Trialexercises should be conducted usingsample questionnaires in order to checkaccuracy and possible systematic errors.

After the data has been entered, there aretwo common quality check methods:

1) (This is the standard and is recom-mended by WFP.) Enter data twiceinto two different databases and com-pare the two resulting databases. Ifdiscrepancies are found between thedatabases, the paper data collectionforms should be consulted to determi-ne the correct value.

2) Alternatively, the data can be enteredinto the computer only once, and theneach record in the database comparedto the paper data collection form tocheck for errors. This is often fasterthan entering the data twice.

Once the data are entered and checked,preliminary analysis should be done toassess the quality of the data. Some of thetechniques used are described below.These analyses can be included in the finalreport to demonstrate to readers the preci-sion and accuracy of the survey results.

Analysing the distributionof children's ages

Z-scores or percentage of median forheight-for-age or weight-for-age requireaccurate ages to within 2 weeks. Inmany cultures, mothers may not knowthe exact dates of birth of their chil-dren. Mothers will often round theirchildren's ages to the nearest year, lea-ding to a disproportionate number ofchildren with ages reported as 12, 24,36 or 48 months of age. As can be seenin Figure 3.6, the mothers of most chil-dren included in this survey roundedtheir children's age to a full year. Evenyoung infants were reported as 12months of age. In such situations, theage data may be so inaccurate as tomake analysis of height-for-age andweight-for-age indices unreliable.

In contrast, during a survey inMacedonia of Kosovar refugees, motherscould report precise dates of birth fortheir children. The distribution of thecalculated ages for these children isshown in Figure 3.7. Although the num-ber of children in each 1-month agegroup is not identical, there does notseem to be a preference for the ages 12,24, 36 and 48 months, as shown in theblack bars. The differences in the num-ber of children in each age group is pro-bably due to random selection of thesurvey sample. These figures demonstra-te that a simple graph of age distribu-tion of the children in the survey samplecan demonstrate whether ages are bia-sed toward whole years. Such a graphprobably should be displayed in the finalreport of any nutrition survey in whichheight-for-age or weight-for-age areimportant outcomes.

96


DESIGNING A SURVEY

97

3CHAPTER

Figure 3.6 Distribution of Children Under 5 Years of Age, by Their Mothers' Report of Their Age, Bardera,Somalia, January 1993

140

120

100

80

60

40

20

0

6 12 18 24 30 36 42 48 54

Num

ber o

f chi

ldre

n

Age in months

Figure 3.7 Distribution of Children Under 5 Years of Age, by Calculation From Mothers' Reported Date of Birth,Kosovar Refugees, Macedonia, May-June 1999

30

25

20

15

10

5

0

12 24 36 48

Num

ber o

f chi

ldre

n

Age in months

CALCULATING THE STANDARDDEVIATION FOR Z-SCORESSloppy measurements lead to random error.Random error means that for any particularbit of data, the error may be in either direc-tion; some measurements will be largerthan the real number, and some will besmaller. If the survey workers rounded eachheight measurement to the nearest 5 cm,some heights would be recorded as lowerthan they truly are and others would berecorded as higher than they truly are. Achild who is actually 72 cm long would berecorded as 70 cm. Another child who is 73cm would be recorded as 75 cm. On theother hand, systematic error means thatmeasurements, on average, are biased inone direction or the other. For example, if

the measuring tape were affixed to theheight board so that its lower end with the“0” mark was 5 cm above the footboard,every child would be measured as 5 cm lon-ger or taller than they actually were. Thiswould produce a systematic error withevery measurement biased upwards.

Random error does not affect the average value,but does lead to a greater width in the distribu-tion of the measurements, which is reflected ina larger standard deviation. Therefore, if surveyworkers are not careful with measurements ofheight or age (i.e., if they introduce randomerror), the standard deviation of the height-for-age Z-score will be greater than it should be,which reduces the precision of estimates (i.e.,the confidence interval gets wider).

The standard deviations of weight-for-height, height-for-age and weight-for-age Z-scores in the reference population are, bydefinition, 1.0. If the measurements used tocalculate the height-for-age Z-score (i.e.,height or age), have substantial randomerror, the standard deviation of the height-for-age Z-scores for the survey sample willbe substantially greater than 1.0. Many nutri-tion computer programs, such as EpiNutfrom Epi Info, calculate Z-scores based onthe measurements of children. Once theseare calculated and added to the data file, anystatistical computer program can calculatethe standard deviation of the Z-scores. Thisstandard deviation should be included in thenutrition survey report so that readers canjudge the amount of random error in themeasurements made during the survey.For example, in the Somalia survey shownin Figure 3.6, the standard deviation for theheight-for-age Z-scores was 1.63.Inaccurate age information leads to randomerrors in the measurement of age and there-fore in the height-for-age Z-scores. On theother hand, the standard deviation ofheight-for-age in the Macedonia survey was1.28. In this survey, the age of each childwas much more accurate, leading to lessrandom error and a lower standard devia-tion for the height-for-age Z-score. A stan-

dard deviation for height-for-age or weight-for-height greater than about 1.4 probablyindicates that measurement of either age,weight or height was not very accurate.

ANALYSE DISTRIBUTION OF DECIMAL OF HEIGHTEven though survey workers should alwaysbe trained to measure and record height tothe nearest millimeter, many tend to roundheight to the nearest full centimeter or one-half centimeter. As a result, a disproportio-nate number of height measurements willend in .0 and/or .5, sometimes referred toas digit preference. An analysis of the deci-mal for the height measurements can tellsurvey managers if this error is common inthe survey data. The distribution of thedecimal for the Somalia survey is shown inFigure 3.8. Note the disproportionate num-ber of children whose height measurementended in 0 or 5. Figure 3.9 shows a similardistribution for a survey done in Mongolia.There is also a preference for rounding theheight measurement to the nearest centime-ter in this survey, but it is not so pronouncedas in the Somalia data. Moreover, there seemsto be no rounding to the nearest 0.5 cm. Suchan analysis should be presented in the surveyreport in order to inform readers about theprecision of the height measurements.

98

Figure 3.8 Distribution of Decimal of Height Measurements for Children Under 5 years of Age, Bardera,Somalia, January 1993

90

80

70

60

50

40

30

20

10

0

0 1 2 3 4 5 6 7 8 9

Num

ber o

f chi

ldre

n

Decimal of height measurement

Calculate the proportion of recordswith extreme Z-scoresInevitably, mistakes will be made in eithermeasuring children, transcribing the results todata collection forms or entering the data intoa computer. Certain values for anthropometricindices are highly improbable or incompatiblewith life and should not be included in theanalysis of Z-scores. For example, it is extreme-ly unlikely that a child would have a weight-for-height Z-score of -6.0. Many computer pro-grams which calculate anthropometric Z-sco-res or percents of median from anthropometricmeasurements will somehow “flag” thosechildren whose Z-scores fall outside certainboundaries, indicating that the calculatedindex is probably the result of a mistake in thedata and is not real. Epi Info™ creates a newfield in the data set when calculating the Z-sco-res from measurements. This field is called

“flag,” and it contains a code number depen-ding on which anthropometric index is out ofbounds. For Epi Info, the “flag” variable isbased on an older definition of extreme valuesand should be used with caution.

A more acceptable method of flagging is tocreate acceptable boundaries for the Z-scoresdepending on the data set. For areas with avery high prevalence, an alternative appro-ach to defining Z-scores is any Z-score morethan 4 standard deviations below or abovethe mean Z-score for that index. For exam-ple, if in a particular survey, the averageweight-for-height Z-score is -2.0, then thosechildren with Z-scores less than -6.0 andgreater than +2.0 should be excluded fromthe analysis. Alternately, fixed boundariescan be used to flag records; Table 3.3 detailsboundaries as recommended by WHO.


DESIGNING A SURVEY

99

3CHAPTER

Figure 3.9 Distribution of Decimal of Height Measurements for Children 6-59 Months of Age, MongoliaNationwide Nutrition Survey, 2001

180

160

140

120

100

80

60

40

20

0

0 1 2 3 4 5 6 7 8 9

Num

ber o

f chi

ldre

n

Decimal of height me asurement

Table 3.3 Boundaries for flagging outlying data

Index

Weight-for-heightHeight-for-ageWeight-for-age

Lower boundary

< -4.0< -5.0< -5.0

Upper boundary

> +5.0> +3.0> +5.0

Source: Physical Status: The Use and Interpretation of Anthropometry - Report of a WHO ExpertCommittee. WHO, Geneva. 1995

Of course, each flagged record with anextreme Z-score should be investigatedto correct errors in data recording or dataentry, and obvious errors should be cor-rected in the computer data file. Thosechildren whose flags are not the result ofobvious errors must be excluded fromthe analysis of the Z-scores. The final sur-vey report should present the proportionof children with extreme Z-scores.

14. ANALYSE THE DATA AND PRESENTTHE RESULTSOnce the survey data are entered into acomputer program, checked, and theaccuracy of measurements confirmed,the data can be analysed to describe chil-dren's nutritional status. In the finalreport, it is customary to present resultsin certain standardized ways.

• Primary results often will be the pre-valence of wasting, stunting andunderweight, both <-2 SDs and <-3SDs. These prevalences are merelythe number of children with any mal-

nutrition and the number of childrenwith severe malnutrition, respective-ly, divided by the number of childrenwith a valid anthropometry index.For example, consider a survey inwhich 397 children have validweight-for-height Z-scores, 24 have aZ-score between -3.0 and -2.0, and 7have a weight-for-height Z-score < -30. The prevalence of wasting wouldbe 7.8 percent [(24 + 7) / 397]. Theprevalence of severe wasting wouldbe 1.8 percent (7 / 397).

• In addition, you should present themean anthropometric Z-scores andtheir standard deviations for all chil-dren without oedema. This numbergives an overall summary of thedegree of acute malnutrition amongchildren in the survey population.

• You also should present a graph sho-wing the distribution of anthropome-try Z-scores among children withoutoedema. Figure 3.10 shows oneexample of such a graph.

100

Figure 3.10 Distribution of Weight-for-Height Z-Scores, Children Under 5 Years of Age, Badghis Province,Afghanistan, March 2002

20

15

10

5

0

Sample

-3.0 -2.0

-6,00

to -5

,0

-4,99

to -4

,5

-4,49

to -4

,0

-3,99

to -3

,5

-3,49

to -3

,0

-2,99

to -2

,5

-2,49

to -2

,0

-1,99

to -1

,5

-1,49

to -1

,0

-0,99

to -0

,5

-0,49

to 0.

0

0,01 t

o 0,5

0,51 t

o 1,0

1,01 t

o 1,5

1,51 t

o 2,0

2,01 t

o 2,5

2,51 t

o 3,0

3,01 t

o 3,5

3,5 to

6,0

NCHS/CDC/WHO reference

Perc

ent

Weight-for-height z-score

• The prevalence, mean Z-score andgraph of the distribution of height-for-age Z-scores also can be presen-ted for chronic malnutrition.

• If hemoglobin or serum retinol aremeasured on survey subjects, thecomparable results for these valuescan be presented (the prevalence ofabnormally low values, the meanvalue, and a graph showing thedistribution of all the values).

• It is very important to calculate con-fidence intervals for the major out-comes measured in a survey and topresent them in the verbal presenta-tions and the final report. This willgive the reader or listener an idea ofthe precision of the estimates. Mostcomputer programs used to analysedata will give confidence intervalsaround prevalence rates; however,many such programs assume thatyour data come from a sample cho-sen by simple or systematic randomsampling. If you have done clustersampling, the confidence intervalscalculated by these programs are notcorrect. For cluster samples, youmust use a computer program whichtakes into account the cluster sam-ple and calculates the appropriateconfidence intervals. Such computerprograms will ask you the name ofthe variable which identifies the clu-ster to which each unit of analysisbelongs. For example, the Analysismodule of Epi Info will calculateconfidence intervals, but they willnot be correct for a cluster survey. InEpi Info for DOS (version 6), youmust use the CSample module tocalculate confidence intervals forcluster surveys. One of the boxes inCSample asks for the PSU, or prima-

ry sampling unit. You should indica-te in this window what variable con-tains the cluster number. In Epi Infofor Windows, you must use the“Complex Sample Frequencies,”“Complex Sample Tables,” or“Complex Sample Means” com-mands. SAS and Stata have complexsample commands and SPSS has anoption complex sampling module.

Confidence intervals for mortalityrates may be more difficult toobtain. If mortality data are collec-ted as recommended in this manual,you should be able to calculate thenumber of deaths and the popula-tion denominator in each house-hold. These numbers are then com-bined for all the households in thesurvey sample and analysed as acluster sample. The total number ofdeaths divided by the total house-hold population gives an estimate ofthe proportion of the populationwhich died during the recall period.Computer programs, such asCSample or Epi Info for Windows,should give a correct confidenceinterval for this proportion thattakes into account the cluster sam-pling. The lower and upper ends ofthis confidence interval then can beconverted to death rates expressedas the number of deaths per stan-dard population size per standardtime period, such as the number ofdeaths per 1,000 per year.


DESIGNING A SURVEY

101

3CHAPTER

102

COMMON MISTAKE #17:Comparing results of different surveys or subgroups within a survey withoutcalculating P values or providing confidence intervals for each estimate.

Explanation: The results of two surveys may look different, but this difference may bedue solely to sampling error. The P value tells you the likelihood that the differenceobserved between two surveys is due only to sampling error. If the P value is 0.05, thenthe likelihood that the difference is due only to sampling error is 0.05, or 5 percent.

If there is no statistical difference between the results of two surveys done in twodifferent populations, then there may be no real difference between these twopopulations. If the two surveys are conducted in the same population at two diffe-rent points in time, then there may have been no real change over time. This isobviously very important when formulating programmatic interventions.Expenditure of time, money and other resources should not be based on differen-ces between surveys that do not reflect a real difference in populations.

Solution: Always judge the statistical significance of any observed difference by either comparing confidence intervals or calculating a P value for the difference.

STOP

GO

COMMON MISTAKE #18:Calculation of confidence intervals or P values without accounting for clustersampling.

Explanation: Most computer programs, such as Epi Info, that are used to calcula-te epidemiologic measures will display confidence intervals and P values. However,these programs usually assume that the sample was selected by simple randomsampling. This may be inaccurate if the sample was chosen using cluster samplingbecause cluster sampling usually results in greater statistical variance, and there-fore less precision, than simple random sampling. Because the variance is greater,confidence intervals will be wider with cluster sampling. Therefore, if cluster sam-pling was done and the computer program does not take this into account, the con-fidence intervals are inaccurate and should not be used.

Nonetheless, many survey managers do not know this and will copy the confiden-ce intervals from the computer screen into the survey report, thus giving the rea-der the impression that the precision of the prevalence estimate is greater than itactually should be. This may be especially important when comparing the resultsfrom two surveys. Using the wrong confidence intervals may lead the reader todecide that the two surveys' results are truly different, when the difference mayonly be the result of sampling error.

Solution: When calculating confidence intervals and P values for surveyresults obtained by cluster sampling, be sure you are using formulas orcomputer software that account for cluster sampling.

STOP

GO


DESIGNING A SURVEY

103

3CHAPTER

COMMON MISTAKE #19:Not reporting the number of children with oedema in the survey report.

Explanation: As mentioned above, malnourished children with oedema have a muchworse prognosis than malnourished children without oedema. The report of a nutri-tion survey should always include the number of children with wasting who haveoedema. This is especially important in an emergency situation or where the preva-lence of protein-energy malnutrition is relatively high since this information will tellthe reader what type of wasting is predominant (marasmus vs. kwashiorkor).

Solution: Be sure to include in the survey report the number, as well as theproportion, of children who had oedema.

STOP

GO

15. DISSEMINATE FINDINGS IN PRESENTATIONS AND REPORTSThe final part of a nutrition survey is pre-senting the findings in oral presentationsand in a final report. Often, distribution ofa preliminary report of the key findings isrecommended as soon as data analysis hasbeen completed because a final report maytake a long time to complete. The results ofthe survey should be presented in a stan-dardized format so as to enable easy com-parisons between the results of differentsurveys. One suggested format is shown inthe outline below and can be used for bothoral presentations and written reports:

Executive summaryThis section of the report should be short(one or two pages), and should include aclearly presented summary of all importantinformation in the report. Be aware that 90percent of readers probably will look at thissection only. Write the summary last, afteryou have finished the rest of the report, andinclude the following information: the geo-graphic area covered, the date and theobjectives of the survey, the methodologyused, the main results and recommenda-tions. One way to quickly and clearly pre-sent the major results is to create a table.

Report introductionThe context in which the survey was car-ried out should be described. What popu-lation was surveyed, at which period andin which geographical area? The introduc-tion should be scene-setting, so thatsomeone who has never been to the areacan understand how the surveyed commu-nity lives, what has happened to them andwhy this survey was done.

Objectives of the surveyThe objectives of the survey should be stated clearly.

MethodologyA straightforward description of themethods employed, including samplingtechniques, is necessary so that readerscan judge the validity of the survey'sresults and have a clear reference for futu-re comparison. Include selection criteriafor inclusion in the survey sample andclear definitions of all the health and nutri-tion outcomes measured in the survey.Describe what measurements were taken,by whom and using what instruments.Details on the training of the survey mem-bers should be included, noting whetherstandardization exercises and a trial/pilot

survey were conducted. Describe how thequestionnaires were designed and pilo-ted. Standard definitions of indicatorsused and cutoff values used to definemalnutrition also should be included.The methods should be in such detail asto enable someone to replicate your sur-vey in the future using only the finalreport as a guide.

ResultsThis section is mainly graphs and tablesillustrating the results of analysis. Theresults often include:1) A description of the response rate,

including the number of subjects orhouseholds chosen for the sample, thenumber from whom data were collec-ted and the reasons for non-response;

2) A description of the survey subjectsfrom whom data were collected;

3) The overall prevalences of the majoroutcomes;

4) Comparison of subgroups within thesurvey sample, such as comparison ofmales and females or different geogra-phic locations, if the sample size per-mits such comparison with any stati-stical power; and

5) A description of potential causal fac-tors investigated in the survey.

LimitationsThis section should discuss the limita-tions that were faced, such as thoseencountered during team selection andtraining, survey design, sampling andanalysis. Examples of this could includean incomplete or outdated samplingframe, unexpected population movement,security or staff constraints or other fac-tors affecting access to the sample popu-lation.

DiscussionThe discussion puts the results back intocontext. The aim of the discussion is to

explain the results seen (for example, prevalence of malnutrition and mortalityrates) in terms of the causes of malnutri-tion - health, care environment, and foodsecurity. Organize your discussion byaddressing the following questions (notan exhaustive list):1) Is the level of malnutrition typical

(referring back to previoussurveys/baseline levels)?

2) Is the level of mortality typical?3) What are the major acute causes of

malnutrition and mortality (takinginto account causes already addressedby other interventions)?

4) What are the prospects for the comingmonths?

5) Who is worst affected?6) What are the chronic causes of mal-

nutrition?7) What does the community recom-

mend?

Much of the information for the discus-sion will come from referring back to theresults section and considering the majorfindings in light of other informationgathered from prior surveys, other typesof assessments, observation, discussionswith community leaders and survey wor-kers, etc.

Conclusions and recommendationsThis section should restate the majorconclusions and make specific, opera-tional recommendations for what isneeded most urgently in the surveyedpopulation.

Reports should be written and submittedin a timely fashion to prevent any delayin the intervention. Reports for emergen-cy nutrition surveys should be availablewithin one month after the survey datacollection has been completed. Baselinesurvey reports may not be needed sorapidly.

104

REFERENCES

Surveys in general

Boss LP, Toole MJ, Yip R. “Assessments of mortality, morbidity, and nutritional statusin Somalia during the 1991-1992 famine”. JAMA 1994;272:371-376.

United Nations Department of Economic and Social Affairs, Statistics Division.Household Sample Surveys in Developing and Transition Countries. Studies in MethodsSeries F, No. 96. United Nations, New York. 2005. Available at:http://unstats.un.org/unsd/hhsurveys/pdf/Household_surveys.pdf

Spiegel PS, Salama P, Maloney S, van der Veen A. “Quality of malnutrition assessmentsurveys conducted during famine in Ethiopia”. JAMA 2004;292:613-618.

Gorstein J, Sullivan K, Parvanta, I, Begin F. Indicators and Methods for Cross-SectionalSurveys of Micronutrients. Micronutrient Initiative and the Centers for Disease Controland Prevention. Ottawa and Atlanta. 2005 (in press).

“Measuring Mortality, Nutritional Status and Food Security in Crisis Situations: SmartMethodology, Version 1, June 2005”. UNICEF and USAID, New York and Washington,D.C. 2005. Available at: http://www.smartindicators.org/.

Sampling

Bennett S, Radalowicz A, Vella V, Tomkins A. “A computer simulation of householdsampling schemes for health surveys in developing countries”. International Journalof Epidemiology 1994;23:1282-1291.

Binkin N, Sullivan K, Staehling N, Nieburg P. “Rapid nutrition surveys: how many clu-sters are enough?” Disasters 1992;16:97-103.

Brogan D, Flagg EW, Deming M, Waldman R. “Increasing the accuracy of the expanded pro-gramme on immunization's cluster survey design”. Annals of Epidemiology 1994;4:302-311.

Milligan P, Njie A, Bennett S.“Comparison of two cluster sampling methods for health surveys in developing countries”.International Journal of Epidemiology 2004;33:1-8.

Magnani R. “Sampling Guide. Food and Nutrition Technical Assistance”. Washington,D.C. December 1997. Available at http://www.fantaproject.org/publications/sampling.shtml.

Design effects

Katz J, Carey VJ, Zeger SL, Sommer A. “Estimation of design effects and diarrhea clusteringwithin households and villages”. American Journal of Epidemiology 1993;138:994-1006

Katz J, Zeger SL. “Estimation of design effects in cluster surveys”. Annals ofEpidemiology 1994;295-301.

Kaiser R, Woodruff BA, Bilukha O, Spiegel PB, Salama P. “Using design effects from pre-vious cluster surveys to guide sample size calculation in emergency settings”. Disasters(in press).


DESIGNING A SURVEY

105

3CHAPTER

As an organization working with aresults-based management approach,WFP is committed to not only strivingto achieve positive outcomes for itsbeneficiaries, but also to using outputand outcome indicators to monitor itsperformance. Such indicators enableus to understand whether the plansand strategies are being implementedin a way that leads to desired projectimpact.

One of the first steps of programmedesign involves formulating objectivesand defining the means to achievethose objectives (what is often calledputting together a “results chain”).Many of WFP's programmes (including

most emergency operations andMother and Child Health/Nutritionprogrammes) have specific objectivesrelated to either reducing or stabilizingthe prevalence of malnutrition or tosaving lives in emergencies. As such,the main outcome indicators of interestare malnutrition or mortality rates,which generally are best collectedthrough surveys as outlined in thismanual.

This section provides guidance on crit-ical issues related to the collection,interpretation and use of survey resultsfor decision-making in the context ofWFP's operations.


Using and Interpreting SurveyResults for Decision-Making

4CHAPTER

Key messages• It is important to review the quality of nutrition surveys before using them for deci-

sion-making. One should not automatically assume that survey data is useful.

• There are many factors to consider when interpreting survey data, including pre-exi-sting information on the prevalence of malnutrition in a given area, country or region;the magnitude of changes over time; the geographical coverage of a survey; the sea-sonality of malnutrition and information on the different potential causes of malnutri-tion in an area.

• When assessing trends of malnutrition or mortality in a given area, a rigorous pro-cess of comparison between the multiple surveys done over time is necessary. Somefactors to consider include: (1) whether the same population/geographic area wascovered by both surveys; (2) whether the sample size was adequate enough to ensu-re that comparisons are possible; (3) whether any differences in the prevalenceobserved over time were statistically significant or programmatically meaningful;and (4) whether any factors have changed over time that might be responsible forleading to the change in prevalence.

108

WHAT INFORMATION IS AVAILABLE TOHELP YOU TO DECIDE WHETHER A NEWSURVEY MAY BE NECESSARY?When designing intervention projectswith an objective related to malnutri-tion or mortality, an essential first stepis to gather and review existing infor-mation to determine whether a newsurvey is truly needed. There are anumber of places to look to determinewhat surveys may be available fromyour country or working area.Consultation with partners includingnon-governmental organizations(NGOs), governments and other agen-cies (particularly UNICEF) may revealthat historical data already exist thatcan be used either to help you to esta-blish a baseline understanding of thesituation or to help to interpret your fin-dings should you choose to do a survey.

IS AVAILABLE INFORMATION SUFFICIENT FOR YOUR NEEDS?Once you have collected available nutri-tion surveys or other information, it iscritical to define the potential uses thatyou may have for the nutritional ormortality data. Measuring trends inmalnutrition or mortality over the cour-se of a project is often the main objecti-ve of nutritional surveys, but it is notthe only potential need that you mayhave for such data. Nutritional dataalso can be useful in making decisionsabout the relative vulnerability of popu-lation groups living in different geogra-

phic areas or for planning beneficiarynumbers for supplementary or thera-peutic feeding programmes. Therefore,when deciding whether a new survey isnecessary, it is important to considerwhat your real needs are.

Different types of surveys may have dif-ferent uses for you. Data from nationalsurveys, while useful for giving you anidea of the general situation of thecountry, generally does not have anadequate sample size on a local level tobe much use for baseline/follow-uppurposes. Local surveys, such as thosedone in specific refugee camps, may bevery useful for your purposes, althoughin larger operations they may only giveyou a relatively small picture of the ove-rall situation.

Clinic surveillance data also may beavailable; while such data is generallyunreliable for getting an accurate ideaof the prevalence of malnutrition, itmay be quite useful for either persua-ding decision-makers to undertake anutrition survey or for understandingseasonal patterns of malnutrition.

Figure 4.1 outlines decision-making stepsfor three likely scenarios in which you: • may use existing survey data; • need to conduct a survey; or • can make decisions without using

survey data.


USING AND INTERPRETING SURVEY RESULTS FOR DECISION-MAKING

109

4CHAPTER

ASSESSING THE QUALITY OF SURVEYSAnother essential part of the process ofdetermining whether a survey is necessaryis to review the quality of any existing sur-veys. The interpretation and critiquechecklist (Checklist 4.1) will help you todecide whether the content and quality ofa survey are sufficient. For the overall critique of a survey report, it may also beworth answering the following questions:• Which organizations conducted

the survey? • How much experience do those

organizations have in nutrition and health?

• How long have they been in the country and in the specific area of interest?

• Why did they conduct the survey?

Knowing the area and having been ableto visit and talk to local people willalso help in interpreting a surveyreport. The report should make sensein the context of what you have seenand recent history. It also should makesense to the people living in the area.The survey should cover the geograph-ic areas and topics about which youneed information.

Even if the survey's quality in terms ofmethodology may appear to be suffi-cient, it may be deficient in other ways.It may not include a food securityassessment, for example, or it may coveronly part of the area where you plan anutrition program. Therefore, you maydecide to do a larger survey with a foodsecurity component.

Figure 4.1 Is a survey necessary?

Survey data available

Use interpretation and critique checklist to review quality of survey

Survey contents and quality sufficient

Collect complementary data on causes of malnutrition and mortality

Survey contents and/or quality insufficient

Conduct survey, usingChapter 3 guidance

Collect complementary data on causes of malnutrition and mortality

No survey data available Survey currently not necessary (e.g. population needs obvious, immediate intervention necessary)

Decision making process

110

8 Reported regularly in Nutrition Information in Crisis Situations (NICS). Geneva: Standing Committee on Nutrition, UnitedNations System. Available at URL: http://www.unsystem.org/scn/Publications/html/rnis.html.

Check list 4.1 Key questions to be considered when interpreting a survey report

Objectives3Have the objectives been stated clearly and are they realistic?

Survey Planning & Implementation Issues3Have the survey area and target group been specified? 3Has the questionnaire been translated into local language and back-translated into the original language?3Has the questionnaire been piloted in the survey area (but with people who were not part of the sample)?3Has the training been long enough (minimum 3-5 days depending on background of personnel)?3Were enough qualified supervisors available to assure quality of measurements and interviews?3Were all team members trained in the same way?3Were the interviewers able to read questions in a standardized way from the questionnaire?

Survey Methodology3Has the sampling frame been adjusted for recent population movements?3Was the sample representative of the target population, i.e., nobody was left out in the sampling approach?3Has the sample size calculation been described in detail, including sample size calculations that are based

on different outcomes?3Is the sample size large enough for appropriate precision? If the sample size calculation has not been described,

did the survey follow international standards, i.e., 30 x 30 cluster sample for nutrition and mortality surveys?(see section on sample size in Chapter 3)

Survey Reporting3Has the proportion of severely malnourished children with oedema been reported?3Are case definitions provided, and do they meet international standards?3Was software used for analysis that allows adjusting for cluster survey design?3Have weight-for-height Z-scores been used to measure malnutrition?3Has oedema been included in the definition of severe acute malnutrition?3Has the survey questionnaire been provided in the report?

Results3Do the results reflect the objectives of the survey?3Does the report contain standard information (i.e., survey area, date of survey, population, survey conducted by,

acute malnutrition, acute severe malnutrition, oedema, measles immunization coverage, vitamin A distribution coverage,women's anthropometric status, crude mortality, under-5 mortality) 8

3Have 95% confidence intervals been reported with prevalences and rates?3Does the report provide detailed information and discussion of causes of malnutrition and mortality?

Discussion3Does the report include a discussion of results, including limitations of the survey?3If results are compared to a baseline, is the quality of the baseline information discussed

(e.g., organization that conducted assessment, methods, one or multiple years, did those years qualify as baseline)?

Conclusions3Are conclusions based on results?3Are conclusions realistic (e.g., a solid interpretation of what the data can provide and what it cannot)?

Recommendations3Are recommendations based on science and best practices and not driven by politics?3Are the recommendations useful, i.e., could they have been made without the study?



111

4CHAPTER

INTERPRETING SURVEY RESULTSOnce you have determined that a givensurvey (or surveys) are reliable, the nextstep is interpreting those survey results.This section outlines several major con-siderations that you may want to use aspart of this process, including:

1. Comparing survey findings againstexisting information;

2. Using the conceptual framework ofmalnutrition as a tool to help inter-pret complementary information andto consider the relative importanceof different causes of malnutrition;and

3. Interpreting trends over time.

COMPARING SURVEY FINDINGSAGAINST EXISTING INFORMATIONThe following tables provide informa-tion by region on the average rates ofmalnutrition (and variation in thoserates) that may be useful for interpre-ting findings of your surveys. Itshould be noted that many regions

have extreme variation from countryto country in the prevalence of mal-nutrition (or mortality rates), andthat it is generally much more usefulto use country-specific data thanthese estimates for comparisons. It isalso important to note that whilethese tables are useful for understan-ding what a “normal” nutritionalsituation is in these regions, it isoften the case that the “normal”situation may be extremely poor, andtherefore not necessarily a good stan-dard to compare against Tables 4.1-4.4 show regional malnutrition andmortality rates recently published bythe United Nations StandingCommittee on Nutrition (SCN)4 andthe United Nation's Children's Fund(UNICEF)5. Because of differences ingeographical coverage and methods,Tables 4.1-4.4 should never be usedas the reference baseline alone, butmay help with interpretation whenused along with a variety of differentinformation collected.

112

* Defined as <-2 standard deviations or Z scores of the weight-for-height median value of the NCHS/WHOinternational reference data † Not available

Table 4.1 Projected prevalence and 95% confidence intervals of acute malnutritionin preschool (0-5 years) children, 2000 and 20054

Region

AfricaEasternMiddle NorthernSouthernWesternAsiaEasternSouth-Central South-EastWesternLatin America and CaribbeanCaribbeanCentral AmericaSouth AmericaOceaniaAll developing countries

Acute malnutrition*(%)

8.37.69.16.24.910.39.22.214.08.94.21.62.51.71.4NA†8.2

95% confidenceinterval

7.3-9.26.3-9.26.6-12.63.5-10.63.3-7.49.0-11.97.7-10.72.1-2.411.3-17.26.5-12.21.9-8.71.1-2.10.09 1.2-2.40.8-2.4NA7.2-9.3

Acute malnutrition*(%)

9.58.711.98.06.610.28.91.813.35.13.91.52.41.61.4NA8.3


8.2-10.76.8-11.18.4-16.74.5-14.04.4-10.29.0-11.67.3-10.51.6-1.910.7-16.53.7-6.91.4-10.40.9-2.10.09 0.9-2.70.7-2.6NA7.2-9.4

2000 2005



113

4CHAPTER

* Defined as <-2 standard deviations or Z scores of the height-for-age median value of the NCHS/WHO internationalreference data † Not available

Table 4.2 Projected prevalence and 95% confidence intervals of chronic malnutritionin preschool (0-5 years) children, 2000 and 20054

Region


Chronic malnutrition*(%)

35.244.437.821.724.632.930.114.839.721.318.713.77.420.411.3NA†29.6


32.5-38.037.6-51.433.7-42.116.1-28.621.5-28.130.2-35.727.1-33.113.9-15.834.4-45.317.0-26.010.9-30.19.1-18.43.8-14.112.5-31.56.5-18.9NA27.5-31.7

Chronic malnutrition*(%)

34.544.435.819.124.332.025.710.034.518.116.111.85.718.09.6NA26.5


31.7-37.437.6-51.433.0-38.613.5-26.520.4-28.628.4-35.722.5-28.99.3-10.729.0-40.514.3-22.57.8-30.37.0-16.52.7-11.510.8-28.44.9-18.2NA24.2-28.7

2000 2005

114

*Defined as <-2 standard deviations or Z scores of the height-for-age median value of the NCHS/WHO internationalreference data † Not available

Table 4.3 Projected prevalence and 95% confidence intervals of underweight in preschool (0-5 years) children, 2000 and 20054

Region


Underweight*(%)

24.229.226.19.713.727.127.99.340.827.411.36.16.19.24.6NA†24.8


21.9-26.424.6-34.321.8-30.84.6-19.49.7-19.024.2-30.324.0-31.78.8-9.933.5-48.523.4-31.85.0-23.74.0-8.13.3-10.85.2-15.72.9-7.4NA22.2-27.3

Underweight*(%)

24.530.625.38.613.626.824.86.536.523.910.65.04.77.93.7NA22.7


22.1-26.825.7-35.821.6-29.33.6-19.59.6-18.823.6-30.320.8-28.86.1-6.929.3-44.419.9-28.53.3-28.93.2-6.82.5-8.74.3-14.02.3-6.1NA20.1-25.4

2000 2005



115

4CHAPTER

USING A CAUSAL FRAMEWORK TO CONSIDER THE RELATIVE IMPORTANCE OF DIFFERENT CAUSES OF MALNUTRITIONA causal framework (see ConceptualModel of Malnutrition) is necessary tounderstand long-term causes of malnu-trition in the population and decide onthe information to be gathered in asurvey. Non-food causes of malnutri-tion and mortality may aggravate thesituation rapidly even if rates are notalarming at the moment.In an individual, malnutrition is theresult of inadequate dietary intake, infec-

tion or a combination of both. The wide-ly used UNICEF conceptual model of thecauses of malnutrition (Figure 4.2)organizes and explains malnutrition bylayers of causes. Malnutrition is causedby more than a simple lack of food; thereare other underlying causes that can con-tribute to adverse nutritional outcomes.The multisectoralcauses of malnutritioninvolve dietary intake, poor health, envi-ronmental factors and care practices. Foran individual to be adequately nour-ished, the underlying causes of malnutri-tion - health, food, care - need to beaddressed in tandem.

Table 4.4 Baseline reference mortality data by region5

Region

Sub-Saharan AfricaMiddle East and North AfricaSouth Asia East Asia and Pacific Latin America and CaribbeanCentral and Eastern EuropeanRegion/CIS and Baltic StatesIndustrialized countriesDeveloping countriesLeast developed countries World

CMR(deaths/10,000/day)

0.440.160.250.190.160.30

0.250.250.380,5

CMRemergencythreshold(2 x crudemortalityrate)

0.90.30.50.40.30.6

0.50.50.80.5

U5MR(deaths/10,000 childrenunder 5/day)

1.140.360.590.240.190.20

0.040.531.030.48

U5MRemergencythreshold (2 x under-5 mortalityrate)

2.30.71.20.50.40.4

0.11.12.11.0

2000 2005

116

Underlying causes of malnutritionThe main underlying preconditions to adequate nutrition are food, health andcare; the degree of an individual's orhousehold's access to these precondi-tions affect how well they are nourished.

Food quantity and qualityFood security exists when, at all times,everyone has access to and control oversufficient quantities of good, quality foodneeded for an active and healthy life. Fora household, this means the ability tosecure adequate food to meet the dietaryrequirements of all its members, eitherthrough their own food production orthrough food purchases. Food productiondepends on a wide range of factors,including access to fertile land, availabil-ity of labor, appropriate seeds and tools,and climatic conditions. Factors affectingfood purchases include household

income and assets as well as food avail-ability and price in local markets. Inemergency situations, other factors -including physical security and mobility,the integrity of markets and access toland - may come into play.

Health and sanitation environmentAccess to good, quality health services,safe water supplies, adequate sanita-tion and good housing are precondi-tions for adequate nutrition. These pre-conditions are affected by the existingprimary health infrastructure, the typesof services offered, their accessibilityand affordability to the population, andthe quality of these services. Key envi-ronmental issues include the degree ofaccess to adequate quantities of safedrinking water, adequate sanitation andadequate shelter. Health and environ-mental factors influence incidence and

Source: the States of the world’s Children 1998 - UNICEF

9 The State of the World's Children 1998 (URL: http://www.unicef.org/sowc98/). New York: UNICEF; 1998.

Child malnutriction,death and disability

Outcomes

Immediatecauses

Underlyngcauses athouseshold/family level

Basic causesat societal level

Inadequate maternaland child-care practices

Poor water/sanitationand inadequate healtservice

Insufficient accesto food

Inadeguatedietary intake

Quantity and quality of actual resources-human, economic and organizational-and the way they are controlled

Inadequate and/or inappropriateknowledge and discriminatoryattitudes limit household accesto actual resources

Political, cultural, religious,economic and social systems,including women’s status, limit theutilization of potential resources

Disease

Potential resources: enviroment,technology, people

Figure 4.2 Conceptual model of the causes of malnutrition, UNICEF 9

4severity of disease. Health and nutri-tion are closely linked in a “malnutri-tion-infection cycle” in which diseasescontribute to malnutrition and malnu-trition makes an individual more sus-ceptible to disease.

Social and care environmentMalnutrition can occur even whenaccess to food and healthcare is suffi-cient and the environment is reason-ably healthy. The social and care envi-ronment within the household and localcommunity also can directly influencemalnutrition. Appropriate childcare,which includes infant and child feedingpractices, is an essential element ofgood nutrition and health. Cultural fac-tors and resources, such as income,time and knowledge, influence caringpractices. The values of the society dic-tate the priority given to the care ofchildren, women and the elderly.Attitudes to modern health services,water supplies and sanitation alsoaffect caring practices.

Immediate causes of malnutritionOn an immediate level, malnutritionresults from an imbalance between theamount of nutrients needed by the bodyand the amount of nutrients being intro-duced or absorbed by the body.Adequacy of food intake relates to:

a) the quantity of food consumed; b) the quality of the overall diet with

respect to various macro- andmicronutrients;

c) the form of the food consumed,including palatability and energy density; and

d) how frequently the food is consumed.

However, malnutrition is not synony-mous with a lack of food. In an individ-ual, malnutrition is the result of inade-quate dietary intake, disease, or both.Health and nutrition are closely linked.Disease contributes to malnutritionthrough a loss of appetite, malabsorp-tion of nutrients, loss of nutrientsthrough diarrhea or vomiting, orthrough altered metabolism.Malnutrition, in turn, makes an individ-ual more susceptible to disease.

While the conceptual framework pro-vides a useful approach for consideringthe causes of malnutrition, its useful-ness depends on the availability ofinformation about those causes. Suchcomplementary information mayinclude quantitative data from sourcesother than surveys and qualitative infor-mation gathered from focus groups,interviews with key persons, and per-sonal observations or observations bynational or international colleagues whoknow the situation well. Sources of suchdata can vary widely and not all typesare equally useful for all types of WFPneeds. Even within the scope of WFPneeds, uses of data can differ dependingon the context of the data collection andobjectives for the use of the informa-tion. A matrix (Table 4.1) providesrough guidance on suggested use of dif-ferent sources of information for a vari-ety of WFP purposes. This is by nomeans exhaustive, nor regulatory.



117

CHAPTER

118

Complementary data may also includethe causal framework or pathway of mal-nutrition and a seasonal calendar:

A seasonal calendar provides informa-tion about harvests and hungry periods,seasonal disease epidemics and otherevents that may affect food insecurity(e.g., food distributions).

All primary and secondary informationshould be checked against other sour-ces of information and confirmed withpartners.

Interpreting trends over timeIn general, when you compare two ormore surveys to assess trends in malnu-trition and mortality in a given popula-tion, these surveys must have usedmethodologies that meet similar qualityand procedural standards. For example,did both samples use representativesampling methods and comparable

measuring techniques? Consider whe-ther the surveys used or collected datafor the same:• definitions of malnutrition (i.e., Z score

[preferred] or percent of the median); • population; • age groups; • geographic area; and • season.

Whenever possible, WFP and partnerorganizations should conduct a baselinesurvey in the area where the nutrition pro-gram or intervention will later be conduc-ted. In acute emergencies, interventionsmay have to start immediately, and a sur-vey will be conducted as soon as is possi-ble. Any initial survey should considersample size requirements for baseline andfollow-up surveys (see Chapter 3).

If conducting a baseline survey is notpossible, other sources for surveys maybe used. Note that national survey

Table 4.1 Matrix of sources of information for WFP purposes

Facility Based Data

Sample Survey

ENA Assessment

VAM Survey

Nutrition Surveillance

National Survey

Operations/ProgramManagement

+ + +

+ + +

+

+

+ + +

+

Targeting/Prioritization

+ +

+ +

+ + +

+ + +

+

+ +

RBM/CorporateReporting

+

+ + +

+

+ +

+ +

+ +

Advocacy

+

+ + +

+ + +

+ + +

+ +

+ + +Popu

latio

n <

>

leve

lIn

divi

dual

leve

l

Note: + + + optimal use, + + of some use, + limited use

4results should not be used as a baselinebecause national prevalence may not berepresentative of smaller subsets of thepopulation. If a national survey is con-ducted with survey methodology thatallows for statistical disaggregation intosubsets, then this disaggregated datamight be useful for these purposes.However, consultation with statisticiansor the Nutrition Service in Rome shouldbe undertaken before such a decision ismade. National surveys that usually haveprovincial-level estimates include MEA-SURE DHS (Demographic and HealthSurveys; DHS)1, UNICEF's MultiIndicator Cluster Surveys (MICS)2, or theWHO Global Database on Child Growthand Malnutrition3.

What are some potential differences ingeographical coverage and methods?Comparing current malnutrition preva-lence in a district in southern Ethiopiawith a baseline estimate for the whole ofEthiopia or eastern Africa has little mea-ning. A current survey result may becompared to a previous survey that wasconducted in a different season, or abaseline survey that has covered severalseasons over one or several years.

Differences in measuring mortality are ano-ther potential problem when comparingsurvey results. Child mortality is expressedby UNICEF and WHO as the probability ofdying between birth and 5 years of age per1,000 live births. Conversion to an age-spe-cific mortality rate for children under 5years, as shown in Table 4.4, requires amathematical transformation. That tran-sformation is based on certain assumptionsthat need to be considered as limitationswhen results are discussed.

If 95 percent confidence intervals are provi-ded, as in Tables 4.1-4.3, overlapping con-fidence intervals alone are not sufficient to

assume that there is not a statistically signi-ficant difference between the baseline andthe current survey. However, they at leastindicate that the results of the two surveysmay not be different. Non-overlapping con-fidence intervals show that the two surveysare significantly different. Always determi-ne the statistical difference between a base-line and follow-up survey by either compa-ring confidence intervals or calculating a Pvalue for the difference. Confidence inter-vals should be presented for all outcomesin your survey, independent of their availa-bility from a previous baseline survey.

How should I interpret changes frombaseline to follow-up?Interpretation of changes between baseli-ne and follow-up(s) depends on themagnitude of the change, trends and thecontext. A doubling of the baseline oftenindicates the presence of an acute emer-gency. However, it is essential to verifythat such an increase is actually “real,”as described above.

The causes of malnutrition are complex.Emergency settings and perceptions ofwhat is “typical” or “atypical” in a givencountry or a given district within a coun-try differ. Such differences make it diffi-cult to develop and make available astandardized, generally agreed uponclassification of severity of malnutritionto match to baseline information.However, some internationally agreedupon cut-off values have been developedto provide a guide for classifying malnu-trition and mortality rates. Tables 4.5 and4.6 show malnutrition classifications andmortality benchmarks, as defined byWHO6 and Sphere7. These classificationsare based on acute malnutrition, chronicmalnutrition, underweight and mortality.Population benchmarks for other outco-mes, such as low birth weight or mater-nal mortality, are not available.



119

CHAPTER

120

Table 4.5 Classification of severity of malnutrition in a community by prevalence of acutemalnutrition, chronic malnutrition and underweight for children under 5 years of age 6

Severity of malnutrition

AcceptablePoorSeriousCritical

Acute malnutrition(%) (weight-for-height)< -2 z scores

<55-910-14≥ 15

Chronic malnutrition (%)(height-for-age)< -2 z scores

<2020-2930-39≥ 40

Underweight (%)(weight-for-age)< -2 z scores

<1010-1920-29≥ 30

Table 4.6 Mortality benchmarks 7

Indicator

Crude mortalityrate

Prevalence ofanaemia

Baseline

0.5/10,000/day

1/10,000/day

Benchmark for alert

1/10,000/day

2/10,000/day

Benchmark for critical emergency

2/10,000/day

4/10,000/day

Table 4.7 Classification of public health significance of anemia based on theprevalence of anemia 8

Category of public health significance

NormalMildModerateSevere

Prevalence of anaemia (%)

≤ 4.95.0-19.920.0-39.9≥ 40.0

Note: The prevalence of iron-deficiency disorder is likely to be 2-2.5 times greater than the prevalence ofanaemia

Note: Average baseline value based on sub-Saharan Africa.



121

4CHAPTER

Table 4.8 Classification of public health significance of iodine deficiency disordersbased on the prevalence of goiter or urinary iodine6



Total goiter rate (%)

< 5.05.0 - 19.920.0 - 29.9≥ 30.0

Median urinary iodine levelin school children µg/L (%)

≥ 100.050.0 - 99.920.0 - 49.9< 20.0

Table 4.9 Classification of public health significance of vitamin-A deficiency inchildren (6-71 months) based on the prevalence of night blindness or serum retinol 9



Night blindness (%)

00 < 1≥ 1 < 5≥ 5

Serum retinol < 0.7 µmol/L(20 µg/dL) (%)

< 2≥ 2 < 10≥ 10 < 20≥ 20

At least 95 percent of children 6months to 15 years of age should bevaccinated against measles and havereceived an appropriate dose of vita-min A supplementation7. All infantsvaccinated between 6-9 months of agereceive another dose of vitamin Aupon reaching 9 months. Routine vac-cination programs ensure the mainte-nance of 95 percent coverage.

INTERPRETING MORTALITY IN LIGHT OF MALNUTRITION RATESMany nutrition surveys collect bothmalnutrition and mortality rates, andit is desirable to interpret each ofthese in light of the other. The cau-sal framework of malnutrition and aseasonal calendar are often useful tohelp do this.

Table 4.10 shows possible combinationsof mortality rates and malnutrition preva-lence and their likely causes. In mostsituations, mortality will increase withhigher malnutrition and morbidity; howe-ver, mortality can rise in the setting ofrelatively low prevalence of malnutritionand acute malnutrition may rise withoutsubstantial increases in mortality.10, 11

There has been a misconception thatrising mortality may plateau as malnou-rished children die at a rate equal to therate that non-malnourished children arebecoming malnourished. Observed pla-teaus in the prevalence of malnutritiononly occur at unusually high levels ofboth malnutrition and mortality.Therefore, this scenario was not inclu-ded in Table 4.10.

122

Malnutrition prevalence and mortalityrates should always be interpreted withcaution, in the context of the overall situa-tion and under review of causes of malnu-trition and mortality.

Causes of malnutrition and mortality inclu-de (see causal framework of malnutritionin Chapter 1):• General poverty: This includes poor

infrastructure, housing, clothing, sani-tation and lack of schooling/educa-tion.

• Loss of entitlements: In an economictheory of the causes of malnutrition,famine and death, four legal “entitle-ments” are described by which hou-seholds obtain food. Those include aproduction-based entitlement, anexchange-based entitlement, anown-labor entitlement, and an inhe-ritance and transfer entitlement (i.e.,households have claims onothers/institutions to assist themaccessing food). If households loseone or more of these entitlements,destitution and death may result.Economic policies, such as an exportban for livestock, may also affectmalnutrition and mortality.

• Ecological stress: This includes highpopulation density, recurrentdroughts, livestock overgrazing, watershortages (quantity and quality),wood fuel shortages and deforesta-tion. These stresses can degrade land,substantially increase the time neededto collect water and wood, and reducefood production.

• Societal changes: Population move-ments can alter cultural norms andhabits, potentially leading to healthrisk behavior or increasing the far-ming and trade of narcotics as cashcrops.

• Poor household food availability andaccessibility: This includes poor

harvest, poor pasture conditions, lossof livestock, high market prices,food unavailability in the market,insecurity and inadequate generaldistributions.

• Poor health status of the population:Contributing factors include lowlevels of measles vaccination and lackof vitamin A supplementation. Othercontributing factors include high pre-valence of diseases such as diarrhea,acute respiratory infections, tubercu-losis, human immunodeficiency virus(HIV), high worm loads in children,and outbreaks of measles, cholera,Shigella dysentery, meningitis, mala-ria and other communicable diseases.The combination of diseases and mal-nutrition greatly exacerbates the riskfor mortality.

• Poor public health system with bar-riers to health care access: This causesand is affected by delays in recogni-zing health problems and deciding toseek care, delays in arriving at appro-priate levels of care, and delays inreceiving adequate care.

• Inadequate social and child care envi-ronment: This often equates to a lackof health and nutrition education.The results can include a lack ofexclusive breastfeeding in infantsyounger than 6 months, lack of time-ly introduction of complementaryinfant/weaning food, and lack of con-tinuation to breast feed childrenduring episodes of diarrhea. Familiesalso may not know how to preparehealthy food or know about basichygiene.

• Dietary habits in the population: Thisincludes the use of herbal medicines,which might interfere with adequatedietary intake, and extensive tea con-sumption, which may inhibit ironabsorption and result in anaemia,even in children.

MAKING DECISIONS BASED ON A COMPREHENSIVE REVIEW OF THE INFORMATIONBased on the processes described above,decision-making largely depends yourjudgement of the situation (if possibleafter traveling to the area, observing thesituation and talking to local people),consultation with partners, a review ofpreviously implemented interventions,predicted seasonal changes (from a sea-sonal calendar and early warningsystems), and the security situation.

There is no magic formula to help youwith decision-making, but there are someprinciples for the systematic review of theevidence. The collected informationshould enable you to decide whether theprevalence of malnutrition and mortalityrates are atypical, thereby justifying anutrition response. Information about therole of food insecurity and other causalfactors, vulnerable groups and geographic

distribution of population needs will helpyou plan a programme or intervention.

The general objective of a nutritionresponse is to stabilize the situation andreduce malnutrition prevalence and mor-tality rates to acceptable levels. If baselinelevels have been relatively high before theresponse, the response should attempt todecrease malnutrition and/or mortalitylevels in line with more acceptable baseli-nes. A comprehensive package of techni-cal information including a detaileddiscussion of the situation will facilitateboth communication with WFP seniormanagement and good decision-making.

When you interpret information, bear inmind:• Quality of secondary information may

be a problem; use the interpretation andcritique checklist to ensure the quality.

• The information collected may befrom different age groups and periods.



123

4CHAPTER

Table 4.10 Possible combinations of mortality rates and malnutrition prevalence and likely causes12

High prevalence ofmalnutrition

Typical prevalence of malnutrition

High rates of mortality

Likely causes:· Acute food insecurity and failure to cope· High levels of infection arising from

displacement or uncontrolled epidemic· Major disruption to care environment

Likely causes:· High rates of infection not typically

associated with malnutrition (e.g., malaria or meningitis epidemic)

· Mortality directly caused by conflict or acute disaster (e.g., earthquake)

· Possible outbreaks of micronutrientdeficiency

Typical rates of mortality

Likely causes:· Acute food insecurity· Disruption to care environment

resulting from damaging copingstrategies

· No major disease outbreaks

Likely causes:· Either no major acute causes

of malnutrition or mortalityresulting from the emergency

· Or causes which have yet tohave an impact on malnutritionand mortality

124

For example, malnutrition prevalenceis usually collected for age 6-59months as a proxy for the whole popu-lation, whereas mortality rates may beavailable for the whole population andfor children under 5 years. Mortalityrates refer to the recall period of thesurvey, whereas the prevalence of mal-nutrition reflects the status at the timeof the survey. Information about acuterespiratory disease and diarrhea isusually collected for the two weekspreceding the survey.

• In some circumstances, children under 5years of age may not be a good proxy forthe total population. For example, in areaswith high HIV/AIDS prevalence, wastingand mortality may be considerably higherin adults than in children. Attention tomorbidity and mortality in age and gendergroups, for which data may not routinelybe available, is essential.

• You may not have information aboutall potential causes of malnutritionand mortality. It can be difficult toassess the relative importance of eachfactor and the timing to best affectchanges. Review the causal frameworkto identify information gaps.

• If you are not able to explain inconsi-stencies, you may decide to either col-lect additional information or base yourdecision-making on the data you have.

SYSTEMATIC REVIEW OF THE EVIDENCEFOR DECISION-MAKINGIn summary, there is no magic formulato help you with decision making. Forexample, the WHO classification of seve-rity of malnutrition in a community(Table 4.5) cannot be used alone fordecision-making. You must also consi-der the context, including complementa-ry factors such as economic and ecologi-cal conditions, food security, or health.Your goal will be to collect differenttypes of information from different sour-

ces to provide a comprehensive picture,and then systematically review the situa-tion. You need to identify strengths,limitations and uncertainties of theinformation, and discuss those in yourreasoning. The following criteria13should help you with the decision-making process:1) Consistency: Consistently high malnu-

trition, micronutrient deficiency, morbi-dity or mortality rates from different sur-veys in your program area or emergencysetting might justify an intervention.

2) Strength: If you find acute malnutrition(> 20 percent) or severe acute malnu-trition rates (> 5 percent), an imme-diate intervention may be justified.

3) Trend: If trend analysis shows a sharpincrease in malnutrition and mortality,an intervention may be justified.

4) Temporal relationship: Although it isnot possible to prove causality in across-sectional survey, the informationcan be used for decision-making.Observed or (preferably) measuredshortcomings in the areas of food,health, nutrition, agriculture, live-stock, livelihoods, or water and sanitation most probably have prece-ded excessive malnutrition, mortalityand other outcomes in the population.Showing increasing trends may suffi-ciently support the evidence.

5) Coherence: Your reasoning should becompatible with existing theory andknowledge. Documenting risks (e.g.,failing harvest, disease outbreaks) andadverse outcomes (e.g., morbidity andmortality, malnutrition) in differentareas also supports the evidence.

6) Interventions: Early interventions mayhave changed the situation if you aredealing with an emergency or imple-menting a baseline survey after WFPinterventions have already begun. Youneed to take this into account in yourevaluation.



125

4CHAPTER

REFERENCES

1. MEASURE DHS. Demographic and health surveys. Calverton, Maryland. URL:http://www.measuredhs.com.

2. UNICEF. End of decade assessment: multiple indicator cluster survey. New York: UNICEF; 2004. Available at URL: http://www.childinfo.org/MICS2/Gj99306k.htm.

3. World Health Organization. Global database on child growth and malnutrition.Geneva: World Health Organization; 2004. Available at URL:http://www.who.int/nutgrowthdb/.

4. United Nations Standing Committee on Nutrition. 5th report on the world nutritionsituation, nutrition for improved development outcomes. Geneva: United Nations;2004. Available at URL:http://www.unsystem.org/scn/Publications/AnnualMeeting/SCN31/SCN5Report.pdf.

5. UNICEF. The state of the world's children 2003. New York: UNICEF; 2003.

6. World Health Organization. The management of nutrition in major emergencies.Geneva: World Health Organization; 2000.

7. The Sphere Project. Humanitarian charter and minimum standards in disaster respon-se. Geneva: The Sphere Project; 2004.

8. World Health Organization. Iron deficiency anaemia. Assessment, prevention, andcontrol. A guide for programme managers. Geneva: World Health Organization; 2001.Available at URL: http://www.who.int/nut/documents/ida_assessment_prevention_control.pd.

9. World Health Organization. Indicators for assessing vitamin A deficiency and theirapplication in monitoring and evaluating intervention programmes. Geneva: WorldHealth Organization; 1996.

10. Yip R, Sharp TW. “Acute malnutrition and high childhood mortality related to diar-rhea: lessons from the 1991 Kurdish refugee crisis”. JAMA 1993; 270:587-590.

11. Young H, Jaspars S. “Nutrition, disease and death in times of famine”. Disasters1995;19: 94-109.

12. Save the Children UK. Emergency nutrition assessment: guidelines for field workers.Plymouth, England: Save the Children; 2004.

13. Hill AB. “The environment and disease: association or causation”. Proc R Soc Med1965;58:295-300.

Ethical issues have to be considered when-ever human subjects are participating insurveys and other assessments. Verbal con-sent to participate in the assessment mustbe obtained from all adult participants andconsent from legal guardians must beobtained for individuals <18 years of age.Specific scenarios where WFP personnelhave to consider ethical issues may include:• Assessments that include the collection

of bodily fluids such as blood or urine(even if methods are noninvasive, i.e.,when the body is not entered by punc-ture or incision).

• Studies that include one or more com-parison groups (e.g., randomized con-trolled trials) are routinely conductedand ethically accepted if they meet cer-tain criteria, such as the inclusion of amechanism to stop the study immedi-ately if an intervention shows signifi-cantly better results than either otherinterventions or comparison groupswithout interventions.

• Referrals for survey participants whoshow signs or symptoms that requireimmediate clinical intervention, e.g.,persons with severe malnutrition. Teamleaders should immediately calculatethe weight-for-height classification ofthe individual and refer participantswith severe malnutrition to a therapeu-

tic feeding center. Participants withmoderate malnutrition should bereferred to a supplementary feedingprogram. Referral mechanisms shouldalso be established for micronutrientdeficiencies, such as anaemia, scurvy,pellagra, or beriberi as well.

The consideration of ethical issues is essentialwhen planning assessments and reviewingsecondary information. You do not want tobase your decisions on data that have beencollected in an unethical way. Usually, nutri-tion and mortality surveys are exempt fromethical review if: (1) data are collected forplanning of interventions and programmes,and (2) the survey does not qualify asresearch because results cannot be general-ized. Nevertheless, you have to ensure thatwhoever is commissioned to implement anutrition survey or other assessment onWFP's behalf is competent to do so. Theymust also have the consent of the host gov-ernment and approval from the country'sethical review board before starting the sur-vey. Also, whenever possible, make surveyprotocols available to local authorities sothey can review the methods and verifythat these conform with the community'sethical values. An example of a WFPresearch proposal to the Nepal HealthResearch Council is presented in Annex 6.


Ethical Issues

5CHAPTER

Key messages• You should not conduct any assessment without considering ethical issues. Consult

with ethical experts among all partner organizations and inform local authoritiesabout your plans in order to allow review of methods and conformity with ethicalvalues in the community. In some cases, you will need permission from the localcountry's health review boards. In others, you will also need to meet with localgovernmental officials to make sure that they agree to your plans.

We recommend being familiar with gen-eral regulations and ethical guidelines,including those of the World MedicalAssociation Declaration of Helsinki1 andthe international guidelines for ethicalreview of epidemiological studies2.Additionally, the National Institutes ofHealth ethics course is easily accessibleon the Internet, gives clear instructionsand provides a certificate after successfulparticipation3.

The guidelines for ethical review of epi-demiological studies address, among oth-ers, the following ethical issues:

• Although verbal consent usually willbe sufficient when WFP and partnersconduct nutrition and mortality sur-veys for programme planning, it maybecome relevant when organizationsare commissioned for an assessmentor research study. Consult with ethi-cal experts among all partner organi-zations in order to determine thesteps that should be taken and informlocal authorities about your plans. Tohelp guide the process, the Office ofHuman Research Protections at theU.S. Department of Health andHuman Services provides aninformed consent checklist forresearch studies (Table 5.1)4.

• One of the main principles inherent inany assessment should be to mini-mize harm to the community, in thesense of bringing disadvantage, andof doing wrong, in the sense of trans-gressing cultural values and socialmores. Harm may occur, for instance,when scarce health personnel arediverted from their routine duties toserve the needs of a survey. Surveyprotocols should address perceivedrisks for participants and include pro-posals to prevent or mitigate them.

• Investigators should make arrange-ments for protecting the confiden-tiality of survey data by, for exam-ple, omitting information that mightlead to the identification of individ-ual subjects, or limiting access to thedata, or by other means. It is cus-tomary in epidemiology to aggregatenumbers so that individual identitiesare obscured.

• You should not enter personal identi-fying information, such as addressesor names, into the data. Survey formsshould be stored in a locked cabinetor room.

• You need to think about benefits forthe community if you plan a survey.Will the results be used to imple-ment an intervention if the resultsshow the need for a response? Willthe results be available shortly afterthe survey? Are funds and trainedpersonnel available to start the inter-vention as soon as possible after thesurvey? The primary objective of thesurvey should be to improve nutri-tion and health care for the popula-tion and train local personnel in epi-demiological and, if applicable, othermethods such as blood tests foranaemia. The community needs tobe informed about the survey and, inmost situations, results should becommunicated as soon as they areavailable.

128


ETHICAL ISSUES

129

5CHAPTER

Table 5.1 Informed consent checklist - basic and additional elements

Basic elementsA statement that the study involves research.

An explanation of the purposes of the research.

The expected duration of the subject's participation.

A description of the procedures to be followed.

Identification of any procedures which are experimental.

A description of any reasonably foreseeable risks or discomforts to the subject.

A description of any benefits to the subject or to others which may reasonably be expected from the research.

A disclosure of appropriate alternative procedures or courses of treatment, if any, that might be advanta-geous to the subject.

A statement describing the extent, if any, to which confidentiality of records identifying the subject will bemaintained.

For research involving more than minimal risk, an explanation as to whether any compensation will begiven, and an explanation as to whether any medical treatments are available if injury occurs and, if so,what they consist of or where further information may be obtained.

An explanation of whom to contact for answers to pertinent questions about the research and researchsubjects' rights, and whom to contact in the event of a research-related injury to the subject.

A statement that participation is voluntary and that either refusal to participate or the subject'swithdrawal from participation at any time will involve no penalty or loss of benefits to which the sub-ject is otherwise entitled.

Additional elements, as appropriateA statement that the particular treatment or procedure may involve risks to the subject (or to the embryoor fetus, if the subject is or may become pregnant), which are currently unforeseeable.

Anticipated circumstances under which the subject's participation may be terminated by the investigatorwithout regard to the subject's consent.

Any additional costs to the subject that may result from participation in the research.

The consequences of a subject's decision to withdraw from the research and procedures for orderlytermination of participation by the subject.

A statement that significant new findings developed during the course of the research, which may relateto the subject's willingness to continue participation, will be provided to the subject.

The approximate number of subjects involved in the study.

REFERENCES

1 National Institutes of Health. “Regulations and ethical guidelines”. Available atURL: http://ohsr.od.nih.gov/guidelines/guidelines.html

2 Council for International Organizations of Medical Sciences. International guide-lines for ethical review of epidemiological studies. Geneva; 1991. Availablethrough the Centers for Disease Control and Prevention at URL:http://www.cdc.gov/od/ads/intlgui3.htm

3 National Institutes of Health. “Researcher computer-based training”. Available atURL:http://ohsr.od.nih.gov/cbt/cbt.html

130

This chapter serves to conclude with howimportant and crucial is to undertake anaccurate assessment and present the find-ings in a comprehensive and standardmanner. Accurate assessment is key toeffective and appropriate interventions. Itis crucial to present your findings in a waythat reflects following issues:

>Have the objectives been clearly stated?>Methodology including sampling

is appropriate?>WFP strategic priority indicators

reported (+ other standard indicators)?

>Causes of malnutrition and mortalityincluded?

>Are confidence limits reported andderived correctly?

For your easy reference we have includedan example of a good survey report inorder to provide guidance and referencewhen writing or reviewing a report. The checklist mentioned in Chapter 4 hasbeen applied to the report “NutritionSurvey in Saharwi Refugee Camps-Tindouf, Algeria” to highlight thestrengths of the survey and the report.


The End Point: Example of a Good Survey Report

6CHAPTER

Application of checklist to Tindouf, Algeria survey report

Objectives3Have the objectives been stated clearly and are they realistic?

Survey planning & implementation issues3Have the survey area and target group been specified?

X Has the questionnaire been translated into local language and back-translated into the original language?

3Has the questionnaire been piloted in the survey area (but with people who were not part of the sample)?

3Has the training been long enough (minimum 3-5 days depending on background of personnel)?

3Were enough qualified supervisors available to assure quality of measurements and interviews?

3Were all team members trained in the same way?

3Were the interviewers able to read questions in a standardized way from the questionnaire?

Survey methodologyX Has the sampling frame been adjusted for recent population movements?

3Was the sample representative of the target population, i.e., nobody was left out in the sampling approach?

3Has the sample size calculation been described in detail, including sample size calculations that arebased on different outcomes?

3Is the sample size large enough for appropriate precision? If the sample size calculation has not beendescribed, did the survey follow international standards, i.e., 30 x 30 cluster sample for nutrition andmortality surveys? (see section on sample size in Chapter 3)

132

Survey Reporting3Has the proportion of severely malnourished children with oedema been reported?

3Are case definitions provided, and do they meet international standards?

3Was software used for analysis that allows adjusting for cluster survey design?

3Have weight-for-height Z-scores been used to measure malnutrition?

3Has oedema been included in the definition of severe acute malnutrition?

X Has the survey questionnaire been provided in the report?

Results

3Do the results reflect the objectives of the survey?

3Does the report contain standard information (i.e., survey area, date of survey, population, surveyconducted by, acute malnutrition, acute severe malnutrition, oedema, measles immunization covera-ge, vitamin A distribution coverage, women's anthropometric status, crude mortality, under-5 mortality)11

3Have 95% confidence intervals been reported with prevalences and rates?

3Does the report provide detailed information and discussion of causes of malnutrition and mortality?

DiscussionX Does the report include a discussion of results, including limitations of the survey?

3If results are compared to a baseline, is the quality of the baseline information discussed (e.g., organi-zation that conducted assessment, methods, one or multiple years, did those years qualify as baseline)?

Conclusions3Are conclusions based on results?

3Are conclusions realistic (e.g., a solid interpretation of what the data can provide and what it cannot)?

Recommendations3Are recommendations based on science and best practices and not driven by politics?

3Are the recommendations useful, i.e., could they have been made without the study?

11 Reported regularly in Nutrition Information in Crisis Situations (NICS). Geneva: Standing Committee onNutrition, United Nations System. Available at URL: http://www.unsystem.org/scn/Publications/html/rnis.html


THE END POINT: EXAMPLE OF A GOOD SURVEY REPORT

133

6CHAPTER



135

6CHAPTER



137

6CHAPTER



139

6CHAPTER



141

6CHAPTER



143

6CHAPTER



145

6CHAPTER



147

6CHAPTER



149

6CHAPTER



151

6CHAPTER



153

6CHAPTER



155

6CHAPTER



157

6CHAPTER



159

6CHAPTER



161

6CHAPTER



163

6CHAPTER



165

6CHAPTER



167

6CHAPTER



169

6CHAPTER



171

6CHAPTER



173

6CHAPTER

1.1


WFP POLICY PAPER BRIEF: FOOD FOR NUTRITION

175

ANNEXES


WFP POLICY PAPER BRIEF: MICRONUTRIENT FORTIFICATION

177

1.2ANNEXES


WFP POLICY PAPER BRIEF: NUTRITION IN EMERGENCIES

179

1.3ANNEXES

AGEThe child's accurate age is required todetermine whether they are eligible to bein the survey and whether the child ismeasured standing or reclining for heightor length; accurate age is also necessaryfor calculating height-for-age and weight-for-age Z-scores. At the time of measure-ment, an age estimate is needed for deci-sions on sampling and for the position onthe measuring board. The preferredmethod of finding the age of the child isby obtaining the exact birth date of thechild, after which the age of the child inmonths is then calculated. If available,the enumerator needs to examine docu-

mentary evidence of the birth date (suchas a birth certificate or immunizationcard). Where there is a general registra-tion of births and where ages are general-ly known, recording age to the nearestmonth is relatively easy. Cross-checkingis necessary when the date of birth isgiven verbally by the mother, as recallerrors are common.

If dates cannot be recalled, use of a localevents-based calendar will assist mothersin recalling the date of birth. A localcalendar should be developed andshould span 5 years (since nutrition sur-veys gather information about children


Anthropometry

2ANNEXES

Simple steps to help in taking measurements

1) Consent: Before measuring the child, you should explain the purpose of both the visitand the survey and obtain the consent of the mother or caretaker. Do not pressureanyone into consent. It's best to measure children after the survey is completedbecause some children may become upset during the measurements and thereforeinterfere with the interview.

2) Two trained people required: At least two trained people are required to measure achild's length/height and weight. Anthropometric measurements should never betaken by one person alone.

3) Instrument placement: You should make sure that you have all the pieces of the mea-suring board and the scale and that they are working properly. You should recalibratethe scales as necessary and install your equipment in a quiet place, on level ground,with adequate light.

4) Weigh and measure one child at a time: If there are numerous eligible children in thehome, weigh and measure one child at a time.

5) Control the child: You have to be firm yet gentle while measuring the child. Firm, so thechild will be correctly positioned on the measuring board and will not move. Gentle, sothe child (and the mother) will be at ease and more likely to cooperate. While measu-ring the child, you can talk to him or her, explaining the procedure, etc.

6) Recording: Use a pencil to record the measurements so that you will be able to correctmistakes. Ideally, the measurer will measure and call out the measurement while a sepa-rate person - the assistant - will record the measurement while repeating it out loud.

under 5 years of age), starting with themonth the survey commences. Annualnational events are important to include,as significant annual landmarks caneasily jog the memory and help to pin-point the timing of a birth. Additionally,large national events should be detailed(such as elections, religious holidays,

harvest seasons, etc.) as well as eventsthat are more local to the region in que-stion. The local calendar should be con-structed before the survey and testedusing the enumerators; thus, it is essen-tial that all staff are comfortable usinglocal calendars to ensure that the fewestmistakes are made.

182

Example of estimation of age using a local calendar

A child is selected for inclusion in the survey. The mother is unable to tell you how old the childis, and there is no vaccination card or birth record to determine the age. At this point, a localcalendar of events is invaluable and should be referred to for further investigation of age. For this example, a local calendar developed for use in the September 2004Nutrition survey in Darfur, Sudan, should be followed.

Surveyor: What year was the child born in?Mother: 2000Surveyor: Was the child born before or after the start of the rainy season? Mother: She was born after the rainy season began.Surveyor: Was she born during Jobraka?Mother: Yes!Surveyor: That means your child was born in September 2000.

If you refer to the sample local calendar (below), you will see that the child was born inSeptember 2000 and is therefore 47 months old.


ANTHROPOMETRIC MEASUREMENT TECHNIQUES AND EQUIPMENT

183

2ANNEXES

Note that the calendar has been created to be used right to left to accommodate the local practice ofreading from right to left.

Figure 1 Local calendar of events for use in age estimation, Darfur, Sudan

2004

7

6

2003

19

18

17

16

15

14

13

12months

11

10Ramadanbegins the22nd

9 Eid alfitter the18th

8

2002

31

30

29

28

27

26

25

24months

23

22

21

20

2001

43

42

41

40

39

38

37

36 months

35

34

33

32

2000

55

54

53

52

51

50

49

48 months

47

46

45

44

1999

5 years

5 years

5 years

5 years

5 years

5 years

5 years

5 years

59

58

57

56

Events

Eid El Istiklal (independence)

Eid Alada (slaughter animalsafter Mecca)

Seyeff (TswigAlbamia) (summercrops)

Shem Elnaseem(Mosim Alhijra)(migration time)

Cultivation of Millet(Alnathafa, AlGoada,Alrimel)

Bdait El Karif (start rainy season)

El-Karif rainy season (Altirab)(planting season)

Cleaning weeds(Fatrat ELHashasha)

Jobraka (harvest of vegetables)

Dard Hassad (harvest)

Fatrain Shita(Hassad) Darad(second harvest)

DahiaEid El-milad;Darad um Nwal (late harvest)

Month

January

February

March

April

May

June

July

August

September

October

November

December

SEXThis variable is easy to determine - byasking the caretaker and direct observation.You need this information to determine yourreference population according to gender.WFP is required to break out report data bygender.

LENGTH OR HEIGHTHeight measurement (for children over 2 years)a) Place the measuring board on a hard flat

surface against a wall, table, tree, stair-case, etc. Make sure the board is notmoving.

b) Ask the mother to remove the child'sshoes and hat, and unbraid any hair thatwould interfere with the height measu-rement. Ask her to walk the child to theboard and to kneel in front of the child.

c) Kneel on your right knee on the child'sleft side. This will give you maximummobility.

d) Place the child's feet flat and together inthe center of and against the back andbase of the board/wall. Place your righthand just above the child's ankles onthe shins, your left hand on the child'sknees and push against the board/wall.Make sure the child's legs are straightand the heels and calves are against theboard/wall.

e) Tell the child to look straight ahead atthe mother, who should stand in front ofthe child. Make sure the child's line ofsight is level with the ground. Place youropen left hand under the child's chin.Gradually close your hand. Do not coverthe child's mouth or ears. Make sure theshoulders are level, the hands are at thechild's side, and the head, shoulder bla-des and buttocks are against theboard/wall. With your right hand, lowerthe headpiece on top of the child's head.

f) When the child's position is correct,read and call out the measurement tothe nearest 0.1 cm.

184

Figure 2 Measurement Techniques for Height (children over 2 years)

Source: How to Weigh and Measure Children: Assessing the Nutritional Status of Young Children. United Nations: 1986

Length (for children under 2 years of age)a) Install the measuring board against a

hard flat surface - preferably on theground. The measuring board has to bestable.

b) With the mother's help, remove the chil-d's shoes and hat, and unbraid any hairthat would interfere with the heightmeasurement.

c) Kneel on the right side of the child sothat you can hold the foot piece withyour right hand.

d) With the mother's help, lay the child onthe board by supporting the back of thechild's head with one hand and thetrunk of the body with the other hand.Gradually lower the child onto the boardwith their head at the fixed end of theboard.

e) Tell the child to look straight ahead atthe mother, who should stand in front ofthe child. Make sure the child's line of

sight is level with the ground. Place youropen left hand under the child's chin.Do not cover the child's mouth or ears.Make sure the shoulders are level, thehands are at the child's side, and thehead, shoulder blades and buttocks areagainst the board/wall. With your righthand, lower the headpiece on top of thechild's head.

f) Place the child's feet flat and together inthe center of and against the back andbase of the board or wall. Place yourright hand just above the child's ankleson the shins and your left hand on thechild's knees. Push against theboard/wall. Make sure the child's legsare straight and the heels and calves areagainst the board or wall.

g) When the child's position is correct,read and call out the measurement tothe nearest 0.1 cm. Remove the footpiece and release your left hand from thechild's shins or knees.



185

2ANNEXES

Figure 3 Measurement Techniques for Child Length (under 2 years)

Source: How to Weigh and Measure Children: Assessing the Nutritional Status of Young Children. United Nations: 1986

WEIGHTUsing an electronic scalea) Put the scale on the floor. The display

window should be blank. The displaywindow will show in kilograms and1/10ths of a kilogram.

b) This scale has no push-button switch.The best way to turn the scale on is byclosely passing one foot over the top ofthe switch window from one side to theother. In 5 seconds, the scale will adjustitself to zero.

c) To measure a child, ask the mother tostep on the scale by herself, without thechild. She should stand still on the scale.

d) Wait for the mother's weight to bedisplayed, then tare (zero-out) the mea-surement.

e) Pass the child to the mother on the scale.The child's weight will be displayed.

f) Ask the mother to step off the scale. Passyour foot across the switch window to

reset the scale before weighing the nextperson. The display window shouldindicate 0.0.

g) If a child is old enough to stand alone,ask him or her to stand still on the scale.Make sure that the child's feet or clothesdo not cover the switch window. Waituntil the child's weight is displayed.Note it, and ask him or her to step offthe scale. Pass your foot across theswitch window to reset the scale beforeweighing the next person. The displaywindow should indicate 0.0.

Using a hanging scalea) Hang the scale from a secure place. Ask

the mother to undress the child as muchas possible.

b) Attach a pair of the empty weighing

186

Figure 4 Measurement techniques for child weight

Figure 5 Measurement techniques for child weight

Source: How to Weigh and Measure Children:Assessing the Nutritional Status of Young Children.United Nations: 1986



187

2ANNEXES

pants to the hook of the scale and adjustthe scale to zero, then remove the pantsfrom the scale.

c) Have the mother hold the child. Put yourarms through the leg holes of the pants(Arrow 1). Grasp the child's feet and pullthe legs through the leg holes (Arrow 2).Make certain the strap of the pants is infront of the child.

d) Attach the strap of the pants to the hookof the scale. DO NOT CARRY THECHILD BY THE STRAP ONLY. Gentlylower the child and allow the child tohang freely. Check the child's position.Make sure the child is hanging freelyand not touching anything. Repeat anysteps as necessary.

e) Hold the scale and read the weight to thenearest 0.1 kg (Arrow 5). Call out themeasurement when the child is still andthe scale needle is stationary. Even chil-dren who are very active, which causesthe needle to wobble greatly, will becomestill long enough to take a reading. WAITFOR THE NEEDLE TO STOP MOVING.

f) Immediately record the measurement tothe nearest 0.1 kg.

MID-UPPER ARM CIRCUMFERENCE (MUAC)a) First, locate the tip of the child's shoul-

der with your fingertips.b) Bend the child's left arm at the elbow

to make a right angle.c) Measure the mid-point between the

elbow and the shoulder using the tapeapplied on the arm.

d) Wrap the tape around the arm at themidpoint mark. Cover the mark withthe tape. Put the end of the tapethrough the slot.

e) Check the tension of the tape around thearm. Do not pull too tightly, but do nothave the tape too loose. The tape shouldalways be flat on the skin.

f) Read the measurement in the win-dow when the tape is in the correctposition.

OEDEMATo measure the presence of oedema, exertmedium pressure for three seconds on theupper part of both feet. If the thumb leavesan indentation, known as pitting, on theupper side of both feet, then nutritional oede-ma is present. Nutritional oedema is alwaysbilateral (present on both feet); therefore,only individuals with pitting on both feet arerecorded as positive for nutritional oedema.

Figure 6 Child mid-upper arm circumference measurement

Source: How to Weigh and Measure Children:Assessing the Nutritional Status of Young Children.United Nations: 1986

Oedemous pitting occurring as a result of malnutrition

188

Recommended equipment for anthropometric measurements

Anthropometric data

Height/Length

Equipment

Measuringboards forinfants/adults

Characteristics

Measures both height andlength of children

A measuring board should belightweight, durable and havefew moving parts.Length/height boards shouldbe designed to measure chil-dren under 2 years of agelaying down (recumbent), andolder children standing up(note: only one board withdual purposes is standard).The board should measure upto 120 cm for children and bereadable to 1/10th of a centi-meter. Each field team shouldhave their own board.

Local Construction: Variousplans exist for the local con-struction of foldableheight/length boards andthey can be made for aroundUS$ 20. It is important thatthe materials are durable andlightweight. The wood shouldbe well seasoned to guardagainst warping. Sealing thewood with water repellentand ensuring the measuringtape is protected from wearwill improve the durability ofthe board. The tape measure should bedurable with 0.1 cm incre-ments. The numbers of thetape measure must be nextto the markings on the boardwhen the measure is gluedto the side of the board.Blueprints for the construc-tion of portable measuringboards are also availablefrom the National Center forChronic Disease Preventionand Health Promotion of theCenters for Disease Controland Prevention, Web site:http://www.cdc.gov/nccdphp/

Providers and contacts

UNICEF*: http://www.supply.unicef.dk/catalogue/Item No. 0114500

UNICEF Supply Division;Telephone: (45) 35 27 3527; Fax: (45) 35 26 94 21;Email: [email protected];Website: www.supply.unicef.dk



189

2ANNEXES

Weight Electronic scale(Uniscale)

Hanging scale

Bebe way

The scale is manufactured bySECA. It is a floor scale forweighing children as well asadults (capacity 150 kg). Itcan measure from 1 kg to 150kg in 100 g divisions, with anaccuracy of +/-100 g. Weightof adult on scale can be sto-red (tared) in memory, allo-wing the weight of baby orsmall child held by adult toshow on scale indicator.

The major advantage of thisscale is the microcomputerchip, which allows it to adjustto zero and weigh peoplequickly and accurately. Thechild may be weighed direc-tly. If a child is frightened, themother can first be weighedalone and then weighedwhile holding the child in herarms, and the scale will auto-matically compute the child'sweight by subtraction.

The portable scale, weighing4 kg, includes a solar cell on-switch and is powered bylong-life lithium battery.Instructions are available inEnglish, French and Spanish.

UNICEF Hanging Scale (ItemNo. 0145555 Scale, infant,spring, 25 kg x 100 g with No.0189000 weighingtrousers/pack of 5): This is aSalter-type spring scale witha capacity of 25 kg and 100-gincrements.Salter Hanging Scale Model235-6S: This is a lightweightscale that has a durable, rust-resistant metal case andan unbreakable plastic face. Its capacity is 25 kg, marked in 100-g increments.

Infant weighing scale for usein growth monitoring andobtaining birthweight.

UNICEFhttp://www.supply.unicef.dk/catalogue/ Item No. 0141015

UNICEF Supply Division;Telephone: (45) 35 27 3527; Fax: (45) 35 26 94 21;Email: [email protected];Website: www.supply.uni-cef.dk.

For more information contact: UNICEF SupplyDivision; Telephone: (45) 3527 35 27; Fax: (45) 35 26 9421; Email:[email protected];Website: www.supply.uni-cef.dk. The price of thisscale is about US$ 30.For more information contact: Salter IndustrialMeasurement, Ltd.;Telephone: (44) 121 5531855. The price is US$ 77.

UNICEFhttp://www.supply.unicef.dk/catalogue/

190

Mid-upper arm circumference (MUAC)

Height/weight/MUAC

Infant scale

MUAC tape

OXFAM Anthro-pometrickit

Infant weighing scale to usewhen taking infant weights.

This insertion tape measuresmid-upper arm circumferen-ce. It is made of stretch-resi-stant plasticized paper.

The anthropometric kit con-tains equipment for measu-ring the weight and height ofchildren to assess their nutri-tional status, and othermaterials needed for nutritio-nal surveys. The kit containsmeasuring and survey mate-rials for two survey teams, ormeasuring equipment fortwo feeding centers.

Item No.0557100Scale,birthweight indicator,2.5kg x 100g


Item No.0145520 Scale,infant,clinic,metric,16kg x 10g


Item No. 145600 Armcircumference insertiontape/pack of 50)

UNICEF Supply Division;Telephone: (45) 35 27 3527; Fax: (45) 35 26 94 21;Email: [email protected];Website: www.supply.unicef.dk


Item No. 0000824

UNICEF Supply Division;Telephone: (45) 35 27 3527; Fax: (45) 35 26 94 21;Email: [email protected];

Website: www.supply.unicef.dk


Weight-for-Height Reference Values in Z-scoresNCHS/CDC/WHO

3.1ANNEXES

Weight in kilograms

Height(centimeters)

50.0 3.4 0.400 2.6 2.2 1.8 50.5 3.4 0.400 2.6 2.2 1.8 51.0 3.5 0.400 2.7 2.3 1.9 51.5 3.6 0.430 2.7 2.3 1.9 52.0 3.7 0.450 2.8 2.4 1.9 52.5 3.8 0.460 2.9 2.4 2.0 53.0 3.9 0.480 2.9 2.5 2.0 53.5 4.0 0.485 3.0 2.5 2.1 54.0 4.1 0.500 3.1 2.6 2.1 54.5 4.2 0.500 3.2 2.7 2.2 55.0 4.3 0.515 3.3 2.8 2.2 55.5 4.4 0.520 3.4 2.8 2.3 56.0 4.6 0.550 3.5 2.9 2.4 56.5 4.7 0.555 3.6 3.0 2.5 57.0 4.8 0.565 3.7 3.1 2.5 57.5 4.9 0.570 3.8 3.2 2.6 58.0 5.1 0.600 3.9 3.3 2.7 58.5 5.2 0.600 4.0 3.4 2.8 59.0 5.3 0.600 4.1 3.5 2.9 59.5 5.5 0.630 4.2 3.6 3.0 60.0 5.6 0.630 4.3 3.7 3.1 60.5 5.7 0.625 4.5 3.8 3.2 61.0 5.9 0.660 4.6 3.9 3.3 61.5 6.0 0.650 4.7 4.1 3.4 62.0 6.2 0.680 4.8 4.2 3.5 62.5 6.3 0.675 5.0 4.3 3.6 63.0 6.5 0.700 5.1 4.4 3.7 63.5 6.6 0.685 5.2 4.5 3.9 64.0 6.7 0.675 5.4 4.7 4.0 64.5 6.9 0.700 5.5 4.8 4.1 65.0 7.0 0.700 5.6 4.9 4.2 65.5 7.2 0.725 5.8 5.0 4.3 66.0 7.3 0.715 5.9 5.2 4.4 66.5 7.5 0.730 6.0 5.3 4.667.0 7.6 0.730 6.1 5.4 4.7 67.5 7.8 0.760 6.3 5.5 4.8 68.0 7.9 0.750 6.4 5.7 4.9 68.5 8.0 0.750 6.5 5.8 5.0 69.0 8.2 0.770 6.7 5.9 5.1 69.5 8.3 0.760 6.8 6.0 5.3 70.0 8.5 0.785 6.9 6.1 5.4 70.5 8.6 0.780 7.0 6.3 5.5

Mean Wt Std Dev Mean - 2 SD

Mean - 3 SD

Mean - 4 SD

Weight in kilograms

192

Height(centimeters)

71.0 8.7 0.770 7.2 6.4 5.6 71.5 8.9 0.800 7.3 6.5 5.7 72.0 9.0 0.800 7.4 6.6 5.8 72.5 9.1 0.800 7.5 6.7 5.9 73.0 9.2 0.800 7.6 6.8 6.0 73.5 9.4 0.830 7.7 6.9 6.1 74.0 9.5 0.830 7.8 7.0 6.274.5 9.6 0.830 7.9 7.1 6.3 75.0 9.7 0.825 8.1 7.2 6.4 75.5 9.8 0.825 8.2 7.3 6.5 76.0 9.9 0.825 8.3 7.4 6.6 76.5 10.0 0.825 8.4 7.5 6.777.0 10.1 0.825 8.5 7.6 6.8 77.5 10.2 0.825 8.5 7.7 6.9 78.0 10.4 0.880 8.6 7.8 6.9 78.5 10.5 0.880 8.7 7.9 7.0 79.0 10.6 0.880 8.8 8.0 7.1 79.5 10.7 0.880 8.9 8.1 7.2 80.0 10.8 0.885 9.0 8.1 7.3 80.5 10.9 0.885 9.1 8.2 7.4 81.0 11.0 0.900 9.2 8.3 7.4 81.5 11.1 0.900 9.3 8.4 7.5 82.0 11.2 0.900 9.4 8.5 7.6 82.5 11.3 0.900 9.5 8.6 7.7 83.0 11.4 0.900 9.6 8.7 7.8 83.5 11.5 0.925 9.6 8.7 7.8 84.0 11.5 0.900 9.7 8.8 7.9 84.5 11.6 0.900 9.8 8.9 8.0 85.0 12.0 1.080 9.8 8.8 7.7 85.5 12.1 1.100 9.9 8.8 7.7 86.0 12.2 1.100 10.0 8.9 7.8 86.5 12.3 1.100 10.1 9.0 7.9 87.0 12.4 1.100 10.2 9.1 8.0 87.5 12.5 1.100 10.3 9.2 8.1 88.0 12.6 1.100 10.4 9.3 8.2 88.5 12.8 1.150 10.5 9.4 8.2 89.0 12.9 1.150 10.6 9.5 8.3 89.5 13.0 1.150 10.7 9.6 8.4 90.0 13.1 1.160 10.8 9.6 8.5 90.5 13.2 1.160 10.9 9.7 8.6 91.0 13.3 1.175 11.0 9.8 8.6 91.5 13.4 1.170 11.1 9.9 8.7 92.0 13.6 1.200 11.2 10.0 8.8 92.5 13.7 1.200 11.3 10.1 8.9


Mean - 3 SD

Mean - 4 SD

Weight in kilograms


WEIGHT-FOR-HEIGHT REFERENCE IN Z-SCORES

193

3.1ANNEXES

93.0 13.8. 1.200 11.4 10.2 9.0 93.5 13.9 1.215 11.5 10.3 9.0 94.0 14.0 1.215 11.6 10.4 9.1 94.5 14.2 1.260 11.7 10.4 9.2 95.0 14.3 1.260 11.8 10.5 9.3 95.5 14.4 1.260 11.9 10.6 9.4 96.0 14.5 1.275 12.0 10.7 9.4 96.5 14.7 1.300 12.1 10.8 9.5 97.0 14.8 1.300 12.2 10.9 9.6 97.5 14.9 1.300 12.3 11.0 9.7 98.0 15.0 1.300 12.4 11.1 9.8 98.5 15.2 1.350 12.5 11.2 9.8 99.0 15.3 1.350 12.6 11.3 9.9 99.5 15.4 1.350 12.7 11.4 10.0 100.0 15.6 1.380 12.8 11.5 10.1 100.5 15.7 1.380 12.9 11.6 10.2 101.0 15.8 1.380 13.0 11.7 10.3 101.5 16.0 1.400 13.2 11.8 10.4102.0 16.1 1.415 13.3 11.9 10.4102.5 16.2 1.415 13.4 12.0 10.5 103.0 16.4 1.440 13.5 12.1 10.6 103.5 16.5 1.450 13.6 12.2 10.7104.0 16.7 1.480 13.7 12.3 10.8 104.5 16.a 1.480 13.8 12.4 10.9 105.0 16.9 1.475 14.0 12.5 11.0 105.5 17.1 1.500 14.1 12.6 11.1 106.0 17.2 1.500 14.2 12.7 11.2 106.5 17.4 1.530 14.3 12.8 11.3 107.0 17.5 1.525 14.5 12.9 11.4 107.5 17.7 1.560 14.6 13.0 11.5 108.0 17.8 1.550 14.7 13.2 11.6 108.5 18.0 1.580 14.8 13.3 11.7 109.0 18.1 1.575 15.0 13.4 11.8109.5 18.3 1.600 15.1 13.5 11.9 110.0 18.4 1.600 15.2 13.6 12.0 110.5 18.6 1.600 15.4 13.8 12.2 111.0 18.8 1.630 15.5 13.9 12.3 111.5 18.9 1.625 15.7 14.0 12.4 112.0 19.1 1.650 15.8 14.2 12.5112.5 19.3 1.680 15.9 14.3 12.6 113.0 19.4 1.660 16.1 14.4 12.8 113.5 19.6 1.680 16.2 14.6 12.9 114.0 19.8 1.700 16.4 14.7 13.0 114.5 19.9 1.700 16.5 14.8 13.1

Height(centimeters)


Mean - 3 SD

Mean - 4 SD

194

115.0 20.1 1.700 16.7 15.0 13.3115.5 20.3 1.730 16.8 15.1 13.4 116.0 20.5 1.750 17.0 15.3 13.5 116.5 20.7 1.760 17.2 15.4 13.7 117.0 20.8 1.760 17.3 15.5 13.8 117.5 21.0 1.775 17.5 15.7 13.9 118.0 21.2 1.785 17.6 15.8 14.1118.5 21.4 1.800 17.8 16.0 14.2119.0 21.6 1.815 18.0 16.2 14.3 119.5 21.8 1.830 18.1 16.3 14.5 120.0 22.0 1.850 18.3 16.5 14.6 120.5 22.2 1.860 18.5 16.6 14.8 121.0 22.4 1.875 18.7 16.8 14.9 121.5 22.6 1.900 18.8 16.9 15.0 122.0 22.8 1.900 19.0 17.1 15.2 122.5 23.1 1.950 19.2 17.3 15.3 123.0 23.3 1.960 19.4 17.4 15.5123.5 23.5 1.975 19.6 17.6 15.6 124.0 23.7 2.000 19.7 17.7 15.7 124.5 24.0 2.030 19.9 17.9 15.9 125.0 24.2 2.050 20.1 18.1 16.0 125.5 24.4 2.060 20.3 18.2 16.2126.0 24.7 2.100 20.5 18.4 16.3126.5 24.9 2.115 20.7 18.6 16.4 127.0 25.2 2.155 20.9 18.7 16.6 127.5 25.4 2.175 21.1 18.9 16.7128.0 25.7 2.200 21.3 19.1 16.9128.5 26.0 2.260 21.5 19.2 17.0 129.0 26.2 2.275 21.7 19.4 17.1 129.5 26.5 2.300 21.9 19.6 17.3 130.0 26.8 2.36 22.1 19.7 17.4 130.5 27.1 2.400 22.3 19.9 17.5 131.0 27.4 2.450 22.5 20.1 17.6 131.5 27.6 2.460 22.7 20.2 17.8 132.0 27.9 2.500 22.9 20.4 17.9132.5 28.2 2.550 23.1 20.6 18.0 133.0 28.6 2.630 23.3 20.7 18.1 133.5 28.9 2.660 23.6 20.9 18.3 134.0 29.2 2.700 23.8 21.1 18.4 134.5 29.5 2.760 24.0 21.2 18.5 135.0 29.8 2.800 24.2 21.4 18.6 135.5 30.2 2.880 24.4 21.6 18.7 136.0 30.5 2.925 24.7 21.7 18.8 136.5 30.9 3.000 24.9 21.9 18.9 137.0 31.2 3.050 25.1 22.1 19.0

Weight in kilograms

Height(centimeters)


Mean - 3 SD

Mean - 4 SD

Weight-for-Height Referencein Percentage of the Median for both Boys and Girls (Supine)


3.2ANNEXES

Length(cm)

49.049.550.050.551.0

51.552.052.553.053.5

54.054.555.055.556.0

56.557.057.558.058.5

59.059.560.060.561.0

61.562.062.563.063.5

64.064.565.065.566.0

66.5

Length(cm)

67.067.568.068.569.0

69.570.070.571.071.5

72.072.573.073.574.0

74.575.075.576.076.5

77.077.578.078.579.0

79.580.080.581.081.5

82.082.583.083.584.0

84.5

Median(kg)

7.67.87.98.08.2

8.38.58.68.78.9

9.09.19.29.49.5

9.69.79.89.910.0

10.110.210.410.510.6

10.710.810.911.011.1

11.211.311.411.511.5

11.6

85%

6.56.66.76.87.0

7.17.27.37.47.5

7.67.77.98.08.1

8.28.28.38.48.5

8.68.78.88.99.0

9.19.19.29.39.4

9.59.69.69.79.8

9.9

80%

6.16.26.46.46.6

6.76.86.97.07.1

7.27.37.47.57.6

7.77.87.97.98.0

8.18.28.38.48.4

8.58.68.78.88.8

8.99.09.19.29.2

9.3

75%

5.75.85.96.06.1

6.26.36.46.56.6

6.76.86.97.07.0

7.27.37.47.47.5

7.67.77.87.87.9

8.08.18.18.28.3

8.48.48.58.68.7

8.7

70%

5.35.45.55.65.7

5.85.96.06.16.2

6.36.46.56.56.6

6.76.86.96.97.0

7.17.27.27.37.4

7.57.57.67.77.7

7.87.97.98.08.1

8.2

Median(kg)

3.23.33.43.43.5

3.63.73.83.94.0

4.14.24.34.44.6

4.74.84.95.15.2

5.35.55.65.74.9

6.06.26.36.56.6

6.76.97.07.27.3

7.5

85%

2.72.82.92.93.0

3.13.13.23.33.4

3.53.63.73.83.9

4.04.14.24.34.4

4.54.64.84.95.0

5.15.25.45.55.6

5.75.96.06.16.2

6.4

80%

2.62.62.72.72.8

2.93.03.03.13.2

3.33.43.53.53.6

3.73.83.94.04.2

4.34.44.54.64.7

4.84.95.05.25.3

5.45.55.65.75.9

6.0

75%

2.42.52.52.62.?

2.72.82.82.93.0

3.13.23.23.33.4

3.53.63.73.83.9

4.04.14.24.34.4

4.54.64.74.85.0

5.5.25.35.45.5

5.6

70%

2.32.32.42.42.5

2.52.62.62.72.8

2.92.93.03.13.2

3.33.43.43.53.6

3.73.83.94.04.1

4.24.34.44.54.6

4.74.84.95.05.1

5.2

Percents of median Percents of median

196

Length(cm)

85.085.586.086.587.0

87.588.088.589.089.5

90.090.591.091.592.0

92.593.093.594.094.5

95.095.596.096.597.0

97.598.098.599.099.5

100.0100.5101.0101.5102.0

102.5103.0103.5104.0104.5

105.0105.5106.0106.5107.0

Length(cm)

107.5108.0108.5109.0109.5

110.0110.5111.0111.5112.0

112.5113.0113.5114.0114.5

115.0115.5116.0116.5117.0

117.5118.0118.5119.0119.5

120.0120.5121.0121.5122.0

122.5123.0123.5124.0124.5

125.0125.5126.0126.5127.0

127.5128.0128.5129.0129.5

130.0

Median(kg)

17.717.818.018.118.3

18.418.618.818.919.1

19.319.419.619.819.9

20.120.320.520.720.8

21.021.221.421.621.8

22.022.222.422.622.8

23.123.323.523.724.0

24.224.424.724.925.2

25.425.726.026.226.5

26.8

85%

15.015.215.315.415.5

15.715.816.016.116.2

16.416.516.716.816.9

17.117.317.417.617.7

17.918.018.218.418.5

18.718.919.119.219.4

19.619.820.020.220.4

20.620.821.021.221.4

21.621.822.122.322.5

22.8

80%

14.114.314.414.514.6

14.814.915.015.115.3

15.415.515.715.816.0

16.116.216.416.516.7

16.817.017.117.317.4

17.617.817.918.118.3

18.418.618.819.019.2

19.419.619.719.920.1

20.420.620.821.021.2

21.4

75%

13.313.413.513.613.7

13.814.014.114.214.3

14.414.614.714.815.0

15.115.215.415.515.6

15.815.916.116.216.4

16.516.716.817.017.1

17.317.517.617.818.0

18.218.318.518.718.9

19.119.319.519.719.9

20.1

70%

12.412.512.612.712.8

12.913.013.113.313.4

13.513.613.713.814.0

14.114.214.314.514.6

14.714.915.015.115.3

15.415.515.715.816.0

16.116.316.516.616.8

16.917.117.317.517.6

17.818.018.218.418.6

18.7

Median(kg)

12.012.112.212.312.4

12.512.612.812.913.0

13.113.213.313.413.6

13.713.813.914.014.2

14.314.414.514.714.8

14.915.015.215.315.4

15.615.715.816.016.1

16.216.416.516.716.8

16.917.017.217.417.5

85%

10.210.310.410.510.6

10.610.710.810.911.0

11.111.211.311.411.5

11.611.711.811.912.0

12.112.212.412.512.6

12.712.812.913.013.1

13.213.313.513.613.7

13.813.914.014.214.3

14.414.514.614.814.9

80%

9.69.79.89.89.9

10.010.110.210.310.4

10.510.610.710.810.8

10.911.011.111.211.3

11.411.511.611.711.8

11.912.012.112.212.3

12.412.612.712.812.9

13.013.113.213.313.4

13.613.713.813.914.0

75%

9.09.19.19.29.3

9.49.59.69.79.7

9.89.910.010.110.2

10.310.310.410.510.6

10.710.810.911.011.1

11.211.311.411.511.6

11.711.811.912.012.1

12.212.312.412.512.6

12.712.812.913.013.1

70%

8.48.58.58.68.7

8.88.88.99.09.1

9.29.29.39.49.5

9.69.79.79.89.9

10.010.110.210.310.3

10.410.510.610.710.8

10.911.011.111.211.3

11.411.511.611.711.8

11.912.012.112.212.3

Percents of median Percents of median


Height-for-Age Index Z-scoresfor Combined Sexes 0-59 Months of AgeNCHS/CDC/WHO

3.3ANNEXES

COMBINED SEXES 0-59 months of age

Age (months)

01234567891011121314151617181920212223242526272829303132

Z scoremedian

50.254.157.460.362.865.066.968.570.071.472.774.075.276.477.578.679.780.781.782.683.684.585.486.285.085.986.788.489.289.890.791.592.2

-4

41.244.547.450.152.454.456.157.959.460.661.963.064.265.266.167.067.968.769.570.270.871.772.473.072.272.973.574.074.675.276.076.577.1

-3

43.546.949.952.755.057.158.860.662.163.364.665.867.068.069.069.970.971.772.673.374.074.975.776.375.476.276.877.478.178.779.580.180.7

-2

45.749.352.455.257.659.761.563.264.766.067.368.569.770.871.872.873.874.775.676.477.278.178.978.679.480.180.881.582.282.983.684.384.9

-1

48.051.754.957.860.262.464.265.967.468.770.071.372.573.674.775.776.877.778.779.580.481.382.282.981.882.783.484.285.085.786.487.287.9

COMBINED SEXES 0-59 months of age

198

Age (months)

333435363738394041424344454647484950515253545556575859

Z scoremedian

93.093.794.495.195.896.597.297.898.599.299.8100.4101.1101.7102.3102.9103.5104.1104.7105.2105.8106.4106.9107.5108.1108.6109.2

-4

77.678.278.979.479.980.681.181.682.282.783.283.684.284.785.185.786.186.787.187.588.288.689.089.589.990.390.8

-3

81.381.982.683.283.784.485.085.586.186.787.287.788.388.889.389.990.390.991.491.892.592.993.493.994.394.895.3

-2

85.686.386.987.588.288.889.490.090.691.291.792.392.993.494.094.595.195.696.196.797.297.798.298.799.299.7100.1

-1

88.689.390.090.791.392.092.793.393.994.695.295.896.497.097.698.298.799.399.9100.4101.0101.5102.1102.6103.1103.7104.2


Weight For Age Reference Values for Boys

3.4ANNEXES

Age (months)

012345678910111213141516171819202122232425262728293031323334353637

Mean weight

3.34.35.26.06.77.37.88.38.89.29.59.910.210.410.710.911.111.311.511.711.812.012.212.412.312.512.712.913.113.313.513.713.914.114.314.414.614.8

Std Dev

0.40.70.91.01.01.01.01.01.01.01.01.01.01.01.11.11.11.11.21.21.21.31.31.31.11.21.21.31.31.31.41.41.51,51.51.61.61.6

-2 SD

2.42.93.54.14.75.35.96.46.97.27.67.98.18.38.58.78.89.09.19.29.49.59.79.810.110.210.310.410.510.610.710.911.011.111.211.311.411.5

-3 SD

2.02.22.63.13.74.34.95.45.96.36.66.97.17.37.57.67.77.87.98.08.18.38.48.59.09.09.19.19.29.39.49.49.59.69.79.79.89.9

- 4 SD

1.61.61.82.22.73.33.94.54.95.35.65.96.16.36.46.56.66.76.86.86.97.07.17.27.87.97.97.97.97.98.08.08.08.18.18.28.28.3

200

Age (months)

38394041424344454647484950515253545556575859

Mean weight

15.015.215.315.515.715.816.016.216.416.516.716.917.017.217.417.517.717.918.018.218.318.5

Std Dev

1.71.71.71.71.81.81.81.81.91.91.91.91.92.02.02.02.02.02.02.12.12.1

-2 SD

11.711.811.912.012.112.312.412.512.612.812.913.013.113.313.413.513.713.813.914.114.214.3

-3 SD

10.010.110.210.310.410.510.610.710.810.911.011.111.211.311.411.511.611.811.912.012.112.2

- 4 SD

8.48.48.58.68.68.78.88.88.99.09.19.29.39.49.59.59.69.79.89.910.010.1


Weight for Age Reference Values for Girls

3.5ANNEXES

Age (months)

012345678910111213141516171819202122232425262728293031323334353637

Mean weight

3.24.04.75.46.06.77.27.78.28.68.99.29.59.810.010.210.410.610.811.011.211.411.511.711.812.012.212.412.612.813.013.213.413.613.813.914.114.3

Std Dev

0.50.60.70.70.80.80.90.90.91.01.01.01.01.11.11.11.11.21.21.21.21.21.21.21.21.21.21.31.31.31.31.41.41.41.41.51.51.5

-2 SD

2.22.83.33.94.55.05.55.96.36.66.97.27.47.67.88.08.28.38.58.68.89.09.19.39.49.69.89.910.110.210.310.510.610.810.911.011.211.3

-3 SD

1.82,22.73.23.74.14.65.05.45.75.96.26.46.66.76.97.07.27.37.57.67.77.98.18.38.48.58.78.88.99.09.19.29.49.59.69.79.8

- 4 SD

1.31.62.02.42.93.33.74.14.44.74.95.25.35.55.65.85.96.06.16.36.46.56.76.87.17.27.37.47.57.67.77.87.97.98.08.18.28.3

202

Age (months)

38394041424344454647484950515253545556575859

Mean weight

14.414.614.814.915.115.215.415.515.715.816.016.116.216.416.516.716.817.017.117.217.417.5

Std Dev

1.51.51.61.61.61.61.61.61.71.71.71.71,71.81.81.81.81.81.81.91.91.9

-2 SD

11.411.511.611.811.912.012.112.212.312.512.612.712.812.913.013.113.213.313.413.513.613.7

-3 SD

9.910.010.110.210.310.410.510.610.710.810.911.011.111.211.311.411.511.511.611.711.811.9

- 4 SD

8.48.58.68.78.88.98.99.09.19.29.29.39.49.49.59.69.69.79.89.89.99.9


4ANNEXES

Province ______________________ District _____________________ Village _________________Cluster number: _________ Household number: _____________Team code: __________ Interviewer code: _______________ Date of interview: ______ / _______

Day MonthHOUSEHOLD DATA1) Does your family now live in your usual place of residence? (circle one) Yes / No / Unk

1a) If NO, how long since the family has lived there? ______ months OR ______ years2) Has anyone in the family received any relief food in the last month? Yes / No / Unk3) What is your main source of water? (circle one) Central piped system / Truck or water seller / Bore hole

Open Well / River or stream / Lake or pond / Other4) Do you use the same source of water now as you did this time last year? Yes / No / Unk5) Results of iodine testing of salt used for last night's food (circle one) Positive / Negative / Not Done

I would like to ask you about each person who lived in this household at the time of [beginning of recall period] andwho lives here now:

Column 4Columns 3 & 4Column 5Columns 3 & 5Column 6Columns 3 & 7

Nutrition and Health Survey Questionnaire

a. Number of current HH members - totalb. Number of current HH members < 5 yearsc. Number of HH members at beginning of recall period - all agesd. Number of HH members at beginning of recall period < 5 yearse. Total number of deaths during recallf. Number of deaths in children < 5 years of age

Tally (if data on household members will not be entered into the computer):

2

Sex (M/F)

3

Current age(in years, <1 year = 0)

4

Present now(Y/N)

5

Present at thebeginning ofrecall period

(Y/N)

6

Current status(1=Alive;2=Dead;

3=Unknown)

1

ID

1234567891011121314151617181920

204

Pers

onno

.fro

mta

ble

1ab

ove.

Preg

-na

ntno

w?

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

No. o

fdo

ses

tetan

usva

ccine

MUA

C(c

ms)

Wei

ght

(kgs

)He

ight

(cm

s)He

mog

lobi

n(g

ms)

Liter

ate?

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Nig

htbl

ind-

ness

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Goite

r

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Y /

N

Num

-be

r of

preg

-na

n-ci

es

Num

-be

r of

live

birth

smm

/yy

Date

of

last

deliv

e-ry

mm

/yy

/ / / / / / / / / /

Livi

ng w

omen

of c

hild

bear

ing

age

(15

- 49

year

s) -

data

col

lect

ion

form


SAMPLE DATA COLLECTION FORMS FOR NUTRITION AND MORTALITY

205

4ANNEXES

Nutrition and health survey - data collection form (child 0-59 months)

Cluster number:______ HH: _____ Child's person number: _____ Mother's person number: _______

Questions for adult caretaker

1) Relationship of respondent to child: ____________Mother Father Grandmother Grandfather Other

2) Is this child's mother alive?________________________________________Yes / No / Unk

3) Sex___________________________________________________________Male / Female

4) Date of birth OR Age in months _____ / _____ / _____ OR ______ monthsDay Month Year .

5) Does this child have difficulty seeing at night or in the evening when other people do not? Yes / No / Unk

6) Since this time yesterday, has this child breast fed? Yes / No / Unk6a) If YES, was breast milk this child's main source of food since yesterday? Yes / No / Unk6b) If YES, how long after birth did this child first breast feed? __________________ hours

7) Since this time yesterday, has this child received anything other than breast milk? Water, tea, or juice /(circle all that are true) Powdered milk or infant formula / Semi-solid or solid food / None of these

8) Since this time yesterday, has this child drunk anything from a bottle with a nipple? Yes / No / Unk

9) Has this child received any vitamin A? Vitamin A is given as drops from a capsule Yes / No / Unk(show example)

10) Since 2 weeks ago, has this child had diarrhea? Yes / No / Unk Diarrhea is 3 or more stools in 24 hours.

10a) If YES, was this child taken to a clinic or hospital for this problem? Yes / No / Unk

11) Since two weeks ago, has this child had fever and difficulty breathing? Yes / No / Unk10a) If YES, was this child taken to a clinic or hospital for this problem? Yes / No / Unk

12) Has this child received measles vaccination? Yes / No / UnkThis vaccine is given by injection.

Examination of child13) Bitot's spots Yes / No14) Angular stomatitis Yes / No15) BCG scar Yes / No16) Bilateral edema Yes / No17) Does this child have a physical deformity making it difficult to obtain an accurate height? Yes / No

Anthropometry and laboratory18) Weight: (kgs) 19) Length/Height: (cms) 20) MUAC: (cms) 21) Hemoglobin:


Random Number Table

5ANNEXES

64612 77930 16137 12927 89071 72799 41537 36124 90640 31518 68866 19304 42847 17249 97332 86300 39716 03893 06408 32722 50198 35604 77895 61969 51985 08141 33488 78995 04992 75339 76698 11509 43552 41494 83724 01956 75786 19758 45947 94834 73412 52071 43503 62873 53324 11284 43196 06348 30008 62652

42295 74036 20944 62432 59331 89684 88553 32377 93850 12720 14980 35863 08297 96342 19765 47025 29892 81190 68117 08072 76350 78339 37830 99947 43444 98453 50998 75554 04195 85201 01581 46405 52672 46305 08886 33547 38993 18768 14469 72645 67238 13884 20162 80008 62569 22205 30546 28072 44837 49459

66570 33762 21469 00199 27172 15397 82047 61497 07638 97270 10557 21230 49179 29167 91844 51682 71808 45604 47827 87184 09219 97504 31797 55465 99417 95123 17753 98301 97544 98741 32543 64753 03363 75921 19893 88730 18290 20197 61643 60201 05689 43380 65162 24128 11352 45001 03769 89504 99057 83269

03507 88301 79068 65814 83846 19277 66548 97374 68215 52775 28225 32562 80334 30146 61413 91111 43080 28520 49848 82813 99646 08072 73891 72968 00687 38170 31~09 05309 49248 05801 26756 07050 27244 13452 53824 42973 53428 95469 10687 17704 25235 65105 57132 92464 29317 60554 06727 88036 74389 67967

25656 67440 05564 71519 49575 64287 00165 16939 41789 66082 33390 91113 08488 81634 16286 46749 73217 41865 19390 67245 43992 57138 00819 15070 20945 25400 57957 71599 16271 57901 13893 92231 60466 90318 37897 66912 90283 37008 36989 78760 66398 01315 02014 70505 34941 76983 61435 54541 97455 39820

31762 31972 63350 36644 33992 44364 85710 21443 77930 38707 30127 40804 64291 59007 77904 18539 75234 65215 67092 5864032105 53327 84967 52173 65105 98585 56590 57180 25674 8445457981 21947 84104 02266 33572 35803 16381 96110 52509 1604956126 26952 92400 94553 96271 66806 89957 86934 47075 94908

13006 34316 09174 78732 96563 29286 02657 02883 18857 37822 71463 03840 20296 13460 48767 73046 59743 77656 04051 18536 85318 60674 67335 63363 48627 83227 35832 12923 73892 07336 88510 93235 41827 12682 46688 41684 97946 93028 99020 15613 00429 98471 73469 59309 02463 11443 64722 09558 33674 17649


Example of Application forApproval of Research Proposal [shortened]

6ANNEXES

PART IAdministrative data sheet

1. Name and title of principal investigator responsible for the proposed research:Mr. X. X, , programme officer

Signature: ………………………….. Date: …………………………………….Postal address:

2. Full name of the institution / University / NGO / INGO associated with the principal investigator (if applica-ble): World Food Programme

Declaration of the head of the institution / University / NGO / INGOIf the proposed research is approved, we will allow him/her to conduct the research in this institution.

Signature: ………………………………… Date: ……………………..

4. Name and title of co-investigator (if any) responsible for the proposed research:Last (Surname) Middle (if any) First name Title (eg., Mr. Ms. Dr.)XY XY Mr.

Designation: public health nutrition officerSignature: ………………………….. Date: ……………………..Postal address (if different from the address given above):

5. Is the research responsive to the health priorities and needs of Nepal?Yes ( X) No ( ) Explain.

6. Is the research sensitive to the Nepali culture and the social values?Yes ( X) No ( ) Explain.

7. Is health insurance being made available to the research participants? If yes, please provide the necessary insurance data. No

8. List the name(s) and institutional affiliation of foreign researcher(s) (other than co-investigator) to assist yourproject in Nepal and abroad (if any)- None

Name Institution(a) …………………………………… ……………………………………

Nepal health research council, Kathmandu, Nepal

210


9. List the name(s) of Nepali researcher(s) (other than co-investigator) or Nepalese institution/hospital/NGO(s)etc. from whom you may seek co-operation (if any) - None

PART IIFinancial data sheet

10. Which funding organization or agency is going to fund your research project?United Nations World Food Programme, Nepal country office

Contact information of funding organization or agency:

Total amount of funds (in US $) allocated for the proposed research project:

Overall gross budgetary breakdown of the research project:Personnel: Equipment: Operational expenses:

Laboratory/Office expenses: …………………….Clinical expenses: …………………….Field expenses:

Data analysis: ………………Others:

PART IIIResearch proposal description sheet

11. Title: Baseline survey in two Makwanpur VDCs (Phaparbari and Dhiyal) prior to implementation of WFP'sMother and Child Health Care (MCHC) activity.

12. Objectives: Collection of baseline data prior to implementation of the MCHC activity.

13. Summary WFP Nepal is planning to implement its Mother and Child Health Care Activity in two VDCs of Makwanpurdistrict (Phaparbari and Dhiyal). Prior to implementation, it is necessary to collect baseline data to be ableto measure the progress of the different indicators during the project lifecycle. The baseline data to becollected is directly related to the logical framework developed for the MCHC Activity.

The long-term objective of the MCHC Activity is to support government efforts to reach the goals set underthe Nutrition and Safe Motherhood Programmes of the Ministry of Health (MoH), namely to improve theoverall health and nutritional status of children, expectant and nursing mothers.

More specifically, it is expected that provision of a fortified blended food and MCHC services to expec-tant and nursing mothers (until 6 months after delivery) and children between 6 months and 3 years ofage will contribute to:


EXAMPLE OF ETHICAL CLEARANCE FORM

211

6ANNEXES


• prevention or reduction of the prevalence of underweight among young children;• reduction in iron-deficiency anaemia among expectant and nursing mothers and young children;• raising awareness and knowledge of expectant and nursing mothers on their health and nutrition and that

of their children; • increased and more regular utilization of community-based and MCHC outreach services

(e.g., growth monitoring).

The indicators to be measured in the baseline survey are consequent with the objectives of the MCHC activity.These indicators include:

• Weight of children 6-36 months old;• Height/length of children 6-36 months old;• Hb level (analysis done in the field using HemoCues of Expectant and Nursing Mothers (ENMs)

and of children 6-36 months old;• Knowledge and practice of appropriate health, nutrition, caring and feeding practices of ENMs

and caretakers of young children;• Awareness and use of outreach clinic services by ENMs and caretakers of young children;• Household history;• Knowledge and practices of appropriate health, nutrition, caring and feeding practices of household

members.

14. Rationale / JustificationIt is essential to conduct a baseline survey, measuring all the indicators chosen to measure the achievement ofthe activity's objectives prior to implementation. The baseline data collected will allow us to adapt the project tothe specific needs of the communities and to its socio-cultural specificity, as well as enabling us to measure theprogress of the different indicators over the project lifecycle.

15. Research design and methodology Research method Qualitative ( ), Quantitative ( ), Combined (x)

Study VariablesThe quantitative data collected will include, amongst others:• Age;• Weight of children 6-36 months old;• Height/length of children 6-36 months old;• Hb level (analysis done in the field using HemoCues provided by WFP) of ENMs and of children

6-36 months old;• Knowledge and practice of appropriate health, nutrition, caring and feeding practices of ENMs and

caretakers;• Awareness and use of outreach clinic services.

The qualitative interviews will include questions on:

• Household history;• Knowledge and practices of appropriate health, nutrition, caring and feeding practices of household

members; Awareness and use of outreach clinic services.

212


Type of study

Descriptive study ( )(Specify ……………...………………………………………………….)

Analytical study ( )(Specify ……………...………………………………………………….)

Experimental study ( )(Specify ……………...………………………………………………….)

Other …Baseline survey: In order to collect the baseline information prior to the implementation of the Motherand Child Health Care Activity in Makwanpur district.

Study site and its justificationTarget populationExpectant and nursing mothers and children of 6 months to 3 years of age in these two VDCs.

Sampling methodsNon-probability Sampling ( )(Specify ……………...………………………………………………….)

Probability sampling ( o)(Specify: systematic sampling method)

Sample Size• Expectant and nursing mothers: 268 (1,500 households to be visited)• Children aged 6-36 months: 203 (560 households to be visited)

Sampling frame (if relevant) and sampling process including criteria for sample selectionA percentage of the total number of households in each of the 18 wards of the two VDCs will be visited, toreach the total number of households for each target group. A list of households will be made in collaborationwith the local authorities. Surveyors will visit every Xth household (to be determined), following a systematicsampling method, and ask simple questions in order to determine if there are any target beneficiaries in thehousehold. If the criterion is met, the full questionnaire will be filled in, anthropometric measurements will betaken if applicable, and a blood sample will be taken and analyzed.

Tools and techniques for data collection • Household questionnaire: Questionnaires have been developed for the expectant and nursing mothers

and the caretakers of the children of 6-36 months. Additionally, there will be a questionnaire for identifying eligible households.

• Qualitative interview: The qualitative part of the baseline will consist of a minimum of 10 semi-directive household interviews. The 10 households to be interviewed will be selected in collaboration with the VDC officials and health staff.

• Test of Haemoglobin level for both ENMs and the children of 6-36 months: The lab assistant hired by the NEW ERA will be involved mainly for collecting blood samples and examining the Hb level using HemoCues, microcuevettes and lancets.



213

6ANNEXES


• Anthropometric measurements: Salter scales will be used for taking weights of the children and a measuring board will be used for measuring heights of the children. The entire fieldwork will be carried out by 5 teams each team consisting of two female interviewers and a lab assistant.

Pretesting the data collection tools (if relevant)

Validity and reliability of the research (if relevant)

Biases (if relevant)

Limitation of the study (if relevant)

16. Plan for supervision and monitoring

17. Plan for data management

18. Plan for data analysis

19. Expected outcome of the research

20. Plan for dissemination of research resultsThe research results are meant exclusively for the internal use of the organization. However, the results canbe made available to any of the interested organizations working in the field of nutrition as appropriate.

21. Plan for utilization of the research findings (optional)The baseline survey will establish a socio-cultural standard for the two VDCs and also provide baselineinformation prior to the project implementation on certain indicators. This baseline information will pro-vide a ground for monitoring and evaluating the project throughout its lifecycle.

PART IVEthical consideration

22. Regarding the human participants:

Are human participants required in this research? If yes, offer justification.Yes, the targeted beneficiaries of the MCHC activity are the expectant mothers and nursing mothers andthe children of 6-36 months age. Therefore, the ENMs and the children of 6-36 months age are the parti-cipants for this baseline survey.

How many participants are required for the research? Explain.A total of 268 expectant and nursing mothers and 203 children (6-36 months) in Phaparbari and Dhiyal VDCs willbe reached during the survey.

What is the frequency of the participant's involvement in the research? Explain.Each of the participants will be interviewed (ENMs and Caretakers) and weighed and measured for the height (onlychildren) once during the survey. Likewise, a blood sample will be collected from each of the participant, both ENMsand children, during the survey. Additionally, 10 semi-directive household interviews will be conducted.

214


Clearly indicate the participants' responsibilities in the research. What is expected of the research participantsduring the research?

• Participants are expected to agree to finger prick in order to collect a drop of blood to be analysed with a HemoCues machine for determination of HB levels

• Participants are expected to attend the interviews• Participants are expected to allow the enumerators to measure their height and weight• Participants are expected to help the survey teams as appropriate

Are vulnerable members of the population required for this research? If yes, offer justification.The expectant and nursing mothers and children of 6-36 months of age are the participants for this survey andsupposedly the vulnerable members of the population. In order to track the achievements towards the above-mentioned objectives, some baseline information on the beneficiaries are required.

Are there any risks involved for the participants? If yes, identify clearly what are the expected risks for thehuman participants in the research and provide a justification for these risks.A few drops of blood will be collected by pricking the fingers of the participants in the survey. Sterile,disposable lancets will be used for finger pricking.

Are there any benefits involved for the participants? If yes, identify clearly what are the expected benefits forthe participants.Following the survey, the MCHC activity will be implemented in these two VDCs. Upon implementationof the activity, the participants will benefit with the following:

• A take-home ration for expectant and nursing mothers (7.5 kg/month), linking food supplementation to the local health services.

• A take-home ration for children 6 months to 3 years (6 kg/month), linking food supplementation to the local health services.

• Deworming tablets for expectant women after the first trimester of pregnancy• Mass counseling on Antenatal Natal Care, Growth Monitoring and other health and nutrition related

issues through the existing local health system

23. Informed Consent Form / Ethical issues:Statements required in the Informed Consent Form include:

A statement that the human participants can withdraw from the study at any time without giving reason and without fear. State clearly how the participants can opt out the study.A statement guaranteeing the confidentiality of the research participants.If required, a statement on any compensation that might be given to the research participant and/or their community.A statement indicating that the participant has understood all the information in the consent form and is willing to volunteer/participate in the research. Signature space for the research participants, a witness, and the date.(Informed Consent Form should be submitted in English and in the language appropriate to the research participants)



215

6ANNEXES


Obtaining the consent

How is informed consent obtained from the research participants?During the survey the enumerators will get the consent from the participants.

Please indicate who is responsible for obtaining informed consent from the participants in this research study. Enumerators

Is there anything being withheld from the research participants at the time the informed consent is beingsought?

No ( x ) Yes ( )If yes, explain

PART VAnnexes24. Annexes should includea. References, b. Data collection instruments including questionnaires, c. Information sheet and informed consent form (if relevant), d. List of abbreviations, Recently updated Curriculum Vitae of principal investigator


Resources

7ANNEXES

A list of useful publications, information sources, and Web links(see reference lists at the end of chapters also).

World Health Organization PublicationsField guide on rapid nutritional assessment in emergencies. Cairo; 1995. Available at URL: http://whqlibdoc.who.int/emro/1994-99/9290211989.pdf.

Global database on child growth and malnutrition (standardized compilation of anthropometric datafrom population-based surveys around the world from 1960 onwards). Available at URL:http://www.who.int/nutgrowthdb.

Guidelines for the inpatient treatment of severely malnourished children. SEARO Technical PublicationNo. 24. Geneva: 2003. Available at URL: http://w3.whosea.org/nhd/pdf/pub24/malnourished_children.pdf.

Infant feeding in emergencies. A guide for mothers. EURO, 1997. Available at URL: http://www.who.dk/nutrition/pdf/breastfeed.pdf.

Iron deficiency anaemia: assessment, prevention and control - a guide for programme managers. 2001.WHO/NHD/01.3. Available at URL: http://www.who.int/nut/documents/ida_assessment_prevention_control.pdf.

Management of severe malnutrition: a manual for physicians and other senior health workers. Geneva;1999. Available at URL: http://www.who.int/nut/documents/manage_severe_malnutrition_eng.pdf.Physical Status: The Use and Interpretation of Anthropometry - Report of a WHO Expert Committee.WHO, Geneva. 1995

The management of nutrition in major emergencies. Geneva; 2000. Available at URL:http://wholibdoc.who.int/publications/2000/9241545208.pdf.

World Food Programme PublicationsFood and Nutrition Handbook. Rome; 2000.

Food and Nutritional Needs in Emergencies. Rome; 2003.

Guidelines for selective feeding programmes in emergency situations (based on WHO's norms andstandards). Rome; 1999. Available at URL: http://www.unsystem.org/scn/archives/rnis26/ch7.htm.

Nutrition Policy Papers http://www.wfp.org/policies/Introduction/policy/

218

Other Publications, Organizations, and Web Sites

Action Contre la Faim. Web site: http://www.acf-fr.org/.

Concern. Web site: http://www.concern.ie/.

Demographic and Health Surveys. Web site: http://www.measuredhs.com.

Food and Nutrition Technical Assistance (FANTA). Web site: http://www.fantaproject.org.

Institute of Child Health, London. Web site: http://www.ich.ucl.ac.uk/.

The Micronutrient Initiative. Web site: http://micronutrient.org

Save the Children. U.K. Web site: http://www.savethechildren.org.uk.

Save the Children. U.S. Web site: http://www.savethechildren.org.

Stoltzfus RJ, Dreyfuss ML. Guidelines for the use of iron supplements to prevent and treat iron defi-ciency anemia. Washington: International Nutritional Anemia Consultative Group; 1998. Available atURL http://www.ilsi.org/file/guidelinesforuseofiron.pdf.

The Sphere Project. Humanitarian charter and minimum standards in disaster response. Chapter 3:Minimum standards in nutrition. Geneva; 2004. Available at URL: http://www.sphereproject.org/han-dbook/hdbkpdf/hdbk_c3.pdf

Tufts University, Feinstein International Famine Center. Web site: http://famine.tufts.edu/.

UNICEF. Multiple Indicator Cluster Surveys. Available at URL: http://www.childinfo.org

United Nations Standing Committee on Nutrition (SCN). Web site: http://www.unsystem.org/scn/.

Valid International. A limited company specializing in improving the quality and accountability ofhumanitarian assistance. Web site: http://www.validinternational.org.


Glossary of Terms

8ANNEXES

Accuracy

Anaemia

Anthropometry

Baseline data

Benchmark

Bias

Body mass index (BMI)

Chronic malnutrition

Cluster sampling

Indicates how close the survey estimates are to the true value.Accurate measurements are essential to obtain accurate esti-mates. Accuracy, contrary to precision, cannot be quantified.

Abnormally low hemoglobin levels; can be caused by lack ofiron, folate, vitamin B12.

The technique that deals with the measurements of the size,weight and proportions of the human body.

Baseline data represent the situation before or at the beginningof a programme or intervention. Survey data may be comparedto baseline data if defined criteria for comparison are met (e.g.,similar methods and coverage).

The prevalence of mortality rate used as a population thresholdin a classification of severity of a situation.

Anything other than sampling error which causes the surveyresult to differ from the actual population prevalence or rate.

Anthropometric measure defined as weight in kilograms dividedby height in meters squared

Chronic malnutrition is an indicator of nutritional status overtime. Chronically malnourished children are shorter (stunted)than their comparable age group.

Cluster sampling requires the division of the population intosmaller geographical units, e.g., villages or neighborhoods. In afirst step, usually 30 units are selected among all geographicalunits. In a second and sometimes third step, households areselected within the units using either simple random sampling,systematic random sampling or the EPI method.

220

Confidence interval

Crude mortality rate or death rate(CMR/CDR)

Cut-off points

Design effect

Expanded Program of Immunization(EPI) sampling method

Fortificant

Global acute malnutrition (GAM)

Growth monitoring

Incidence rate

Computed interval with a given probability (e.g., 95%) that the true valueof a variable (mean, proportion, rate) is contained within the interval.

Mortality rate from all causes of death for a populationFormula:

(Number of deaths during a specified period)(Number of persons at risk of dying during that period)

The point on a nutritional index used to classify or screen individuals'anthropometric status.

Cluster sampling results in greater statistical variance (see defini-tion below) than simple random sampling because health outcomestend to be more similar within than between geographical units (seecluster sampling). To compensate for the resulting loss in precision,the sample size calculated for simple random sampling must bemultiplied by a factor called “design effect”: a measure of howevenly or unevenly the outcome (for example wasting, stunting,anaemia or mortality) is distributed in the population being sampled.

A proximity sampling method in which households are selectedaccording to their proximity or distance from the previouslysampled household.

The micronutrient compound that is added to a food either sin-gularly, or as a part of a vitamin-mineral premix.

GAM includes all children suffering from moderate and severe malnutri-tion; percent of children under 5 who have low weight-for-height mea-sured by -2 z-scores and with or without oedema.

Observation of child growth over time by periodic measurementof weight-for-height.

The number of new events, e.g., new cases of a disease in adefined population within a specified period of time. The nume-rator is the number of new cases occurring during a given timeperiod while the denominator is the population at risk.

X time period


GLOSSARY OF TERMS

221

8ANNEXES

Indicators

Infant mortality rate (IMR) (per 1,000)

Iron deficiency

Low birth weight

Malnutrition

Mean

Measurement error

Median

Mode

Monitoring

Morbidity

At the population level, it is a measure used to describe the proportionof a group below a cut-off point.

A measure of the yearly rate of deaths in children under 12 months. Thisis often cited as a useful indicator of the level of health in a community.;To calculate it, the number of deaths of infants under 12 months of agein a given year is divided by the total number of live births in that yearmultiplied by 1,000.

Shortage, insufficiency of iron in the body.Iron deficiency is one of the most common causes of anaemia.

Birth weight of less than 2500 grams.

State in which the physical function of an individual is impaired to thepoint where he or she can no longer maintain adequate bodily perfor-mance processes, such as growth, pregnancy, lactation, physical workand resisting and recovering from disease.

Measure of central location commonly called the average. It is compu-ted by adding all the individual values in the group and dividing by thenumber of values in the group.

Measurement error may lead to two types of error: (1) randomerror, resulting in a larger standard deviation of Z scores forweight-for-height or other measures of malnutrition, and (2)systematic error, if measurements, on average, are biased inone direction or the other.

Measure of central location which divides a set of observations into twoequal parts.

The most frequently occurring value in a set of observations.

The systematic and continuous assessment of the progress of an inter-vention over time.

A condition related to a disease or illness

222

Normal distribution

Nutritional index (indices plural)

Oedema

Outcome

P-value

Percentile

Precision

The symmetrical clustering of values around a central location.The arithmetic mean, the mode and median are identical in anormal distribution.

Derived by relating a child's measurement with the expected valueof a child of the same height (or age) from the reference population.

An accumulation of excessive fluid in extracellular fluid in the body;a distinguishing characteristic of kwashiorkor when bilateral.

Wasting and mortality are examples of outcomes measured in surveys.

If you want to know whether there is a significant difference bet-ween two survey estimates, frequently a statistical test is appliedand a P value calculated. The P value is the probability that the twoestimates differ by chance or sampling error.

The set of numbers from 0 to 100 that divide a distribution into 100parts of equal area, or divide a set of ranked data into 100 classintervals with each interval containing 1/100 of the observations.

Indicates how similar the survey estimates are to each other if asurvey is repeated over and over. Precision is decreasing withincreasing sampling error (see definition). Precision or samplingerror, contrary to accuracy, can be quantified (e.g., in a confidenceinterval).


GLOSSARY OF TERMS

223

8ANNEXES

Prevalence

Protein-energy malnutrition (PEM)

Recall period

Reference population

Sample

Sampling error

Sampling frame

Sampling interval

Sample size

Sampling unit

Sampling universe

Proportion of a population with a disease or condition of interest ata designated time.

Clinical disorders of malnutrition; the two more severe nutritionaloutcomes are marasmus and kwashiorkor.

A defined period in the past used to calculate mortality rates.

The WHO/NCHS/CDC reference values are based on two largesurveys of healthy children, whose measurements represent aninternational reference for deriving an individual's anthropome-tric status.

A part of the sampling universe that should be selected at randomto guarantee a sample representative of the sampling universe (seedefinition below).

Sampling error is the degree to which a sample might differ from thewhole target population, e.g., how well it represents a target popu-lation or sampling universe (see definition). Sampling error can bequantified (e.g., in a confidence interval).

The list of all the sampling units from which you choose the sample.

The sampling interval is the total number of sampling units in thepopulation divided by the desired sample size.

The size of the sample calculated based on objectives of the surveyand statistical considerations.

The unit that is selected during the process of sampling; dependingon the sampling process, the sampling unit can be a person, house-hold, district, etc.

The entire group of sampling units who are eligible to be includedin the survey sample; this population should match the populationfor which you are trying to estimate the outcomes measured in thesurvey (e.g., all children under 5 years old).

224

Severe acute malnutrition (SAM)

Simple random sampling

Standard deviation

Standard error

Stunting (chronic malnutrition)

Surveillance

Survey

Systematic random sampling (SRS)

Vulnerable groups

Wasting (acute malnutrition)

SAM includes all children suffering from severe malnutrition; per-cent of children under 5 who have low weight-for-height measuredby -3 Z-scores and with or without oedema.

The process in which each sampling unit is selected at random oneat a time from a list of all the sampling units in the population.

A measure of dispersion or variation. It is the most widely usedmeasure of dispersion of a frequency distribution. It is equal to thepositive square root of the variance. The standard deviation is asummary of how widely dispersed the values are around the cen-ter mean.

Standard error is a measure of sampling error or precision, howe-ver, the most common method used when presenting survey resultsis confidence intervals.

Growth failure in a child that occurs over a slow cumulative processas a result of inadequate nutrition and/or repeated infections; mea-sured by the height-for-age index. Stunted children are short fortheir age and may look younger than their actual age; it is not pos-sible to reverse stunting.

Surveillance is a regular and continuous collection of informationfor use in analysis, interpretation and decision-making aboutactions or policies in question.

An assessment of health and nutrition outcomes that requires arandom sampling process. Surveys are cross-sectional becauseresults are representative only for the day of survey or (for mortali-ty) during the recall period.

A methodology which selects a sampling unit at random, thenselects every nth household thereafter, where “n” equals the sam-pling interval.

Categories of people who are at heightened risk; common vulnera-ble groups are women, children, elderly and displaced peoples.

Growth failure as a result of recent rapid weight loss or failure togain weight; wasting is measured by the weight-for-height index.Wasted children are extremely thin; readily reversible once condi-tions improve.


GLOSSARY OF TERMS

225

8ANNEXES

Under-five mortality rate (U5MR)

Underweight

Variance

Wasting (acute malnutrition)

Z-score

Probability of dying between birth and exactly five years ofage expressed per 1,000 live births; or the (number of deathsamong children under five years divided by the number of chil-dren under 5 in that population) X 1,000.

U5MR is a critical indicator of the well-being of children.

Percentage of children under the age of five with weight-for-age below -2SD from median weight-for-age of referencepopulation.

A measure of the dispersion of an outcome in a sample. Ahigher variance for the same outcome in cluster sampling thansimple random sampling reduces the precision of the measu-red outcome (see design effect).

Percentage of children under the age of five suffering frommoderate or severe wasting (below -2SD from median weight-for-height of reference population.

Score expressed as a deviation from the mean value in termsof standard deviation units; the term is used in analysing con-tinuous variables such as heights and weights of a sample.


Q & A Sheet

9ANNEXES

1. Q: Is mid-upper arm circumference (MUAC) acceptable as an indicator of nutritional status in our survey?

A: No. We do not recommend that MUAC be used to measure the prevalence of malnutrition in young chil-dren in a population. MUAC is more suitable as a screening tool for determining admission to supplemen-tary feeding programs and for rapid assessments. If you are working with a partner agency that requires thatMUAC be included in a population-based nutrition survey of children, it is essential that weight-for-heightindices are also included. WFP should use weight-for-height instead of MUAC to measure acute malnutri-tion in children. (Refer to Common Mistake #3 for more information)

2. Q: How do I measure nutritional status of pregnant women?

A: Pregnant women are often targeted by WFP's mother and child nutrition programs because of theiradditional nutritional needs. As a result, WFP seeks to measure the nutritional outcomes associatedwith these programs.

During pregnancy, women experience significant weight gain, particularly during the second and third tri-mesters. Unfortunately, the variable nature of this weight gain and the difficulty of establishing gestationalage, means that body mass index (weight in kilograms/height in meters2) alone is not an appropriate indi-cator to use for assessing the nutritional status of pregnant women.

The proportion of newborns with low birth weight (<2.5 kg) is currently being piloted as one of the strategicobjective outcome indicators associated with MCHN programs reaching pregnant women. Although manyfactors can lead to a baby being born with low birth weight, mother's under-nutrition is a major cause, par-ticularly in developing countries. As such, this indicator is a good proxy indicator for women's nutritional sta-tus during pregnancy. However, it should be noted that this indicator is best collected through programmemonitoring rather than through population-based surveys due to reporting bias. For programmes providingfortified foods (such as CSB/WSB) or iron supplements, the prevalence of iron deficiency anaemia is ano-ther indicator that is being piloted by WFP.

In population-based surveys, mid-upper arm circumference is sometimes used to assess the nutritio-nal status of pregnant women because it is relatively stable during pregnancy. However, the disad-vantages are: (1) it is less responsive than weight to short term changes in food consumption or healthstatus; (2) if collected through surveys it can be difficult to get a sufficient sample size to get a relia-ble estimate; and (3) different cut-offs are used by different organizations to define who is “malnou-rished” (18.5 cm and 22.5 cm are often used). In situations where weight gain is measured over timeand gestational age is estimated, it is conceivable that the percentage of women who gain weight ator above expected weight gain rates could be used as an indicator. However, such settings are ratherunusual for WFP to work in.

228

3. Q: What indicators can I use to assess the nutritional status of lactating women?

A: Following childbirth, women still retain additional weight, which is often lost over the course of the few monthsfollowing delivery. The amount and rate of weight loss during lactation depends on many factors, including thewoman's nutritional status prior to pregnancy, the amount of weight gained during pregnancy and whether or notthe woman is breastfeeding. Unfortunately, little published data exist on the rate of post-partum weight loss indeveloping country populations. In absence of guidance from technical agencies on whether or not body massindex is a reliable indicator of the nutritional status of lactating women, WFP is piloting the prevalence of lacta-ting women with BMI<18.5 as an outcome indicator for WFP's MCHN programs. Other outcome indicators, suchas the prevalence of women who are exclusively breastfeeding, are sometimes used as well.

4. Q: Are there any cut-off points for the prevalence of low body mass index which indicate that an emergen-cy situation exists?

A: No. To date, WHO has published guidance to help interpret the situation based on the prevalence ofvarious indicators of child under-nutrition. Information on adult body mass index, however, is increasinglycollected through nutrition surveys.

5. Q: Do I always have to use 30 x 30 cluster survey?

A: No. Cluster surveys should be carried out in large, geographically dispersed populations where no accu-rate list of households is available and where households cannot be visited systematically. If the need for acluster survey methodology is established, do not automatically plan on doing a cluster survey of 900 basicsampling units in a 30 cluster by 30 unit formation. You should calculate the sample size which provides thedesired statistical precision and use this sample size to determine the size of your clusters. (Refer to com-mon mistake #6 for more information).

6. Q: Is the design effect of a cluster survey always 2?

A: No. The design effect is a measure of how evenly or unevenly an outcome is distributed in the popula-tion. Therefore, the design effect varies from survey population to survey population, and differs dependingon the outcome of interest. For example, the design effect for malnutrition is usually in the range of 1.5-2.0,although it can be higher if there is reason to believe that the malnutrition outcome of interest is distribu-ted unevenly in the population (for example, higher levels of wasting in the lowlands included in the surveysample compared with that of the highlands). Some outcomes, such as mortality, can be quite different indifferent parts of the population and, as a result, their design effects can be quite high. The best source forestimates of design effect are prior surveys done in the same or similar populations. (See page 70 for moreinformation.)

7. Q: Should I present results using percentage of the median or Z-scores?

\A: Z-scores are the standard and the preferred mode of presenting anthropometric indicators in nutritionsurveys. Percentage of the median can be presented in addition to Z-scores in the survey report if there isa specific need for this alternative expression (such as when results will be used as a programmatic tool inselective feeding programs).


Q & A SHEET

229

9ANNEXES

8. Q: Can I use clinic data to measure and report on the prevalence of malnutrition in a population?

A: No. These data are not truly representative of the population of interest because they are a self-selec-ted and potentially biased sample. The principle of random probability selection is not implemented whenclinic data is collected; therefore, clinic data do not give an accurate estimate of the prevalence of malnu-trition in the larger population.

9. Q: Can I report UNICEF's figure for under-5 mortality from the State of the World's Children as the “base”rate against which to compare changes in mortality due to emergencies?

A: The UNICEF under-5 mortality rate and the age-specific mortality rate for children under 5 years of ageare very different measures of mortality and should not be confused or compared. An age-specific mortali-ty rate is expressed as the number of deaths in an age group divided by the number of individuals in thatage group who are at risk of death. For example, if the age-specific death rate for children under 5 yearsof age is 35 deaths per 1,000 children under 5 per year, this means that, in a population of 1,000 children,35 will be expected to die over the course of one year. This is a true mortality rate because it is expressedas the number of deaths per the number of people at risk of death over a certain time period.

The UNICEF under-5 mortality rate is expressed as the number of children who die before their fifthbirthday per 1,000 live births. It tells you how many children who are born alive will die before theirfifth birthday. This gives the risk during the entire five years between birth and the fifth birthday. Asa result, it is usually almost five times higher than an annual age-specific mortality rate for childrenunder 5. However, because the UNICEF under-5 mortality rate is based on live births and not theactual population of children under 5 years of age, even dividing the UNICEF rate by 5 cannot provi-de a direct comparison.

For example, Ethiopia has an under-5 mortality rate of 169 per 1,000 live births - that is, the expected rateof death relative to a rate of live births. That is very different from a survey-derived mortality rate whichreflects an absolute number (per 10,000 per day).

See page 40 in the manual for additional explanation.

10. Q: Can I use the same reference population for all populations, regardless of race or ethnicity?

A: Different studies carried out on children have shown that up to the age of 5, growth in children of dif-ferent populations is very similar (Habicht, 1974). The difference in growth becomes apparent on the onsetof adolescence; therefore it is commonly agreed that the same reference population can be used for allpopulations.

11. Q: If my nutrition baseline results were calculated using the NCHS/CDC/WHO reference population, shouldmy follow-up survey use the NCHS/CDC/WHO reference or the new WHO 2005 reference population?

A: If the primary purpose of the second, or follow-up, survey is to measure the difference in preva-lence between the baseline and the follow-up, then it is essential that the same reference popula-tion be used to calculate results. Therefore, if the 1977 reference was used in the baseline survey,

230

the 1977 reference should be used to calculate the results for the follow-up survey as well.Alternately, the data from the baseline survey could be re-analysed using the new WHO referencepopulation. Then the data from the follow-up survey could be analysed using this new reference andcompared to the baseline.

12. Q: I collected data on nutritional status through a VAM “Baseline” (situation analysis). Is this sample ade-quate for measuring outcomes of MCH programs?

A: Typically, it is not. Although VAM situation analyses are intended to reflect the situation of the overallsampled area prior to programme implementation, the sample is typically not large enough to be able torepresent specific working areas of nutrition programmes. As such, unless other information exists on thebaseline prevalence of key nutritional indicators, a separate baseline survey is usually required.

13. Q. Is it still necessary to collect monitoring data on beneficiary outcomes for MCHN programmes if we areplanning to do baseline and follow-up surveys of the overall population?

A: Absolutely. Population-based surveys provide a critical role in assessing the extent to which foodinterventions have achieved outcomes in the general population where those interventions are opera-ting. Nonetheless, indicators collected through programme monitoring play an even more important role,because: (1) they are collected more frequently and can therefore be used for rapid decision-making andfine-tuning the programme , and (2) they provide an insight on what is happening to those who are recei-ving the intervention (in contrast to population-based surveys which normally collect information frombeneficiaries and non-beneficiaries living in the same area). Examples of monitoring information inclu-de beneficiary numbers, regularity of receipt of programme services (including food), as well as outco-me indicators specific to beneficiaries (such as recovery rates or dropout rates for beneficiaries enrol-led in supplementary or therapeutic feeding programmes).

Monitoring information plays an essential role in interpreting changes in nutritional status (or lackof change) observed at a population level. If positive changes are observed, monitoring informationcan be used to help determine if the programme was responsible for this improvement and, if itwas, why the programme was effective. Such information can help to build a case for expansion ofthe programme. If no change was observed or if the situation deteriorated over time, informationfrom monitoring can help to explain why (for example, was it due to low coverage or was it due toa lack of inputs?).

14. Q: I have data on crude mortality rates from refugee camps collected through surveillance. Is this accepta-ble for corporate reporting?

A: Generally, yes. Many camps, particularly well-established ones, have quite reliable mortality surveillan-ce systems, which may even provide more accurate information about mortality than retrospective surveys.Regardless, mortality rates from routine surveillance systems can be subject to many biases, the mostimportant of which is incomplete reporting of deaths. Never accept such data at face value; you shouldalways try to get some idea if death reporting is complete, if the population estimate used to calculate themortality rates is accurate, etc.


Q & A SHEET

231

9ANNEXES

15. Q: Can I understand the causality of malnutrition from a cross-sectional survey?A: Results from a nutrition survey, in combination with complementary information, can help interpretpotential causes of malnutrition. However, survey results cannot establish definitive causality; results mustbe interpreted in context and with caution (see chapter 4, step 4 onwards).

16. Q: Where can I learn about how to use nutrition information in the context of emergency needs asses-sments (ENA) or vulnerability analysis and mapping (VAM)?

A: Analysis of the nutritional situation plays a critical role in the process of needs assessments, both inemergency settings and in development contexts. Guidance on how to use nutritional information in the con-text of emergency needs assessment is available in chapter 6 of the WFP Emergency Food SecurityAssessment Handbook (available at http://home.wfp.org/oen/EFSA/EFSAIndex.htm). VAM has also develo-ped thematic guidelines related to Nutrition and Health which are available athttp://vam.wfp.org/main/them_guid.jsp.

17. Q: Should we collect information about causes of death as part of a mortality survey in an emergency?A: If data on the causes of death are unavailable from other sources, this information can be collectedduring a cross-sectional survey, but the questions should probably be limited to those on injury/trauma andother causes of mortality that are well known and/or that have local terms. (See page 49 for more details.)

18. Q: Can nutrition surveys be used to collect mortality data? A: Yes, but the nutrition survey should be explicitly designed and conducted with attention to this additio-nal objective. Sample size should be calculated including mortality as one of the multiple outcomes (seepage 70). Additionally, in order to correctly estimate mortality, the sample selected for the survey should berepresentative of all households in the population. Mortality information should not be collected from asample which only contains households which have a child under 5 years of age (see common mistake #9).

19. Q: Where can I get assistance identifying a consultant to help me do a survey?A: Often there are specialized technical agencies (international or national NGOs, UNICEF, MOH, etc.)in the country of operation that can provide you with support or collaborate to design and conduct thesurvey in partnership. If this is not available or is not sufficient, please contact [email protected] forfurther support.

Graphic Design: EnergylinkPrinted on November 2005

a manual: measuring and interpreting malnutrition · measuring and interpreting malnutrition and...

Documents