The Gerald J. and Dorothy R. Friedman School of Nutrition Science and Policy FOOD POLICY AND APPLIED NUTRITION PROGRAM

DISCUSSION PAPER NO. 13

Allocative Responses to Scarcity: Self-Reported Assessments of Hunger Compared with Conventional Measures of

Poverty and Malnutrition in Bangladesh

Patrick Webb, Jennifer Coates, and Robert Houser

September 2002

Corresponding Author: mailto:Patrick.Webb@tufts.edu Discussion papers provide a means for researchers, students and professionals to share thoughts and findings on a wide range of topics relating to food, hunger, agriculture and nutrition. They contain preliminary material and are circulated prior to a formal peer review in order to stimulate discussion and critical comment. Some working papers will eventually be published and their content may be revised based on feedback received. The views presented in these papers do not represent official views of the School. The discussion paper series is available on line at http://nutrition.tufts.edu/publications/fpan/. Please submit drafts for consideration as FPAN Discussion Papers to Patrick.Webb@Tufts.edu.

- - 2

Allocative Responses to Scarcity: Self-Reported Assessments of Hunger Compared with Conventional

Measures of Poverty and Malnutrition in Bangladesh

Patrick Webb Jennifer Coates Robert Houser

Summary1 This paper presents preliminary results from research aimed at assessing the validity of alternative measures of food insecurity. It focuses on: a) links between food security status as defined through self-reporting by households themselves, versus interviewer ratings, and comparator indicators of food access, poverty and nutritional status, b) changes in status over time (for a sub-sample of 125 households surveyed first in the Winter/Spring of 2001 and again in the Spring of 2002), and c) insights gained from more in-depth interaction with the sub-sample households that have influenced the module adaptation and validation process. The research finds that a viable set of around 11 questions from the ‘self-reporting’ hunger module appears to work well both in characterizing the problems experienced by households in Bangladesh and in identifying households along a continuum of food stresses. Those questions correlate well not only with interviewer ratings but also with a range of comparator indicators commonly used in the analysis of poverty, malnutrition and food insecurity. There is a high degree of concordance between male and female interviewer ratings, as well as between interviewer assessments of change in household conditions between the two rounds of data collection and households’ own assessments of change (versus stability). While many of the variables tested are strongly correlated with the conditions under consideration no single indicator serves well in defining all

1 The research on which this paper is based is supported by the Food and Nutrition Technical Assistance (FANTA) Project of the Office of Health and Nutrition of the Bureau for Global Programs Field Support and Research at the U.S. Agency for International Development, under terms of Cooperative Agreement No. HRN-A-00-98-00046-00 awarded to the Academy for Educational Development (AED). The opinions expressed herein are those of the authors and do not necessarily reflect the views of the U.S. Agency for International Development or of the Academy for Educational Development.

- - 3

aspects of food insecurity, be it anthropometry, expenditure, food groups consumed, or caloric adequacy. This confirms the need for composite variables that distinguish between outcomes and processes often generically and simplistically characterized as ‘food insecurity’ or ‘poverty’ or ‘malnutrition’. Further statistical analyses (parametric and non-parametric, including Rasch) are needed to gain an understanding of how the determinants of these related but different conditions overlap and where they do not.

- - 4

1. Overview of the Research Agenda

This research tests the extent to which qualitative questionnaire approaches devised for use in the United States during the 1990s (with Tufts and Cornell involvement) can be adapted and enhanced for applications in diverse developing country settings. If the ‘direct measure’ approach can successfully track key benchmark indicators (such as income, assets or nutritional status), while also being sensitive to changes in household economic status associated with project interventions, the approach may in future serve for reporting on food security activities (Zeller et. al. 2001). The Tufts field study is being carried out in parallel with operational activities of World Vision (Bangladesh)’s Food Security Enhancement Initiative (FSEI), a DAP supported by USAID since the end of 2000. Tufts’ role is to explore objective and subjective, quantitative and qualitative measures of food insecurity among a sub-sample of WV’s participant households, plus a control group. The aim is to interact with these households at several points over several years, eliciting information from them about food and dietary perceptions and practices, as well as other quantitative indicators of socioeconomic and nutritional status. These data serve to augment WV’s own baseline and mid-term surveys, thereby allowing for direct comparison with current monitoring and evaluation methods, and also to determine whether the new food security questions are sufficiently sensitive to register impacts from the operational intervention itself. This report presents insights from in-depth qualitative interaction with a sub-sample of the Round 1 households, findings from an analysis of interactions among variables from the first round of collected data, and initial results from a re-survey of the sub-sample households using the all of the instruments from Round 1. 2. Approach to the In-Depth Study The sub-sample of 125 households was selected (from the 606 household sample) in early 2002 to participate in qualitative explorations of the core module approach. One aim was to explore nuances in interpretations by the respondents while also identifying draft questions that did not perform well in Round 1 due to respondent confusion, a lack of variability in responses, or poor inter-item correlation. Sub-sample households were asked to help in improving the module itself by engaging in ‘debriefing sessions’ either singly or in focus groups. During such interactions respondents were able not only to provide feedback on the appropriateness of language used and questions asked, but also to suggest alternative wording and questions. The following criteria were applied in sub-sample selection: a) Each food security category (as determined by the female enumerator rating from the baseline study) was to be equally represented; b) The sub-sample was to be drawn from locations in which the distribution of households across the three categories is roughly equivalent;

- - 5

c) The number of household clusters should be minimized so as to allow for in-depth interaction. This contrasts with Round 1 where the sample was selected to optimize wide geographical representation. These criteria lent themselves to a purposive sampling procedure, wherein eligible upazilas were pre-identified and all baseline households located in the upazila clusters were revisited. The final sub-sample comprised 7 household clusters located in Moksedpur, Phulpur, Shatikira, and Faridpur districts in the North, Center and Southern regions. A number of instruments were developed or modified for the sub-sample study:

a. A broad quantitative questionnaire (which includes 2 days of diet recall) b. The revised ‘hunger module’ c. A debriefing questionnaire in which respondents elaborate on their

interpretation of hunger module questions (including differences by gender) d. Group Informant Rating (GIR) and Focus Group Discussion (FGD) guides e. One female FGD with participants from each Food Security rating category f. A Community Questionnaire

A qualitative questionnaire guide was developed during a three-day “formative research trip” to a village in Halughat, a World Vision intervention upazila from the baseline study in the North of Bangladesh. With the assistance of senior DATA managers GIR and FGD exercises were pre-tested with male and female participant groups. A facilitator guide was subsequently elaborated. Two pre-tests (one for the quantitative instruments and a separate one for the qualitative guides) were conducted in Saturia upazila prior to the formal training. The training of 14 experienced enumerators (6 male/female pairs plus two supervisors) was conducted over a period of 15 days. The training focused on underlying concepts of the food security module, the appropriate delivery of questions, and on how to debrief after the formal interview. Four enumerators with prior experience in qualitative research were trained to facilitate separate focus groups and group rating exercises. The supervisors were designated to administer community-level questionnaires on top of their other duties. Two days of pilot testing were required to ensure enumerator proficiency with the multiple instruments and determine the most appropriate sequencing of components. Male/female enumerator teams joined forces to complete one village at a time. This approach alleviated some of the enumerators’ burden by allowing supervisors to fill in gaps where necessary and to rely on each other for logistical support. The teams resided together during the survey which also allowed for fruitful exchange of insights while providing time for cross-checking of completed questionnaires. The Field Coordinator and two senior DATA managers spent time supervising in 3 of the 4 target upazilas.

- - 6

3. Insights on the Meaning of Questions and their Interpretation Since the ‘direct measure’ of food insecurity is based on self-reported behavior it is often assumed that respondents’ perceptions of the experience of food insecurity are fully reflected in their answers. As Brock (1999) puts it, less tangible aspects of food insecurity “emerge from the words people use to describe their experiences.” According to Rahman (1991), capturing perceptions is critically important to broader development since “the perception of a community about itself and its ability to deal effectively with vulnerabilities...can make a significant difference in [its] ability to improve its material base or social institutions.” However, the assumption that perceptions are adequately captured either through self-reporting or even through observed behavior is difficult to test. Stress, anxiety and ‘insecurity’ are often assessed by medical professionals in hospitals or refugee camps in terms of the impact on nutritional outcomes of acute ‘life events’ such as death in the family, major sickness, or conflict trauma (Schneider and Hebuterne 2000; Spiegel and Salama 2000; IFRCRCS 2001).2 Anthropologists and psychologists have long recognized that perceptions affect both personal interpretations of past experiences and reported anticipated outcomes (Jochim 1981; Deyo 2000; Block and Webb 2001). This is important where a self-reporting tool to measure food stress is concerned. Yet the assessment of perceptions in the context of chronic hunger or food insecurity is in its infancy. As Lozano et. al. (1999) put it, “there is little research examining the relation between hunger and attitudes.” According to Tierney (1999) such research is important because “the beliefs people hold about risk are typically used in social sciences to explain behavioral outcomes, such as the actions people take to protect themselves against hazards. However, such perceptions might more usefully be studied as dependent variables; that is, by focusing on where ideas about risk come from in the first place.” While the current research is not designed to assess these psychological dimensions in great depth it was recognized even before Round 1 that careful cognitive testing was needed to only to ensure clarity in the interpretation of responses but also to understand how respondents interpret hunger within a larger context of anxiety, stress, and insecurity--the ‘unseen’ elements of hunger not easily proxied by observable parameters. Thus, following Alaimo et. al. (1999) and Derrickson and Brown (2002) interviewer teams engaged respondents in the sub-sample in formal and more informal conversations to explore meanings of individual questions as well as respondents’ understanding of the answers that they gave. This extended cognitive testing has been crucial to the reformulation of certain items and the dropping of others.3 2 Engle et. al. (1996) suggest that ‘vulnerability’ itself can be defined as an “individual’s predisposition to develop…behavioral ineffectiveness or susceptibility to negative development outcomes that can occur under high-risk conditions.” 3 Deeper insights in this direction are expected in later phases of the research. Translation is already underway of local ‘hunger diaries’ and fuller explanatory notes written in the margins during interviews by the enumerators. In addition, the nuances of how perceptions and attitudes affect observed behaviors will be further explored through interaction with the sub-sample households.

- - 7

Numerous issues have arisen as a result of the in-depth interaction, many of which will require further consideration. A few of the more important issues are raised here: 1. Food Cultures. Great care is needed in ensuring that religious or cultural practices do

not distort responses to questions that should be appropriate to all households. That is, local food habits/preferences need to be well understood in order to be able to interpret the validity of questions relating to food behaviors. In one survey village Moslems consider the consumption of crustaceans to be an extreme sign of food distress (carrying connotations of ‘unclean’ habits), while some non-Moslems consume crustaceans on a daily basis without any sense that this is an inferior food. There is no apparent (visual) difference between such households so misinterpretation of food norms can hugely distort the interpretation of responses relating to food groups consumed, total value of foods, or ‘unusual’ foods consumed.

2. Food Preferences. Closely linked to food culture issue are food preferences that make it difficult to identify ‘icon foods’ (the consumption of which might be a marker of distress). While so-called ‘inferior’ foods are often used in intervention programs as a means of reducing leakage to less poor households it is becoming clear that the concept of ‘inferior’ can be quite localized. For example, sweet potato (mishti alu) and wild water lily (shaluk) were identified in the pre-testing phase as foods typically only eaten by people facing food stress. In fact, almost 10 percent of households reported consuming sweet potato at least once a year because they like the taste (2 percent reported eating it at least a couple of times each month). By contrast, almost 100 percent of households denied ever eating water lily even during times of major flood or famine. Other potential distress markers also failed once their local specificity was assessed in more detail. For example, a river plant called Gom Bhaja is only consumed if one lives close to a river where it can be harvested and therefore not easily used as a food marker beyond such localities. The same applies to Bonn Kochu (wild taro) a “famine food” suggested by Bruce Currey (1984). While 95 percent of respondents in this sample said that they “never” consumed Bonn Kochu, 5 percent report at least sometimes eating it, but only when it is available—a constraint that relates to location, season and price substitution effects. In other words, as stand-alone markers none of these foods works well. No single item has yet been found that could be used to represent food distress for the entire sample, all of the time.

3. Dietary Balance. While it is already known that the term “balanced meal” (used in

the US food security module) does not have an easy translation in many developing country contexts an attempt was still made to construct a question aimed at assessing respondent understanding of food quality and diversity. In Jatia village, Phulpur upazila, a focus group was conducted with households representing the “food insecure without hunger” category as rated by enumerator the prior year. While several of the participants defined their perception of a good (‘full stomach meal’) in terms of satiety or quantity of food consumed relative to other meals in the day, others considered a meal incomplete (hence not a ‘full stomach meal’) if it did not include meat, fish or dal. Furthermore, there were differences across food security categories. For example, households from the ‘food secure’ group stated that a full

- - 8

stomach meal has to include ‘big fish’, the insecure without hunger group argued that a good meal can include fish of any size, while the hungry households proposed that a meal would be good if it contained dried fish. In other words, in this (relatively wealthy) village dietary quality appears to be contained within the construct of food adequacy or sufficiency.

4. Dietary Diversity. Considerable time was spent in the attempt to define a ‘food

groups consumed’ question by adapting the Indian and Chinese food pyramids and basing questions around a visual representation of Bangladeshi coins (some of which show food items on a silver coin—chicken, fish, fruit and a vegetable). So far these attempts have had limited success because of the difficulty of communicating ideas of nutritional balance in the context of widespread energy deficiency. While many households understand that a diversity of foods is ‘desired’ this may be because it reflects wealth rather than nutrition knowledge. More research is needed on food substitution, as well as on diversity within (not just across) food groups in order to construct questions about desired consumption paths once a caloric threshold has been reached.

5. Hunger versus want. Use of common terms does not guarantee common

interpretation. Just as Studdert et. al. (2001) found that the term “hunger” in a questionnaire was considered embarrassing and almost offensive in parts of Indonesia, the term ‘hunger’ can have multiple meanings in Bangladesh. Extended interaction with sub-sample households allowed for an exploration of nuances in terms such as akla and durvickha, suggested by Currey (1984) as different terms for “hunger scarcity” (the first relating to a short-term problem, the second to a more chronic, structural condition). For example, according to households in the Haluaghat region there is no difference between the terms—locally they are synonymous. They use both terms to mean short shocks (such as droughts or floods) and a different word to define lean times unrelated to shocks (obhab, meaning ‘want’ or penury linked to health problems or lack of employment income). In other localities the terms are used roughly as described by Currey (1984). Thus, the temptation should be avoided to build generic terms relating to the outcome under consideration (hunger) into the questionnaire since those words themselves contain different meanings to different people.

6. The trajectory of food insecurity. According to Sarlio-Laehteenkorva and Lahelma (2001) “the association between past and present economic disadvantage [and] food insecurity…is poorly understood.” There are three main reasons for this: first, most studies of food insecurity and hunger are cross-sectional in nature; second, recall of past behavior is notoriously difficult to measure and validate; third, disentangling recent experience from current status is problematic—especially where questions about food security are based on recall over 12 months.

Interaction with the sub-sample households allows for an exploration of this key issue: are respondents answering module questions based on what they feel today, what they believe was the case over the last year, or how they think things are

- - 9

becoming? Already there is a sense that most households do respond as required—averaging conditions over the previous 12 months regardless of whether things are likely to improve in coming months. However, the effect of short time-bound shocks (positive or negative) on 12 month recalls requires further investigation. For example, 6 households interviewed in R2 (of the 125) had ‘lumpy’ non-food expenses that greatly exceeded not only their reported R1 disbursements but also those of all other households in the sub-sample. Closer scrutiny revealed that this resulted mainly from one or two large expenses, such as a wedding, dowry, purchase of a new rickshaw, a motorcycle, or veterinary treatment for a sick cow. These big ticket items represented 90 percent (on average) of non-food expenditure and up to 95 percent of their total expenditure, including all outlays for food. The potential for windfall, investment or catastrophic expenses to distort trends has to be examined more closely; since there is no ‘normal’ baseline year where food security is concerned the relative starting point (prior conditions) for differing households has to be controlled for. Whether the food security trajectory can be characterized as a line running from A to B, or as part of an arc (be it converse or concave) matters greatly to expected outcomes.

7. The Gendered Perspective. It has been argued by Monello and Mayer (1967) that “men appear to experience hunger [differently] than women…in a more specific physical way.” It is also suggested that men describe hunger differently than women (Rime and Giovanni 1986; Macht 1999). This has important implications for a self-reported statement about hunger that seeks to reflect conditions for entire households. The US module, for example, was formulated with the explicit purpose of “understanding hunger from the perspective of women who had experienced it and to construct and evaluate indicators to measure hunger directly in similar populations”, namely, women (Radimer et. al. 1992). Most cross-cultural applications of the US core module approach have also only interviewed women (Welch et. al. 1998; Maxwell et. al. 1999; Derrickson and Anderson 1999; Studdert, et. al. 2001). This raises a question about potential gender (or status) bias in module responses. Does what mothers (not even ‘women’ but ‘mothers’) say about household experiences accurately reflect what others in the same household perceive and experience?

This is an important issue that will be elaborated on in coming months. However, it is possible to report the following:

a) The overall the rate of concordance in answers (men and women responding independently with the same answer to the same question) was high: male and female respondents agreed 81 percent of the time;

b) There was more divergence of opinion in food secure households (23

percent disagreement) than in hungry households (10 percent disagreement). It seems that the behavioral options that are open to a given household the more chance there is of disagreement since both spouses may be engaged in many types of income diversification, social capital building or coping strategies. It may also reflect an increasing

- - 10

independence of women (more work outside the home, separate incomes) in more prosperous households (an hypothesis to be tested);

c) The closest agreement in responses related to questions about group

experiences (such as number of meals eaten on a typical day—almost perfect agreement). The largest divergence in responses relates to questions about cash (purchase of snacks, borrowing money, using money for food instead of something else); this is probably because such actions are not always visible to everyone and may not even be discussed with the spouse because such decisions may not fall in the female domain;

d) There is also large disagreement between men and women on what

constitutes ‘good food’ (bhalo mondo) which suggests that this difference needs to be clarified before the next round—do women believe they are preparing fine dishes while men secretly wish for a different menu? This in itself would have implications for a choice of questions about ‘balance’ and quality in the diet.

8. Cultural assumptions about maternal buffering. There remains some uncertainty

about whether a “child not eating for entire days” is recognized in local terms as the ‘worst case scenario’. While the US module takes it for granted that adults accept that restricted food consumption for children is worse than for adults, and to be avoided at all costs, this is a tricky assumption to export. Recent studies of intra-household food distribution raise questions about the universal applicability of concepts such as maternal or ‘equity’ in food distribution. For example, Wasito et. al. (2002) report that during the 1997/98 crisis in Indonesia, “the nutritional status of mothers remained relatively unaffected while the nutritional status of children deteriorated.” While in China (Wei et. al. 2002) and Myanmar (Thwih and Yhoung-Aree 2002) it has been argued that a “contributions rule” is more applicable in explaining food allocations than ‘nutritional need’ or ‘equity’; that is, individuals who make greater contributions to the family receive a higher percentage of the family’s food.” Both such possibilities have surfaced during discussions with survey households in Bangladesh.

For example, when asked “who normally gets the most full-stomach meals”, some families answer “I know that I must keep food for my children even if I have no food at all”, but others say “the son who pulls the rickshaw”, or “those who go outside to work get the most”. One female respondent stated that “if our husbands don’t get food how would they get for the next meal?” This is not to suggest that feeding children is a low priority for any mother or that adults will selfishly feed themselves while starving their child. However, children skipping meals may be one, among many, adjustments that households make to ensure their survival, especially where that survival depends on the continuing ability of an income earner to earn that income. Would a ‘contributions rule” change the nature or sequence of questions in the hunger module? This has to be answered in the coming year.

- - 11

9. A Food Security Continuum? Do local people understand the conceptual construct that categorizes households into groups along a continuum from of a state of food insecurity with hunger to one of food security?4 There are indications that such a conceptualization may not be fully shared by all households concerned. In-depth discussion suggests a characterization of insecurity as a many-headed beast—one that sees gains in one area of life simultaneous to losses in another. For example, while floods can destroy certain assets and inhibit income streams, they are also widely seen as enhancing long-term productivity (through the silt deposits on farm-land) and potentially enhancing security (where char river banks shift in favor of landless people).5 Similarly, losses in food consumption can sometimes be accepted for a while if balanced by gains in access to assets or other investments.

Interestingly the idea of a continuum is not typically compared against a ‘threshold’. Few households feel that once they cross a certain point they become ‘secure’; the fact that a single death in the family can plunge a family from self-reported food security into insecurity reemphasizes the fragility of life in an environment characterized by multiple risks to life and livelihood. What is more, some households in focus groups did express the idea of relative insecurity (“what makes me different from my neighbor?”) rather than a sense of absolute insecurity based on the notion of ‘a day without food’. For example, during one group informant rating exercise in Haluaghat participants were asked why they had categorized households as they had done. In several cases, the fact that a family had a daughter nearing marriageable age (hence soon needing a dowry) or had a quickly growing family was sufficient cause to rank a household as more insecure today than they had been a year earlier. In one case a household was rated a more food secure because the daughters of the household had recently been married. In another group exercises participants had some trouble in deciding whether a household was food secure or a category lower. They recognized that the household head owned his land and also had a paying job. However, many participants argue that he was not really secure because he had only recently obtained the job (with a local PVO) and only two years earlier had been forced to mortgage some of his land. The final consensus of the group was to classify that particular household head as insecure—his recent past carrying more weight in their minds than his current status.

Thus, while the idea of a composite index of food insecurity is attractive, whether a continuum best describes food security rankings and whether rankings are locally understood in absolute terms (against a threshold) rather than relative (to all other households in the community) remain open questions.

4 It might be pointed out that ‘food security’ is not alone in being defined in terms of relative states along a continuum. For example, according to WHO (2001) “iron status can be considered as a continuum from iron deficiency with anaemia, to iron deficiency with no anaemia, to normal status with varying amounts of stored iron.” This conceptualization is similar to that of food security with, versus without, hunger. 5 New lands are created as well as destroyed each year in the large rivers and estuaries. Where land is scarce and labor bondage the norm, the creation of new farms after floods can be a boon to the formerly landless despite the threat that these land may also one day disappear again.

- - 12

10. Supply Constraints Inhibit Access to Resources. A draft question about “borrowing money at high interest rate” was dropped earlier when it became clear that “high interest” is a relative concept. However, it has since become clear that the definition of interest rate is perhaps not the biggest problem with such a question. Two items in the hunger module will probably have to be dropped from R3 of data collection; namely, questions asking about borrowing money from moneylenders (whatever the interest rate), and selling or mortgaging assets in order to buy food. It was found that there is very little variability in responses to such questions: roughly 99 percent of respondents said they ‘rarely’ or ‘never’ do either. But this is not because they did not want to borrow or sell. The problem with these questions relates to constraints on the supply side.6 Wealthy households are unlikely to seek to borrow at high interest or sell assets. But poorest households desperate to borrow or sell they are unlikely to be able to—borrowing is constrained because they lack collateral and few lenders will lend to them, while selling assets is constrained either because they have few assets to sell or few assets of value to a buyer. To be useful a module question must be universal such that all households have the same potential to answer it. It is now clear that these two questions do not meet that requirement.

The next section presents preliminary empirical findings from the first round of data collection and explores the links between ‘hunger module’ questions and a range of other indicators of food and nutrition security. 4. Links between Food Security Status (Round 1) and Comparator Indicators There has been a growing effort in recent years to make food security ‘indicators’ more reflective of the multifaceted nature of development problems. Building on the broad acceptance gained during the 1970s of the Physical Quality of Life Index (Morris 1979) and its successor, the Human Development Index (UNDP 2001), measures of poverty and food insecurity have increasingly rejected unitary proxies and embraced more nuanced composite measures. For example, Morris et. al. (1999) suggest a 10-item expenditure composite as a proxy for income and a ‘rural asset score’ (based on ownership of various categories of consumer durables and productive assets) as a viable proxy for wealth. Tschirley et. al. (2000) also test a set of 39 variables (that includes assets owned, diversity of commodities produced and sold, and income sources) that can act as a combined proxy for total household income. Similarly, Zeller et. al. (2001) use a multidimensional composite for ‘poverty’ that combines 34 measures of human capital, dwelling conditions, value of assets and food deprivation. In each case, the authors argue that composite variables are conceptually more credible than single indicators and that they correlate well with chosen benchmarks. But this raises the question of what a ‘benchmark’ actually represents. Just as income, expenditure and wealth are common proxy measures for a conceptual construct called

6 Of course this is similar to the issues of households only consuming certain types of ‘big fish’ is they live close to big water bodies.

- - 13

‘poverty’ there are many possible proxies and comparators for ‘food insecurity’.7 The food security concept is built around food availability, access and utilization, as well as the potential risks that can compromise all three (Webb 2001). Each element of the concept can be proxied individually using any number of generic indicators (Figure 1) or in terms of combined variables. Based on an analysis of data from 5 countries Haddad et. al. (1994) argue that “relatively simple indicators perform well in locating the food and nutrition insecure”.8 For example, Zeller et. al. (2001) define ‘the proportion of clothing expenditures in total household expenditure’ as a benchmark for poverty, arguing that clothing expenditure is one component that remains stable over time while increasing proportionally with household spending. By contrast, Christiaensen and Boisvert (2000) measure food vulnerability against an absolute calorie consumption threshold. Maxwell et. al. (1999) compare their indices of food ‘coping strategies’ with per capita expenditure, caloric consumption shortfall (<80% of 2,230 kcal/AEU/day), and the presence of stunting in children under 3 years of age. Ali and Delisle (1999) relate coping strategies to dietary adequacy and food stores in the household. Chung et. al. (1997) define their “benchmark indicators of food insecurity” as caloric inadequacy (<70%), severe stunting in children under 5 years, and clinical measures of micronutrient deficiencies.9 And while Morris et. al. (1999) do not seek to measure composites against a benchmark they assess how well a 10-item set of expenditure items mirrors total expenditure on all items (the assumption being made that expenditure is a valid proxy for income). The important point here is that most of these ‘indicators’ or ‘benchmarks’ could be placed on either side of the analytical equation; they can (and often do) serve either as determinants of the condition under consideration, or as representing the condition. It is widely accepted that the absence of any gold standard measure of food insecurity argues for a cross-referencing of indicators and methods (a convergence of evidence approach) rather than use of a single benchmark used to proxy a (non-existent) gold standard.10 That is the approach adopted here. A range of indicators that ‘perform well’ in defining at least parts of the food insecurity problematic are included in the analysis below.

7 Zeller et. al. (2001) argue that hunger is an “unambiguous measure of poverty”…without elaborating on how ‘hunger’ can be unambiguously assessed. 8 The authors define “perform well” in terms of ‘overlap’ of more than 33.3% (the percent of households in the lowest expenditure tercile which also fall into the lowest calorie adequacy tercile), and an absolute t-value greater than 1.96 (signifying an overlap is significantly different from 33.3%). 9 It should be pointed out that in their study of northern Iran Omidvar et. al. (2002) recently found that “no relationship was observed between serum retinol and other dietary variables.” The authors note that “serum retinol levels did not differ according to BMI or for anemic versus non-anemic subjects.” This study joins the accumulation of evidence that clinical and anthropometric measures of nutrition are only weakly related to other outcomes of concern to policymakers, and hence should be treated with great care when used to proxy poverty or food insecurity.

Figure 1- A conceptual framework of food security and generic indicator categories (adapted from Webb et

Food Security

Food Availability Food access

Resources • Natural • Physical H

Production • Farm • Nonfarm

Income • Farm • Nonfarm

Consump• Food • Nonfoo

GENERIC INDICATOR CATEGORIES Resources Production Income Consump Natural Total area cultivated Total income Total expen Rainfall level, stability Irrigated area Crop income Food share Soil quality Area in fallow Livestock income Non-food e Water availability Access to, and use of, inputs Wage income Food Store Forest resource access Number of cropping seasons Self-employment income Dietary div Crop diversity index Producer prices Meals per d Physical Crop yield stability Market, road access Dependenc Livestock ownership Food production level Migrant remittances Gender bia Productive assets owned Cash crop production Number of earners access, ca ‘Commons’ resources access Sources of non-farm income Price Terms of Trade Food taboo Physical constraints Credit/borrowing levels Private barter/transfers Breastfeed (eg. Landmines, closed borders) Functional Landlessness Public transfer share of Monetary v total income consume Human Wealth markers (eg. saris Household Gender of household lead or tin roofs owned, Food Subst Age, education, literacy levels spending on clothing/shoes) ‘Crisis’ foo Social capital markers Household life cycle stage

. al. 1993)

tion

d

Nutrition • Child • Adult

Food utilization

tion Nutrition diture Anthropometry

in spending Micronutrient xpenditure levels d Morbidity ersity Excess Mortality ay consumed Fertility y ratios Health service s in food usage re Access to s clean water

ing behavior Sanitation alue of foods Coexisting obesity

d and undernutrition caloric adequacy Ratio of low/normal itution adult BMIs d consumption

- - 15

Self-Reported Food Insecurity and Hunger A set of 30 questions was tested in the first round of data collection (606 households), and most of the same questions were applied a second time to a sub-sample of 125 households one year later11 As noted above some module questions were dropped, some were slightly changed in order to make their meaning clearer, but the majority was the same across both rounds. A Principal Components Analysis (PCA) was conducted on these 30 or so items. Three factors were derived from the rotated matrix--groupings of question responses (or ‘items’) that correlate well with each other and are distinct from other sets of items. The first factor of 11 items has a high reliability coefficient (Chronbach Alpha) of 0.89, and explains 47 percent of sample variance.12 The 11 items are as follows:

1. Obliged to eat wheat instead of rice (when rice would have been preferred) 2. Needed to borrow food in order to meet social obligations (to serve a meal to

guests or relatives) 3. Took food (usually rice or lentils in kind) on credit from a local store 4. Worried frequently about where the next meal would come from 5. Needed to purchase rice frequently (because own production or purchased

stores ran out) 6. The family ate few meals per day on a regular basis 7. The respondent adult personally cut back on the amount of food consumed

each day (because there was insufficient food in the household) 8. Needed to borrow food from relatives or neighbors to make a meal (making

ends meet on a day-to-day (hand-to-mouth) basis) 9. The main working adult sometimes skipped entire meals (due to an

insufficiency of food in the household) 10. There were times when food stored in the house ran out and no cash to buy more 11. The adult respondent (where not the main working adult) personally skipped

entire meals due to a lack of food in the household. A few clarifications are in order. First, these 11 questions were drawn by the principal components analysis from the full set of 30 or more items tested—they were not applied as a discrete set in the sequence listed above. Second, as hoped for the 11 items cover a range of elements of the food security concept. For example, while some questions relate to a lack of food in quantity (food stores depleted, restrictions on how many meals can be consumed each day, adults reducing food consumption or skipping meals), others relate to food preferences or quality (lack of choice in grain consumption), issues of social acceptability or stigma (taking credit in-kind from shop keepers, being obliged to borrow food to meet social obligations), and anxiety or insecurity (worrying about where the next meal will come from).

11 The questions were derived through an elaborate pre-testing process reported in Webb et. al. (2001). 12 This Alpha value is almost identical to the 0.9 level reported by Radimer et. al. (1992), and compares well with the 53 percent of variance explained by the first factor in their PCA. Kendall et. al. (1995) report a somewhat lower Alpha of 0.84 for their item measure of household food insecurity.

- - 16

Third, none of the questions relate specifically to children. In fact, the only question that competes for a spot in the first factor relates to “infants skipping entire meals”. This is such an extreme eventuality that only 3 percent of households answered this question, compared with less than 1 percent in the US and 6 percent in a separate sample of households from Bangladesh (Table 1). It should be underlined that while the households in the current sample are poor and food insecure by any international standards they were not purposively selected to represent the worst case or ‘ultra poor’ conditions.13 The issues involved in defining a separate food security scale for children are addressed by Nord and Bickel (2002), and the role of child-specific questions in Bangladesh under conditions of more extreme distress will be explored further in the next phase of research.

Fourth, while strongly inter-related as a group the individual items have differing degrees of statistical correlation on a one-to-one basis. The weakest correlation (although still statistically significant at the p<0.01 level) is between ‘borrowing food to serve guests’ and ‘having to eat wheat instead of rice’, a correlation of 0.29. The majority of other items are correlated at a level exceeding 0.5, rising to 0.73 in the case of the association between ‘respondent adult skipping entire meals’ and ‘food stores running dry and there being no cash to purchase more’. Despite these strong two-way correlations it should be remembered that the set as a whole works much better than any individual items in describing, and hence identifying, food insecure households (more on this below). Fifth, a second factor derived by the PCA included only three items: namely,

• The family frequently ate ‘big fish’ (such as carp or hilsha). This is seen as a sign of wealth although it has to be noted that habitation close to large rivers increases the prevalence of this behavior.

• The family typically eats meat as part of an ordinary meal (i.e. not including meals prepared specially for festivals or holy days).

• The family frequently prepares bhalo mondo, or ‘good’ food (in the local understanding of meals that are rich and satisfying).

All 3 items relate to food quality. They work well as a set in defining food insecure households, but much less well than the first 11 items already discussed. A third factor included items that do not function well, such as borrowing cash from moneylenders or selling or mortgaging valuables. As noted above, these reflect behaviors that require willingness on both sides of the transaction—the fact of not selling assets may mean either ‘no need to sell’ or being ‘unable to find a buyer’. The underlying processes are quite different.

13 The 600 Bangladeshi households referred to in Webb et. al. (2002) and in Nord et. al. (2002b) are different from the 606 households referred to here. The former represent a purposively sampled group of ‘absolute poor’ (mainly female-headed) households associated with the BRAC/WFP micro-credit activity. Those are distinct from the households involved in World Vision interventions and the current FANTA research. While the World Vision activities are largely located in regions considered to be food insecure not all households in such districts experience the same degrees of deprivation. It should be recalled that Radimer’s (1992) sample of 32 women in New York State was not a random sample for the US population; the women were identified through welfare agencies as people relying to some degree on public assistance.

- - 17

Table 1. A comparison of responses to food security questions in the U.S. and Bangladesh _________________________________________________________________________

United States (2000) Bangladesh (2001)1 World Vision BRAC/WFP

_________________________________________________________________________ (Percent)

Anxiety Issues * Worried food would run out and no money to buy more 15.1 36.3 71.9 Adult Issues * Adult skipped meals 5.4 23.0 52.8 * Adult did not eat entire day 1.0 1.2 2.3 Child Issues2

* Child skipped meals 0.6 3.4 6.2 * Child did not eat entire day 0.2 0.3 0.0 ________________________________________________________________________ Source: Nord et. al. (2002), Tufts/World Vision/FANTA survey data (2002).

1. The 606 World Vision households are those included in the current analysis—poor but not purposively selected to represent the truly destitute. By contrast, the 600 BRAC/WFP households represent participants in a micro-credit and food aid intervention explicitly targeted towards the ultra poor (see Webb et. al. 2002).

2. In the US measure a child is defined a 0 to 17 years. In the Bangladesh surveys a child was defined as being 0-5 years. This difference is important in interpreting the outcomes.

- - 18

Interviewer Rating of Household Status A first ‘comparator’ against which to assess the 11 item set is an interviewer assessment of household food security status recorded at the end of long interaction with key members in each family. While interviewer ratings can be considered ‘subjective’ it is worth recalling that practitioner ratings are widely used in medical and psychological fields either as stand-alone assessments or as reference points against which to compare self-assessments of pain, wellbeing or ill health (Green and Reid 1996). For example, psychological studies of personal ‘satisfaction’ face the same problem of being unable “to test systematically respondents’ self-ratings against objective indicators”, and therefore rely either on scale constructs (using item response theory) or professional evaluation as an external benchmark (Bowling and Windsor 2000). The early validation work on the US module relied heavily on interviewer ratings or expert opinion as a basis for assigning US households to different categories of food insecurity with, or without, hunger (Radimer et. al. 1992; Kendall et. al. 1995; Frongillo 1999). During the two-week long training process it the interviewers (most of whom are professional enumerators with considerable experience) developed a shared understanding of the intent of food security module as a whole, as well as of the significance of individual questions. While reviewing the ideal interview approach trainers emphasized the nuances of question wording and the importance of eliciting accurate responses. The enumerators were requested to carefully reconstruct a 12-month recall period for each question guiding the respondent through the year using locally-relevant temporal markers, such as the boro harvest, Ramadan or Kurbani Eid. While time-consuming, pre-tests based around the construction of timelines (and using piles of jujube fruits to represent behavioral frequencies) suggested that it was essential for accuracy.14 Enumerators were also trained to assess, and provide a rating of food security for, each household. The rating was recorded after completion of both the quantitative questionnaire and the hunger questions. Enumerators were instructed to base their assessment on the totality of information gained (verbal, visual, measured) during their time with each family. The rating categories (food secure, food insecure without hunger, and hungry) were defined during training and a lengthy discussion of the rationale for placement in one of these categories took place during pre-test debriefings and feedback sessions. Enumerators were also asked to provide a written rationale on each questionnaire supporting their rating. Supervisors and the field manager sought to assure that male and female enumerators working together in the same household made independent rating decisions. This was important since the degree of consistency of such ratings across genders needed to be explored. Since men mostly interviewed men, and women interviewed women, a question arose about potential bias by gender of enumerator. In fact, a remarkable degree of agreement was seen between male and female interviewers in their separate 14 Without memory joggers respondents were sometimes prone to careless answers or to reporting their current situation as opposed to conditions during the prior 12 months.

- - 19

assessments of food security status. Overall 85 percent of households were classified within the same category by male and female interviewers. For example, males classified 42.3 percent of the 600 households as food secure, while the rate was 42.4 percent according to the female enumerators. However, one caveat is important. While the degree of concordance was impressive where the ‘food secure’ rating is concerned greater divergence was apparent at the other end of the scale. That is, more households were classified as being hungry by female interviewers (over 30 percent) than by men (22 percent). A female bias towards rating households as ‘hungry’ was also found by Bergeron et. al. (1998) in Honduras. This may be because female interviews tend to focus on food preparation and consumption issues rather than a discussion of assets, income sources or other questionnaire elements. It may also be that village women share more information with each other on an informal, ongoing basis than men do. Conversely, it could also relate to differing concepts of the functional versus structural implications of food security by gender. Potential gender bias from the interviewers’ side warrants further consideration, especially since the US (and other) self-reported modules tend to be narrowly focused on the experiences and concerns of women only. Table 2 shows the rate of responses to the 11 items according to interviewer assessments of food security status. It is clear that ‘eating wheat despite a preference for rice’, and ‘borrowing food to make special meals to serve guests’ are behaviors common to many households—both food secure and insecure. While some (not all) food secure households engage in those behaviors at times (to get through difficult periods) they do not report many of the other 9 practices in the list. Food insecure households, by contrast, engage in a wide range of behaviors reflecting increasing severity of the insecurity. Again it should be reiterated that not all households report every behavior—it is the combined set of 11 items that together identifies households with similar characteristics. For example, Table 3 illustrates how the 11 items compare with the female interviewer rating for ‘hungry’ households and with per capita expenditure. All correlations are statistically significant at the .01 level. The 11 item responses to the hunger classification are strongly correlated (ranging from 0.4 to >0.7) while the cross-correlation of each item against the set as a whole reaches 0.82. However, while the 11 items are all significantly correlated with per capita expenditure (a commonly used indicator of poverty), they are only weakly so—suggesting that expenditure level is not a strong proxy for the way that households in this sample perceive food insecurity. This brings us to the issue of associations among proxies of food insecurity and proxies of poverty. Table 4 compares mean values of a number of commonly used proxies against 3 different food security categorizations based on, a) the 11 item module, b) the interviewer rating, and c) a combined variable derived from caloric adequacy and food share in total expenditure. The results suggest that in a simple comparison of means there is considerable overlap across the food security categories for many indicators, including dependency ratio, food expenditure per capita, land area cultivated, sari ownership, roofing, unique foods, and food groups consumed. That said, the relationships do not hold well for all indicators. For example, being a female headed household serves as a

- - 20

Table 2. Number of households affirming a response to items in the food security questionnaire, by food security status (female interviewer impression) __________________________________________________________________________ Item Food Secure Food Insecure

Households Households1

__________________________________________________________________________ 12. Eat wheat when prefer rice 174 14 13. Borrowed food to serve guests 64 38 14. Took food on credit 7 43 15. Worried about next meal 5 37 16. Need to buy rice frequently 2 32 17. Eat few meals per day 2 34 18. Adult cuts back on food 0 36 19. Borrow food to feed family 0 33 20. Adults skip meals 0 31 21. Stored food runs out 0 28 22. Main income earner skips meals 0 16 Total Households (in each category) 254 346 __________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002).

1. This category combines both households classified as’ food insecure without hunger’ and those classified as ‘hungry’, based on the rating given by female interviewers.

- - 21

Table 3. Comparison of statistical correlations between responses to individual questions from the 11 item set of the food security module, household classification as ‘hungry’ using female interviewer ratings, and pre capita expenditure ____________________________________________________________________________ Hungry FS112 Per Capita Households1 Expenditure3 ____________________________________________________________________________

(Correlation Coefficients)

1. Eat wheat when prefer rice .485 ** .566 ** -.298 ** 2. Borrowed food to serve guests .421 ** .262 ** -.214 ** 3. Took food on credit .470 ** .500 ** -.250 ** 4. Worried about next meal .562 ** .735 ** -.314 ** 5. Need to buy rice frequently .601 ** .756 ** -.359 ** 6. Eat few meals per day .689 ** .636 ** -.356 ** 7. Adult cuts back on food .704 ** .820 ** -.346 ** 8. Borrow food to feed family .636 ** .729 ** -.363 ** 9. Adults skip meals .671 ** .571 ** -.319 ** 10. Stored food runs out .680 ** .733 ** -.409 ** 11. Main income earner skips meals .728 ** .709 ** -.303 **

Total Households (in each category) 254 5964 600 __________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). ** Correlation is significant at the .01 level (2-tailed).

1. Households categorized as ‘hungry’ according to the rating given by female interviewers.

2. The 11 item factor from the food security module questions 3. From R1 of data collection 4. There were 4 households that did not offer and answer to any of these 11 items.

Table 4: Food Security Status by Common Comparator Means Per 11 item module questions Per Interviewer Rating Per Combined Variable1 Food Secure Food Insecure Food Secure Hungry Food Secure Food Insecure Female Heads (% households) 2.4 7.0** 2.5 9.3** 3.4 5.6 Dependency Ratio 0.52 0.57* 0.51 0.55 0.49 0.6** Food expenditure (p.c.) 414.0 334.0** 454.9 307.3** 393.3 365.9* Monthly expenditure (taka) 6889.0 3120.0** 8690.0 2648.1** 6779.9 2601.8** Land (decimals per capita) 0.2 0.1** 0.3 0.1** 0.2 0.1** Spending on clothes and 86.0 45.9** 107.0 39.2** 87.8 33.5** shoes (taka/month/cap.) Saris (number owned) 6.0 3.1** 7.2 3.0** 5.7 3.3** Tin Roofing Sheets 1.3 0.6** 1.4 0.5** 1.2 0.7** Sanitary Latrine (%) 48.7 18.8** 58.7 19.2** 43.4 24.9** Non-productive assets 40.3 18.6** 48.0 18.0** 37.1 21.6** Landless (%) 46.9 86.2** 35.4 86.3** 52.5 81.2** Food Share in total spending 46.8 60.3 42.4 62.2 42.4 73.4** Unique Foods Consumed 15.4 12.5** 16.4 12.0** 15.0 12.9* Food Groups Consumed 7.7 6.7* 7.9 6.5** 7.6 6.8 Food Stores (kg) 821.0 272.1* 912.5 356.7* 762.8 785.3 _____________________________________________________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). * Significant at .05 level (t-test, chi square or Mann-Whitney non-parametric test as appropriate) ** Significant at .01 level (t-test, chi square or Mann-Whitney non-parametric test as appropriate) 1. Combining <80% caloric adequacy and >60% expenditure allocated to food.

- - 13

indicator for food insecurity according to the 11 item module and the female interviewer rating, but is not significant in relation to the combined variable (columns 5 and 6). Conversely, while dependency ratio is significant using the combined variable it is only weakly so using the 11 item module, and is not significant according to interviewer rating. Differences in the statistical significance of outcomes across the 3 approaches to defining food security also apply to indicators such as food share in total expenditure, food groups consumed and the amount of food stored in the household. In other words, significant correlations cannot be assumed among the many commonly used measures of food insecurity. Tables 5 through 8 provide cross correlations for indicators of poverty and food insecurity compared against female interviewer ratings and household responses to the 11 item set. Several important points can be highlighted. First, where ‘poverty’ is concerned Table 5 indicates that household expenditure (absolute level) and expenditure tercile (relative level) are strongly correlated with each other, and with the food security comparators; indeed these household-level measures of poverty correlate more strongly with the food security parameters than do per capita expenditure (Table 4). Second, while a number of poverty icons such as total expenditure on clothing and shoes and land ownership, are consistently and positively correlated with most of the other measures of poverty they do less well than a variable measuring the ‘number of non-productive assets owned’. The latter variable (a simple count of items such as fans, jewelry, TVs, refrigerators), appears to perform well across the board and is significantly correlated with income/expenditure measures as well as the food security comparators. Table 6 provides a range of Spearson correlations among common food adequacy and access variables. There are strong and consistent relationships between the interview rating, the 11 item module and most comparators. In terms of the female interviewer rating the strongest associations are with total cost of food consumed (imputed subsequent to the interview—not at the time), total food expenditure per capita, and number of unique foods consumed (stronger even than the number of food groups consumed).15 The same variables are correlated with the 11 item set, although absolute food expenditure per capita does not perform as well. However, it should be noted that correlations with the 11 item set in Table 6 are considerably stronger than those reported by Maxwell et. al. (1999) for Ghana. For example, Maxwell’s (1999) highest correlation between a food coping index and food share in total expenditure was 0.19 compared with 0.41 in the current sample. Interestingly, indicators such as the ratio of calorie availability to need (based on an ex ante calculation of household level requirements from the demographic details) and meeting at least 80 percent of caloric requirement (against a cut-off set at 2,200 kcal/cap/day) were significantly, but more weakly correlated.

15 Dunford (2000) found a positive but very weak correlation between the number of food groups consumed in the World Vision Bangladesh sites and income source, farm activities, and area planted.

- - 13

Table 5. Comparison of Statistical Correlations among Poverty Proxies and Module Items __________________________________________________________________________________________________________ FS rating1 FS112 Total Exp. Land Clothes Prod. Other HH Exp. Tercile Owned Exp. Assets Assets __________________________________________________________________________________________________________

(Correlation Coefficients) Food Security Rating 1.00 FS11 (item module) .730** 1.00 Total HH expenditure .465** .436** 1.00 Expenditure Tercile -.446** -.421** -.941** 1.00 Land Owned .506** .572** .425** -.395** 1.00 Spending on clothes/shoes .383** .363** .459** -.420** .275** 1.00 Production Assets .346** .375** .323** -.300** .481** .230** 1.00 Non-Prod. Assets .616** .615** .638** -.600** .590** .409** .415** 1.00 _________________________________________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). ** Correlation is significant at the .01 level (2-tailed).

1. Household categorization according to female interviewer assessments. 2. The 11 item set from the food security module questions

- - 14

Table 6. Comparison of Spearman Correlations among Module Items and Food Access/Security Proxies Relating to Food Consumption __________________________________________________________________________________________________________ FS rating1 FS112 Food Food Food Food Unique Calorie Meet 80% Share3 Exp./cap Value4 Groups Foods Ratio5 Need6 __________________________________________________________________________________________________________

(Correlation Coefficients) Food Security Rating 1.00 FS11 .730** 1.00 Food Share in Exp. .376** -.405** 1.00 Food Expenditure/capita -.482** .160** .119** 1.00 Value of Foods Consumed/capita -.533** .516** -.293** .445** 1.00 # Food Groups -.392** .350** -.142** .318** .471** 1.00 # Unique Foods Consumed -.421** .421** -.190** .351** .557** .805** 1.00 Calorie Ratio -.281** .233** -.043 .327** -.056 .318** .355** 1.00 Meet 80% of Needs -.202** .145** .023 .271** -.067 .220** .216** .701** 1.00 __________________________________________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). ** Correlation is significant at the .01 level (2-tailed).

1. Household categorization according to female interviewer assessments. 2. The 11 item set from the food security module questions 3. Food as a share of total household expenditure 4. Imputed value (cost) of all foods consumed (home produced as well as purchased) per capita 5. Aggregate calorie availability within the household as a share of imputed ‘need’ (according to household demographics) 6. Did the household exceed 80 percent of caloric requirement based on 2200 kcal/capita/day.

- - 13

Where nutritional outcomes are concerned Table 7 confirms that which has been shown elsewhere; namely, that anthropometric status correlates poorly with conventional food security and poverty indicators (Maxwell et. al. 1999; Chung et. al. 1997; Zeller et. al. 2001). For example, while wasting in children 0 to 12 years is significantly correlated both with stunting and underweight none of these is strongly correlated either with the interviewer assessment or the 11 item module responses. Indeed, the variable defined as ‘any child malnutrition’ (the presence of any child classified as wasted, stunted or underweight) is not statistically correlated in this sample with comparators such as number of food groups consumed (-.064, p<.184), or meeting 80 percent of caloric need (-.007, p<.888); nor it is significantly correlated with a ‘food secure’ category derived from answers to the 11 item module (2-tailed test p<.088). Similarly, while stunting in children under 12 is weakly correlated with clothing expenditure (-.15, p<.003), land ownership (-.13, p.013) and non-productive assets (-.154, p,.001), it is not statistically correlated with other factors such as expenditure tercile, unique foods consumed, calorie ratio or share of food in total expenditure. Indeed Table 8 suggests that the pattern of nutritional outcomes is too complex for malnutrition to be used as a simple (single) proxy for either food security or poverty. While adult BMI levels are significantly lower in hungry households than in food secure households (although there is no significant difference between adult men and women in these households). Child anthropometry varies considerably by age and gender with no clear pattern of increasing severity linked to the food security categorization. While there are of course links among nutritional outcomes, expenditure patterns, and food consumption restrictions anthropometric measures are unpredictably correlated with other comparators. But what of combined variables? Table 9 suggests that the composite food security assessment and 11 item module overlaps more strongly with comparators of poverty, food access and nutrition outcomes than comparators do among themselves with each other. That is, the interviewer-driven food security rating has strong associations with a range of variables, including expenditure tercile, ownership of non-productive assets, unique foods consumed, and share of expenditure on clothing. The same group is significantly correlated (albeit at different levels) with the 11 item set. The idea of composite variables lies at the root of work by Maxwell et. al. (1999) who developed a set of elaborately weighted indices derived from self-reported ‘food-seeking strategies’ in urban Accra (building directly on the US hunger module questions). Similarly, Ahiadeke et. al. (2002) elaborated on coping strategies during periods of stress in Accra, finding that a single indicator of ‘reliance on purchased street foods’ was highly correlated with negative food security outcomes, including income high food share in total expenditure. More importantly, they also considered a combined variable approach by identifying which households were both calorie inadequate (<80% household adequacy per adult equivalent unit) and spending more than 50 percent of total expenditure on food. The combined variable allowed for classification of households into four groups along a continuum from ‘food secure’ to ‘food insecure’. The authors found that no household in the poorest expenditure quintile (in the urban Accra sample) fell into the ‘food secure’ group, while none of the richest households were in the ‘food insecure’ group. The combined variable was therefore proposed as a viable benchmark for food insecurity.

- - 14

Table 7. Comparison of Spearman Correlations among Module Items and Nutritional Outcomes for households containing children under 12 years of age

_____________________________________________________________________________ FS rating1 FS112 Wasting3 Under- Stunting weight _____________________________________________________________________________

(Correlation Coefficients) Food Security Rating 1.00 FS11 .772** 1.00 Any Wasting -.051 -.007 1.00 Any Underweight -.057 -.119* .329** 1.00 Any Stunting -.153** -.137** .549** .050 1.00 _____________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). ** Correlation is significant at the .01 level (2-tailed). * Correlation is significant at the .05 level (2-tailed).

1. Household categorization according to female interviewer assessments. 2. The 11 item factor from the food security module questions 3. Child anthropometry relates to children aged less than 12 years at -2 SDs of reference

population. The ‘any malnutrition’ variables are calculated on a household basis and thus do not reflect means across individuals.

- - 15

Table 8. Nutritional Outcomes by Food Security Category (Female Interviewer Assessment) and by 11 item module (mean values) _________________________________________________________________________________ 11 Item Module Rating1 Female Interviewer Rating Food Food Secure Insecure Hungry Secure Insecure (No hunger) _________________________________________________________________________________ Adult Female BMI 19.8 18.4 20.1 18.8 18.3 Male BMI 19.6 18.5 19.8 18.9 18.4

Children (boys) (Z scores) Weight-for-Age 2-5 years -2.1 -2.2 -2.1 -2.0 -2.2 5-10 years -1.9 -2.1 -1.8 -1.9 -2.3 Weight-for-Height 2-5 years -1.5 -1.4 -1.6 -1.3 -1.5 5-10 years -1.3 -1.5 -1.3 -1.3 -1.6 Height-for-Age 2-5 years -1.5 -1.9 -1.3 -1.7 -1.9 5-10 years -1.5 -1.7 -1.4 -1.6 -1.9 Children (girls) Weight-for-Age (Z scores) 2-5 years -2.1 -2.2 -1.9 -2.3 -2.1 5-10 years -1.8 -2.1 -1.9 -1.7 -1.9 Weight-for-Height 2-5 years -1.2 -1.4 -1.1 -1.3 -1.4 5-10 years -1.3 -1.5 -1.4 -1.4 -1.5 Height-for-Age 2-5 years -1.8 -1.9 -1.9 -2.0 -1.6 5-10 years -1.4 -1.4 -1.5 -1.3 -1.6 _________________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002).

1. Based on a median split in the distribution of households according to the number of affirmative answers given to the 11 questions.

- - 16

Table 9. A Spearman Cross Correlation Matrix of Selected indicators of Poverty, Food Insecurity and Malnutrition ________________________________________________________________________________________________________________________ FS rating1 FS112 Any Exp. Clothes/ Non-Prod. Unique Food % Meet Stunting3 Tercile Shoes Assets Foods of Exp. 80% RDA ________________________________________________________________________________________________________________________

(Correlation Coefficients) Food Security Rating 1.00 FS11 .730** 1.00 Any Stunting -.153** -.137** 1.00 Expenditure Tercile -.446** -.421** .080 1.00 Spending on Clothing .383** .363** -.147** -.420** 1.00 Non-productive Assets .616** .615** -.154** -.600** .409** 1.00 Unique Foods Consumed .421** .421** -.081 -.440** .282** .505** 1.00 Food in Share of Total Exp. -.376** -.405** .110* .586** -.412** -.426** -.190** 1.00 Meet 80% caloric need .179** .145** .026 .025 .114** .063 .216** .023 1.00 _______________________________________________________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002). ** Correlation is significant at the .01 level (t-test, chi square or Mann-Whitney non-parametric test as appropriate). * Correlation is significant at the .05 level (t-test, chi square or Mann-Whitney non-parametric test as appropriate).

1. Household categorization according to female interviewer assessments. 2. The 11 item factor from the food security module questions 3. For children aged less than 12 years at -2 or -3 SDs of reference population.

- - 17

That proposition was explored with the current data. A combined variable was constructed to identify households that are both calorie inadequate (not meeting 80 percent household caloric adequacy on a per capita basis) and allocating more than 60 percent of total expenditure on food. That composite results in 30.2 percent of the sample (181 households) being defined as ‘hungry’, compared with 30.3 percent (182 households) defined as ‘hungry’ using the female interviewer rating for the same sample. This surprising concordance warrants further examination. In coming months these data will be used to assess, a) do other combined variables work in identifying households that are characterized by negative coping behaviours and negative outcomes?; and b) can combined variables be ‘tested’ for their potential explanatory power through more complex statistical analysis? In terms of other combined variables another avenue to explore is the construction of a composite ‘food insecurity’ variable based explicitly on the conceptual model described in Figure 1. That is, combining variables that relate directly to:

• food availability (say, the monetary value of all foods consumed per capita, share of food in total expenditure, and food group diversity),

• food access (total household expenditure level, current level of food in store, frequency of rice purchase),

• food utilization (a ‘hygiene’ variable (type of latrine * wash hands with soap after defecation?), and source of water),

• vulnerability (the elements that raise exposure to, or protect from, external shocks anywhere in the food system—possibly including non-productive asset stock (a measure of wealth), land owned (production security), and an index of income sources (income diversification).

There is considerable scope for exploring the components of the food security as interacting terms that generated more insight when combined as a whole than when used as individual parts. Where more complex statistical analyses are concerned, both parametric and non-parametric approaches (including Rasch, CART and others) are warranted to establish the structure of statistical associations and the relative value of one set of comparators in food security analysis over others.16 Table 10 presents preliminary output from regressions exploring the determinants of food security status according to the 11 item module. Even in this rather simple, cross-sectional model achieving an adjusted R square exceeding 0.45 with robust coefficients (and 83 percent correct prediction of outcomes) is encouraging. However, while some variables (such as landlessness) remain significant in all models tested (only one is reported here) others drop in and out of significance depending on co-variables and on how the dependent variable is specified. For example, ‘number of food groups consumed’ is only statistically significant when household size and/or dependency ratio are excluded from the model. Similarly, type of latrine used is significant only if ‘non-productive assets owned’ and ‘landlessness’ are left out. If latrine is out type of water source comes close to significance, but when the two are included water source is not significant.

16 Non-linear factor analysis based on item response theory (such as Rasch modeling) can assess the internal consistency of questions with a view to constructing a scale or ranking not linked to an ‘external’ benchmark (Nord et. al. 2002b). It is planned to test that particular approach with the current data in coming months.

- - 18

Table 10. Results of OLS regression exploring the determinants of food security according to the 11 item question set. Dependent Variable: Categorization of ‘food secure’ according to the 11 item module set (continuous) ___________________________________________________________________________ Variable Coefficient t-statistic Significance ___________________________________________________________________________ Landless (<0.5 acres) - .197 - 5.553 .000 Log of value of all food consumed/capita .208 5.534 .000 # Food groups consumed .055 1.544 .123 Food share in total expenditure - .131 - 3.734 .000 Log of ratio of calories consumed/required .153 4.459 .000 Log of non-productive assets owned .302 7.272 .000 Dependency ratio (<10 + >60) .051 1.618 .106 Adjusted R squared .452 N 590 ___________________________________________________________________________ Source: Tufts/World Vision/FANTA survey data (2002).

- - 19

What is more, the significant variables in this model are ownership of land, the log of per capita, food share in total expenditure, low household calorie adequacy, and non-productive assets owned. Not significant are number of food groups consumed (number of unique foods consumed fared worse still), and dependency ratio (which comes closer to significance than household size, age or gender of household head, number of income earners and other parameters relating to worker/non-worker dependency relationships). However, there appears to be differences in the determinants depending not only on how ‘food security’ is determined but also on whether one seeks to explain ‘being food secure’ rather than being ‘hungry’. For example Tables 11a and 11b present two regressions based on the female interviewer rating. In Table 11a takes ‘being categorized as hungry’ as the dependent variable. In this analysis food groups and dependency ratio become significant, along with other variables which retain significance from the logistic regression. Yet in Table 11b, which takes ‘being rated as food secure’ as the dependent variable, food groups and dependency ratio are no longer very significant In other words, food groups consumed may be a viable proxy for identifying hungry households but it may be less useful in identifying non-hungry households—arguably because in a generally food-deficit environment once a household exceeds a given level of energy per capita the addition of one more food group makes only a small difference at the margin. Yet among hungry households the addition of one more food group on a regular basis may represent a significant change in their wellbeing. Seeking appropriate interpretations of these interesting findings, along with further elaboration of the determinants of different vectors of food insecurity, will be a priority in coming months. 5. Changes in Food Security Status and Comparator Indicators over Time (R1-R2) The next big question to consider in such analyses of determinants will be their stability over time. In addition to exploring in-depth qualitative questions with the sub-sample of 125 households a second round of data was collected (one year after R1). When comparators are linked to interviewer and the 11 item ratings two important findings can be highlighted. First, almost 60 percent of households remained in the same food security category (as assessed by interviewer) across the two rounds. There was strong agreement in this by gender of interviewer--female interviewers placed 59.2 percent in the same category compared with 58.9 percent according to male interviewers—not a statistically significant difference. There was similar agreement in terms of extreme changes; that is male and female interviewers agreed independently that 2 households ‘fell’ by more than one place across the groupings (from ‘food secure’ to ‘hungry’), while 1 climbed more than two placings (from ‘hungry’ to ‘food secure’). The household that had been classified as ‘hungry’ in R1 but appeared to be ‘food secure’ in R2 reported that it was now faring considerably better (stating that this year it was ‘better off’ than last year). The main reasons were off-farm income activity (offering a bicycle-taxi service) and the fact that the family had acquired additional land for cultivation (another 0.5 acre). The perception that things had improved was confirmed through the quantitative questionnaire which found that the household’s share of expenditure devoted to food had fallen by 24 percentage points between R1 and R2.

- - 20

Table 11a Results of logistic regression exploring the determinants of ‘hunger’ according to the female interviewer rating. Dependent Variable: Categorization of household as ‘hungry’ according to female interviewer rating (binary) __________________________________________________________________________________ Variable Coefficient Odds (95% CI) Significance B Ratio __________ Being Landless (<0.5 acres) .619 - 1.857 (1.13, 3.05) .015 Log of value of all food consumed/capita -1. 628 .196 (0.10, 0.37) .000 High # Food groups consumed - .224 .799 (0.67, 0.95) .010 High food share in total expenditure .023 - 1.023 (1.01. 1.04) .000 Log of ratio of calories consumed/required -1. 012 .364 (0.20, 0.67) .001 Log of non-productive assets owned -1. 328 .265 (0.16, 0.45) .000 High dependency ratio (<10 + >60) - .659 .517 (0.31, 0.86) .012 Nagelkerke R square 0.51 Cox & Snell R square 0.36 N 590 __________________________________________________________________________________ Table 11b Results of logistic regression exploring the determinants of ‘food secure’ according to the female interviewer rating. Dependent Variable: Categorization of household as ‘food secure’ according to female interviewer rating (binary) __________________________________________________________________________________ Variable Coefficient Odds (95% CI) Significance B Ratio __________ Being landless (<0.5 acres) -1 .065 .345 (0.20, 0.61) .000 Log of value of all food consumed/capita 1. 990 7. 313 (3.75,14.26) .000 High # Food groups consumed .051 1. 052 (0.89, 1.24) .544 High food share in total expenditure - .026 .974 (0.96, 0.98) .000 High ratio of calories consumed/required 1. 640 5.156 (2.66,10.01) .000 Log of non-productive assets owned 2. 035 7.652 (4.43,13.23) .000 High dependency ratio (<10 + >60) .599 1.821 (1.10, 3.00) .019 Nagelkerke R square 0.64 Cox & Snell R square 0.47 N 590 __________________________________________________________________________________

- - 21

By contrast, the two households classified as ‘hungry’ in R2 when 12 months earlier they had been ‘food secure’ both experienced “big shocks”. In one case, the main income-earner had fallen sick and was now disabled. In the other, the male household head died leaving the household without its primary income earner. In this environment of chronic insecurity the loss of household headship and/or income earning has a devastating impact even in the short-term. Both households reported having to mortgage various assets during the past year and the two families were currently spending nearly 70 percent of their total budget on food--an increase in one of case of 42 percentage points over one year ago. These examples highlight the critical links between food security and poverty, but they also underline the importance of vulnerability within the food security equation—the risk of shocks can unsettle a previously food secure household in a very short period of time. The second major point to be made with regard to differences between rounds is that temporal dimensions of food insecurity are rather complex—a snap-shot in time may capture relative levels within a single time period but it cannot adequately capture the dynamics of change through time. While 60 percent of households remained in the same category during the 12 months 40 percent did not which confirms the considerable dynamics that need to be understood in any assessment of food insecurity. Some 15-20 percent of households moved between two food security categories (say, from food secure to hungry), while slightly more than 20 percent showed signs of improvement (moving up one group). This compares reasonably well with households’ own assessments of the past year in which 26 percent reported that nothing much has changed, almost 40 percent felt that their conditions had improved and 34 percent saw themselves as ‘worse off’ than a year earlier. Importantly, more than 83 percent of households showing signs either of stability or improvement according to interviewer ratings also considered themselves to be faring better this year compared with last. Similarly, 70 percent of households doing as badly as, or worse than, a year ago (according to interviewer ratings) report the same through self assessment. Where there was a big change in the number of responses to the 11 item module between R1 to R2 (suggesting increasing severity of food stress) this was significantly correlated (0.553 at p<0.001) with an interviewer ‘downgrade’ in those households’ category placing. A next step is to consider how well interviewer ratings and the 11 item set parallel changes in comparator outcomes. At this stage it can be reported that there was stability in most of the parameters in the 12 months under consideration. For example, based on interviewer rating of food security categories there was no significant change between Rounds 1 and 2 in the number of unique foods consumed, monthly food expenditure, total household expenditure or the share of food in total household expenditure. Yet, there was a weakly significant decline in total monthly non-food expenditure levels. This does not discount that fact that a few households have exceptionally large ‘lumpy’ expenditures in R2 (as discussed above), including weddings (which represent major expenditures even for very poor households), the purchase of productive assets (such as a rickshaw which will generate an income stream) or investments such as sinking a deep well (used for access to cleaner water or for pump irrigation). What is more, when food security status is categorized according to the self-reporting module (using a median split in responses to items) the decline in non-food expenditure as measured by

- - 22

spending on clothing and shoes is more strongly significant for both food secure and food insecure groups (Table 12). Interestingly there also appears to be a small but significant decrease in the percent of households cultivating less than 0.5 acres (functionally landless) in the food secure group mirrored by a quite large increase in landlessness in the food insecure group. The cause of such a large shift in land access over a 12 month period needs to be explored further. All other elements remained roughly stable over the two periods. A third important point is that, this picture reinforces the impression gained from interviews that a majority of households remained in the same general food security category for both periods, while a few households suffered a sharp decline (perhaps these are among the households becoming functionally landless in the intervening period, and reporting a significant decline in non-food expenditure to maintain spending on food essentials (food share in total expenditure)). By contrast the fortunes of households already in (or recently joining) the food secure group appear to have improved slightly, as represented by declining share of food in total expenditure and some households gaining cultivable land. The fact that food secure groups report significantly lower food stores in hand compared with a year before is somewhat puzzling. 6. Conclusions and Next Steps The analysis of R1 and R2 data continues, and it continues to give rise to exciting findings that sometimes confirm expectations and sometimes challenge underlying assumptions of the research. A number of conclusions and next steps can be proposed:

1. Thanks to the huge effort that went into defining locally appropriate concepts and questions a viable set of 11 or so questions from the ‘self-reporting’ hunger module appears to work well both in characterizing the problems experienced by households in Bangladesh and in identifying households along a continuum of food stresses.

2. Those questions correlate well not only with interviewer ratings but also with a range

of comparator indicators commonly used in the analysis of poverty, malnutrition and food insecurity. There is a high degree of concordance between male and female interviewer ratings, between interviewer ratings and the self-reported 11 item set, among the 11 questions and other comparators, and between interviewer assessments of change in household conditions between R1 and R2 and household’s own assessments of that change (or stability).

3. While many of the variables tested are strongly correlated with the conditions under

consideration no single indicator serves well in defining all aspects of food insecurity--not anthropometry, not expenditure, not food groups consumed, not spending on food. This confirms the need for composite variables that distinguish between outcomes and processes often generically characterized as ‘food insecurity’ or ‘poverty’ or ‘malnutrition’. Further statistical analyses (parametric and non- parametric, including Rasch) are needed to gain an understanding of how the determinants of these related but different conditions overlap and where they do not.

- - 23

Table 12: Comparator Means for Rounds 1 and 2, by Food Security Status

Food Secure1 Food Insecure Round 1 Round 2 Round 1 Round 2 Comparator Dependency Ratio 0.5 0.4* 0.6 0.6 Food expenditure (p.c.) 418.9 447.7 315.1 312.1 Monthly expenditure (taka) 6843.1 6416.9 3164.0 2669.6 Spending on Clothes and Shoes 80.2 34.5** 53.9 16.6** (take/month/cap.) Saris (number owned) 7.4 6.6 3.5 2.7 Landless (%) 38.0 36.4** 87.0 97.5** Food Share in total spending 46.9 39.9* 58.6 57.5 Unique Foods Consumed 15.8 15.5 12.2 12.9 Food Groups Consumed 7.6 7.5 6.3 6.7 Source: Tufts/World Vision/FANTA survey data (2002). 1 Food security status according to the self-reported module set in Round 1 divided by median split. * Significant at .05 level (t-test and Mann-Whitney non-parametric test) ** Significant at .01 level (t-test and Mann-Whitney non-parametric test)

- - 24

4. The 11 item set of questions on hunger covers the entire range of conceptual

underpinnings of food insecurity, but further analysis is needed of how questions and responses are perceived by households in different contexts. That is, the psychological aspects relating to item responses require some elaboration. To what extent do household conditions at time of interview unduly influence how respondents answer question about the entire preceding 12-month period? Does the angle of the household’s trajectory (improving or worsening conditions) matter to respondents’ and interviewers’ ratings of food security? On which questions or issues do responses differ most between male and female respondents, and why?

Further sensitivity analysis is required on the dietary intake questions, the nature of ‘dependency ratios’ (life cycle and family composition issues rather than simple adult-child relationships), revealed demand for food quality (the extent to which micronutrient deficiencies are understood and factored into food preferences) and the role of both human and social capital. Being ‘Moslem’ was found to be a highly significant predictor of food security, but not of ‘hunger, even controlling for income and other status variables—that may suggest a ‘social connectedness’ factor that needs to be explored. Similarly, being a female headed household is not, in this sample, significantly correlated with being a ‘hungry’ household (consistent with recent findings from Ghana).

- - 25

References Ahiadeke, C., M. Armar-Klemesu, C. Levin, D. Maxwell, S. Morris, and M. Ruel. 2002. Achieving Urban Food and Nutrition Security for the Vulnerable in Greater Accra. WHO Multicountry Study on Improving Household Food and Nutrition Security for the Vulnerable. Mimeo. Ali, M. and H. F. Delisle. 1999. A Participatory Approach to Assessing Malawi Villagers’ Perception of their own Food Security. Ecology of Food and Nutrition.

38: 101-121. Alaimo, K., C. Olson and E. Frongillo. 1999. Importance of Cognitive Testing for Survey Items: An Example from Food Security Questionnaires. Journal of Nutrition Education 31: 269-75. Bergeron, G., S. Morris and J.M Medina Banegas. 1998. How Reliable are Group Informant Ratings? A Test of Food Security Rating in Honduras. Discussion Paper No. 43. Food Consumption and Nutrition Division. Washington, D.C.: International Food Policy Research Institute. Bickel, G. et. al. 2000. Guide to Measuring Household Food Security: Revised 2000. Office of Analysis, Nutrition and Evaluation. United States Department of Agriculture. Mimeo. Block, S. and P. Webb. 2001. The dynamics of livelihood diversification in post-famine Ethiopia. Food Policy. 26 (4): 333-67 Bowling, A. and J. Windsor. 2000. Towards the Good Life: A Population Survey of Dimensions of Quality of Life. Journal of Happiness Studies. 2: 55-81. Brock, K. 1999. “It’s not only wealth that matters—it’s peace of mind too”: a review of participatory work on poverty and ill-being. Brighton, Sussex: Institute of

Development Studies. Mimeo. Christiaensen, L. and R. Boisvert. 2000. On Measuring Household Food Vulnerability: Case Evidence from Northern Mali. Washington, D.C.: International Food Policy Research Institute. Mimeo. Chung, K., L. Haddad, J. Ramakrishna, and F. Riely. 1997. Identifying the Food

Insecure: The Application of Mixed-Method Approaches in India. Discussion Paper No. 22. Washington, D.C.: International Food Policy Research Institute.

Currey, B. 1984. Issue in evaluating food crisis warning systems. Food and Nutrition Bulletin. 6 (2). http://www.unu.edu/ unupress/food/8F062e/8F062E03.htm Derrickson, J. and A. C. Brown. 2002. Food Security Stakeholders in Hawai’i: Perceptions of Food Security Monitoring. Journal of Nutrition Education and Behavior. 34: 72-84. Derrickson, J. and J. Anderson. 1999. Face Validity of the Core Food Security Module

with Asians and Pacific Islanders. Jou. Nutr. Education. 32 (1): 21-30. Deyo, R. A. 2000. Pain and Public Policy. The New England Journal of Medicine. 342 (16): 1211-13. Dunford, B. 2000. Baseline Survey Report (October-November 2000). Food Security

Enhancement Initiative: FY 2000-FY 2004. World Vision Bangladesh. Mimeo. Engle, P., S. Castle and P. Menon. 1996. Child Development: Vulnerability and Resilience. Discussion Paper 12. Food Consumption and Nutrition Division. Washington, D.C.: International Food Policy Research Institute,

- - 26

Frongillo, E.A., Jr. 1999. Validation of Measures of Food Insecurity and Hunger. Jou. Nutrition 129 (2S): 506S-09S.

Green, C.W. and D. H. Reid. 1996. Defining, validating and increasing indices of happiness among people with profound multiple disabilities. Journal of Applied Behavior Analysis. 29: 67-78. Haddad, L., E. Kennedy, and J. Sullivan. 1994. Choice of Indicators for Food Security and nutrition monitoring. Food Policy. 19 (3): 329-43. Hoddinott, J. and Y. Yohannes. 2002. Dietary diversity as a food security indicator. Report to the Academy of Educational Development. Mimeo. Hoddinott, J. 1999. Choosing Outcome Indicators of Household Food Security.

Technical Guide 7. Washington, D.C.: International Food Policy Research Institute. IFRCRCS (International Federation of Red Cross and Red Crescent Societies). 2001. Psychological Support. Geneva. Jochim, M.A. 1981. Strategies for Survival: Cultural Behavior in an Ecological Context. New York, NY: Academic Press. Kendall, A., C. Olson and E. Frongillo. 1995. Validation of the Radimer/Cornell

Measures of Hunger and Food Insecurity. Jou. Nutrition. 125: 2793-2801. Louk, K. R., E. Sharfer, R.B. Scharfer, and P. Keith. 1999. Comparison of Dietary Intakes of Husbands and Wives. Journal of Nutrition Education. 31: 145-52. Lozano, D.I., S.L. Crites and S.N. Aikman. 1999. Changes in Food Attitudes as a Function of Hunger. Appetite. 32: 207-18. Macht, M. 1999. Characteristics of Eating in Anger, Fear, Sadness and Joy. Appetite. 33: 129-39. Maxwell, D., C. Ahiadeke, C. Levin, M. Armar-Klemse, S. Zakariah and G.M. Lamptey.

1999. Alternative food-security indicators: revisiting the frequency and severity of ‘coping strategies’. Food Policy. 24: 411-29.

Monello, L.F. and J. Mayer. 1967. Hunger and Satiety Sensations in Men, Women, Boys and Girls. American Journal of Clinical Nutrition. 20 (3): 253-61. Morris, M.D. 1979. Measuring the Conditions of the World’s Poor. New York, NY: Pergamon Press/Overseas Development Council. Morris, S., C. Carletto, J. Hoddinott and L.J.M. Christiaensen. 1999. Validity of Rapid Estimates of household Wealth and Income for Health Surveys in Rural Africa. Discussion Paper 72. Food Consumption and Nutrition Division. Washington, D.C.: International Food Policy Research Institute. Nord, M. and G. Bickel. 2002. Measuring children’s Food Insecurity in U.S. Households, 1995-99. Food Assistance and Nutrition Research Report No. 25. Washington, D.C.: United States Department of Agriculture. Nord, M., N. Kabbani, L. Tiehen, M. Andrews, G. Bickel and S. Carlson. 2002a. Household Food Security in the United States 2000. Food Assistance and Nutrition Research Report No. 21. Washington, D.C.: USDA. Nord, M., B. Kuhn, A.K. Satpaty, N. Raj, P. Webb and R. Houser. 2002b. Comparing Household Survey-Based Measures of Food Insecurity Across Countries: Case Studies in India, Uganda, and Bangladesh. Paper prepared for the International Scientific Symposium on Measurement and Assessment of Food Deprivation and Under-nutrition, 26-28 June, 2002, Rome, Italy.

- - 27

Omidvar, N., M. Ghazi-Tabatabie, G.G. Harrison, S. Eghtesadi, S-A. Mahboob and M. Pourbakht. 2002. Development and validation of a short food-frequency questionnaire for screening women of childbearing age for vitamin A status in northwestern Iran. Food and Nutrition Bulletin. 23 (1): 73-82. Radimer, K.L., C.M. Olson, J.C. Greene, C.C. Campbell and J-P Habicht. 1992.

Understanding Hunger and Developing Indicators to Assess it in Women and Children. Journal of Nutrition Education. 24 (1): 36S-45S.

Radimer, K.L., C.M. Olson, and C.C. Campbell. 1990. Development of indicators to assess hunger. Journal of Nutrition. 120: 1544-48.

Rahman, A. 1991. Missing Dimensions of Poverty: The Issues of Vulnerability and Insecurity. ADAB News (Association of Development Agencies in Bangladesh). Jan-March: 3-19. Rime, B. and D. Giovanni. 1986. The physiological patterns of reported emotional

states. In K. Scherer, H. Wallbott and A. Summerfield (eds.) Experiencing emotion: A cross-cultural study. Cambridge: University Press, pp. 84-97.

Rose, D. and D. Tschirley. 2000. A Simplified Method for Assessing Dietary Adequacy in Mozambique. Research Paper No.36. Directorate of Economics. Ministry of agriculture and Fisheries. Maputo: Mozambique.

Sarlio-Laehteenkorva and Lahelma. 2001. Food Insecurity is Associated with Past and Present Economic Disadvantage and BMI. Jou. Nutrition 131: 2880-84. Schneider, S.M., and X. Hebuterne. 2000. Use of Nutritional Scores to Predict Clinical Outcomes in Chronic Diseases. Nutrition Reviews. 58 (2): 31-38. Spiegel, P. and P. Salama. 2000. War and Mortality in Kosovo, 1998-99: an epidemiological testimony. The Lancet. 355: 2204-9. Studdert, L., E. Frongillo and P. Valois. 2001. Household Food Insecurity Was Prevalent in Java during Indonesia’s Economic Crisis. Jou. Nutrition.131: 2685-91. Tierney, K.J. 1999. Towards a Critical Sociology of Risk. Sociological Forum. 14: 215-42. Tine, S. 1992. Perception of poverty and income satisfaction: An empirical analysis of

Slovene households. Jou. Economic Psychology. 13: 57-69. Tschirley, D., D. Rose and H. Marrule. 2000. A Methodology for Estimating Household Income in Rural Mozambique Using Easy-to-Collect Proxy Variables. Research Report No. 38. Directorate of Economics Research Paper Series. Maputo, Mozambique: Ministry of Agriculture and Fisheries/Michigan State University. UNDP. 2001. Human Development Report. New York: Oxford University Press. Webb, P., J. Coates, and R. Houser. 2002. Does Microcredit Meet the Needs of all Poor Women? Constraints to Participation Among Destitute Women in Bangladesh. Food Policy and Applied Nutrition Program. Discussion Paper No. 3. Friedman School of Nutrition Science and Policy, Tufts University. Webb, P., J. Coates and R. Houser. 2001. A Report on the formulation of the core food security module and experiences in its implementation in Bangladesh. A report to the Academy for Educational Development on the Food Insecurity Measurement and Validation Study. Mimeo. Webb, P., E. Richardson and J. von Braun. 1993. Conceptual Framework for Famine Analysis: A Household Income Approach. Report to USAID. Washington, D.C.: IFPRI. Mimeo. Webb, P. 2001. Land Degradation in Developing Countries: What is the Problem? Int. Journal of Agricultural Resources, Governance and Ecology. 1 (2): 124-36.

- - 28

Welch, K.J., N. Mock and O. Netrebenck. 1998. Measuring Hunger in the Russian Federation Using the Radimer Hunger Scale. Bulletin of the World Health Organization. 76 (2): 143-8. WHO (World Health Organization). 2001. Iron Deficiency Anaemia: Assessment, Prevention and Control. Document WHO/NHD/01.3. Geneva: WHO/UNU/UNICEF. Zeller, M., M. Sharma, C. Henry and C. Lapenu. 2001. An Operational tool for Evaluating Poverty Outreach of Development Policies and Projects. Discussion Paper No. 111. Washington, D.C.: IFPRI.