Mark Nord
US Department of Agriculture
Washington, DC, USA
Anoop Kumar Satpathy and Nikhil Raj
Jawaharlal Nehru University
New Delhi, India
Patrick Webb and Robert Houser
Tufts University
Medford, MA, USA
Introduction
Reliable and adequately detailed information about the food security of a nation’s population is important for the development of policies and programmes to increase food security and reduce food insecurity and hunger. The United States (US) Food Security Measurement Project has developed and implemented survey-based methods to help provide this information for households in the United States. Research is under way to explore the feasibility of adapting these methods for use in other countries. This paper examines data collected from three such studies in low-income populations in India, Uganda and Bangladesh, and assesses scales developed from these data.
The US household food security survey module and food security scale
The food security status of each household is assessed by their responses to 18 questions about food-related behaviours, experiences and conditions that are known to characterize households in the United States having difficulty meeting their food needs. Responses are combined into a scale using non-linear statistical methods based on the Rasch measurement model. Based on their food security scale scores, households are also classified into three categories - food secure, food insecure without hunger and food insecure with hunger - for monitoring and statistical analysis of the food security status of the population.
Adapting the US food security measurement methods for use in other countries
Adapting these measurement methods for use in other countries may be as simple as translating the questions in the module into another language, but more likely will require substantial qualitative research work including focus groups and cognitive testing of proposed questions and statistical analysis of pilot survey data. The Bangladesh survey module was developed using these qualitative research methods, while the Orissa and Kampala survey modules were developed by translating a subset of questions in the US module and conducting a modest amount of cognitive testing to assure that the questions were understandable.
A strength of this measurement method is that scales adapted to two different contexts, with several unique items, can be brought into a common metric so that prevalence rates of food insecurity and hunger can be meaningfully compared between countries. This requires that at least three or four items be common to the two surveys and judged to measure equivalent levels of severity.
For use in very low-income populations, additional attention may need to be given to incorporating the dimensions of frequency and duration of food deprivation into the measure.
Data and methods
Separate analyses were conducted of data collected from low-income populations in three studies: a study of child labour in Orissa, India; a study of participants in an income-generation programme in Bangladesh; and a general household survey in Kampala, Uganda. The analyses focused on internal validation of food security scales developed from each dataset based on the statistical relationships among the items in each scale. Statistical methods based on the Rasch measurement model (a non-linear factor analysis based on item response theory) were used to assess the items in each module and combine them into a food security scale. Each analysis examined overall item discrimination, item-fit statistics and relative severity of the item to address four issues:
(1) Which of the candidate items should be included in a food security scale?
(2) How should the metric be adjusted so that the scale is equivalent in meaning to the US scale? (Comparison with the United States is examined for illustrative purposes. Comparability with scales from other countries could be achieved using the same techniques.)
(3) Where should thresholds be set so as to make prevalence rates of food insecurity and hunger comparable with those based on the US scale?
(4) How reliable is the scale?
Summary of findings
In all three cases, reasonably reliable scales were constructed from the food security items in the survey modules. This was true even for the two surveys that essentially just translated questions from the US module and conducted modest amounts of cognitive testing prior to fielding the surveys. The statistical assessments indicate that within each study, the items in the preferred scale measure a common underlying phenomenon and do so with sufficient sensitivity to provide reasonably precise and reliable measurement of household food security. Two of the scales can be benchmarked with some confidence to the US scale, based on sets of equivalent items, so that prevalence rates measured in the various countries can be meaningfully compared. The Bangladesh scale can also be set approximately equivalent to the US scale, although with less precision and confidence.
To develop a scale for widespread or multiple uses in a country or unique population, it is worth while to begin with extensive qualitative research to develop a survey module that is thoroughly grounded in the food-related experiences of the culture and the natural language used in that culture to discuss food conditions. The excellent fit statistics of the Bangladesh scale are evidence of the efficacy of this approach. However, findings from the other two surveys suggest that for single-use surveys or limited applications, careful translation of questions from the US module combined with a modest amount of cognitive testing may provide acceptable results.
Other findings and observations:
Questions should be distributed across the entire range of severity that is of interest for the purposes of the study.
It should be emphasized that these analyses assessed only the internal characteristics of the scales, based on the interrelationships among the multiple items in each scale. External validation both to alternative measures of food insecurity and to expected outcomes of food insecurity are still needed before widespread use of these scales can be recommended with confidence.
For further information, visit the Economic Research Service World Wide Web briefing room, Food Security in the United States, at: http://www.ers.usda.gov/briefing/foodsecurity.
Patrick Webb, Jennifer Coates and Robert Houser
Friedman School of Nutrition Science and Policy, Tufts University
Medford, MA, USA
Summary
The aim of this research is to test the extent to which questionnaire approaches devised for use in the United States during the 1990s (with Tufts and Cornell involvement) can be adapted and enhanced for application in developing country settings. If the “direct measure” approach can track key benchmark indicators (such as income, assets or nutritional status) successfully, while also being sensitive to changes in economic status associated with project interventions, in future the approach may serve for reporting on food security activities. The study is carried out in partnership with World Vision/Bangladesh’s Food Security Enhancement Initiative. More than 600 households in rural Bangladesh will be followed over several years.
The researchers spent considerable time in villages seeking appropriate concepts, words and questions for the construction of a survey questionnaire on reported experience. The importance of in-depth ethnographic work focused on cognitive testing of questions and their interpretations cannot be overemphasized. Interviewer teams engaged respondents in formal and more informal conversations to explore meanings of individual questions as well as respondents’ understanding of answers given. During two rounds of data collection, some 40 draft questions were tested.
A principal-components analysis was conducted on these questions, resulting in a “factor” of 11 questions that appears to work well both in characterizing the problems experienced by households in Bangladesh and in identifying households along a continuum of food stresses:
(1) obliged to eat wheat instead of rice (when rice would have been preferred);
(2) needed to borrow food to meet social obligations (e.g. to serve a meal to guests);
(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 often (because own production or purchased stores ran out);
(6) the family ate few meals per day on a regular basis;
(7) the respondent adult cut back on amount of food consumed (owing to lack of food);
(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 (owing to an insufficiency of food in the household);
(10) there were times when food stored in the house ran out, and there was no cash to buy more;
(11) other adults (not the main working adult) personally skipped entire meals.
The factor of 11 items has a high reliability coefficient (Cronbach alpha of 0.89). The items cover a range of elements of the food security concept. 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 shopkeepers, being obliged to borrow food to meet social obligations), and anxiety or “insecurity” (worrying about where the next meal will come from).
The set of 11 questions correlates strongly not only with interviewer assessment of food security rankings but also with a range of comparator indicators commonly used in the analysis of poverty and food insecurity. For example, significant correlations are found with household expenditure (absolute level) and expenditure tercile (relative level) as well as a number of poverty icons such as total expenditure on clothing and shoes, landownership and especially “number of non-productive assets owned”. The set is also strongly correlated with measures of food access and adequacy, including food share in total expenditure and number of unique foods consumed. These correlations are stronger than those reported by Maxwell et al. (1999) for Ghana. However, where nutritional outcomes are concerned, anthropometric status correlates poorly with the 11-question set (as expected).
There is a high degree of concordance between male and female interviewer ratings. Overall, 85 percent of households were classified within the same category by male and female interviewers. However, 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). Men and women in the same households gave the same answer to identical questions roughly 80 percent of the time. While this shows a high agreement, the degree of disagreement increases as food security increases and there are wide divergences of opinion on some of the individual questions by gender of respondent, which requires further exploration.
Some 60 percent of households remained in the same food security category (as assessed by interviewer) across the two rounds. There was a 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. There was a similar agreement in terms of extreme changes; that is, male and female interviewers agreed independently that two households “fell” by more than one place across the groupings (from “food secure” to “hungry”), while one climbed more than two placings (from “hungry” to “food secure”).
So far, the research offers strong encouragement to the idea that this questionnaire approach is viable for the assessment of food insecurity and of the of impact of anti-hunger initiatives in this developing country.
Acknowledgements
The research is made possible through the support provided to the Food and Nutrition Technical Assistance (FANTA) Project by the Office of Health and Nutrition of the Bureau for Global Programs Field Support and Research at the US Agency for International Development (USAID), 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 USAID or of the AED.
References
Maxwell, D., Ahiadeke, C., Levin, C., Armar-Klemse, M., Zakariah S. & Lamptey, G.M. 1999. Alternative food-security indicators: revisiting the frequency and severity of “coping strategies”. Food Policy, 24: 411-429.
John Seaman
Save the Children UK
London, UK
In 1992 Save the Children UK (SC UK) entered a collaboration with the FAO Global Information and Early Warning System (GIEWS) to see if an approach could be developed that would allow GIEWS to estimate the effect of a shock, such as production failure caused by drought, on people’s ability to acquire food. GIEWS’ interest was in an approach that could be used at the national level to prioritize populations requiring further investigation.
The technical challenge was to see if it was possible to develop a methodology to estimate people’s entitlement (Sen, 1981)[29] that could:
(1) discriminate between populations of different livelihoods (e.g. primarily pastoral, agricultural) at a useful level of geographical disaggregation;
(2) discriminate between poorer and richer people within those populations;
(3) take into account changes in the value of entitlements, e.g. the collapse of the terms of trade between livestock and food;
(4) deal with multiple simultaneous shocks, e.g. changes in price, production and market access; and
(5) be used by non-specialist staff at reasonable cost while providing outputs useful to decision-makers.
These criteria suggested the use of a quantitative economic model based on information obtained using rapid data collection methods. Information is obtained from secondary sources, key informants and (chiefly) primary investigation at the household level. The basic unit of analysis is the household, which is defined as a consumption unit. Populations of households are defined in economic terms often following existing agro-economic zones.
The key steps in data collection (the definition of populations, the selection of sample sites and the collection of information at each site) are broadly similar to those of any survey (Save the Children UK, 2000). One major difference is that site selection is usually purposive, not random, chiefly in the interests of speed and cost. Another difference is that household interviews are conducted with groups of people drawn from locally defined wealth groups (in terms of livestock, land, labour or some combination of these). Group interviews allow discussion of a “typical household” within the different wealth groups. The data collected include a household budget (sources of income as food and cash, expenditure including food purchase and non-food basic items such as taxes, clothes, soap, education, etc.); the markets used for each traded commodity; and the availability of wild foods and non-market redistribution through gift, reciprocity and exchange. Information is collected for a reference year - typically a recent non-extreme year. Definitions and techniques are standardized. The balance of household budgets, their consistency with physiological requirements and the balance of intragroup exchange (e.g. of labour) provide a check on data quality. Variation in findings within wealth groups is retained in the form of ranges, e.g. maize production of 30-40 percent of household food income might reflect variation between one wealth group at two sites in one population.
The dataset therefore summarizes a population in terms of the characteristics of “typical households” across poor, middle and better-off wealth groups, and the relevant features of the economic context to which these relate.
Data analysis
Analysis is done with a simple (largely arithmetic) mathematical model. For a single area, calculations may be done by hand or on a spreadsheet. Purposefully designed software is used for larger datasets. The analysis is in two stages.
(1) Estimating the direct effect of a shock (e.g. crop failure) on the income of households in each defined wealth group. For example, if the reference income from maize production is 50 percent, and the estimate of crop failure is 50 percent, household income would be reduced by 25 percent.
(2) Working through the possible ways in which the household might be able to overcome this deficit. This involves calculating the use of food stocks, the possibility of an increased consumption of wild foods, increased availability of gifts and the increased sale of livestock and labour. The terms of trade between household assets and food are usually calculated from an estimate of future prices, although we have also experimented with simple market models to predict patterns of future price change.
A household economy approach (HEA) analysis aims to develop a quantified argument or hypothesis about the most likely effect of a given shock on household food access. The advantages of the approach are:
(1) The assumptions must be declared.
(2) A position is taken on household tradeoffs between food and non-food goods, e.g. to acquire sufficient food, a household may sell capital assets or reduce expenditure on education.
(3) It accommodates information of uncertain quality by allowing different scenarios to be developed using different starting assumptions, e.g. on shocks.
(4) It also allows interventions to be tested, e.g. price stabilization, asset preservation.
(5) Output is a hypothesis that can be used to predict observable events that should occur if the hypothesis is to be sustained, e.g. if increased livestock sales are predicted, this should also be observed.
The explanatory nature of output appears to be attractive to users. HEA has been widely applied by WFP, FAO, Famine Early Warning Systems Network, SC UK chiefly in east and southern Africa at a large scale (e.g. operation lifeline Sudan, Darfur, Somalia and Malawi) and for local assessments.
HEA and other food security measurement methods
In many settings, HEA can be used broadly to predict “wasting”, e.g. weight-for-height, where household energy intake is estimated or predicted to be low. SC UK is currently collecting cases where both datasets are available. The use of the household as the basic unit and the further simplification of the “typical household” limits the value of the data for quantitative description. However, HEA yields much useful information on income quality, seasonality (including typically gender variation in labour use), markets and wild foods that is largely unavailable from existing sources. HEA provides useful information on the degree to which households depend on production (often low) for food, information that has been employed in some locations by FAO in the interpretation of crop data.
References
Save the Children UK. 2000. The household economy approach: a resource manual for practitioners. London.
Sen, A. 1981. Poverty and famines: an essay on entitlement and deprivation. Oxford, UK, Clarendon Press.
Edward A. Frongillo and Siméon Nanama
Cornell University
Ithaca, NY, USA
Development organizations and other institutions need to measure household food insecurity for planning, targeting, monitoring and evaluation. Existing measures of food availability often are inadequate and should be augmented by measures of access to food (Wolfe and Frongillo, 2001). This study aimed to develop and validate a direct, experience-based measure of food insecurity in northern Burkina Faso that included the component of access to food, and to gain an understanding of how to develop such measures in an efficient manner. Three studies were conducted.
First, a questionnaire-based tool built on previously reported and locally known experiences of food insecurity was created and tested using individual and group interviews. It was then used in a survey by the non-governmental organization Africare to assess food security of rural households. A cluster sample of 420 production units from 26 villages was obtained. The resultant food insecurity score, which varied substantially among and within villages, was compared with socio-economic and anthropometric data using correlation and linear regression at village and production-unit levels to assess accuracy (Frongillo, 1999). Food insecurity was consistently associated as expected with socio-economic variables such as total and per capita income, livestock and equipment ownership.
Second, an in-depth qualitative study was conducted with household heads and women. The themes in the two interview guides were: identification and demographic information, production and decisions about food, cooking and eating patterns, perception on food quality, daily concerns, income sources and utilization, long-term strategies to escape food insecurity and coping mechanisms. Two villages were selected to have slight differences in language and culture. In each village, five households were selected to include secure and insecure, simple and complex, and polygamous and monogamous households. Two teams of two interviewers conducted 36 interviews (minimum of two per household); one interviewer in a team did the interviewing, and the other took notes. Data analysis proceeded in several steps to identify themes, classify households, create a table of food insecurity categories, identify questions to add or delete to the initial questionnaire, and develop and revise answer choices.
Third, a longitudinal study was initiated to provide quantitative data on changes over time in households in food insecurity, household economic situation and related factors. Data were collected on 126 simple and complex households from nine villages. These data allow examination of changes in the experience of household food insecurity twice annually across both the best and worst seasons for food and evaluation of the ability of the experience-based tool to differentiate changes in households over time. The first two (of five planned) waves were completed in July 2001 and January 2002. Across these two waves, measured food insecurity lessened as expected from the pre- to the postharvest period. Multiple linear regression was used to assess the association of change in food insecurity with changes in other variables, controlling for initial status. Change in food insecurity was associated with change in variables such as hectares farmed with animal traction, number of traction animals and farm equipment owned, number of poultry, amount of staple and cash crop harvested, and adult weight. Change in food insecurity depended on initial food insecurity such that those with a worse initial food insecurity had a greater improvement, but change in food insecurity generally did not depend on the initial status of the other variables.
The associations with variables known to be strongly related to food insecurity, both at one time and over time, demonstrate the validity of the questionnaire as a simple tool that could be used in this setting by both governmental and non-governmental organizations to assess, evaluate or monitor household food insecurity. An analogous experience-based tool can likely be developed in most other settings. About 15 items will probably be sufficient in most settings to assess directly the experience of household food insecurity, with some items probably being similar across settings and other items being different and highly specific.
This research reaffirms the value of gaining in-depth understanding of the experience of household food insecurity. This approach (rather than translating questions from other sources) will likely lead to suitable questionnaire-based tools to directly measure the experience of household food insecurity. Typically, about 20 in-depth interviews purposively selected to capture the range of the experience of food insecurity would be needed. Interview data can be analysed efficiently through a series of steps: summarize across interviews to identify themes, summarize each interview, perform (by researchers and/or key informants) ranking or sorting of households to identify what discriminates households, and perform field and cognitive testing. A generic interview and analysis guide outlining these steps is in preparation.
Acknowledgements
This research is part of the FANTA Project funded by the Office of Health and Nutrition of the Bureau for Global Programs Field Support and Research at the USAID and awarded to the AED. We are grateful to participating villagers, Africare and its staff, research staff in Burkina Faso, and FANTA and Cornell colleagues.
References
Frongillo, E.A. 1999. Validation of measures of food insecurity and hunger. J. Nutr., 129(Suppl. 2): 506S-509S.
Wolfe, W.S. & Frongillo, E.A. 2001. Building household food security measurement tools from the ground up. Food Nutr. Bull., 22: 5-12.
John Hoddinott and Yisehac Yohannes
International Food Policy Research Institute
Washington, DC, USA
Household food security is an important dimension of well-being. Although it may not encapsulate all dimensions of poverty, the inability of households to obtain access to enough food for a productive healthy life is surely a characteristic of deprivation. Thus, devising an appropriate measure of household food access is useful for several reasons: to identify the food insecure; to characterize the nature of their insecurity (for example, seasonal versus chronic); to monitor changes in their circumstances; and to assess the impact of interventions. However, obtaining detailed data on dimensions of household food security - such as 24-hour recall data on food intake - can be time-consuming and expensive, and requires a high level of technical skill in both data collection and analysis.
The juxtaposition of the value of indicators of food security, together with the difficulties in obtaining detailed information, is the motivation for this paper. Dietary diversity is defined as the number of different foods or food groups consumed over a given reference period. One dimension of household food security is household food access - a measure of the population’s ability to acquire available food during a given period. This paper explores whether dietary diversity can act as a proxy indicator of household food access under a variety of circumstances, including in poor and middle-income countries, in rural and urban areas, and across seasons. Field experience indicates that questions on dietary diversity are relatively straightforward for respondents to answer, are not considered intrusive and do not impose burdensome demands on time or recall; asking these questions typically takes less than 10 minutes per respondent. But while dietary diversity is clearly simpler to collect than data on caloric availability from seven-day recall of food acquisition or 24-hour recall of individual food intakes, in order to be appropriate as a proxy measure, it is necessary to show that dietary diversity is strongly correlated with more conventional measures of household food access.
The paper presents evidence on this issue drawing on survey data from ten countries: Bangladesh, Egypt, Ghana, India, Kenya, Malawi, Mali, Mexico, Mozambique and the Philippines. These datasets encompass both poor and middle-income countries, rural and urban areas, data collected in different seasons and data on caloric availability obtained using both recall on food acquisition and 24-hour recall on individual food intake. To be confident that our results were not driven by the use of a particular method or variable, we examined associations between dietary diversity (defined as the number of unique foods consumed in the previous seven days) and household per capita consumption, household per capita caloric availability, household per capita caloric availability from “staples” and household per capita caloric availability from “non-staples”. Additionally, we explored the associations between the number of unique food groups consumed in the previous seven days and these variables, using linear regression techniques. We also checked for the robustness of results by calculating three other measures of association: correlation coefficients (Pearson and Spearman), contingency tables, and Receiver Operator Curves.
On average, a 1 percent increase in dietary diversity is associated with a 1 percent increase in household per capita consumption, 0.7 percent increase in household per capita caloric availability, 0.5 percent increase in household per capita caloric availability from staples and 1.4 percent increase in household per capita caloric availability from non-staples. Eliminating the extreme estimates, a 1 percent increase in dietary diversity is associated with households experiencing: 0.65-1.11 percent increase in household per capita consumption; 0.37-0.73 percent increase in household per capita caloric availability; 0.31-0.76 percent increase in caloric availability from staples; and 1.17-1.57 percent increase in caloric availability from non-staples. These associations are found in both rural and urban areas and across seasons; they do not depend on the method used to assess these associations and are equally strong when using the number of unique food groups consumed as the measure of dietary diversity rather than the number of unique foods. There is also an association between dietary diversity and caloric availability measured at the individual level. Across all the country datasets examined, the magnitude of the association between dietary diversity and per capita caloric availability at the household level increases with the mean level of caloric availability. Accordingly, dietary diversity would appear to show promise as a means of measuring food security and monitoring changes and impact, particularly when resources for such measurement are scarce.
|
[29] In this context,
entitlement essentially refers to the sum of current food holdings and the
exchanges value of labour and other assets in terms of food. |
