Food Policy 72 (2017) 1–6 Contents lists available at ScienceDirect Food Policy journal homepage: www.elsevier.com/locate/foodpol Food counts. Measuring food consumption and expenditures in household MARK consumption and expenditure surveys (HCES). Introduction to the special issue ⁎ Alberto Zezzaa, , Calogero Carlettoa, John L. Fiedlerc, Pietro Gennarib, Dean Jolliffea a World Bank, USA b Food and Agriculture Organization of the United Nations, Italy c International Food Policy Research Institute, USA A R T I C L E I N F O A B S T R A C T JEL classification: This introductory paper presents the results of an international multi-disciplinary research project on the C81 measurement of food consumption in national household surveys. Food consumption data from household Q18 surveys are possibly the single most important source of information on poverty, food security, and nutrition I32 outcomes at national, sub-national and household level, and contribute building blocks to global efforts to Keywords: monitor progress towards the major international development goals. Food consumption The paper synthesizes case studies from a diverse set of developing and OECD countries, looking at some of Household surveys the main outstanding research issues as identified by a recent international assessment of 100 existing national Questionnaire design household surveys (Smith et al., 2014). The project mobilized expertise from different disciplines (statistics, economics, food security, nutrition) to work towards enhancing our understanding of how to improve the quality and availability of food consumption and expenditure data, while making them more valuable for a diverse set of users. The individual studies summarized in this paper analyze, both theoretically and empirically, how different surveys design options affect the quality of the data being collected and, in turn, the implications for statistical inference and policy analysis. The conclusions and recommendations derived from this collection of studies will be instrumental in ad- vancing the methodological agenda for the collection of household level food data, and will provide national statistical offices and survey practitioners worldwide with practical insights for survey design, while providing poverty, food and nutrition policymakers with greater understanding of these issues, as well as improved tools for and better guidance in policy formulation. 1. Background why repurposing HCES for poverty, food security and trends in social, economic, and human development. and nutrition is a smart idea Data on food consumption are needed, for example, to construct and monitor the indicator and targets required to assess progress towards Food constitutes a core component of several of the most widely the achievement of Sustainable Development Goals 1 and 2 (end pov- used welfare indicators in the domains of food security, nutrition, erty and end hunger, respectively). Similarly, its measurement is crucial health, and poverty. Accounting for about 50 percent of the household to assess and guide the FAO’s efforts to monitor and address its mandate budget (USDA, 2011), it makes up the largest share of total household to eradicate hunger, food insecurity and malnutrition. It is also the expenditure in low-income countries, on average. Low levels of access primary data input for monitoring progress of the World Bank’s twin to food are an important factor contributing chronic undernutrition, goals of eliminating extreme poverty and boosting shared prosperity. which is now estimated to plague 793 million individuals worldwide Reliable food consumption data are also required by national and local (FAO, 2015). The collection of high quality food consumption data is governments, as well as non-governmental organizations, to guide their therefore central to assessing and monitoring the well-being of any analysis, programming and policymaking. The lack of food consump- human population, and is a concern of national governments, interna- tion data or its mismeasurement may result in the misallocation of tional agencies, and anyone interested in understanding the levels of funding and may compromise the design, monitoring and evaluation of ⁎ Corresponding author. E-mail address: azezza@worldbank.org (A. Zezza). http://dx.doi.org/10.1016/j.foodpol.2017.08.007 0306-9192/ © 2017 Published by Elsevier Ltd. This is an open access article under the CC BY 3.0 license (http://creativecommons.org/licenses/BY 3.0/IGO/). A. Zezza et al. Food Policy 72 (2017) 1–6 programs and policies. data. To date, the nutrition community’s role has been overwhelmingly The last two decades have witnessed unprecedented progress in the that of a passive user of HCES data from surveys that have already been production and dissemination of household consumption and ex- conducted. Many HCES shortcomings, however, stem from design and penditure data across the developing world. In 1990, the World implementation issues. If the nutrition community—with its unique Development Report published by the World Bank was based on data skills and experiences—were to get more proactively involved in the from only 22 countries, and no country had more than one survey design, implementation and analyses of HCES, they could be strength- (Jolliffe et al., 2014). Today, there are at least 99 countries with con- ened substantially as a tool for evidence-based food and nutrition sumption or expenditure information, and most of them have multiple programming and policymaking (Fiedler et al., 2017). surveys, adding to a total of more than 687 surveys (Ferreira et al., In this paper we summarize a set of purposely assembled studies 2012). The number of countries with no poverty data (which is pri- that provide a useful perspective on the challenges and opportunities marily estimated from food consumption data) over a 10-year period for promoting a set of science-based guidelines for the food data com- declined from 33 percent to 19 percent since the 1990s, whereas the ponent of HCES. The guidelines, if endorsed and widely adopted, would share of countries with 3 or more data points over a 10-year period promote survey harmonization and increase HCES value for money by increased from 27 to 41 percent over the same period (Serajuddin et al., making these surveys more relevant to a wider set of potential users 2015). that includes nutritionists and food security analysts. The main com- Despite this progress, there are still 29 countries without a single monality linking these studies is that they compare alternative ap- survey between 2002 and 2011 – and another 28 only have one survey proaches to data collection (from existing datasets as well as from data in that time frame – that would enable estimating national poverty purposely collected for methodological studies), in order to identify the figures (Serajuddin et al., 2015). Without such data it is impossible for implication of survey design for measurement and analysis, and trans- these countries or for international development actors to analyze late those approaches considered as “best” into recommendations for trends and progress (or lack thereof) in poverty eradication, something scalable approaches in future survey design efforts. that has prompted the World Bank President to pledge to assist all IDA countries in conducting at least one such survey every three years. At 2. Putting new methodological research to work for survey design the same time, the UN Statistical Commission has established an Inter- Secretariat Working Group on Household Surveys at its forty-sixth The work program summarized in this paper has been sparked by a session “to foster coordination and harmonization of household survey desk review of the reliability and relevance of the food data collected in activities across agencies and member countries” (United Nations, national household consumption and expenditure surveys, which was 2014). These initiatives will result in a surge in the number of house- jointly led by the International Household Survey Network (IHSN), FAO hold surveys in developing countries in the coming years, underscoring and the World Bank (Smith et al., 2014). That assessment identified the the urgent need for more rigorous guidance on survey design. multiple purposes these household surveys serve, proposed methods to Depending on their primary objective, the surveys collecting in- assess the reliability and relevance of survey questions, and applied formation on household consumption or expenditure take different these methods to 100 household surveys from low- and middle-income forms, including Household Budget Surveys (HBS), Income and countries (a sample that resulted from selecting the most recent na- Expenditure Surveys (IES), or ‘multi-purpose’ or ‘integrated’ household tionally representative household survey from each developing country, surveys, like the Living Standards Measurement Study (LSMS) surveys. with the only condition of having enough documentation). The as- We refer to this family of surveys, which are usually nationally and sub- sessment points to many areas where survey design and questionnaires nationally representative, as Household Consumption and Expenditure can be significantly improved, among which five were selected as key Surveys (HCES). themes for the research projects we summarize in this paper. They are While the variety of HCES purposes naturally translates into dif- the following: ferent designs, the dramatic increase in the number of household sur- • Choosing diary or recall surveys, and the appropriate re- veys in developing countries has been associated with a proliferation of ference period. Citing concerns about memory loss when collecting approaches and methods used in the collection of food data that is not very detailed consumption or intake data, nutritionists generally favor only due to their different purposes or country-specific considerations. shorter recall periods (e.g. 24-hour recall), whereas expenditure surveys While there exist international guidelines and recommendations for the commonly use recall of one week or more in order to be better able to design and implementation of each of the distinct types of HCES sur- capture “usual” behavioral patterns. The impacts of recall period de- veys, they are specific to each type of survey, are generally not pre- cisions on the quality of the data for different uses are far from being scriptive, lack coherence and usually leave much flexibility to national fully understood, and some of the papers we summarize address ques- survey statisticians. Consequently, we observe heterogeneity in tions related to quantifying the tradeoffs involved in having a longer methods, even within the same type of survey, both across countries, as recall period and increased memory loss. To assess usual consumption, well as within countries over time. how many times should data be collected from households and for what While HCES typically are not designed for the purpose of addressing observation or “reference” period? What difference will extending re- food and nutritional information gaps, they are increasingly being used ference periods and conducting repeat visits actually make to estimates for this purpose. There are many reasons for the increase in use of the of poverty and inadequate nutrient intakes? HCES surveys for this end, including that they contain a wealth of in- • Food consumed away from home (FAFH) and cooked/pack- formation about food acquisition and consumption; are being done with aged meals. FAFH and prepared foods represent an increasing share of increasing frequency in an increasing number of countries (Serajuddin food consumption, and will continue to be so as GDP per person grows, et al., 2015); have large samples, statistically representative at subna- and food systems evolve. This is an area where many national surveys tional levels; and are much less costly than other dietary assessment could be improved, but where evidence on the robustness of alternative data sources because these multi-purpose surveys are already being methods is weakest. A sub-set of the papers look at the implication and conducted and paid for by other government agencies (Fiedler, 2013). methods for capturing FAFH, whether eaten in commercial or public While there has been a surge of interest and HCES analyses of nu- establishments (e.g. restaurant, schools). trition and food security issues, the potential of this particular type of • Measuring individual versus household consumption. The repurposing of HCES has yet to be realized for several reasons. First, food consumption/expenditure modules of HCES capture household there is a lack of awareness of public nutritionists and food policy level information. Yet, food and nutrition policies and programs often analysts about what these data contain. Second, there is a need for require information about which foods and nutrients are consumed by further research and action to improve the quality and utility of these which groups of individuals, and in what quantity. While individual 2 A. Zezza et al. Food Policy 72 (2017) 1–6 dietary intake data are more appropriate for meeting these information indications on the effects of these key survey design choices on the needs, HCES are more widely available and conducted more regularly resulting data, but it also identifies areas where more research and than individual-level dietary assessments. Furthermore, most dietary validation is needed. In what follows, we present a summary of the surveys do not assess the intake of all household members, making it main findings, organized around the key survey features listed in Sec- difficult to plan programs, such as fortification programs, that are in- tion 2 above. The reader is encouraged to turn to the individual papers tended to benefit more than just one demographic group. Until in- for a more complete understanding of the issues and findings of each dividual-level dietary data collection becomes routinely available, un- paper. derstanding whether and how household-level data can be used to approximate actual individual food and nutrient consumption is a 3.1. Assessing the quality of recall and diary surveys and informing the worthwhile undertaking. Some of the papers summarized here propose choice of their reference period methodologies for deriving individual level estimates from household data. The first finding is that recall surveys tend to return higher con- • Measuring food acquisition versus measuring food consump- sumption values (whether in monetary or caloric terms) than diaries, tion. The term food consumption is interpreted in many ways. For with the difference between the two declining with the length of the economists and poverty analyst the focus is on the amount of money recall period. This finding is common to the papers by Conforti et al. spent to acquire food; for food security analysts, it is on the amount of (this issue) based on a cross country analysis of dietary energy con- food available for consumption; whereas nutritionists are primarily sumption (DEC) estimates, to an analysis of experimental Niger data by concerned with the quantities of foods actually eaten. Food data were Backiny-Yetna et al. (this issue) and to the analysis of Canadian data in initially collected in HCES simply to construct the consumer price in- Brzozowski et al. (this issue) and is consistent with the evidence from dices or to inform national accounts. As a result, the food data collected the Tanzania experiment reported in Beegle et al. (2012). referred primarily to items the household acquired through purchases The paper by Friedman et al. (this issue) based on the same Tan- during the reference period. Over the years, food items procured zania survey experiment, sheds some light on how different survey through own-production, barter, gifts and payment-in-kind were in- design options are affected by different types of reporting error. They troduced into these surveys to better capture food acquisition in rural find that relatively short recall modules (such as the 7 day) are affected areas. These surveys aimed at capturing food that was acquired by the on the one hand by an underestimate of the incidence of consumption, household with the intention that it would be consumed. With time, particularly for infrequently consumed items, accompanied by an surveys have been increasingly focused on food items actually con- overestimate of the value of consumption, conditional on positive sumed by the household and the various sources from which food was consumption (due to telescoping error). The recall error appears to be acquired. One issue for methodological research to examine, therefore, larger for less frequently consumed items, and on short recall periods. is the extent to which there are systematic differences in food data Adopting a ‘usual month’ approach to recall does not appear to solve collected using acquisition type surveys versus consumption type sur- the problem, as it results in an overestimate of the incidence of con- veys versus combinations of these two types of surveys. sumption particularly for goods that are not regularly consumed, and • Length and specificity of survey food lists. For many analytical an underestimate of consumption values for staples. Importantly, the purposes, survey food lists need to be sufficiently detailed to accurately ‘usual month’ approach also imposes a significantly higher burden on capture consumption of all major food groups making up the human the respondent and results in longer interviews. Bounding of the recall diet. There are trade-offs involved in the design of a survey food list, period with an earlier visit, and further prompting of food items are including its length and specificity that are not well understood. Some possible avenues for improving the quality of recall data, but more of the papers provide evidence to help survey design practitioners and methodological work is needed to test that hypothesis. analysts to quantify those tradeoffs and to highlight their implications Diaries remain a viable option where the conditions are such that for policy analysis. they can be properly implemented, but steps can be taken to improve their accuracy by explicitly prompting about the consumption of in- 3. New and emerging evidence dividual household members, particularly for foods and meals that may be consumed outside the households (more on this below). Repeating Starting from where Smith et al. (2014) left off, the World Bank and diaries in successive visits (Troubat and Grünberger, this issue), or in- FAO decided to join forces to assemble rigorous empirical evidence to creasing their length (Backiny-Yetna et al., this issue; Brzozowski et al., inform key decisions related to survey design choices for national this issue; Engle-Stone et al., this issue), appears on the contrary not to consumption and expenditure surveys, using data from a variety of add, and may even reduce the quality of the resulting data (including settings and types of data sources, with the aim to contribute to filling a for nutrition analysis), while increasing cost, and is therefore to be significant gap in the literature and inform policy. discouraged. The research that we summarize in what follows brings together Worryingly, some of the observed measurement error patterns ap- empirical studies on the implications of different survey design options pear to be systematically associated with household characteristics, for the measurement and estimation of different indicators and para- with measurement error higher in low resource households in Tanzania meters of importance to several development domains. The data used in (Friedman et al., this issue) and in selected regions in urban Mongolia these studies include nationally-representative data and detailed case (Troubat and Grünberger, this issue), or significantly associated with studies across a range of countries from several developing regions, as income and household demographics in Canada (Brzozowski et al., this well as from high-income countries. The results draw on expertise from issue). a range of disciplines and institutional backgrounds, as authors are nutritionists, food security and poverty analysts, economists, and sta- 3.2. Measuring food away from home (FAFH) tisticians from national statistical offices, international organizations and academia. Several of the studies exploit the peculiar features of The secular increase in FAFH as a percentage of total food con- existing national datasets which collected data in a particular way that sumption has long been demonstrated (Smith, 2013) and national sta- allowed comparing data associated with different survey design tistical offices the world over are struggling to find new tools to keep up choices, by either within or between subject comparisons. One of the with the challenge of measuring this component of households’ food studies utilizes data from a randomized methodological experiment. expenditure and consumption. Three papers in this volume deal spe- The list of studies this synthesis draws on is provided in Table 1. The cifically with measuring FAFH. evidence presented in the forthcoming volume provides some very clear Borlizzi et al. (this issue) use national data from Brazil to show how 3 A. Zezza et al. Food Policy 72 (2017) 1–6 Table 1 List of papers in the special issue. Authors Country Data Focus 1. Conforti et al. Global 81 national surveys Impact of survey design issues on estimates of dietary energy consumption (DEC) 2. Friedman et al. Tanzania Experimental Impact of survey design approach on frequency and value of consumption expenditure by (type of) food item 3. Sununtnasuk and Fiedler Bangladesh National. Includes both HH level and individual 24HR data Comparison of estimated AME from HH data with results from 24HR recall 4. Engle-Stone et al. Bangladesh National. Includes 7 consecutive 2-day recall periods Impact of extending the length of the recall period on estimates of nutrient inadequacies 5. Troubat and Grünberger Mongolia National, urban. Includes one-month recall, and a diary of Comparison of diary and recall, and value added of longer 3*10 days, and questions on food stocks recall and repeated visits for estimates of DEC 6. Kirlin and Denbaly USA National data, innovative in many respects (coverage of Lessons learned from the application of innovative FAFH, assistance programs; use of hand-held scanners) approaches to data collection 7. Farfán et al. Peru National. Includes detailed FAFH module, including at the Impact of accounting for FAFH on the estimated incidence individual level and profile of poverty 8. Borlizzi et al. Brazil National. Detailed information on FAFH including meals at Impact of accounting for school feeding on average and school distribution of DEC 9. Coates et al. Ethiopia, Oromiya and SNNP regions (Ethiopia), 15 districts in Comparison of estimated AME from HH data with results Bangladesh Bangaldesh. HH dietary data and individual level intake from 24HR recall estimates 10. Brzozowski et al. Canada National. Diary and recall data Impacts of adoption of diary vs recall on extent and features of measurement error in estimate of food expenditures 11. Louzada et al. Brazil National. HH acquisition and individual food intake Possibility of capturing important dietary features (consumption of ultra-processed foods) in HCES 12. Fiedler and Yadav India National. HH questionnaire on FAFH, and individual Impact of adopting a MAFH questionnaire on indicators of questionnaire on MAFH food consumption and food insecurity 13. Backiny-Yetna et al. Niger Experimental. Niamey and Tillabéry districts. Includes 7-day Impact of questionnaire design (length of recall, diary) on and usual month recall, and 7-day diary questionnaires frequency and value of food expenditure failing to account for school meals inflates the degree of inequality in on the potential to use household level HCES data to make inferences the distribution of calorie consumption. They therefore recommend about individual level nutrient intake and adequacy in individuals. that surveys should collect data on actual quantity and quality of food Sununtnasuk and Fiedler (this issue) using national data for Bangla- consumed in schools and in other large national feeding programs, in- desh, find that in 91 percent of cases, estimates of energy and nutrient cluding by integrating the survey data collection with administrative intake based on the gold-standard (for nutrition analysis) individual 24 information on the meal contents when that is available. Similarly, hour-recall were identical to those estimated from household data as- Farfán et al. (this issue) show how poverty analysis needs to take FAFH suming that food is distributed to household members in proportion to into account as once it is incorporated it can shift both the consumption their share of the household’s total Adult Male Equivalent (AME). expenditure distribution and the poverty line in ways that cannot be Qualitatively similar results are reported by Coates et al. (this issue) determined a priori. Using data from Peru they find that accounting for using a different dataset for Bangladesh and data from Ethiopia. Both FAFH has opposite implications on the extent of severe and moderate studies warn however that while population based estimates using the poverty levels, and has a significant impact on the profile of the poor. AME approach are reasonably accurate, they tend not to be reliable for The paper by Fiedler and Yadav (this issue) opens one avenue for specific at-risk groups in the population, such as children under three further methodological validation by showing how the introduction of a years of age. According to Coates et al. (this issue) adjusting for par- new module collecting information on Meals Away from Home takers and activity levels does not significantly improve the accuracy of (MAFH), with household-member specific responses, in the India Na- the AME prediction. While HCES cannot replace individual nutrition tional Survey Sample Round 68 substantially reduced measurement intake surveys, they can provide useful information for nutritionists. error in this component of food consumption expenditure. The paper by Nutritionists should be brought on board at the survey design stage to Kirlin and Denbaly (this issue) while not exclusively focused on FAFH, ensure that- to the extent possible, -survey implementation reflects their also makes the point that in the US, accounting for FAFH-including food particular insights and needs. acquired for free-is essential in order to understand the determinants of One specific aspect where capturing information at the individual demand for food, one of the many objectives for which HCES data is levels is especially important is FAFH. The main ‘food preparer’ can be used to investigate. reasonably informed of household members’ food consumption at the house, but less so when it comes to food consumed (especially if also prepared) away. The findings of Fiedler and Yadav (this issue) do, 3.3. Individual vs household however, report on an important innovation in India’s main national survey where complementing household- and individual-based in- While concepts like poverty and food security can be articulated and formation on meals away from home substantially reduced the extent of studied at the level of the household, nutrition is an eminently in- measurement error. This is, therefore, an approach that holds promise dividual outcome. That is the reason why the papers in the volume that for reducing measurement error if it is adopted in other countries: ad- look at methods for estimating individual level outcomes from HCES ditional testing in different settings would be desirable. have a specific nutrition focus. That is not to say of course that intra- household resource allocation questions are not important for poverty and food security analyses as well (Chiappori et al., 2015) and there 3.4. Acquisition vs consumption have been studies recently attempting to estimate individual con- sumption shares from HCES data (Dunbar et al., 2013). One important, often overlooked, difference in the design of HCES is The results of the papers by Coates et al. (this issue) and that while some ask respondents to report about how much food was Sununtnasuk and Fiedler (this issue) are both reasonably encouraging acquired (i.e. purchased, received as gift or other in-kind transfer, or 4 A. Zezza et al. Food Policy 72 (2017) 1–6 produced or collected by household members) during a given period, the volume reviewed here try to shed more light on these issues. other surveys ask about the food the household consumed (often asking The experimental work undertaken by Friedman et al. (this issue) separate questions on the source of acquisition for the food that was compares the use of lists of 58 and 17 food items, and a list of 11 broad consumed). Some surveys ask the household to report acquisition from food categories (as opposed to specific food items). Their findings purchases and combine that with questions on consumption from own provide strong evidence that shortening the list to such an extent in- production and transfers. For surveys implemented over a full year the troduces considerable measurement error (particularly in the case of difference between these methods should not matter as one would not the food categories) without a correspondingly significant reduction in expect to observe large changes in food stocks at the household level. In interview time. practice the differences might be substantial, particularly when data are Nutritionists are interested to the length and composition of food collected only at one particular time of the year (Conforti et al., this lists also to the extent that they can signal trends in the consumption of issue). food items or food groups that are of particular interest due to their The evidence based from a regression analysis using a sample of 81 nutritional benefits or for their association with dietary related issues national surveys presented in this volume by Conforti et al. (this issue) such as overweight and obesity.1 Louzada et al. (this issue) use a par- is that acquisition (and to an even larger extent acquisition/consump- ticularly unique survey that combines household acquisition data with tion surveys) tend to return higher caloric counts than surveys asking individual intake data to study whether HCES hold promise of allowing households to report food consumption. Conforti et al. (this issue) also to study the consumption of ‘ultra-processed foods’. These are food find that while the coefficient of variation of acquisition surveys is on items that are relatively dense in their content of sugars, total fats, average also higher, as one would expect a priori, the difference be- saturated fats and trans fats, and poor in vitamins and other important tween survey types disappears when other survey features and country micronutrients, and have been shown to be associated with obesity, characteristics are controlled for. The authors note how even if small on diabetes and other diet related diseases. They find a substantial average, the differences in dietary energy consumption (DEC) asso- agreement between the individual intake data and the HCES data, ciated with collecting consumption instead of acquisition data may particularly in terms of relative (percentage) energy consumption from result in large increase in the measures of undernourishment and have ultra-processed foods. HCES do, therefore, hold potential to be used in therefore substantial implications for analytical applications. the analysis of food consumption of such specific components of the Troubat and Grünberger (this issue) use a survey of households in diet, and clearly food lists will have to be crafted with an attention to urban Mongolia to compare average DEC from acquisition data to a those items that are more relevant for a country’s nutrition policy. measure of consumption derived from acquisition augmented with In the future barcode scanners hold promise of being integrated in stock variations. They find that since in a majority of cases ending survey operations to identify specific food items, but the challenges of stocks were recorded in days of the month away from pay days, they such operations are substantial even in countries with the most ad- tended to be underestimated resulting in consumption being sig- vanced statistical services, such as the United States. In this volume, nificantly higher than acquisition. When considering only households Kirlin and Denbaly (this issue) report on the issues encountered in the for which beginning and ending stock data were collected at compar- implementation of the national Food Purchase and Acquisition Survey. able times, the difference disappeared, suggesting stocks can be esti- A substantial amount of post-data collection processing was found to be mated with recall surveys, but attention needs to be paid to correctly necessary in order to identify specific food items in the data, and re- spreading the interview time over the survey period. sulted in unexpected delays and increased costs. Future rounds of the survey will provide more indications of how such problems can be 3.5. Food lists avoided and how scanners may be better integrated in survey opera- tions. Trade-offs in the design of the food lists included in a food con- sumption (or acquisition) questionnaire are quite evident. A short list 4. Conclusions will result in fewer foods and transactions, with an impact likely to be more serious for households with a more varied diet. Adding ‘too many’ The research summarized in this paper presents new evidence on food items to the list will on the other hand result in a burden of re- the impact that survey design choices may have on the quality and spondents and enumerators in return for an irrelevant amount of in- relevance of the food consumption data collected in HCES. The survey formation collected for the few households that will report consuming design lessons derived from this body of work include: or purchasing the additional items. Any quest for an ‘optimal’ length of a food list with global applicability is however not likely to succeed as 1. There is enough evidence about how some survey design options are diets are so different across the planet that the length and composition detrimental to data quality while not providing sufficient benefits in of any list will need to be country specific. other domains (e.g. cost savings) to justify that loss of accuracy or Smith et al. (2014) put forward two criteria and a few rules of precision. Such survey practices–including the use of a “usual thumb to guide the design of food lists. They call for survey designers to month” recall period, and neglecting to collect information about ensure the comprehensiveness and specificity of food lists. With compre- school-based and other widespread feeding programs–should be hensiveness they mean that data should be collected on all of the types discontinued. of food and beverages that make up a modern human diet. The rules of 2. Trade-offs exist across some survey design options, for which there thumb they employ on that aspect are that (a) 14 basic food groups is no clear right or wrong approach. The research presented here can must be represented by at least one item in the survey questionnaire; help making those trade-offs clear to survey designers and data and (b) at least 40 percent of products should be processed food items. analysts and inform their decisions. Among the most important such With specificity they mean that survey food lists should be sufficiently considerations are: taking into account common shopping habits detailed to accurately capture consumption of all major food groups and balancing the capturing of foods less frequently purchased, the making up the human diet. They suggest two rules of thumb to employ length of the recall period and memory loss; the degree of specificity in this respect: (a) a minimum number of food items should be included and length of the food list and the capturing of nutritionally distinct in each of the 14 basic food groups; and (b) no more than 5 percent of the food items listed in the questionnaire should span more than one basic food group. While motivated by clear objectives, these criteria 1 Products that are or can become the target for food fortification programs should also and rules of thumb are not based on empirical evidence on how specific be singled out in food list as HCES can be used to assess the (potential) coverage of such choices in food list design affect data collection. Some of the papers in programs. 5 A. Zezza et al. Food Policy 72 (2017) 1–6 and nutritionally significant food items; and the length of the food Coates, Jennifer, Lorge Rogers, Beatrice, Blau, Alexander, Lauer, Jacqueline, Roba, Alemzewed, 2017;al., this issue. Filling a dietary data gap? Validation of the Adult list and how it affects interview time and survey implementation Male Equivalent method of estimating individual nutrient intakes from household- costs. level data in Ethiopia and Bangladesh. Food Policy (this issue). 3. For some survey design options the empirical evidence is not yet Conforti, Piero, Grünberger, Klaus, Troubat, Nathalie, 2017;al., this issue. The impact of survey characteristics on the measurement of food consumption. Food Policy (this sufficient to formulate definite conclusions regarding how best to issue). capture some information. Investments in methodological research Louzada, da Costa, Laura, Maria, Levy, Renata Bertazzi, Martins, Ana Paula Bortoletto, should prioritize these domains, with the collection of data on food Claro, Rafael Moreira, Steele, Euridice Martinez, Junior, Eliseu Verly, Cafiero, Carlo, Monteiro, Carlos Augusto, 2017;al., this issue. Validating the usage of household food away from home, being a priority. acquisition surveys to assess the consumption of ultra-processed foods: evidence from 4. HCES data do hold the potential for a wide range of users (econo- Brazil. Food Policy (this issue). mists, food security analysts, statisticians, nutritionists) for im- Dunbar, Geoffrey R., Lewbel, Arthur, Pendakur, Krishna, 2013. Children’s resources in collective households: identification, estimation and an application to child poverty portant, policy relevant analyses. A multidisciplinary approach to in Malawi. Am. Econ. Rev. 103 (1), 438–471. survey design both at the national and international levels can help Engle-Stone, Reina, Sununtnasuk, Celeste, Fiedler, John L., 2017;al., this issue. to ensure that these surveys are designed with an eye to their Investigating the significance of the data collection period of household consumption multiple potential uses, thus increasing the informational value for and expenditures surveys for food and nutrition policymaking: analysis of the 2010 Bangladesh household income and expenditure survey. Food Policy (this issue). money that comes with implementing an HCES. Farfán, Gabriela, Genoni, María Eugenia, Vakis, Renos, 2017;al., this issue. You are what 5. Even though there might not be enough empirical evidence to re- (and where) you eat: capturing food away from home in welfare measures. Food solve all survey design puzzles, and countries might have good Policy (this issue). Ferreira, Francisco H.G., Chen, Shaohua, Narayan, Ambar, Sangraula, Prem, Dabalen, reasons to adopt different approaches, an international effort to Andrew L., Serajuddin, Umar, Yoshida, Nobuo, et al., 2015. A Global Count of the systematize lessons learned from methodological research into Extreme Poor in 2012: Data Issues, Methodology and Initial Results. Policy Research practical guidelines for survey designers would be very useful. Such Working Paper. The World Bank, Washington, DC, October 3, 2015. Fiedler, J.L., 2013. Towards overcoming the food consumption information gap: a report would help to (a) ensure national statistical offices receive strengthening household consumption and expenditures surveys for food and nutri- consistent, science based advice for designing surveys to capture tion policymaking. Global Food Secur. 2 (1), 56–63. data on food consumption/acquisition, and (b) increase the har- Fiedler, John L., Yadav, Suryakant, 2017Fiedler and Yadav, this issue. How can we better capture food away from home? lessons from india’s linking person-level meal monization of surveys across countries and over time. household-level food data. Food Policy (this issue). Food and Agriculture Organization (FAO), 2015. The State of Food Insecurity in the Acknowledgments World 2014. FAO, Rome. Friedman, Jed, Beegle, Kathleen, De Weerdt, Joachim, Gibson, John, 2017;al., this issue. Decomposing response error in food consumption measurement: implications for The authors, as Guest Editors of the Special Issue, would like to survey design from a randomized survey experiment in Tanzania. Food Policy (this thank all those who contributed to this project: the authors, the peer issue). reviewers, and participants in the November 2014 FAO Workshop on Jolliffe, D., Lanjouw, P., Chen, S., Kraay, A., Meyer, C., Negre, M., Prydz, E., Vakis, R., Wethli, R., 2014. A Measured Approach to Ending Poverty and Boosting Shared Improving the Relevance and Reliability of Food Data from Household Prosperity: Concepts, Data, and the Twin Goals. The World Bank. Consumption and Expenditure Surveys. The project benefitted from Kirlin, John A., Denbaly, Mark, 2017Kirlin and Denbaly, this issue. Lessons learned from funding from the Global Strategy to Improve Agricultural and Rural the national household food purchase and acquisition survey in the United States. Food Policy (this issue). Statistics. Serajuddin, Umar, Uematsu, Hiroki, Wieser, Christina, Yoshida, Nobuo, Dabalen, Andrew L. 2015. Data Deprivation: Another Deprivation to End. Policy Research Working References paper; No. WPS 7252. World Bank Group, Washington, D.C. Smith, L.C., 2013. The Great Indian Calorie Debate: Explaining Rising Undernourishment during India’s Rapid Economic Growth. IDS Working Paper 430. Backiny-Yetna, Prospère, Steele, Diane, Yacoubou Djima, Ismael, 2017;al., this issue. The Smith, L., Dupriez, O., Troubat, N., 2014. Assessment of the Reliability and Relevance of impact of household food consumption data collection methods on poverty and in- the Food Data Collected in National Household Consumption and Expenditure equality measures in Niger. Food Policy (this issue). Surveys. IHSN Working Paper No. 008. Beegle, K., De Weerdt, J., Friedman, J., Gibson, J., 2012. Methods of household con- Sununtnasuk, Celeste, Fiedler, John L., 2017Sununtnasuk and Fiedler, this issue. Can sumption measurement through surveys: experimental results from Tanzania. J. Dev. household-based food consumption surveys be used to make inferences about nu- Econ. 98 (1), 3–18. trient intakes and inadequacies? A Bangladesh case study. Food Policy (this issue). Borlizzi, Andrea, Del Grossi, Mauro Eduardo, Carlo, Cafiero, 2017;al., this issue. National Troubat, Nathalie, Grünberger, Klaus, 2017Troubat and Grünberger, this issue. Impact of food security assessment through the analysis of food consumption data from survey design in the estimation of habitual food consumption a study based on urban household budget and expenditure surveys: the case of Brazil’s Pesquisa de households of Mongolia. Food Policy (this issue). Orçamento Familiares 2008/09. Food Policy (this issue). United Nations, 2014. Report of the World Bank on Improving Household Surveys in the Brzozowski, Matthew, Crossley, Thomas F., Winter, Joachim K., 2017;al., this issue. A Post-2015 Development Era: Issues and Recommendations Towards a Shared Agenda. comparison of recall and diary food expenditure data. Food Policy (this issue). UN Economic and Social Council E/CN.3/2015/10, New York. Chiappori, Pierre-André, Meghir, Costas, 2015. Intra-household inequality. In: Atkinson, United States Department of Agriculture, Economic Research Service, 2011. International Anthony B., Bourguignon, François (Eds.), Handbook of Income Distribution, vol. 2B, Food Consumption Patterns, 2005. USDA/ERS, Washington, DC. pp. 1369–1418. 6