Policy Research Working Paper 10018 Determining the Caloric Content of Food Consumed away from Home An Application to the Construction of a Cost-of-Basic-Needs Poverty Line Kristen Himelein Poverty and Equity Global Practice April 2022 Policy Research Working Paper 10018 Abstract Food purchased and consumed away from home is a grow- equal composition in the food baskets consumed inside and ing share of household expenditure in developing countries. outside the home and uses data from a consumption exper- Therefore, measuring the monetary value and estimating the iment in the Marshall Islands to estimate a “multiplier” to caloric equivalent of these meals are increasingly important increase the per-calorie cost to allow for these expenses. The for the accurate calculation of a cost-of-basic-needs poverty methodology generates reasonable estimates of meal-spe- line. The standard approach uses the per-calorie cost of the cific and overall multipliers. Although the impact of their food consumed at home to estimate the caloric equivalent application is minimal in this case, it may be larger in con- of food purchased and consumed away from home, but it texts with higher shares of food purchased and consumed does not include an allowance for the overhead or profit away from home in total consumption. of the food seller. This paper retains the assumption of This paper is a product of the Poverty and Equity Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at kristen. himelein@gmail.com. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Determining the Caloric Content of Food Consumed away from Home: An Application to the Construction of a Cost-of-Basic-Needs Poverty Line Kristen Himelein 1 JEL: C81, C83, E21, I32 Keywords: cost-of-basic needs poverty lines, food away from home, consumption, Marshall Islands Acknowledgments: Funding provided by the Government of the Marshall Islands and the Australia Department for Finance and Trade. The author is grateful to Michael K. Sharp, Sergio Olivier, and Maria Gabriela Farfan Betran for comments on earlier drafts. All remaining errors are those of the author. 1 Kristen Himelein is in the Poverty and Equity Global Practice of the World Bank and can be contacted at khimelein@worldbank.org. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1. Introduction The objective in calculating a cost-of-basic-needs monetary poverty line is to determine and price a bundle of goods that meet the basic consumption needs of a household (Ravallion, 1994). Though these calculations have been an active area of research for decades, much of the work has focused on determining the appropriate threshold value for the food energy intake or setting the non-food component, while the calculation of the food energy intake itself was treated as a simple accounting exercise based on quantities from the survey and calories per 100 grams from auxiliary data. These calculations, however, have become more difficult as food away from home (FAFH) has become a larger share of the consumption basket, due to both changing dietary patterns and improved efforts in surveys to accurately capture this information. This paper describes a method developed using data from an experimental survey conducted in the Republic of the Marshall Islands (RMI) to translate spending into calories for the purpose of the calculation of a poverty line. The method builds on the standard method proposed by Subramanian and Deaton (1996) in which calories for FAFH are assumed to equal the cost per calorie estimated across the foods consumed within the household when only expenditure data is collected. The approach, however, ignores the share of FAFH spending that is captured by the food vendor either to cover expenses or as profit. This paper therefore calculates a simple “multiplier” to account for those costs in the conversion of FAFH spending into calories. The remainder of this paper is organized as follows. The next section further motivates the work by discussing the relative importance of food consumed away from home in the consumption aggregate, how this information is usually used in the construction of a cost-of-basic-needs poverty line, and the potential impact of changes to the amount and share of FAFH. Section 3 describes the 2018 Marshall Islands consumption experiment and the RMI 2019/2020 Household Income and Expenditure (HIES) data set, section 4 presents the analytical approach and results, and section 5 concludes with a discussion of the limitations of this study and recommends areas for future research. 2. Background Food consumed away from home is an increasing share of dietary energy and nutrition in both developed and developing countries. In the United States, FAFH increased from 44 percent of total food spending in 1987 to just over 50 percent in 2010, representing the first time in which the majority of food spending was outside the household (Saksena et al, 2018). In developing countries, an increasing share of meals are also consumed outside the household. In the Arab Republic of Egypt, the share of meals away from home increased from 20 percent in 1981 to 46 percent in 1998, and in India, the share of households reporting consuming at least one FAFH meal in the previous month increased from 23 percent in 1994 to 39 percent in 2010 (Smith et al, 2014; Smith 2012). Spending has increased accordingly. Claro et al (2014) found that food consumed away from home in Brazil increased 25 percent between 2002/3 and 2008/9 to a total of 28 percent of total food spending. Farfán et al (2017) found similar increases between 2006 and 2013 in Peru, which increased by 21 percent over this period to a total of 27 percent. You (2014) estimated a mean annual increase in per capita food away from home spending of 9.5 percent from 2002 to 2011 in China, where the total share reaching 22 percent, and De Brauw and Herskowitz (2020) found that the mean FAFH consumption increased by about 17 percent from 2011 to 2016 in Nigeria. Beyond restaurant and take-away consumption, free meals have also been identified as an important source of calories, with Borlizzi et al 2 (2017) finding that accounting for calories consumed by children through free school meal programs halved the estimated prevalence of childhood undernourishment. In addition to accurately capturing mean consumption, properly capturing and analyzing FAFH is also important to inequality measurement as the share of FAFH in the consumption basket is not consistent across the wealth distribution. Farfán et al (2017) find that though the absolute amount of spending on food away from home in Peru increases with higher total consumption, the share of spending on food away from home to total consumption is highest for those at the bottom of the distribution. De Brauw and Herskowitz (2021) found that 8.9 percent of spending for the lowest decile was on FAFH in Nigeria compared to a mean of 13.2 percent and 18.4 percent in the highest decile, and Claro et al (2014) similarly found the FAFH share to be increasing as incomes increased. The increasing prevalence of FAFH in the diets of individuals and households underscores the importance of collecting this important consumption item completely and accurately. According to the criteria established by Smith et al (2014) to assess the adequacy of FAFH measures, even though 90 percent of the surveys they reviewed collected some type of FAFH, only 42 percent met the minimum criteria to be considered reliable. Nearly one-quarter of countries reviewed in this work used only one question to capture FAFH for all household members. In their recent work in Vietnam, Farfán et al (2019) found the one question per household method underestimated food consumption by 33 percent, and that the cost-effect best practice to FAFH consumption data was a bounded household-level recall assisted by a simple worksheet over individual diaries. There is limited literature on the impact of choices around how FAFH is collected and incorporated into poverty analysis. Farfán et al (2017) found that failure to adequately account for FAFH leads to distortions in the poverty rates, which come from underestimating food expenditures and either under or overestimating the value of the poverty line. In Peru, they find an overestimation of moderate poverty by 16 percent, with the gains in expenditure offsetting the resulting shift in the poverty line, and an underestimation of extreme poverty by 18 percent, which occurs due to higher per-calorie costs from FAFH compared to home-prepared food increasing the cost of the food basket component of the poverty line. In the case examined here, the impact is more straightforward as spending is constant and therefore there is no expenditure effect, only an increase in poverty due to an upward shift of the poverty line. Thinking of a simplified example in which a basket that is 75 percent food at home and 25 percent food away from home has a cost of 100 local currency units (LCU) and provides 3,000 calories with no multiplier, split between 2,250 food at home calories costing 75 LCUs and 750 FAFH calories costing 25 LCUs. If the calorie threshold is 2,500 calories, the cost of the basket would scale to 83.33 LCUs and this would be the food component of the poverty line. If instead a multiplier of 1.2 was applied, which says that 20 percent of the spending on FAFH actually does not go to food but instead to profit and overhead, and therefore the FAFH spending buys 20 percent fewer calories, or 625 calories. With the multiplier, the 100 LCUs buys only 2,875 calories, and therefore when scaled to the 2,500 calorie threshold, the food component of the poverty line is 86.96 LCU. This result is intuitive since increasing the cost per calorie of FAFH to allow for profit and overhead increases the cost of the food basket. In this example, the 20 percent increase in FAFH calories leads to a 4.4 percent increase in the cost of the basket, but this result would vary depending on the relative share of FAFH in the basket and the size of the multiplier. The increased value of the food component would also likely lead to a higher value for the non-food component as it shifts the reference population used to calculate the non-food component towards better-off households, which generally have a higher non-food share in their total consumption. 3 3. Data This analysis uses two data sets collected in RMI to examine the impact of incorporating a multiplier in the construction of a poverty line. The first is used to calculate the multipliers and is an experimental data set by the Economic Policy, Planning and Statistics Office (EPPSO) of the Government of the Marshall Islands between July and October 2018. The experiment was designed to study the impact of switching the collection method for consumption data and the mode of collection from paper-based to tablet-based questions. The sample design was therefore stratified by geography and treatment arm. The geographic stratification divided the country into three areas: Majuro – the capital and main urban center; Ebeye – the second largest urban area located on the Kwajalein atoll; and finally the rural outer islands. The design selected four urban clusters in each of Majuro and Ebeye and four clusters on the outer islands, but did not attempt to generate nationally representative estimates, rather targeted the most populous areas of the country as well as the rural contrast. In Majuro and Kwajalein, clusters were selected randomly with probability proportional to size from the universe of census enumeration areas with more than 80 households at the time of the 2011 census. In the rural outer islands, the most populated enumeration areas of Ailinglaplap, Namdrick, Jaluit, and Wotje were selected. Households in the selected EAs were randomly selected following a listing operation. See Sharp et al. (2019) for a further description of the experimental design. The total sample size of the experiment was 792 across 5 arms, but only 2 arms collected data appropriate for this analysis, the highly monitored paper and electronic diaries. The protocols for these arms instructed respondents to maintain a diary of all food and common non-food expenditure, with interviewers visiting every two days to monitor compliance, and in the case of the electronic diaries, transcribe the entries. These arms also included both information on partakers (or non-household members sharing meals with the household) and meal-specific information on FAFH collected at the individual level. The sample size available for this analysis therefore was limited – only 120 households across 6 atolls – with 38 households having purchased at least one breakfast outside the household, 58 having purchased at least one lunch, and 30 having purchased at least one dinner. The second data set is a larger, nationally representative data set used to understand the impact on poverty statistics from incorporating a multiplier into the analysis. The 2019/2020 Household Income and Expenditure Survey (HIES) was conducted by EPPSO between July 2019 and June 2020 with a total sample size of 873 households. As this was the first HIES conduct since 2002, it was necessary to construct a new poverty line, which was done jointly by the World Bank and EPPSO, with the results being published in the Pacific Poverty Assessments Report (World Bank, forthcoming). The HIES survey includes 45 categories of food consumption, including 7 that would be classified as FAFH spending: breakfast away from home; lunch away from home; dinner away from home; snacks away from home; hot drinks away from home; bottled water; non-alcoholic beverages away from home; “sale of cooked dishes by restaurants for consumption off their premises” (take away meals); and a generic category for “catering services provided by restaurants, cafes, etc.” The poverty analysis of the RMI HIES data first compiled a consumption aggregate using food consumption, including pricing for home production and meals received as gifts; and non-food consumption, including health spending, education spending, imputed value of housing rent, and use value of durable goods. The aggregate is then adjusted for spatial price differences using a Törnqvist deflator. 4 The first step in constructing the cost-of-basic needs poverty line is to determine the caloric requirements for an adult to be a healthy and active participant in society. This threshold varies substantially across countries depending on the level of urbanization and prevalence of subsistence agriculture and other labor intensive activities, and is different than the FAO’s average minimum daily energy requirement. RMI set the threshold value at 2,100 calories per capita or 2,382 calories per adult equivalent per day. Once the minimum calories are defined, a food bundle that reflects the consumption patterns of those living near the poverty line must be defined. In this case, a basket of 45 food items based on the consumption of households in the second and third deciles based on real per adult equivalent consumption was used, covering more than 99 percent of food expenditure, which is then scaled to the threshold calorie requirements. This information is used to construct the food poverty line and the “Ravallion lower” method (Ravallion, 1998) is then used to compute the non-food component. 2 The sum of the food poverty line and the associated non- food component is the cost-of-basic-needs poverty line. Food away from home spending enters into the equations in the construction and scaling of the basket. For non-FAFH items, it is straightforward to compute the cost per calorie using the Pacific Nutrient Database (SPC, UOW and FAO, 2020), but those conversion tables do not cover FAFH meals. For the FAFH items, which comprised 22.7 percent of total spending and include the item with the second largest share of expenditure “lunch away from home” at 8.6 percent of total spending (Table 1), the standard approach following Subramanian and Deaton (1996) is to apply the same cost per calorie calculated from the items for which conversions are available to calculate the calories for the FAFH items, assuming the same types of food are eaten within and outside the household. This paper makes a similar assumption on FAFH meal composition, but examines integrating an additional multiplier to scale the spending down to allow for a margin for the food seller as well as cost recovery for the rent and salaries required to run a business. Once the caloric value of the bundle has been calculated, the amounts are aggregated and scaled to 2,382 calories per adult equivalent per day to calculate the food poverty line and the non-food component is calculated and added. 4. Analytical approach, results & implications for poverty measurement The multiplier is the ratio of the average cost of a single meal consumed outside the household to the average cost of the same type of meal consumed within the household. This analysis requires more detailed information on household consumption than is collected in a standard HIES survey. The diary arms of the consumption experiment include disaggregated information on the household spending on food; the number, cost, and type of meals (breakfast, lunch, dinner, hot drinks, soft drinks, and snacks) consumed outside the household by household members; and the number and type of meals (breakfast, lunch, and dinner) consumed within the household by non-household members aged 0-5, aged 6-15, and aged 16 and older. The diary also included an individual-level question on the location of each meal (breakfast, lunch, dinner, hot drinks, snacks, and soft drinks) with the answer choices (1) away from home (paid), (2) away from home (free), (3) at home, and (4) did not eat, which allows for the construction of a reliable estimate 2 The “Ravallion lower” methodology is a simple non-parametric approach to construct the non-food component of the poverty line. The calculations first estimate the non-food consumption of households whose total expenditures lie within 1 percent above or below the food poverty line. This calculation is then repeated for 2 percent above or below, 3 percent, etc., up to 10 percent. These 10 values are then averaged to obtain the final non-food poverty line, which is added to the food poverty line to obtain the final cost-of-basic-needs poverty line. 5 of the per adult equivalent number of meals. With this information, it is possible to calculate the number of per adult equivalent breakfasts, lunches, and dinners consumed within the household and the total household spending on food; the number of and spending on per adult equivalent breakfasts purchased outside the household; the number of and spending on per adult equivalent lunches purchased outside the household; and the number of and spending on per adult equivalent dinners purchased outside the household. To calculate the ratio of the average cost of a single meal consumed outside the household to the average cost of the same type of meal consumed within the household, it is possible to compare the ratio of the coefficients from two linear regressions. For consumption within the home, the model is: ( ) = + + (1) Where ( ) is the log total food spending in a given household, including beverages; is a vector of the number of per adult equivalent meals for breakfast, lunch, dinner, hot drinks, soft drinks, and snacks; is a vector of dummy variables for the atoll of the household and an indicator variable for which arm of the experiment the household was in; and is the error term. For consumption outside of the home, the model is: ( ) = + + (2) Where ( ) is the spending on meal i consumed outside the household, is the number of per adult equivalent meals of type i consumed by the household, and and are as above. Equation (2) is limited to the sample of household that have non-zero spending on a given meal. There is no intercept as in a standard Ordinary Least Squares model as it is not possible to have non-zero spending if food or meals were consumed. Comparing the ratio of the exponentiations of the from equation (1) with the corresponding meal equation (2) regression, the coefficient for breakfast was 1.33, lunch was 1.10, and dinner was 1.52. These results are reasonable for the cultural context where breakfast away from home is often donuts or other quick fried foods, which are inexpensive and simple to make and sold on roadsides or other outdoor markets; lunch has more variety and is the most common meal eaten away from home, but meals are often simple and competition in street-side and market dining leads to low margins; dinners are more complex meals and more likely to be eaten in formal restaurants, and therefore have the highest margins. In addition to the meal-specific coefficients, the analysis was repeated with the spending and number of meals aggregated to determine the impact of losing information, but saving time by using a simpler questionnaire design. In this case, the ratio of the exponentiations is 1.24. These results are lower than the only example of a multiplier having been used in region previously where a constant value of 1.5 was used in the analysis of the 2009/10 Papua New Guinea (PNG) HIES. The full regression results for both models are presented in Table 2. The analysis of this paper builds from the official poverty line to explore the impact of different choices on the use of multipliers on the poverty level. As there are no changes to the computation of the consumption aggregate, there are no changes in the ranking of households, only to the poverty line and therefore by extension, the number of households which fall below it. To understand the impact of the multiplier, three sets of poverty lines are calculated. First, the methodology replicates that which was used to calculate the official poverty line, using a multiplier of 1. In the second approach, the aggregated multiplier of 1.24 is used for the six types of FAFH, and in the third approach, the meal-specific multipliers of 1.33, 1.10, and 6 1.52 are used from breakfast, lunch, and dinner, respectively, and 1.24 is used for snacks and the other two types of undifferentiated FAFH spending. Despite the relatively high share of FAFH spending in the consumption aggregate, the application of the multiplier has little impact in the overall poverty numbers. Using the default multiplier of 1, the food poverty line is 1,333.30, with a non-food component of 606.73, and a total cost-of-basic-needs poverty line of 1,940.03, which corresponds to a headcount poverty rate of 8.6 percent. 3 Using the aggregated multiplier of 1.24, the food poverty line is 1,342.77, with a non-food component of 610.86, and a total cost-of-basic- needs poverty line of 1,953.63, which corresponds to a headcount poverty rate of 8.7 percent. Using the differentiated multipliers, the food poverty line is 1,337.40, with a non-food component of 609.07, and a total cost-of-basic-needs poverty line of 1,946.47, which corresponds to an identical headcount poverty rate of 8.7 percent to the aggregated multiplier. Finally, using the 1.5 multiplier previously used in PNG, the food component is 1,346.23, with a non-food component of 619.44, a total cost-of-basic-needs poverty line of 1,965.67, and a poverty headcount of 8.7 percent. Overall, the difference from the lowest line with a multiplier of 1 and the highest line with a multiplier of 1.5 is only 1.3 percent. 5. Conclusions and further research The main conclusion of the above analysis seems to be that the choice of multiplier, or indeed if one is used at all, seems to matter little for poverty analysis, and that survey practitioners are better served investing scarce questionnaire real estate into collecting more complete and more accurate FAFH data. There are, however, several important caveats to the conclusion that there is minimal impact of the multiplier. First, all of the above analysis relies on the assumption that food consumed away from home is similar in composition to food consumed within the household and that the per-calorie cost and impact of the multiplier are consistent across the distribution. In rural areas and in certain developing countries, this assumption might be reasonable, but it becomes more difficult to justify in urban areas or for middle income contexts. If the calories per meal outside the home are substantially higher, as would be in the case of highly processed or fried foods, then the cost per calorie would be lower than food within the home, even when accounting for commercial mark-ups. In that case, households would actually be consuming more calories for a given expenditure and the poverty rate could be lower if poor households had substantial spending on FAFH. Additionally, the per-calorie cost may vary across the distribution. This paper attempted to examine the impacts by quartile but unfortunately the sample size was too small to be disaggregated. Therefore, an imperative next step for this research would be a better understanding of FAFH meal composition and calorie counts by subpopulations. Second, this study represents only one data point and there was at least one substantial issue in that data collection with the recording of stocks that could impact the conclusions. As discussed in Sharp et al. (2019), there was substantial net destocking found in the diary data collection, which would lead in-home consumption to be over-estimated and the ratios to be underestimated. To ensure the robustness of the multiplier values, further analysis in other countries and contexts is needed. Finally, even assuming that the multiplier is accurately calculated, different characteristics of the spending profile could lead to great impact on the poverty measures. If a higher share of expenditure is spent on 3 This paper uses a different vintage of the data from the final RMI poverty numbers and therefore the figures do not match exactly to the official statistics. The differences, however, are immaterial to the conclusions of the paper. 7 FAFH, in particular if that spending is concentrated in populations around the poverty line, the impact would be greater. Similarly, if a country context has higher margins for FAFH, or higher multipliers overall, the impact would be greater, as would it be if the poverty line was in a steeper part of the cumulative distribution function, for which small changes would have bigger impacts on the headcount figure (Figure 1). Therefore, while the results presented here on the use of the multiplier are encouraging for the robustness of previous analysis done with Subramanian and Deaton (1996), further work is needed to reach a definitive conclusion. Figure 1. Cumulative distribution function and poverty line from RMI 2019/2020 HIES 1 .8 Cumulative Probability .6 .4 .2 0 0 5000 10000 15000 20000 total per adult equivalent expenditure (deflated) 8 References Borlizzi, A., Delgrossi, M. E., & Cafiero, C. (2017). National food security assessment through the analysis of food consumption data from Household Consumption and Expenditure Surveys: The case of Brazil’s Pesquisa de Orçamento Familiares 2008/09. Food Policy, 72, 20-26. Claro, R. M., Baraldi, L. G., Martins, A. P. B., Bandoni, D. H., & Levy, R. B. (2014). Trends in spending on eating away from home in Brazil, 2002-2003 to 2008-2009. Cadernos de saude publica, 30, 1418- 1426. de Brauw, A., & Herskowitz, S. (2021). Income variability, evolving diets, and elasticity estimation of demand for processed foods in Nigeria. American Journal of Agricultural Economics, 103(4), 1294-1313. Farfán, G., Genoni, M. E., & Vakis, R. (2017). You are what (and where) you eat: capturing food away from home in welfare measures. Food Policy, 72, 146-156. Farfán, M. G., McGee, K., Perng, J. T. T., & Vakis, R. (2019). Poverty Measurement in the Era of Food Away from Home: Testing Alternative Approaches in Vietnam. World Bank Policy Research Working Paper, (8692). Ravallion, M. (1994). Poverty Comparisons. Chur, Switzerland: Harwood Academic Press. Ravallion, M. (1998). Poverty lines in theory and practice (Vol. 133). World Bank Publications. Saksena, M. J., Okrent, A. M., Anekwe, T. D., Cho, C., Dicken, C., Effland, A., Elitzak, H., Guthrie, J., Hamrick, K, S., Hyman, J., Jo, Y., Lin, B.H., Mancino, L., McLaughlin, P.W., Rahkovsky, I., Ralston, K., Smith, T.A., Stewart, H., Todd, J., & Tuttle, C. (2018). America’s eating habits: food away from home (No. 281119). United States Department of Agriculture, Economic Research Service. Sharp, M., Buffiere, B., Himelein, K., & Gibson, J. (2019). Effects of Data Collection Methods on Estimated Household Consumption and Poverty, and on Survey Costs: Evidence from an Experiment in the Marshall Islands. Paper Presented at the IARIW-World Bank Conference on New Approaches to Defining and Measuring Poverty in a Growing World, Washington DC, November. Smith, L. C. (2015). The great Indian calorie debate: Explaining rising undernourishment during India’s rapid economic growth. Food Policy, 50, 53-67. Smith, L. C., Dupriez, O., & Troubat, N. (2014). Assessment of the reliability and relevance of the food data collected in national household consumption and expenditure surveys. International Household Survey Network. Pacific Community, University of Wollongong, and Food and Agriculture Organization. 2020. Pacific Nutrient Database. https://microdata.pacificdata.org/index.php/catalog/755 Subramanian, S., & Deaton, A. (1996). The demand for food and calories. Journal of political economy, 104(1), 133-162. World Bank. (forthcoming). Pacific Poverty Assessments Report 2021. World Bank Publications. You, J. (2014). Dietary change, nutrient transition and food security in fast-growing China. In Handbook on Food. Edward Elgar Publishing. 9 Table 1: Food basket items and shares for 2019/2020 RMI HIES Food item Share Food item (cont.) Share Rice in all forms 8.8% Bread 1.6% * Lunch Away from Home 8.6% * Breakfast Away from Home 1.5% Fresh, chilled or frozen fish 7.7% Edible oils (excludes cod or halibut liver oil) 1.5% Fresh, chilled or frozen meat of chicken 6.1% Ripe banana 1.5% Pasta products 4.9% * Dinner Away from Home 1.3% * Sale of cooked dishes by restaurants for consumption off 4.5% Green coconut 1.2% their premises Canned Cornbeef 3.5% Cooking banana 1.1% Maize, wheat, barley, oats, rye and other cereals in the form 3.5% Eggs 1.0% of grain, flour or meal Other preserved or processed fish and seafood and fish and 3.3% Dried, salted or smoked meat and edible offal, 0.7% seafood-based products, e.g. canned fish, caviar... e.g. sausages, salami, bacon, ham, pâté. Fresh, chilled or frozen fruit (excludes vegetables 3.3% Fresh, chilled or frozen meat of swine 0.6% cultivated for their fruit such as cucumbers and tomatoes) Other preserved or processed meat or meat-based products, 3.1% Orange 0.6% e.g. canned meat and pies (excludes lard and other animal fat) Sauces 3.0% Dairy milk 0.6% * Hot Drinks Away from Home 2.8% Biscuits 0.6% Coffee 2.7% Brown coconut 0.5% Breadfruit 2.6% Syrups and concentrates for the preparation of 0.5% beverages * Snacks Away from Home 2.4% Fresh, chilled or frozen seafood, e.g. crustaceans, 0.4% molluscs and other shellfish, sea snails. Tinned Mackerel 2.0% Cereal preparations, e.g. cornflakes, oatflakes and 0.4% other cereal products, e.g. tapioca, sago and other starches Fruit juices 2.0% Cream 0.4% Other Tinned Meat 1.9% Tea 0.4% Other bakery products, e.g. quiches, pizzas, pies (excluding 1.7% Other food products, e.g. homogenized baby food 0.3% meat pies, fish pies and sweet corn) * Catering services provided by restaurants, cafés, etc. in 1.6% Salt 0.0% places providing recreational, cultural, sporting or entertainment services Sugar, unrefined or refined, powdered, crystallized or in 1.6% Mineral or spring waters; all drinking water sold 0.0% lumps in containers Soft drinks 1.6% Note: * indicates food away from home items. Source: Pacific Poverty Assessment, Republic of the Marshall Islands chapter 10 Table 2: Full regression results from in-home and FAFH analysis in-home FAFH FAFH FAFH in-home FAFH coef/se coef/se coef/se coef/se coef/se coef/se Number of adult equivalent meals 0.015*** 0.234*** (0.001) (0.045) Number of per adult equivalent: breakfasts -0.053*** 0.231*** (0.020) (0.038) lunches 0.026 0.124*** (0.022) (0.023) dinners 0.074*** 0.495*** (0.008) (0.099) hot drinks 0.012 (0.026) snacks 0.008 (0.032) Number of soft drinks 0.014 (0.009) Atoll (reference: Majuro) Ailinglaplap 1.820*** 0.925** 1.857*** 0.111 (0.271) (0.426) (0.160) (0.239) Namdrick 2.179*** 0.508*** 0.763*** 2.126*** 0.046 (0.150) (0.034) (0.125) (0.164) (0.240) Jaluit 1.930*** 0.611*** 1.192*** 0.727*** 2.132*** 0.812*** (0.199) (0.126) (0.044) (0.187) (0.145) (0.200) Kwajalein 2.072*** 0.954*** 1.955*** -0.077 2.541*** 2.145** (0.165) (0.177) (0.304) (0.538) (0.254) (1.080) Wotje 2.048*** -0.356 0.527** 1.865*** 0.200 (0.487) (0.312) (0.248) (0.147) (0.234) Experimental arm (reference: highly monitored CAPI diary) Highly monitored PAPI diary 1.318*** 0.564* 0.891*** 0.735** 1.140*** 0.147 (0.287) (0.331) (0.243) (0.338) (0.297) (0.470) Number of observations 120 38 58 29 120 120 note: *** p<0.01, ** p<0.05, * p<0.1 11