The Effect Of Income On Demand For Food In Poor Countries: Are Our Food Consumption Databases Giving Us Reliable Estimates?

Abstract

Cross-section estimates of income elasticities for food staples in the aggregate based on quantity information from household food expenditure surveys are often in the 0.3-0.6 range. It is shown that differences in per capita calorie intakes across income groups implied by these income elasticities are grossly inconsistent with observed differences in bodyweights. Moreover, time series data frequently indicate that national per capita consumption of food staples in the aggregate remains constant even as per capita gross national product rises substantially. Cross-section income elasticity estimates are biased upwards due to the failure of food expenditure surveys to undertake an accurate accounting of food transfers from high to low income groups, biases which are not generated by survey techniques which record food intakes directly.

Generated Summary

This research article explores the reliability of food expenditure surveys in estimating demand relationships for individual foods, particularly in the context of developing countries. The study focuses on the analysis of income elasticities for food staples, investigating the potential biases inherent in data collected through food expenditure surveys compared to 24-hour recall surveys. The research approach involves comparing calorie demand patterns derived from food expenditure and 24-hour recall surveys conducted in Kenya and the Philippines. The methodology includes examining the relationship between calorie intakes and bodyweights to assess the relative accuracy of the survey data. The scope encompasses an examination of income elasticities and their implications for nutritional assessment, emphasizing the importance of accurate data in policy analysis related to nutrition.

Key Findings & Statistics

• Cross-section estimates of income elasticities for food staples in the aggregate based on quantity information from household food expenditure surveys are often in the 0.3-0.6 range.
• In Table VII-2, an average foodgrain income elasticity of 0.5 is reported for rural populations as a whole for 1963-64.
• For four subgroups within the lowest rural expenditure group the foodgrain elasticity ranged from 0.7 to 1.2.
• The calorie–income elasticities for the disaggregate expenditure data set using food quantities was approximately a factor of three times as large as that generated by the 24-hour recall data (see Table 2).
• Had only aggregate information on expenditures been available for cereals from the expenditure survey, this factor would have been larger still.
• In Table 1, C2 and C4 increase more rapidly with income as compared with C1 and C3, respectively, because of the inclusion of food fed to guests and hired workers, which increases as a proportion of total calories as income increases.
• The Kenyan estimate are based on an unbalanced panel of 1,501 observations from 496 households. The Philippine estimates are based on a balanced panel sample of 1,792 observations from 448 households.
• The Philippine data were collected over four rounds at four-month intervals from July, 1984 through July, 1985 in Bukidnon province, for 448 households.
• If the (quantity) food staple (in the aggregate) income elasticity is estimated to be (say) 0.30, then the calorie-income elasticity might be in the range of (say) 0.35-0.40.
• Note in Table 7 that most rice and wheat income elasticity estimates exceed 0.3.
• Very roughly, Table 8 indicates that between 1970 and 1990 calorie consumption rose between one and three percent for every 10 percent increase in income in these countries.
• The estimated per capita calorie intake for Indonesia of 2605 for 1988-90 is a figure applying to children as well as adults. To convert this per capita figure to an adult equivalent requires division, conservatively, by a ratio of 0.8, giving 3256.
• The rural percentage increases with income are much higher than urban percentage increases.

Other Important Findings

• Differences in per capita calorie intakes across income groups implied by these income elasticities are grossly inconsistent with observed differences in bodyweights.
• Time series data frequently indicate that national per capita consumption of food staples in the aggregate remains constant even as per capita gross national product rises substantially.
• Cross-section income elasticity estimates are biased upwards due to the failure of food expenditure surveys to undertake an accurate accounting of food transfers from high to low income groups, biases which are not generated by survey techniques which record food intakes directly.
• The study indicates that income often correlates with errors in the measurement of ‘leakages’ (the difference between a household’s food purchases and food intakes) in food expenditure surveys, leading to upwardly biased estimates.
• The results from the Kenyan and Philippine data suggest that the food expenditure information leads to gross overestimates of the income elasticities for food staples.
• The study indicates a general tendency for food expenditure surveys to generate much higher calorie-income elasticities than 24-hour recall surveys, leading to elasticities that are inconsistent with observed bodyweight gains across income groups.
• Income-group-specific estimates of calorie availability from expenditure surveys of national populations in six Asian countries are presented, which generally show implausibly high increases in calorie consumption as income increases.
• The study suggests that the ultimate solutions would seem to lie in a careful restructuring of how food intakes are accounted for in the survey process itself.
• The analysis will show that percentage differences in calorie intakes across income groups should be somewhat less than percentage increases in bodyweights, with the implication that percentage differences in bodyweights across income groups constitute an upper bound to estimates of percentage differences in calorie intakes.
• The study argues that the food expenditure survey data overstate the effect of income on calorie consumption for two primary reasons: First, households which randomly over-estimate (under-estimate) food expenditures by construction over-estimate (under-estimate) calorie consumption and by construction over-estimate (under-estimate) total expenditures; and second, there is a severe under-estimation of meals served to non-household members (guests, hired laborers) by high-income groups, and an under-estimation of food eaten outside the household by low-income groups.
• The study finds that the calorie-income elasticities generated using the 24-hour recall data are biased upwards.

Limitations Noted in the Document

• The proposed resolution of the bias problem does not directly involve either a theoretical or an econometric ‘fix’.
• The study acknowledges that the Kenyan and Philippine data sets are somewhat unique in that both food expenditure and 24-hour recall surveys were conducted for the same households.
• The data sets provide some clues. For example, the problem seems to center on measurement of food staple consumption. For the Philippine data set, the largest discrepancies occur for households that hire the most labor in the season in which hiring of labor is heaviest.
• The study emphasizes that there are systematic errors in many food expenditure surveys, but precisely what the underlying problem or problems are (which may vary by cultural and economic setting) are not clear.
• It is difficult to prescribe specific guidelines for amending existing food expenditure data sets for policy analysis in the short run.

Conclusion

The central argument of this research is that food expenditure surveys, a common tool in development economics for understanding food demand, may provide unreliable estimates due to inherent biases. The study uses data from Kenya and the Philippines to highlight discrepancies between food expenditure surveys and 24-hour recall surveys. These discrepancies, particularly concerning income elasticities for food staples, raise concerns about the accuracy of policy recommendations based on such data. The study underscores the importance of recognizing these biases, especially when analyzing nutrition-related issues and food security. A key observation is that the calorie intake data suggests that food staple consumption remains virtually constant as income increases, while the calorie availability data indicate that food staple consumption increases substantially with income. The research implies that using food expenditure information leads to gross overestimates of the income elasticities for food staples. The analysis suggests that the best solution lies in a careful restructuring of how food intakes are accounted for in the survey process itself, possibly shifting to a food intake accounting system or by asking very detailed questions in food expenditure surveys about food fed to non-household members and food eaten away from the household. The study suggests that food recalls are much more difficult and expensive to administer and process than food expenditure surveys and the need for highly motivated and well-trained enumerators. The research concludes that food expenditure data sets that exhibit strong discrepancies between percentage increases in calorie availability and bodyweights should not be used for nutrition research and they are of questionable value for research on poverty. The article suggests that researchers and policymakers should be aware of these potential biases in food expenditure data to make better informed decisions in the context of food consumption and nutrition analysis.