Measurement error in recall surveys and the relationship between household size and food demand

Variation in household survey design and implementation is used to obtain evidence of nonrandom measurement error in recall surveys of household expenditure. These surveys, which are used especially in developing countries, appear to have measurement errors in food expenditures and in food budget shares that are correlated with household size. These correlated errors may be part of the explanation for a puzzling pattern of falling food demand with rising household size in poorer countries.