Reducing gender bias in household consumption data: Implications for food fortification policy

Household Consumption Expenditure Survey (HCES) data are increasingly used to inform nutrition policy around the world, most prominently for food fortification programs. However, they risk providing incorrect and gender-biased estimates of dietary intakes. We use both 7-day HCES and 24-hour dietary recall (24HR) data on all members of 5604 households in rural Bangladesh to disentangle the two main sources of error: 1) mismeasurement of household consumption, and 2) intra-household allocation assumptions used to individualize household consumption. We show that, relative to 24HR, HCES overestimate household-level quantities and underestimate women’s share of household foods. Errors from modeling the potential benefits and risks of fortification depend on the food – better measurement is needed for foods consumed episodically (e.g. wheat flour or sugar) or in small quantities (e.g. salt and oil). Beyond mean bias, we find poor and heteroskedastic agreement between HCES and 24HR methods, which is more driven by mismeasurement of food quantities than the application of flawed assumptions about food allocation – at least in the Bangladeshi context. We demonstrate a novel generalizable method for improving HCES intake estimates by drawing on the advantages of both HCES and 24HR data. Using a small sample of 24HR data to generate context- and food-specific quantity and allocation corrections, we can almost eliminate mean bias. With further validation, we hope our proposed method can be used to ensure that HCES estimates account for locally-specific measurement error and gender norms, and that nutrition policy based on these data will be safer and more gender-sensitive.