Respondent Biases in Household Surveys

Abstract

Respondent biases introduce measurement error into household statistics. There are two components of respondent bias in household surveys: asymmetric information between proxy respondent and the individual on whom they report and aggregation bias when a proxy respondent reports on a household-level outcome across multiple individuals. We estimate the effects of respondent biases in a randomized control trial in Burkina Faso. Survey treatments vary who reports for the household including a self-report, the household head and a randomly chosen adult proxy. We find no effects of respondent type on total landholdings reported for the household, but statistically significant effects of area cultivated by random proxy reports relative to self-reported land data (11% of the standard deviation). Household heads over-report production of cereals, but under-report labor relative to self-reported labor. Female respondents report lower levels of fertilizer and higher frequencies of crops including legumes and vegetables that women traditionally produce relative to male respondents. Across all production variables, respondent biases are not solely attributable to asymmetric information. Aggregation bias is an important component of proxy misreporting for household production variables which can be addressed through survey design choices. These causal estimates of respondent biases underscore the importance of proxy reporting protocols in surveys and improved reporting of proxy descriptive statistics in empirical work.