Whom to ask? Testing respondent effects in household surveys

The common practice in household questionnaires of surveying the most knowledgeable household member can lead to inaccurate data if they have limited information. Using survey experiments in Paraguay and Uganda, we investigate whether there are discrepancies in intra-household reporting on income and consumption when multiple household members are interviewed. We use data from 4,100 households where we randomly vary whether the survey is administered to one spouse only, both spouses together or both spouses separately. We do not find meaningful systematic differences in the mean or distribution of household income and consumption and conclude that the magnitude of respondent effects for these variables is unlikely to bias most empirical analyses. However, a within-household analysis reveals large, but mostly unsystematic, reporting discrepancies. Taken together, the results indicate that respondent selection may matter for obtaining accurate information for a given household, but not for aggregate analysis of households.