Do labor statistics depend on how and to whom the questions are asked? Results from a survey experiment in Tanzania

Labor market statistics are critical for assessing and understanding economic development. However, widespread variation exists in how labor statistics are collected in household surveys. This paper analyzes the effects of alternative survey design on employment statistics by implementing a randomized survey experiment in Tanzania. Two features of the survey design are assessed - the level of detail of the employment questions and the type of respondent. It turns out that both features have relevant and statistically significant effects on employment statistics. Using a short labor module without screening questions induces many individuals to adopt a broad definition of employment, incorrectly including domestic duties. But after reclassifying those in domestic work as 'not working' in order to obtain the correct ILO classification, the short module turns out to generate lower female employment rates, higher working hours for both men and women who are employed, and lower rates of wage employment than the detailed module. Response by proxy rather than self-report has no effect on female labor statistics but yields substantially lower male employment rates, mostly due to underreporting of agricultural activity. The large impacts of proxy responses on male employment rates are attenuated when proxy informants are spouses and individuals with some schooling.