Abstract:
|
Survey errors can severely affect survey accuracy and the validity of population statistics derived from them. For example, up to six out of ten cash welfare recipients are missing in the most important household surveys. However, the nature and causes of these errors remain a puzzle. This makes it difficult for data producers to reduce the extent of errors and for data users to assess the validity of survey data and address the resulting inaccuracy. This paper studies different causes of survey error using government transfers as a case study. We use high quality validation data from three major U.S. surveys. We find that survey design and post-processing as well as misreporting by respondents affect survey errors systematically. In terms of survey design, imputation for missing data can be particularly problematic as they induce substantial error at the individual level. Our results on respondent behavior confirm several theories of misreporting, e.g. that errors are related to salience of the issue, respondent's cooperation, forward and backward telescoping, event recall, and stigma of social programs.
|