CID Speaker Series: Thomas Crossley

Abstract: Survey data on incomes remain a key data source for measuring living standards and inequality but are known to suffer from reporting errors. Linked administrative data have been used to assess measurement error, but even when available such data often cover only specific components of income (for example, earnings) and/or only provide individual (not family or household) level data. We implement and test a design for identifying reporting errors directly in the survey data collection process. The key feature is an Editable Summary Screen that allows individuals and couples to review previous answers and implied totals, and then to confirm or correct. We first present a description of the resulting corrections. The corrections imply, for example, that period errors and problems with joint receipt within couples are fairly prevalent, but that misplaced decimals and source misclassification errors are nonexistent. There are corrections in a range of income sources, with revisions to benefit reports the most common, and revision to self-employment income the largest in mean absolute value. Downward revisions are more common than upward revisions, but the latter are larger in absolute value. We then consider how the confirmations and corrections can be incorporated in a formal econometric model to deliver improved inferences about the income distribution.