On the Extent, Correlates, and Consequences of Reporting Bias in Survey Wages

We study the extent, correlates, and consequences of reporting bias in survey wages using German linked survey-administrative data (SOEP-CMI-ADIAB). Survey wages differ systematically from administrative records: mean survey wages are 7% lower, and discrepancies follow a mean-reverting pattern. Individual characteristics explain little of the reporting bias, whereas firm context explains most of the variation. Since neither data source alone is sufficient, we construct a hybrid wage that combines their respective strengths. Measurement choice matters, but in ways that depend on how administrative top-coding is handled across the full wage distribution: when wages are outcomes, censoring administrative wages at the assessment limit understates returns to education by 4-11% and the gender wage gap by up to 23%, while imputation reverses the bias for returns to education. When wages are regressors, wage-satisfaction gradients are 9-28% steeper with survey than with administrative wages below the assessment limit, a pattern inconsistent with classical attenuation bias and pointing to non-classical, context-dependent misreporting. We provide guidance for choosing between administrative, survey, and hybrid wages depending on the application, with lessons that extend to any setting where self-reported wages are collected alongside top-coded administrative records.