We Might Both Be Wrong - Reconciliation of Survey and Administrative Earnings Measurements

Abstract

This paper investigates measurement error using a German linked survey administrative dataset. By contributing to a small branch of the literature that allows the possibility of erroneous earnings from both sides of a linked survey-administrative dataset, this study provides a replication of two previously proposed versions of a representation of different combinations of (correct and incorrect) survey and administrative earnings in a finite mixture model. Additionally, the paper investigates the robustness of the results with respect to the empirical choice of what tolerance for the differences between to earnings measures should be accepted to consider them close enough to represent the true earnings. While the estimated distributions of true earnings and error types are relatively robust to the choice of the accepted tolerances, substantial differences in the estimated latent class probabilities are elicited, which highlights the importance of robustness checks for the choice of a tolerance level with alternative values above and below the one chosen in any future study.