Andrews, Martyn J. Schank, Thorsten Upward, Richard
Year of Publication:
IAB-Discussion Paper 3/2004
Methods for the analysis of linked employer-employee data are not yet available in standard econometrics packages. In this paper, we make the fixed-effects methods developed orginally by Abowd, Kramarz, Margolis and others more accessible, where possible, and show how they can be implemented in Stata. To illustrate these techniques, we give an example using German linked data. There is a caveat: when the number of plants is prohibitively large and the investigator wants to estimate the correlation between the worker and firm unobserved heterogeneities, the regression-based techniques discussed are not feasible. We also report an estimate of the correlation of zero. In this version of the paper, we replace our earlier Two-Step estimator by a Classical Minimum Distance estimator.