How much should we trust estimates of firm effects and worker sorting?
-Empirical Micro Seminar
PCPSE Room 100
A large number of studies use matched employer-employee data to estimate the wage model with worker and firm effects introduced by Abowd, Kramarz, and Margolis (1999, AKM). These studies tend to draw two conclusions. First, firms play an significant role in wage determination, with a typical finding that about 20 percent of the variance of wages is attributable to variation in firm premia. Second, the correlation between firm and worker effects tends to be positive but relatively small, indicating modest sorting of high-wage workers to high-wage firm. In this paper, we provide empirical evidence on the sensitivity of these conclusions to statistical bias issues. Using data from the US and several European countries, and making use of fixed-effects and random-effects econometric methods for bias reduction, we re-examine the contribution of workers, firms, and worker-firm sorting to wage inequality. We find that bias matters. In contrast with AKM estimates, corrected estimates suggest that firm premia explain around 10 percent of log-wage dispersion, and that sorting is substantial.