Must-Read: : Sampling Bias in "Firming Up Inequality": "Firming Up Inequality [Song, Price, Guvenen, and Bloom (2015)]...
...estimates the extent to which increasing inequality in the distribution of earnings from labor is caused by rising within-rm vs. between-rm inequality. But its statistical sampling from the Social Security Master Earnings File (MEF) is biased in a way that reduces inequality in the sample relative to the population, articially limiting the scale of the phenomenon the paper purports to investigate. This note explains the two biases the authors introduce into the paper: rst, they draw one 1/16th random sample of the MEF, which has become increasingly skewed over the period they study. Second, they winsorize individual income at the top 0.001%, under-representing earnings of the highest earners. Both procedures pose a particular danger to studying highly unequal distributions. Simulation results of their sampling technique show that the two biases are substantial and relevant to empirical research on inequality using both administrative and survey data.
