Examining the impact of topcodes adjustment on studies of gender earnings inequality in the US: evidence from March CPS data

Topcoded survey data has been widely used for studies of gender earnings inequality. How robust are these studies to the choice of topcode adjustment methods? We examine this issue using the public-use March Current Population Survey data from 1979 to 2018 in the United States. In general, using more informative data and more sophisticated parametric procedure implies greater gender earnings inequality. Regression and mean decomposition analyses of gender earnings gap are quite robust to the choice of topcode correction methods. Nevertheless, topcode correction methods do matter significantly for studies of the relationship between gender earnings gap and some covariates (education) but not others (occupational characteristics). For researchers lacking access to internal data, the recommended topcode correction approach is SW/PC method. This method combines the benefits of using public data with replacement values that replicate the less topcoded internal data distribution while integrating parametric correction for the remaining topcoding bias.