Closing the gender pay gap: can machine learning help

Abstract

The goal of this work project is to analyze US data on wage earnings, combining machine learning and econometric methods, to understand the factors contributing to the continued existence of the pay differential between men and women. The post-double LASSO method employed in this paper allows me to systematically select a large number of controls, including interactions and second-order polynomials. Since 2009, the total gender pay has declined by 6 percentage points, but the portion that can be explained has not declined significantly. In 2019, women earned 23 percent less than men and only 7 percentage points of that gap can be explained by differences between men and women in the relevant controls. Occupational segregation by gender accounts for the majority of the explained portion in both years, but is more important in 2019 than in 2009.