[Paper Review] Closing the U.S. gender wage gap requires understanding its heterogeneity

Philipp Bach, Victor Chernozhukov+1 more|arXiv (Cornell University)|Dec 11, 2018

Retirement, Disability, and Employment20 citations

TL;DR

This study uses high-dimensional double lasso regression on 2016 American Community Survey data to quantify heterogeneity in the U.S. gender wage gap across 642,229 full-time workers. It reveals that the wage gap varies significantly by marital status, motherhood, race, education, occupation, and industry, with married and motherhood status imposing the largest penalties—up to 12 percentage points higher than for never-married women—challenging the use of average gap estimates for policy design.

ABSTRACT

In 2016, the majority of full-time employed women in the U.S. earned significantly less than comparable men. The extent to which women were affected by gender inequality in earnings, however, depended greatly on socio-economic characteristics, such as marital status or educational attainment. In this paper, we analyzed data from the 2016 American Community Survey using a high-dimensional wage regression and applying double lasso to quantify heterogeneity in the gender wage gap. We found that the gap varied substantially across women and was driven primarily by marital status, having children at home, race, occupation, industry, and educational attainment. We recommend that policy makers use these insights to design policies that will reduce discrimination and unequal pay more effectively.

Motivation & Objective

• To move beyond average gender wage gap estimates by quantifying how the gap varies across individual and socio-economic characteristics.
• To identify the key drivers of heterogeneity in the gender wage gap using a large, representative U.S. dataset.
• To provide evidence-based insights for designing targeted policies that address specific subgroups most affected by wage inequality.
• To overcome limitations of traditional subgroup comparisons by using high-dimensional models that control for multiple confounding variables simultaneously.

Proposed method

• Applied high-dimensional wage regression using the 2016 American Community Survey (ACS) data on 1% of the U.S. population.
• Used double lasso (double selection) to select relevant predictors and interactions from a large set of socio-economic variables, reducing overfitting and bias.
• Modelled the gender wage gap as a function of two-way interactions between individual characteristics (e.g., marital status, children, race, education, occupation, industry, hours worked).
• Estimated the gap for each full-time woman in the sample and visualized heterogeneity using quantile plots and effect plots with 95% confidence bands.
• Stratified analysis by educational attainment (bachelor’s degree or higher vs. high school diploma or lower) to compare patterns across education levels.
• Reported joint statistical significance of predictors to assess robustness of heterogeneity effects across subgroups.

Experimental results

Research questions

• RQ1 How does the gender wage gap vary across different socio-economic characteristics such as marital status, motherhood, race, and education?
• RQ2 To what extent do occupational and industrial characteristics moderate the magnitude of the gender wage gap?
• RQ3 How do the effects of human capital variables (e.g., education, experience) on the wage gap differ across educational subgroups?
• RQ4 What is the joint impact of multiple intersecting identities (e.g., married, mother, non-White) on the gender wage gap?
• RQ5 How does the use of high-dimensional models with double lasso improve the identification of true heterogeneity compared to traditional subgroup comparisons?

Key findings

• The average gender wage gap was 17% for those with a high school diploma or lower and 14% for those with a bachelor’s degree or higher, but these averages mask substantial heterogeneity.
• Married women with a spouse present faced a gender wage gap that was 9 to 12 percentage points larger than for never-married women, all else equal.
• Women with at least one child aged 18 or younger at home experienced a 5-percentage-point larger wage gap than childless women, indicating a persistent 'motherhood penalty'.
• The wage gap was significantly smaller for non-White women compared to White women, with this racial differential remaining robust after controlling for education, experience, and other factors.
• Among those with a bachelor’s degree, the wage gap was significantly larger in 10 college majors, including natural sciences, social sciences, and business, compared to the baseline (education administration and teaching).
• In the high school degree subgroup, the wage gap increased with weekly working hours, while in the bachelor’s degree subgroup, the effect of hours worked on the gap was less pronounced, suggesting greater job flexibility for highly educated women.