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.
Introduction: Beyond the Average
Most public discourse on the gender wage gap centers on a single "average" figure—often cited as women earning approximately 80 cents for every dollar earned by men. While this metric highlights a broad systemic issue, it acts as a blunt instrument that obscures significant variation. The experience of gender inequality is not uniform; it is shaped by a complex interplay of socio-economic factors including education, marital status, and occupation. This paper addresses the "curse of dimensionality" inherent in studying these interactions by applying modern machine learning techniques to identify the specific drivers of the wage gap for different groups of women.
Methodology: Inference through Double LASSO
To analyze the heterogeneity of the wage gap without falling into the trap of over-complexity, we employ a high-dimensional wage regression combined with the Double LASSO (Least Absolute Shrinkage and Selection Operator) estimator.
Unlike standard LASSO, which is primarily used for prediction, Double LASSO is optimized for inference. It operates in two stages:
- Variable Selection for Wages: Identifying factors that predict earnings.
- Variable Selection for Gender: Identifying factors that predict gender.
This dual approach ensures that the final gender gap estimate is not biased by the accidental exclusion of important confounding variables. Our model includes thousands of two-way interactions (e.g., "Married" $\times$ "Finance Industry") to capture nuanced effects. We further refine these estimates using a "Post-LASSO" approach, re-estimating selected variables with Ordinary Least Squares (OLS) to eliminate shrinkage bias, and utilize a multiplier bootstrap for robust confidence intervals.
Data and Stratification
The study draws on the 2016 American Community Survey (ACS), utilizing a representative 1% sample of the U.S. population. The final dataset consists of 642,229 full-time, year-round employees aged 25 to 65. To better understand the role of human capital, we stratified the data into two primary subgroups:
- Individuals with at most a high school diploma.
- Individuals with a bachelor’s degree or higher.
Key Findings: The Spectrum of Inequality
Our analysis reveals that the traditional 14–17% wage gap estimates only apply to a small fraction of the population. Instead, the gender penalty exists on a broad spectrum.
The Education Buffer and Marital Penalty
Education serves as a significant mitigator of the wage gap. While over 90% of women with only a high school education face a statistically significant wage penalty, this figure drops to 40% for those with a bachelor's degree.
Marital status emerged as one of the most potent drivers of the gap. Married women (with a spouse present) face a penalty 9 to 12 percentage points larger than their never-married counterparts, even after controlling for other variables. Interestingly, for highly educated women, the "motherhood penalty" appears time-persistent; having a child aged 4 or younger correlates with a slightly smaller gap (by 2 percentage points) than having older children, suggesting that the penalty may compound as children age.
Quantifying the Gap
The distribution of the wage gap varies significantly by education level:
- High School Subgroup: Faces a median wage gap of at least 29%. The distribution is tight, indicating a severe and consistent penalty across this group.
- Bachelor's Subgroup: Faces a median wage gap of approximately 9%. The distribution is much broader, reflecting high variation driven by specific college majors and industries.
Discussion and Policy Implications
The findings suggest that "one-size-fits-all" policies are insufficient for addressing gender pay inequality. The largest penalties are concentrated in specific sectors, such as finance, insurance, and real estate. Conversely, a small elite (approximately 4%) of college-educated women actually earn significantly more than comparable men.
Future policy interventions should be granular and industry-specific. For instance, increasing temporal flexibility in high-penalty fields like finance could be more effective than broad mandates. While Double LASSO provides a robust framework for inference, further research is needed to disentangle causality from correlation and to account for unobservable characteristics like individual negotiation skills or firm-level culture.
Summary updated