💰 quantitative finance

Does the Gender Wage Gap Originate at Labor Market Entry? Evidence from South Korea

This paper demonstrates that South Korea's large gender wage gap does not originate at labor market entry, where the disparity is minimal among recent graduates, but instead widens significantly during the prime-age workforce due to post-entry factors and selection effects.

Original authors: Dongwoo Kim

Published 2026-08-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Dongwoo Kim

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Great Pay Puzzle: Where Does the Gap Start?

Imagine you are trying to figure out why two runners, who started a race with the exact same shoes and training, end up at very different speeds by the finish line. In the world of economics, this is the mystery of the gender wage gap: why do women, on average, earn less than men? For decades, economists have debated when this gap appears. Does it happen right at the starting line, when people first get hired? Or does it creep in slowly over time, like a slow leak in a tire, as careers progress? To solve this, researchers look at labor market entry (the moment a person gets their first job) and sample selection (the tricky idea that we only see the wages of people who actually got hired, not the ones who didn't). If the people who don't get hired are different in hidden ways, it can make the wage gap look bigger or smaller than it really is. Understanding when the gap starts is crucial because it tells us where to fix the problem: do we need to change how we hire, or do we need to change how we promote and support people after they are hired?

The South Korean Mystery: A Race with a Twist

South Korea is the perfect place to test this mystery because it has the largest gender pay gap in the entire OECD (a group of wealthy nations), sitting at a massive 29%. It's like a race where the finish line is miles apart for the two teams. You might guess that in such a place, the gap must be huge right from the starting line. But a new study by Dongwoo Kim from Simon Fraser University and Korea University asks a surprising question: Does this giant gap actually start when young people graduate and get their first jobs?

To find out, the author looked at data from recent college graduates in South Korea between 2008 and 2019. He used a special set of mathematical tools (like a high-tech magnifying glass) to separate the "real" wage gap from the "fake" gap caused by who gets hired. He also checked if the gap was different for people at the bottom of the pay scale versus the top.

Here is what the study found, and it's a bit of a plot twist:

1. The Starting Line is Surprisingly Fair
When the author looked at recent college graduates who were working full-time, the raw gap looked big (about 12%). But once he adjusted for things like age, school grades, and the type of job they got, the gap shrank dramatically. The "real" conditional gap was only 4.3%. Even more interesting, this small gap actually got smaller over time, dropping from 5.0% in the early years to just 3.0% by the end of the decade.

2. The "Who Gets Hired" Trap is a Red Herring
A common worry is that maybe women who don't get hired are the ones who would have earned the most, making the gap look smaller than it is. The author tested this using several different advanced methods (including machine learning and statistical bounds). The result? It didn't matter. Whether he corrected for who got hired or not, the number stayed almost exactly the same. The "selection" problem was negligible at the entry level.

3. The Gap is a "Glass Ceiling" at the Top
The study also looked at where the gap existed across the pay scale. It found that for the lowest-paid jobs (the bottom 10%), the gap had effectively vanished. But for the highest-paid jobs (the top 10%), the gap remained around 5%. It's like a glass ceiling that only exists for the people aiming for the very top of the ladder, even right at the start of their careers.

4. The Real Problem Happens After the Starting Line
This is the big reveal. If the gap is only 4.3% at the start, but the national average is 29%, where did the rest come from? The author then looked at older workers using a different dataset (KLIPS) that tracks people over their whole careers. He found that as people get older, the gap explodes. For workers aged 24 to 35, the corrected gap is about 8%, but for workers aged 24 to 55, it balloons to 21%.

The study concludes that the massive gender wage gap in South Korea is not generated at labor market entry. Instead, it is generated after people start working. The gap widens severalfold as workers age, driven by factors like career interruptions (often due to having children), different promotion rates, and a pay system that rewards long, unbroken tenure.

What This Means
The paper suggests that while fixing hiring practices is good, it won't solve the massive 29% gap on its own. The real battle is fought later in life. The study points out that South Korea has a unique "M-curve" for female employment, where women often leave the workforce during their prime child-rearing years and struggle to return. Because the gap grows so much after entry, the author argues that policies need to focus on retention and progression—making it easier for women to stay in the workforce, return after having children, and get promoted—rather than just focusing on getting them hired in the first place. The starting line is fair; the track just gets much harder for women as the race goes on.

Technical Summary: "Does the Gender Wage Gap Originate at Labor Market Entry? Evidence from South Korea"

1. Problem and Motivation
South Korea exhibits the largest gender pay gap in the OECD (29.3% in 2023), a disparity roughly 2.6 times the OECD average. While aggregate gaps conflate initial differentials with cumulative lifecycle effects (career interruptions, promotion differentials, and occupational sorting), it remains empirically difficult to determine when this disparity originates. This paper investigates whether Korea's extreme aggregate gap stems from labor market entry or emerges subsequently. The study uses South Korea as a "hard test" case: given the country's strong gender norms, substantial female employment gap, and sharp occupational segregation, if significant entry-level gaps or selection biases exist, they should be detectable here. Conversely, if entry-level gaps are negligible even in this setting, it suggests that the aggregate disparity is driven by post-entry mechanisms.

2. Data and Sample Construction
The primary analysis utilizes the Graduates Occupational Mobility Survey (GOMS), a nationally representative repeated cross-section of individuals graduating from postsecondary institutions (two-year colleges and four-year universities) in the preceding academic year. The study covers 12 waves (2008–2019), focusing on graduates aged 35 or younger to isolate entry-level conditions.

Sample: The final analysis sample comprises 202,650 observations, with 123,632 employed in full-time wage work.
Variables: The dependent variable is the log hourly wage. Key covariates include individual characteristics (age, GPA, marital status), educational background (institution type, major, region), and job characteristics (firm size, sector, industry, region).
Comparison Data: To trace the lifecycle evolution of the gap, the author utilizes the Korean Labor and Income Panel Study (KLIPS) (2008–2019), covering the prime-age workforce (ages 24–55).

3. Econometric Methodology
The central econometric challenge is sample selection: wages are only observed for the employed. If unobservables driving employment correlate with wages, Ordinary Least Squares (OLS) estimates conditional on employment are biased. The paper employs a battery of methods to address this without relying on strong exclusion restrictions, which are difficult to defend in the graduate labor market.

Semiparametric Selection Model (Kim and Lee, 2026): This is the primary estimator. It relaxes the joint normality and linear index assumptions of the classical Heckman model. Identification is achieved through the nonlinearity of the selection process (specifically, the interaction of covariates in the selection probability) rather than exclusion instruments. The model assumes the selection bias enters the wage equation only through the scalar selection probability $p_0(X)$ .
Alternative Estimators:
- OLS: Baseline estimates on the employed subsample.
- Heckman MLE: Parametric selection model assuming joint normality.
- Debiased Machine Learning (DML-RF): A locally robust estimator (Pan and Zhang, 2024) using random forests for the first stage and Neyman-orthogonalized moment conditions.
- Bounds Approaches: Age-conditional Lee (2009) bounds and Honoré-Hu (2020) bounds to provide identification without point estimates under weaker assumptions.
Decomposition: Oaxaca-Blinder decompositions are applied to the selection-corrected coefficients to separate the gap into endowments (observable characteristics) and residuals (unexplained factors and selection effects).

4. Key Results

Entry-Level Gap is Small and Selection-Correction is Negligible:
- The raw entry-level wage gap is approximately 12%. However, after controlling for demographics, education, and job characteristics, the conditional gap shrinks to 4.3% (pooled 2008–2019), declining from 5.0% (2008–2011) to 3.0% (2016–2019).
- Methodological Agreement: The semiparametric Kim-Lee estimator, the DML-RF estimator, and the Heckman model yield virtually identical results (differing by less than 0.005 log points). The selection correction adds almost nothing to the OLS estimate, indicating that differential selection into full-time employment does not materially bias entry-level gap estimates for recent graduates.
- Drivers of the Raw Gap: The reduction from 12% to 4.3% is driven primarily by demographic controls (specifically age, reflecting the 2.3-year age gap due to mandatory military service for men) and job sorting (firm size and sector).
Heterogeneity Across the Distribution:
- Quantile regression reveals a "glass ceiling" pattern at entry. The gap is statistically indistinguishable from zero at the 10th percentile but remains around 5% at the 90th percentile (2016–2019). This suggests the entry-level penalty is concentrated in high-paying positions.
Lifecycle Divergence (KLIPS Evidence):
- In contrast to the entry-level findings, the gap widens significantly over the lifecycle. In the prime-age workforce (ages 24–55), the selection-corrected gap is roughly 21% for all workers and 17% for college graduates.
- Selection Becomes First-Order: While selection bias is negligible at entry, it becomes a "first-order adjustment" in the prime-age workforce, particularly as the female employment "M-curve" (decline during child-rearing years) takes hold. The selection wedge increases from near zero at entry to approximately 0.10 log points in the prime-age sample.
- Distributional Shift: Unlike the entry-level glass ceiling, the prime-age gap is broad and "middle-heavy," largest around the median rather than the top.

5. Contributions and Significance
The paper makes three primary contributions:

Empirical Finding: It establishes that South Korea's massive aggregate gender wage gap does not originate at labor market entry. The conditional entry-level gap is small (approx. 4.3%) and robust to extensive selection corrections.
Methodological Application: It demonstrates the feasibility of applying semiparametric selection models (Kim and Lee, 2026) and debiased machine learning methods to repeated cross-sections of recent graduates, a setting where longitudinal wage histories (required for other correction methods like Blau et al., 2024) do not yet exist.
Policy Implications: By isolating the entry-level gap as small and selection-robust, the paper argues that policies focused solely on hiring or entry-stage equality address a limited margin of the problem. The evidence suggests that the bulk of Korea's gender disparity is generated after entry through mechanisms such as career interruptions (specifically the child penalty), the seniority-based wage system, and differential promotion. Consequently, policy attention should shift toward retention, parental leave design, affordable childcare, and reforms to seniority-based pay structures to mitigate the post-entry divergence.

The paper concludes that the small, corrected entry-level gap serves as the low point of a lifecycle profile that steepens with age, driven by post-entry institutional factors rather than initial labor market discrimination or selection bias at the point of entry.

The Great Pay Puzzle: Where Does the Gap Start?

The South Korean Mystery: A Race with a Twist

More like this