JOB MARKET PAPER
JOB MARKET PAPER
Skin Tone and Racial Inequality in the US: Evidence from the 20th Century
with Danny Onorato
Empirical work on racial inequality typically relies on a binary Black–white comparison, treating racial categories as uniform and masking meaningful variation in how individuals are treated within groups. Using newly collected WWII draft registration records containing information on skin tone for the near universe of men born between 1898–1929, linked to the 1940 Census, we document large intra-racial disparities by skin color. Black men with light skin earn about 37% more than their darker-skinned peers. This unconditional gap is economically significant—if all Black men earned what light-skinned Black men did, the Black–white earnings gap in 1940 would be roughly 18% smaller. We explore several mechanisms behind these disparities. Comparing brothers raised in the same household, family fixed effects eliminate essentially the entire education gap, yet darker-skinned brothers still earn about 10% less (9 log points), roughly a quarter of the unconditional gap in the sibling sample, pointing to a labor market penalty distinct from the categorical disadvantages tied to racial classification. We then consider the role of discriminatory institutions. While Jim Crow exposure depresses educational attainment for Black men overall, we find no differential impact by skin tone, consistent with the one-drop rule's uniform legal treatment. Finally, we follow men from 1940 to 1950 and find that among men in the same 1940 occupation, lighter-skinned men were more likely to move into skilled work. Exploiting WWII-driven variation in labor market tightness, we find that shocks that reduced Black–white inequality had no detectable differential effect by skin tone. Our results show that skin tone inequality reflects both inherited disadvantage and distinct discriminatory forces that sort Black men along color lines. These forces likely differ from those driving broader racial inequality.
WORKING PAPERS
The 1980s Recession and U.S. Labor Market Inequality
with Danny Onorato
Almost 60 percent of the sharp rise in 90–10 wage inequality since 1979 happened between 1980 and 1985, amidst the deep recessions of the early 1980s. This paper studies how this recession contributed to the abrupt transformation of the wage distribution over this period. We construct a new longitudinal employer-employee dataset for 5.3 million workers from Census administrative records and exploit variation in exposure to the recession across local labor markets to estimate the effects on workers’ labor market trajectories. We find that the recession persistently reduced earnings and employment, with the largest effects for workers at the bottom of the earnings distribution. Under a less severe recession counterfactual, we estimate that the rise in wage inequality between 1979 and 1988 would have been up to 20 percent smaller. We find no evidence that these effects reflect accelerated technological change or other lasting shifts in relative labor demand. Instead, we find that the recession persistently reduced unionization, and extrapolating our estimates suggests that the recession can account for 42 percent of the national decline in union density over the decade. Decomposing the recession’s impacts on aggregate inequality, we find that 35 to 40 percent is accounted for by its effects on unions, through both the reallocation of workers to non-union jobs and weaker union wage spillover effects. Our results suggest that the recession meaningfully contributed to the post-1980s rise in wage inequality and demonstrate how temporary aggregate shocks can have permanent impacts on inequality by reshaping wage-setting institutions.
From Callbacks to Careers: Do Audit Studies Predict Workforce Diversity?
with Danny Onorato, Adina D. Sterling, and Emmanuel Yimfor
Do callback gaps from audit studies translate into real employment differences? We match firms from Kline, Rose and Walters (2022)’s large-scale correspondence experiment to LinkedIn employment records. Firms discriminating more against Black applicants in callbacks employ fewer Black workers. The employment gap is three times what callback gaps alone would predict. This amplification is not explained by local labor supply: controlling for local Black population shares does not attenuate the relationship. Instead, Black workers at high-discrimination firms have smaller qualification gaps relative to White coworkers and better outside options when they leave, consistent with discrimination continuing at the interview and offer stages. This relationship concentrates among the top quartile of discriminatory firms, is stronger for employees hired during the audit period and at entry-level positions, and replicates with larger magnitudes when restricting to the exact occupations targeted by the audit study. Fewer than 3% of Black workers we identify from profile pictures have the distinctive names used in the experiment, yet callback gaps predict employment for the broader Black workforce. However, callback gaps do not predict post-hire outcomes: Black workers face promotion penalties across all firms, but these disparities are uncorrelated with callback gaps. These findings validate correspondence experiments as measures of hiring barriers, suggesting that targeting the most discriminatory firms could reduce hiring disparities.
PUBLICATIONS
Conviction, Incarceration, and Recidivism: Understanding the Revolving Door
with John Eric Humphries, Aurelie Ouss, Megan Stevenson, and Winnie van Dijk
Quarterly Journal of Economics (2025)
[publisher's link] [pdf] [NBER wp] [Non-technical (Cato)]
Noncarceral conviction is a common outcome of criminal court cases: for every individual incarcerated, there are approximately three who were recently convicted but not sentenced to prison or jail. We extend the binary-treatment judge IV framework to settings with multiple treatments and use it to study the consequences of noncarceral conviction. We outline assumptions under which widely-used 2SLS regressions recover margin-specific treatment effects, relate these assumptions to models of judge decision-making, and derive an expression that provides intuition about the direction and magnitude of asymptotic bias when a key assumption on judge decision-making is not met. We find that noncarceral conviction (relative to dismissal) leads to a large and long-lasting increase in recidivism for felony defendants in Virginia. In contrast, incarceration (relative to noncarceral conviction) leads to a short-run reduction in recidivism, consistent with incapacitation. Our empirical results suggest that noncarceral felony conviction is an important and overlooked driver of recidivism.
SELECTED WORKS IN PROGRESS