Returns in the Labor Market: A Nuanced View of Penalties at the Intersection of Race and Gender in the US

Abstract

There have been decades of research on wage gaps for groups based on their socially salient identities, such as race and gender, but little empirical investigation on the effects of holding multiple identities. Using the Current Population Survey, this study provides new evidence on intersectionality and the wage gap in the US. This article makes two important contributions. First, there is no single “gender” or “race” wage penalty. Second, the evidence suggests that holding multiple identities cannot readily be disaggregated in an additive fashion. Instead, in a comparison of Black and White workers across gender, this study documents that the penalties associated with the combination of two or more socially marginalized identities interact in multiplicative or quantitatively nuanced ways. Further, the findings demonstrate that the presence of an additional intersectional penalty for Black women persists across time.

HIGHLIGHTS

• When it comes to earnings, Black women face distinctive penalties for holding their race and gender identities simultaneously.

• The intersectional wage gap persists across time and during both tight and slack labor markets.

• The unexplained portion of the wage gap has contracted from 1980–2017; however, it remains large and significant.

• Intersectional analysis provides a useful framework to disentangle nuances in the labor market.

KEYWORDS:

• Intersectionality
• stratification economics
• discrimination
• wage gap
• race
• gender

JEL Codes:

• J71
• Z13
• J16

ACKNOWLEDGMENTS

This study was made possible with the generous support of the Nathan Cummings Foundation.

Notes

1 See Browne and Misra (2003) for a review of intersectionality in sociology and Rosette et al. (2018) for a review from an organizational behavior perspective.

2 We solely examine the impact of binary identities associated with race and gender in the US. However, the same technique is sufficiently general that is can be extended to the examination of a larger set (three or more) of socially salient identities.

3 Many aspects of skills acquisitions are associated with other forms of discrimination that exist external to labor market, ensuring that equality of opportunity between Black and White people remains a longstanding myth (Katznelson 2005; Chetty et al. 2014; Rothstein 2017), but economists typically have focused on the presence (or absence) of discrimination within the labor market. We do not discuss AFQT scores in this paper. For a discussion on how they relate to the literature on the racial wage gap, see Darity and Mason (1998).

4 For a discussion of employer discrimination prior to passage of the Civil Rights Act of 1964, see Darity and Mason (1998) and Hamilton (2000).

5 Arguably, researchers may want to omit occupation and industry from wage regressions that attempt to quantify discrimination (see Goldsmith, Hamilton, and Darity 2007). We do not want to control for occupation and industry because part of the lower pay for women and Black workers occurs through barriers to specific occupations in the first place. For example, a “glass ceiling” means it is harder for women to enter higher-paid managerial positions. Controlling for the glass ceiling (occupation) underestimates the true penalties women face.

6 We exclude the unemployed and workers out of the labor force, which does introduce selection bias unless interpretation is extrapolated for only those in the labor market. See Goldsmith, Hamilton, and Darity (2007) for analyses of racial and skin color wage disparities that demonstrate substantially larger wage penalties when the unemployed are incorporated.

7 While we use the 2017 CPS, the data are for last year’s earnings and labor market participation. We also test our main results against two prior six-year intervals, looking at 2005 (pre-recession) and 2011 (weak labor market), as well as the 1990 and 1980 CPS.

8 When deciding which covariates to include in regression models, there are often trade-offs between mitigating the potential for omitted variable bias in relation to endogeneity bias. For instance, we opt to include occupation and industry controls so as to reduce concerns that uncontrolled productivity-linked characteristics associates with occupation and industry sorting bias our coefficients. However, a reasonable concern is that the process of sorting into occupation and industry is stochastically related to wage generation, in which case, we introduce endogeneity. To address this, we also check if our results are robust to dropping industry and occupation controls.

9 The CPS only allows people to identify as “male” or “female.”

10 We did not successfully replicate Kim (2009). When looking at all workers, Kim found an intersectional penalty with no industry/occupation controls and when controlling for just occupation. However, when looking just at full-time workers, as we do, Kim did not find an intersectional penalty when controlling for industry or occupation. Kim does not control for industry and occupation in the same regression, as we do. In an attempt to replicate Kim’s work, we round the existence of an intersectional penalty for the full-time sample when controlling for industry or occupation. The authors could not obtain Kim’s code or data, however, and instead relied on the methods outlined in Kim (2009) and applied them to the 2002 CPS.

11 While there have been important shifts in industry and occupation across time, such as White women seeing wage increases due to their entry in better paying occupations (Misra and Murray-Close 2014), research on intersectionality continues to note this an area of great interest (Rosette et al. 2018).

12 Following Binder and Bond (2019), we elect not to implement any top-coding adjustments on the CPS data for our primary analysis; however, we test our results after excluding wage outliers, trimming workers from the analysis who earn below $2.50 an hour or above $175 an hour.

Additional information

Funding

This work was supported by Nathan Cummings Foundation.

Notes on contributors

Mark Paul

Mark Paul is Assistant Professor of Economics and Environmental Studies at New College of Florida.

Khaing Zaw

Khaing Zaw was Research Associate at the Samuel DuBois Cook Center on Social Equity at Duke University. She is now a member of the statistical research team at Facebook

William Darity

William Darity Jr. is the Samuel DuBois Cook Professor of Public Policy, African and African-American Studies, and Economics and the Director of the Samuel DuBois Cook Center on Social Equity at Duke University.