Race and Gender Differences in Wages: The Role of Occupational Sorting at the Point of Hire

Abstract

Previous work suggests that occupational sorting and differences in starting salary play a large role in race and gender wage gaps. This study uses unique data from the human resources department of a financial company to examine the role of occupational sorting in race and gender differences in initial salary offers. While this company exhibits large race and gender differences in salary offers, controlling for occupational differences accounts for all of the race effects and reduces the gender effect to substantive insignificance. These findings underscore the importance of occupational sorting mechanisms in creating race and gender differences in wages.

ACKNOWLEDGMENTS

I would like to thank Trond Petersen and Roberto Fernandez for their generous comments and numerous insights. This article also benefited from the feedback of Henry Brady, Tonya Dewey, Neil Fligstein, Mike Hout, Lars Jarkko, Peter Kivisto, Julie Kmec, and several anonymous Sociological Quarterly reviewers. Previous versions of this article were presented at the 2004 RC28 summer meeting in Rio de Janeiro and the 2004 ASA annual meeting in San Francisco. Research for this article was conducted while supported by a National Science Foundation Graduate Research Fellowship.

NOTES

Notes

¹ While this archetypical process is the model for the company studied here, it is likely that there is variation in the hiring processes used in different branches and occupations, particularly in the number of interviews.

² While generalization is difficult in all social science, this is especially true of studies surrounding labor market inequalities, as the case study data necessary for fine-grained analyses do not lend themselves readily to generalization. For the sake of brevity, in reviewing the findings of other studies, I do not detail their limitations; however, it is important to note that all of these studies occur in specific empirical contexts and should not be viewed as context-free.

³ Examples of the 866 detailed occupation codes include portfolio monitor analyst II, mail clerk II, client relations rep I, HR help desk specialist, and telemarketing specialist I.

⁴ This occurs when an applicant successfully applies to more than one job opening. There were 120 applicants who received 2 job offers, and 27 applicants who received 3 job offers.

⁵ That is, given that the market salary rate is established before the race and gender of the successful applicant are known, differences in the market rates cannot be the result of the decision to pay a successful applicant more on the basis of their race or gender. However, although occupational sorting processes are the proximate cause of race and gender differences in the market salary rates, at the macro level, the process of setting market rates is likely responsive to valuative discrimination. Thus, while in this analysis I take the market salary rates as given (because from the perspective of applicants and hiring managers they are), it is important to remember that they are not set in a vacuum, and that at the macro level, they are likely responsive to the demographics of the people with that particular occupation.

⁶ Strictly speaking, the coefficients reported pertain to log unit differences; proportions can be obtained by exponentiating the coefficients and subtracting one. When coefficients are small in absolute value (between .1 and –.1), they approximate the relative change in the dependent variable.

⁷ For example, model 1 shows that black women receive salary offers that are 42.8 percent (exp [0.558] – 1) less than white men, while model 2 shows that, controlling for applicant source, they receive offers that are 39.3 percent (exp[–0.5] – 1) less than white men.

⁸ It is worth noting that the constant term in Model 4 is quite different from those in the other models. The constant should be interpreted in conjunction with the effect of log market salary rate to indicate that people on average do not earn their market rate salary. That the constant is negative while the effect of log market rate salary is greater than one reflects the fact that people receiving offers for jobs with higher market rates receive salary offers that are a higher percentage of their market rates.

⁹ The difference between Models 4 and 5 is because of the larger number of men and whites receiving offers for jobs that have larger market salary rates, as jobs with larger market salary rates tend to pay more as a percentage of the market rate. This is confirmed by the fact that results from a model (not reported) including fixed effects for the 80 pay groups yielded coefficients similar to the 866 occupation fixed effects. This suggests that there is something associated with being in a market salary rate group that is not captured by the group's market salary rate.

¹⁰ At the macro level, it is still possible that there is some kind of valuative discrimination at play. It is even possible that the firm's market salary rates are influenced by their knowledge of what kind of groups are likely to apply, rather than what other companies are paying. It seems more likely that the market and society as a whole devalues these occupations. An analysis of how it is that firms establish their market salary rates could be very informative about this process.

¹¹ It is unclear whether it is even possible to obtain both the employer-side data and detailed information about why potential applicants chose not to apply necessary to explore this research agenda. The study that comes closest is Fernandez and Sosa (2005), which matches firm hiring data to census data from the area surrounding the firm.