This study utilizes a unique data set containing detailed labor market information for individuals employed by the state of North Carolina over ten years. The inclusion of the state's promotion decisions and each worker's annual performance evaluations allow for an empirical test of the presence of statistical discrimination that has not been possible in previous studies. Statistical discrimination is said to exist if, in the absence of perfect information about a worker's productivity, an employer uses characteristics of that worker's (race or gender) group to substitute for individual-specific information missing from a signal of the worker's productivity. If the signal is "noisier" for blacks or women than it is for white men, equally productive workers may be paid differently. If, however, performance evaluation serves as a suitable proxy for individual productivity and if statistical discrimination exists at the point of hiring, then the significance of observable characteristics (race and gender) in explaining wage or promotion disparities should diminish with tenure.

Results suggest that blacks may be statistically discriminated against at the time of hire, but that men and women are not treated differently. This outcome elicits further investigation into how statistical discrimination affects other labor market outcomes which differ across groups, including wage disparities and the rate at which promotions are realized. Equations explaining the dynamic employment process approximated by the wage and salary grade, performance evaluation, promotion, and quit probabilities of North Carolina state employees are estimated jointly to allow for correlation in unobserved permanent and time-varying heterogeneity factors affecting each outcome.

When controlling for measures unique to these data, such as performance evaluation and promotion history, women and black men are not shown to be significantly more or less likely to be promoted than white men. After controlling for unobserved heterogeneity, endogeneity of relevant explanatory variables, and performance and promotion history, wage gaps effectively disappear. Consistent with the result that statistical discrimination may exist at the time of hire, tenure interacted with race and/or gender produces negligible or insignificant coefficients in all equations, suggesting that after controlling for performance, race and gender do not independently influence employment outcomes.