i (2)
Turning to the question of most interest, the estimated wage shortfalls for blacks and Hispanics in the OLS estimates of the levels specification are -.185 and -.041 respectively, with the latter insignificant. Instrumenting for the performance rating causes the differential for blacks to fall by more than half, to -.081, and that for Hispanics to fall to zero. Similar results occur for the log specifications, although the changes are a bit smaller. Like for men, these reductions in the wage shortfalls for black and Hispanic women are consistent with a substantial part of these shortfalls being attributable to statistical discrimination rather than taste discrimination. The Hausman tests indicate that the change for Hispanic women is not statistically significant, while for black women the p-values for the test of the null of pure taste discrimination are . 11 in the levels specification, and .14 in the log specification.

Overall, then, for black and Hispanic men and for black women there is some evidence that imperfect information is partly responsible for the lower starting wages they receive, compared with white workers with identical performance ratings. The point estimates of the shortfalls in starting wages experienced by black and Hispanic workers fall substantially once account is taken of statistical discrimination via an instrumental variables procedure, generally by more than half for blacks, and disappearing altogether for Hispanic men. The null hypothesis that these starting wage differentials are solely attributable to taste discrimination is rejected—for the specifications that fit the data—at the .07-. 14 significance level, evidence that is not overly strong, but which nonetheless suggests that the evidence against the null of pure taste discrimination should not be dismissed.