Statistical Discrimination Versus Pure Measurement Error in the Performance Rating

The evidence to this point is consistent with statistical discrimination being partly responsible for the lower starting wages of minority workers relative to equally-productive white workers. However, as discussed earlier, some caution is in order because the same contrast between the OLS and IV estimates could arise if there is no information problem—i.e., employers know a worker’s productivity when the worker is first hired—but the performance rating measures this known productivity with error. In this scenario, the results would have no implications with respect to labor market information, but would nonetheless tend to weaken evidence from studies claiming discrimination based on results in which race or sex differences in wages persist when error-ridden proxies for productivity are included in wage regressions.

There is some information that can be used to assess these alternative interpretations of the data. Specifically, the MCSUI data includes information on whether there was a probation period for the most recently-hired employee. Presumably, a probationary period is used when it is difficult to assess the worker’s productivity prior to hiring. Thus, using the same reasoning as discussed in Section II in relation to the test for different quality of labor market information about different demographic groups, the ratio aoLS/aIV can be compared for the subsample subject to a probationary period and the subsample not subject to one. If in fact labor market information plays a role in driving down the OLS estimate of a relative to the IV estimate, this ratio should be lower for those who are subject to a probationary period.