In a recent article in this Review (Kuhn, 1987), Peter Kuhn seeks to assess the relative importance of statistical and other types of evidence in determining women's perceived level of discrimination. The author's point of departure is what he characterizes as a surprising relationship. He finds a robust, but statistically insignificant, negative correlation between the conventional measure of salary discrimination (statistical in Kuhn's terminology) facing a female worker and the probability that the worker will report discrimination on a confidential survey. In other words, workers facing higher levels of measured salary discrimination are less likely to perceive discrimination. Kuhn explains this finding by hypothesizing the existence of some other evidence (4nonstatistical evidence), which, in addition to the salary differential, leads women to recognize and report sex discrimination in their employment. If the quantity of nonstatistical evidence is low, according to Kuhn, women may not report discrimination even if the statistically measured salary differential is large. Kuhn assumes that the quantity of nonstatistical evidence, which is observed by the worker, but not the researcher, is a function of the worker's observed human capital characteristics. Finally, Kuhn evaluates the effect of the nonstatistical evidence using the worker's observed characteristics as a proxy for the unobserved nonstatistical evidence. According to Kuhn's findings, the unobserved variable is a more important determinant of the worker's decision to report discrimination than is the level of statistically measured wage discrimination. Kuhn uses these results to question the importance of recent Supreme Court decisions allowing statistical measures of discrimination to be admitted into evidence in civil sex discrimination suits. As he concludes that such evidence is not the primary factor influencing workers' reporting behavior, allowing these measures to be presented in court should not lead to a large increase in litigation. In this comment, we argue that Kuhn's result, that workers facing a high level of measured discrimination are less likely to report the discrimination, may be explained by incorporating the quality of information, employers' preferences, and the costs of discrimination into the model. We derive conditions under which the negative correlation between measured and reported salary discrimination may be seen as the result of a rational balancing by employers of the benefits and costs of discrimination. Briefly, employers are more likely to discriminate when employees have less accurate information and the probability of reporting, and hence detection, is low. We also offer statistical evidence from the academic labor market to illustrate this alternative hypothesis.
Stored only when both papers are in the corpus — 41 citations exist upstream.