Relatively consistent evidence emerges from most of the estimations. For both men and women, most of the IV estimates for specifications that are not rejected by the data indicate a positive return to early job stability, in contrast to the OLS estimates that indicate essentially no return. Interestingly, these IV results are contrary to the hypotheses with which this study began. The presumption was that OLS estimates of the effects of early job stability were biased towards finding a positive effect of early job stability because of omitted job match quality that is positively associated with this stability; the IV results, as just noted, instead indicate that the OLS estimates of the effects of early job stability are biased downward. Thus, the evidence requires an alternative explanation. This section discusses a few possible alternative explanations of the results, and considers the consistency of the evidence with each of them.
One possibility is that the early unemployment rates are invalid instruments, owing to a correlation between the unobservable in the wage equation and the instruments.26 The positive relationship between early unemployment rates and early job stability apparent in the first-stage estimates could arise because hires that occur in slack labor markets tend to be better matches, as employers can be more selective in their hiring decisions. In this case, the instrument is positively correlated with the match-specific component of the error, ц, biasing the IV estimate of the effect of early job stability upward.