YOUTH LABOR MARKETS IN THE U.S: Discussion and Conclusions 2

However, the IV estimate of the effect of early job stability should be biased upward more than the OLS estimate only if the positive correlation between the regressor (S) and the match quality component of the error term in equation (1) is exacerbated. This would require, for example, that individuals with high predicted S based on high early unemployment rates are in good matches to a greater extent than are those with high observed S (which may be influenced by a number of factors not limited to early labor market conditions). But if the observed variation in S also reflects self-selection of those with good matches into stable early jobs, it is difficult to see why the bias would not, on net, be lessened-although perhaps still present-in the IV estimates.

Therefore, other explanations of the IV results-maintaining for now the assumption that the instruments are valid-are considered. What is required, in particular, is a source of negative correlation between early job stability and adult wages. Such a negative correlation could arise if the unobserved component ц of the wage equation is primarily an omitted fixed individual effect-independent of the job match-that is positively correlated with adult wages but negatively correlated with early job stability. One reason the latter correlation might arise is if the returns to job search rise faster with this unobservable than do the costs of job search. The indirect opportunity costs of search are presumably higher for higher-wage individuals, although these opportunity costs may be quite low if search occurs while employed. However, the returns to job shopping may be considerably higher for higher-wage individuals, if the returns are characterized as roughly proportional to the current wage. Such higher returns would generate a negative correlation between the error term and S in equation (1), rather than a positive correlation, as the higher-wage individuals engage in relatively more job shopping. In this case the instrumental variables estimate of the effect of early job stability-in which variation in this stability is exogenously driven by labor market conditions, and not endogenously driven by unobserved components of the individual’s productivity-would be more positive than the OLS estimate, because the latter is biased downward.