The estimated coefficients of unemployment rates for each of the five post-schooling years vary in sign, with the results suggesting that unemployment rates in the early part of this period increase longest tenure attained, while unemployment rates in the fourth and fifth years reduce it, although the latter effect is smaller. As will become clear, the general finding throughout the tables is that high unemployment rates during the early years in the labor market increase early job stability, consistent with the predominant effect of slack youth labor markets being to deter search or encourage job stability for other reasons. Column (3) reports the corresponding IV estimates of the wage equation. Contrary to expectations of downward bias in the OLS estimate of the effect of longest tenure, the IV estimate of the effect of this variable rises (to .08), and is nearly statistically significant at the ten-percent level. review

The last few rows of the columns report results of specification tests. First, the F-statistic for the instruments in the first-stage regression is 11.4, indicating that small sample biases are not an issue (Bound, et al., 1995), consistent with the large change in the coefficient estimate upon instrumenting. Second, the Hausman test comes close to rejecting exogeneity of longest tenure attained in the wage equation.20 Third, the instruments pass Newey’s overidentification test, although this test may not be particularly meaningful here because one might argue that there is only one instrument-the early unemployment rate-which has been arbitrarily divided into five years. Finally, the instruments fail the overidentification test when the minimum unemployment rate on the current job is excluded from the wage equation, as discussed above in relation to Beaudry and DiNardo (1991).