Counterfactual distributions of wages via quantile regression with endogeneity

Counterfactual decompositions allow the researcher to analyze the changes in wage distributions by discriminating between the effect of changes in the population characteristics and the effect of changes in returns to these characteristics. In this paper, counterfactual distributions are derived by recovering the conditional distribution via a set of quantile regressions, and correcting for the endogeneity of schooling decisions using a control function approach. Our proposal enables us to isolate the effect on the wage distribution of changes in both the conditional and unconditional distribution of schooling and changes in the distribution of unobserved ability. This methodology is used to analyze the sources of the changes in wage distribution that took place in the United States between 1983 and 1993, using proximity to college for different parental background as instruments. Our results show that the change in the distribution of ability had a negative effect on wages at the low quantiles, which almost compensates the positive effect of the change in the schooling distribution over this period. It is also found that the impact on wages of changes in the conditional distribution of unobserved ability is larger than the impact of changes in the conditional distribution of distance to college.