Model Validity Checks Wage and Selection Equation Estimates

specialties require the longest training periods. Longer training periods apparently deter newly minted, but older, MDs. 28 Though the malpractice coefficient reported in the specialty selection models is common across all specialty branches, the model allows changes in malpractice risk within a specialty to affect specialty selection probabilities as long as those changes are not common across specialties. However, the results imply that doubling mal- practice risk within a specialty has a near zero effect on all specialty selection proba- bilities. In some sense, this is not particularly surprising, since the existence of insurance markets against malpractice risk should blunt the effect of such risk on spe- cialty choice. This result should be of some policy interest since some have implicated increases in malpractice risk as a cause of specialist shortages. 29 My results indicate that if this conclusion is true, the mechanism cannot be that malpractice risk deters young doctors from entering high risk professions. Finally, though mothers’ and fathers’ education is sometimes statistically signifi- cant in predicting wages, the qualitative directions of these effects show no obvious pattern except that they are all small.

C. Model Validity Checks

In this section, I use the results to check two assertions that I have made. First, do the factors, v v and 1 2 , represent components of unobserved ability, as I argue in Section III? Second, is medical school debt strongly correlated with specialty choice, as I argue in Section V? If the factor-loading terms in the wage equations represent the returns to unob- served abilities, then if the factor loads have the same signs and magnitudes in two different specialties, those specialties must reward the same abilities. In Table 3, the pattern of factor-loading signs shows that the first ability is rewarded posi- tively in IM and negatively in the other specialties. The second unobserved abil- ity is positively rewarded in FP and IM and has a quantitatively small effect on wages in the other specialties. This grouping provides empirical support for the notion that a different set of unobserved abilities is required to be a generalist than is required to be a specialist, and supports the interpretation of the factors as unobserved abilities. Table 3 shows unequivocally that medical school debt is an important and statisti- cally significant predictor of specialty selection. Consider two doctors—one with 20,000 of medical school debt, one with 30,000 of debt. The doctor with the 20,000 debt is 14 percent and 9 percent less likely to FP and IM, 4 percent more likely to pick IM Subspecialties, and equally likely to pick Surgery and Radiology. The qualitative effect of debt groups together the generalist branches, and separately groups together the more specialized branches. The Journal of Human Resources 132 28. One reason that training has a small effect is that age at graduation from medical school picks up much of the effect that the years of training variable would normally have—recall the linear relationship between age at graduation, training, and experience in Equation 10. The training variable is identified by variations in residency length within my specialty categories. 29. For a typical example, see Galewitz 2003.

VIII. Counterfactual Lifetime Skill Premiums