This project has evolved as we�ve become concerned that the effect of deployment on the earnings of reservists involves a more complicated selection problem than we previously accounted for.� Although some reservists are not choosing their own treatment, the military's decisions about deployment may be correlated with the reservists' earnings, and some reservists appear to volunteer for deployment.� Furthermore, there are good reasons to expect at least observable heterogeneity in the effects of activation.� Reservists in certain military occupations, for example, may be activated more frequently than others, and those occupations might be more likely to receive special pays as compensation for dangerous or unpleasant work when they are activated.

Previous work, including our own, has assumed that any selection issues can be dealt with by using a simple difference-in-difference approach.� We�re now using propensity score estimates that relax the strong assumptions of this approach, while also considering the sensitivity of our estimates to unobserved variables.