Doing More with Less: Improved Targeting of Social Programs for Maximum Impact

Social programs are increasingly asked to do more with less, but how is this possible? In this paper we consider one such program, supported employment, which is designed to increase employment among adults with intellectual disabilities. We estimate a model which allows for heterogeneous benefits from participation which in turn is allowed to affect the individual's decision to participate. We find that the average treatment effects for the population exceed that of the treated group. The contribution of this paper is to develop alternative schemes for targeting program resources and to measure any gains that result. In our simulations, we find employment under the current program is 1.38%, and this could be increased to an upper bound of 17.1% by an omniscient program administrator who can perfectly target those who gain most. While we assume that program administrators know more about individual program participants than we do, we can consider an administrator who has only the information available to the econometrician. In this case, targeting gains based only on observable characteristics would lead to 12.4% employment. Surprisingly, a simple rule that only requires administrators to predict employment success when treated (based on observables) will achieve almost the same results.