Partial program evaluation with observational data: The effect of treatment on drug addiction

A partial identification methodology is proposed in which self-selection into treatment is motivated by the expectation that treatment is beneficial. Identification is partial because it is not possible to establish empirically that treatment is effective. However, it is possible to establish that treatment is ineffective or even harmful. The evaluation methodology combines control function estimation with differences-in-differences in which the treatment effect is qualitative. The methodology is illustrated using administrative data on drug addicts in Israel in which some addicts were treated and others were not. Controlling for observed heterogeneity, we find that there is no significant difference in the change in drug use frequency between the treated and the untreated. Therefore, although it was not possible to demonstrate that treatment is beneficial, it is possible to establish that it is not beneficial. This type of negative evaluation is most probably useful in many other contexts.