The evaluation of community-based interventions : group-randomization, limits and alternatives

The context of community-based interventions presents formidable problems for any evaluation analysis. Group-randomized studies do possess ideal properties in theory, but in practice, group-randomization might not be a feasible alternative at all or group-randomized studies might be contaminated. Thus, the decisive advantage of randomized controlled trials, that they and only they provide for a completely convincing identification strategy in the presence of observable and unobservable confounders, is lost. There are alternative strategies for the identification of treatment effects also in the case of unobservable confounders, however, although they specifically require unverifiable a priori information to be available. Moreover, when using non experimental data, one can often easily extend sample size at low cost, and thus estimate parameters very precisely; therefore, for any particular situation the relative attractiveness of experimental and non-experimental approaches should be explored.