On the basis of the baseline data collected for the evaluation of the Bolivian Social Investment Fund (SIF) this paper assesses (1) the benefit incidence of the SIF and (2) the quality of the evaluation design. We find that the benefits in education are most equally distributed over the population, the investments in health and sanitation favor those relatively well off. For the education component of the SIF, control groups of schools which will not receive benefits have been included in the survey. In one region theseschools where selected on the basis of matched comparison on the basis of observed characteristics, in the other region by means of randomization. We compare control and treatment groups and conclude there is a systematic bias in favor of treatment schools in the first region. We propose to use instrumental variables to control for the non-random selection. With the pre-intervention data we can test whether an instrument is valid. We find that among several candidates the number of NGOs (non governmental organizations)in the community is a valid instrument. Next, we investigate the possible loss of efficiency in the estimate of the impact due to the non experimental control group design.