Peer Effects and Relative Performance of Voucher Schools in Chile

The assessment of the advantages and disadvantages of vouchers has been hindered by the lack of sufficient empirical evidence. The Chilean education voucher system was established at a national scale and has data for more than 15 years. The empirical literature developed to evaluate the voucher system in Chile faced methodological and/or data limitations up until late 1999, since there was no individual data available, and papers used the school as a unit of study. Additionally, the studies lacked good information on the socioeconomic characteristics of the students. The most recent literature uses individual data and introduces the correction for selection bias, but do not take into account that some public schools receive additional resources from the government. In the first section of this paper we control for the amount of per capita funds received by the public schools from the government, and find that when public and private voucher schools receive similar per capita subsidies, the effect of treatment on the treated (where treatment is attendance to a private voucher school) is large in magnitude and statistically significant. Some fear that this result may be the consequence of sorting and peer effect, and not of the effectiveness of private voucher schools. To analyze the importance of peer effects on the previous results, in the second section we estimate new treatment parameters controlling for peer group characteristics. If the positive treatment effect estimated earlier were exclusively the result of the sorting process and peer effect, this new treatment parameter should be zero. This hypothesis is rejected. Even when we condition on peer group characteristics, we find a treatment parameter that is positive, large in magnitude and statistically significant, when public and private voucher schools receive similar per capita subsidies. Hence, papers that have asserted that positive treatment effects are due to the peer effect and/ or sorting are proved wrong