Avaliação de impacto de uma política pública em educação: análise do programa de intervenção pedagógica / alfabetização no tempo certo no período 2008/2012

Abstract

This study evaluates an education public policy in the state of Minas Gerais in the period 2008-2012, called "Programa de Intervenção Pedagógica – PIP”. We used the databases of SIMAVE and administrative records from the State Department of Education of Minas Gerais - SEE/MG to assess the impact of that public policy in raising the level of proficiency of students in the 3th and 5th year of elementary school, measured by the external evaluation conducted at the state level. We used the method of differences and differences and the method of propensity score. The results show that the schools in the 3rd year of primary education that received the educational intervention (state schools of Minas Gerais) had a mean treatment effect in the intervention of 23,32 points in the grade to 0,35 standard deviations with a confidence level of 95%, using the method of the first nearest neighbor weighted by the calculated propensity score. The results obtained by the method of differences in differences weighted by different ways of pairing on propensity score were very similar to each other. This indicates the robustness of the results, which was expected due to the relatively large size of sample, which should thus converge. The impact achieved by the method of differences in differences weighted by pairing the kernel method was 23,44 points in the grade to 0,35 standard deviations and the nearest neighbor with replacement was 26,24 points in the score to 0,40 standard deviations. On the other hand, when using the Radius method the impact was 26,22 points in the grade to 0,40 standard deviations. The schools in the 5th year of primary education that received the intervention had a mean effect in the intervention of 0,6 points in Portuguese proficiency to 0,03 standard deviations using the first nearest neighbor method weighted by the calculated propensity score, not significant at a 95% confidence level. The impact achieved by the method of xii differences in differences weighted by the weights generated by pairing the kernel method was 1,4 points in the Portuguese Language proficiency grade to 0.06 standard deviations, not significant at a 95% confidence level. The impact achieved by the method of differences in differences weighted by the weights generated by the nearest neighbor method with replacement was 1,3 points in Portuguese proficiency to 0,06 standard deviations, not significant at a 95% confidence level. The impact achieved by the method of differences in differences weighted by the weights generated by pairing using "Radius" or "calibration" by imposing a distance of 0,001 was 1,8 points in Portuguese proficiency to 0,08 standard deviations, significant at a 90% confidence level.