Estimando o impacto do Programa Bolsa Presença através de métodos de inferência causal: uma abordagem baseada em métodos de ponderação para intervenções fixas e tempo-dependentes, com análise de heterogeneidades em subgrupos

Abstract

Measuring the impact of public policies requires identifying the specific effect of an intervention on a given outcome, isolating it from external and confounding influences. This study proposes an evaluation of the Bolsa Presença program, an initiative by the government of Bahia aimed at reducing school dropout rates and improving the academic performance of students in the state public education system through the provision of financial support to families in vulnerable situations. To this end, statistical methodologies for causal inference are employed, capable of handling the complexity of time-dependent interventions. In particular, weighting methods based on propensity scores, such as IPW (Inverse Probability of Treatment Weighting) and OW (Overlap Weighting), are applied to adjust for differences between treated and untreated students over time. Considering that effects may vary across different population profiles, the study incorporates a causal subgroup analysis to identify heterogeneities in program impacts, addressing both baseline interventions and time-dependent interventions through the use of marginal structural models (MSMs). These models allow capturing not only effects that vary at different stages of the intervention but also variations across population profiles through subgroup causal analysis. In addition to the empirical analysis with real Bolsa Presença data, the study presents the theoretical foundation of the methodologies used and illustrates their application through simulation studies, reinforcing the robustness and applicability of these methods in complex public policy evaluation contexts. In the simulations, considering the baseline intervention, the methods demonstrated superior performance in scenarios with balanced treatment prevalence. When accounting for time-dependent interventions through MSMs, IPW stood out for greater accuracy in estimating overall average effects, while OW exhibited lower bias and mean squared error in subgroups. The application results indicated positive effects of the Bolsa Presença program on students’ academic performance trajectories and school retention, also revealing the presence of heterogeneity in impacts, suggesting that effects differ across the various categories of the analyzed subgroups.