Ensaios sobre economia aplicada: doações eleitorais, compras públicas, análise de políticas afirmativas e reprovação no ensino superior

Abstract

This thesis encompasses three unrelated essays in applied microeconomics. The first one assesses the impact of electoral donations on possible favoritism in public procure- ment. We use longitudinal data from companies and service providers for the municipal administration of Paraíba during the period from 2004 to 2016, whose estimates of impact on contract values were carried out by the differences in differences estimator, with control for the specific heterogeneity of companies and service providers with subsamples to correct self-selection bias. The main results of the research validate the hypothesis that political campaign funding by private agents generates a return for donors of elected candidates, on average, of 42% in the contracted values, where this return rate is higher for the companies than for the service providers. In turn, the second essay evaluates the effects of an unreserved affirmative action in higher education on dropout and academic performance educational. For this, we used in- formation from students who were admitted were admitted the Federal University of Paraíba (UFPB), in the years 2010 and 2011. The adopted methodology consisted of two steps: (i) first, we use three matching techniques, Propensity Score Matching (PSM), Mahalanobis Distance Matching (MDM) e Classification Tree Analysis (CTA), in order to evaluate the effects of the intervention on performance, captured by the relative Coeficiente de Rendimento Acadêmico (CRA), (ii) then, we use longitudinal data from the students, contemplating the years from 2011 to 2018, to estimate models of Cox proportional risk duration, weighted by the PSM, in order to evaluate the effect of the student being a quota holder on the probability of survival in the UFPB. The results indicate that the existence of the quota system reduced the performance level of students, regardless of the matching model employed, especially in the distribution that captures the best relative CRA averages. The estimation of the survival analysis models points out that the unlikely probability of non-quota students is lower than that of quota students, which allows us to conclude that the latter tend to persist more in higher education. Finally, the third essay proposes to identify the risk of failing higher education students using Machine Learning (ML) algorithms. Based on the administrative records of the UFPB and Plataforma Lattes, for the period 2010-2016 of the discipline of differential and integral calculus I, we verify that the models with the best performance of forecasting were Penalized Methods Lasso and Logistic Regression. From the modeling on training data (2010-2014), the results show that of the 1,532 observations that make up a new data set (2015 and 2016), the frequency of students with status (failed and approved) correctly predicted by Accuracy was 67 % in both models. In turn, 72.5 % of students were correctly predicted to fail (Sensitivity). These findings confirm that ML algorithms can be viable instruments to assist preventive pedagogical and managerial actions aimed at reducing the failure rates in higher education.