The PhD Program is pleased to invite you to attend the Dissertation Defense of Fernanda M. Araujo Maciel. Those interested in receiving a copy of the dissertation paper in advance are invited to contact Fernanda directly.  As a result of the coronavirus spring semester PhD Defenses will be conducted remotely via Zoom.  Please see the remote connection information below.

DISSERTATION DEFENSE ANNOUNCEMENT

FOR THE DEGREE OF PHD IN BUSINESS

PhD Candidate:  Fernanda M. Araujo Maciel

Title:  The Analytics of Vulnerable Populations in Brazil

Date & Time:  Monday, April 13^th, 2020, 2:00pm – 4:00pm

Location:  Remote connection via Zoom

Remote Connection:  https://bentley.zoom.us/j/5085085678

Remote Connection Instructions:  https://goo.gl/6RxFTd

Dissertation Committee Chair:  Dominique Haughton, Professor of Mathematical Sciences, Bentley University

Committee Members:  Dhaval M. Dave, Stanton Research Professor of Economics, Bentley University; Jennifer L. Priestley, Professor of Statistics and Data Science, Kennesaw State University.

ABSTRACT

Conditional Cash Transfer (CCT) programs became a popular measure to alleviate poverty in Latin American countries. The Brazilian CCT program, called Bolsa Família, is the largest social welfare program in the country, covering a quarter of all Brazilian households. The objective of the program is to reduce poverty and malnutrition, while providing low-income families access to public services, such as health, education, and social assistance. Since the program is based on conditions of maintaining health and schooling for children, my dissertation comprises two studies that examine the impact of Bolsa Família on educational outcomes and unhealthy behaviors of its participants. The third chapter investigates direct and indirect associations between negative body image perception, depression, and risk behaviors among adolescents in Brazil. Taken together, these three studies contribute to the literature on the public policy of vulnerable populations, endogeneity, machine learning applications, and causality estimation.

In the first paper, I evaluate the impact of Brazil's Bolsa Família program on three schooling indicators. In particular, I investigate what happens to the probability of dropping out of school, grade progression, and grade repetition when they participate in the Bolsa Família cash-transfer program. Prior literature has explored the impact of this program on educational outcomes, but these studies use methods that do not address endogeneity. To properly examine the effect of this program, this paper presents the Instrumental Variables method that controls for reverse causality and omitted variables. Consistent with the literature, my results show that the rate of dropping out of school decreases if the household participates in the program, but when addressing endogeneity, this effect is not significant. Surprisingly, in contrast with what is documented by the literature, I find that the rate of grade progression decreases among participants. Finally, I estimate that program participants increase the chance of grade repetition, which was not previously studied. These findings are important to understand the effectiveness of the program in maintaining children in school, contributing to the literature of public policy and causality estimation.

In the second paper, I investigate if participants of Bolsa Família are increasing their unhealthy consumption, measured by the expenses with ultra-processed foods, alcohol, and smoking products. For these analyses, I use the Propensity Score Matching (PSM) method. Different from the existing methodology, I incorporate a machine learning approach for better predictability of the propensity score. I present a comparison between Random Forest, Gradient Boosting, Support Vector Machines, Neural Networks, and Logistic Regression in propensity score estimation. This study is important to estimate if participants of the program are using the cash-transfer to purchase unhealthy goods, which could potentially make these people worse off in the long run. My results show that program participants purchase more food and increase expenses with snacks, such as cookies and out-of-home pastries, but they are not purchasing more unhealthy products than non-participants. This paper also contributes to the literature on machine learning models for econometrics estimation.

In the third paper, I evaluate direct and indirect associations between negative body image perception, depression, and risk behaviors among adolescents in Brazil. Literature has shown associations between these factors among adolescents, however, few studies analyze the Brazilian population. In this paper, I estimate the effects using Directed Acyclic Graphs (DAG), a model that is based on a network of conditional independent nodes (Causal Markovian Condition), instead of theory. My results show similarities to the studies in the literature, validating DAG as a method of identifying directed links between variables. This model also finds associations not yet studied in the literature, shedding light on the vulnerability of Brazilian adolescents and their propensity to risk behaviors.