Essays on Applied Econometrics

Abstract

This dissertation consists of three essays in applied econometrics that explore experimental, quasi-experimental, and observational methodologies to study political behavior, infrastructure-driven labor market change, and price elasticity of demand estimation. The first chapter examines the causal impact of spatial proximity on legislative behavior by exploiting the randomized allocation of offices in the Brazilian Chamber of Deputies. The analysis finds that legislators assigned to neighboring offices are significantly more likely to vote alike in contested decisions—an effect amplified when at least one legislator is a policy expert, such as a committee member. The results provide empirical support for cue-taking theories of legislative decision-making, suggesting that informal physical proximity enhances the influence of expertise, particularly in closely divided votes. The second chapter evaluates the labor market impacts of Brazil’s broadband expansion policy using a regression discontinuity design (RDD) on about 2,000 municipalities covered by microwave radio technology. The findings reveal heterogeneous effects: while the policy stimulates job creation among low-educated workers and in commerce-related sectors, it simultaneously reduces hours worked and wages, especially for highly educated individuals, skilled occupations, and women in services. These findings highlight the inherent trade-offs of digital infrastructure policies, wherein thesame technological improvements that foster inclusiveness and job growth can also precipitate labor market disruptions, possibly through automation and substitution effects. The third chapter applies a new approach to estimating price elasticity of demand by combining double machine learning (DML) with multimodal embeddings derived from product descriptions and images. This combination offers two key advantages. First, the embeddings capture rich, high-dimensional signals of product quality that are often unobserved but crucial in shaping both prices and consumer demand. By leveraging visual and textual features, the method provides a data-driven way to control for latent quality differences across products. Second, DML allows for flexible, machine-learning-based estimation of complex relationships—such as how price and demand depend on covariates—while still delivering valid causal estimates. Together, these tools offer a powerful solution to the endogeneity problem in demand estimation, substantially reducing bias from unobserved quality. Together, these essays demonstrate how causal inference can be empirically applied to diverse data environments to uncover the mechanisms shaping legislative peer effects, labor market outcomes, and consumer behavior.