Three essays on applied microeconometrics with spatial effects

Abstract

This Thesis consists of three independent essays on applied microeconometrics. The first chapter applies theoretical and empirical tools of spatial econometrics to analyze the determinants of residential water demand function for the city of Fortaleza (Brazil).We estimated three econometric models, which have as explanatory variables the average/marginal price, the difference, income, number of male and female residents and the number of bathrooms, under different spatial specifications: the Spatial Error Model (SEM), the Spatial Autoregressive model (SAR), and finally, the Spatial Autoregressive Moving Average model (SARMA), which is the model that best fitted the data. Results suggest that not controlling for spatial effects is a key specification error, underestimating the effect of almost all variables in the model. Sometimes, these differences can be as high as 24.66 % and 13.32 % for price elasticity in the Average Price and the McFadden models, respectively. In the second chapter we estimated willingness to pay (WTP) for a first order stochastic reduction on the risk of robbery, for the city of Fortaleza (Brazil). Inspired by Cameron and DeShazo (2013), we develop a simple choice model that nests a process of contingent valuation (CV) among lotteries and estimate it by both parametric maximum likelihood and geographically weighted regression (GWR). For the global model (i.e., without spatial effects), we estimated an average WTP of R$ 23.35 per month/household, and an implicit value of a statistical robbery approximately equal to R$ 11,969 per crime avoided. For the local model (GWR), we implement a protocol to calculate a surface of WTP using Kriging techniques. The results suggests that although peripheries present lower willingness to pay, as long as we go inwards there is plenty of heterogeneity on its spatial distribution for risk reductions. In the third chapter we analyzed how undergraduate students’ academic performance is affected by theirs classmates, by means of a “discontinuity design”. With data from Ceará Federal University (UFC), we employed regression discontinuity design (RDD) to estimate the difference between entering in the first semester class or second semester class. Due to the great courses availability, we assign each course into one of four categories depending on its admitted students’ results at the entrance exam. Then, we proceed the estimation exercise using a multi-treatment effect model. Results show that students who were ranked just above the cutoff (first semester class) had an academic performance 2% smaller (-0.19) than students who were ranked just below the cutoff (second semester class). Moreover, we found non-linearities in this effect, as well as Sacerdote (2001) and Zimmerman (2003), with intervals between 0.5 to -0.18.