Nonparametric vs parametric binary choice models

In France, despite access to safe public drinking water, and in spite of its excessively high price compared to tap water, 42 % of the population still regularly drink bottled water. Using scanner data on French consumption combined with raw water quality and other environmental data, and using a nonparametric kernel estimator of the conditional PDF, we show that poor raw water quality is an important factor driving the decision not to drink tap water. The estimated effect is found to be stronger for low-income households. Significant direct impacts of socioeconomic and demographic households' characteristics, as well as the role of cultural/regional factors are revealed. The aim of this paper is threefold. First, we employ a fully nonparametric model of a conditional PDF comprised of a binary response (choice) variable and continuous and discrete explanatory variables. Second, we address the issue of the performance of this nonparametric estimator relative to the parametric Probit specification which is dominant in applied settings and evaluate these estimators in a variety of ways. Third, we provide a detailed discussion of the results focusing on environmental insights provided by the two estimators, emphasizing how particular patterns detected using the nonparametric estimator are masked by the parametric specification.