Name: Sandra Elizabeth Flores
Institution: IME, USP, Brazil
E-mail: sefari@usp.br
Co-authors: Marcos O. Prates, Jorge L. Bazán e Heleno B. Bolfarine
Abstract:
Social and educational data sets have in common bounded response variables as percentages, proportions, or rates. This response variable includes for example rates of poverty or rates of achievement by cities, municipalities or provinces. Recently, new regression models have
been proposed to this type of data set, they are among others the beta, simplex or kumaraswamy models. However, commonly this kind of data set presents spatial dependence among its units. For instance, provinces are organized in states and to assume independence between provinces
inside the state remove relevant spatial relations in the group of neighboring provinces.
In this work we propose a spatial bounded distribution regression model, using the bounded distributions cited above, with an efficient estimation approach. The spatial relations in the units are included in the model using a random variable with several dependence structures under a
Bayesian inferential method.
Finally, a simulation study using model comparison criteria suggests some advantages of using the spatial dependence structure and a real data application illustrates the performance of the estimation method adopted and the applicability of the proposal model.
Keywords and phrases: Bayesian inference, Bounded distribution, Proportions, Spatial models.