Modeling Spatial Effect on Travel Mode Choice Using a Synthetic Spatially Correlated Data Set

Abstract

Urban dynamics can be characterized more effectively by considering spatial aspects in studies. This paper, using a synthetic spatially correlated data set, aims to model the spatial effect on travel mode choice based on geostatistics precepts. A method was proposed based on three main steps. The first step consists of building synthetic spatially correlated data, using the intrinsic spatial dependence on travel demand data and mathematical principles of bilinear interpolation. The following two steps correspond to the modeling approach. The Exploratory Spatial Data Analysis stage aimed to attest the existence of spatial autocorrelation of the data set using two indicators: Moran and G-SIVAR (Global Spatial Indicator Based on Variogram). The Confirmatory Spatial Data Analysis stage proposed the calibration of two Binomial Logit models. The first model includes only the original database variables (nonspatial model). The second one is analogous to the original but added to spatial covariates obtained by geostatistical concepts (spatial model). A 15% increase in cross-validation hit rates is achieved when spatial variables are included. This paper presents three significant research contributions: (1) The methodological procedure to model spatial effect on travel mode choice; (2) The proposal of spatial covariates based on geostatistical assumptions; and (3) The suggestion of a simple procedure to propose a simulation of a spatially correlated database.