Locational error in the estimation of regional discrete choice models using distance as a regressor

Abstract

In many microeconometric studies distance from a relevant point of interest (such as a hospital) is often used as a predictor in a regression framework. Confidentiality rules, often, require to geo-mask spatial micro-data, reducing the quality of such relevant information and distorting inference on models’ parameters. This paper extends previous literature, extending the classical results on the measurement error in a linear regression model to the case of hospital choice, showing that in a discrete choice model the higher is the distortion produced by the geo-masking, the higher will be the downward bias in absolute value toward zero of the coefficient associated to the distance in the models. Monte Carlo simulations allow us to provide evidence of theoretical hypothesis. Results can be used by the data producers to choose the optimal value of the parameters of geo-masking preserving confidentiality, not destroying the statistical information.