Discrete choice modeling with anonymized data

Abstract

This paper presents an approach to estimate mode-choice models from spatially anonymized revealed preference travel survey data. We propose an algorithm to find a feasible sequence of activity locations for each individual that minimizes the maximum error of each trip’s Euclidean distance within the activity chain. The synthetic activity locations are then used to create unchosen alternatives within the choice set for each individual. This is followed by the mode-choice model estimation. We test our approach on three large-scale travel surveys conducted in Switzerland, Île-de-France, and São Paulo. We find that our methodological approach can reconstruct activity locations that accurately match trip Euclidean distances but with location errors that still provide location protection. The discrete mode-choice models estimated on the synthetic locations perform similarly, in terms of goodness of fit and prediction, to the ones obtained from the observed activity locations.