Accounting for Distance-Based Correlations Among Alternatives in the Context of Spatial Choice Modelling Using High Resolution Mobility Data

Abstract

Accounting for similarity among alternatives is of paramount importance for having unbiased estimates and for ensuring behaviourally accurate substitution patterns when making demand forecasts. Capturing similarity in a spatial context is a challenging task and the literature has failed to provide a clear answer as to how the presence of similar alternatives would influence the demand for a specific destination, i.e. acting as complements or competitors. The basic approach of relying on Nested Logit models that discretise space into a number of disjoint nests containing alternatives of the same geographical area ignores the influence of alternatives belonging in other areas/nests. We argue that such an approach will lead to uncaptured correlations in a spatial context, since as according to Tobler’s First law of Geography "everything is related to everything else". On the other hand, relying on more complex error structures quickly leads to computational issues, while also relying on non-trivial analyst assumptions. In the present paper, we propose an alternative approach, where a Cross-Nested Logit (CNL) modelling framework with a flexible correlation structure is used, where space is treated as continuous and the allocation parameters are distance-based. The proposed structure is applied in the context of stand-alone destination choice models, as well as joint models of mode and destination choices of shopping trips. A smart-phone panel survey dataset with high resolution location traces from Leeds, UK, is used to benchmark the improvements of the proposed model against traditional ones. Results indicate that in addition to the improvements in model fit, the proposed CNL specification is able to uncover interesting findings about individual mobility behaviour which can be leveraged to make better planning decisions.

Keywords: distance-based correlation, mode-destination choice models, disaggregate shopping behaviour