Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
Recent developments in discrete choice modelling have enabled the specification of models that can accomodate inter-alternative correlation and inter-agent taste heterogeneity. Nevertheless, to date, these two phenomena tend to have been addressed independently in the litterature despite the fact that such independent treatements are in principle vulnerable to serious misspecifications errors, resulting in confounding between these two effects. However, the empirical evidence for such confounding is still rather limited. The aim of this paper is therefore to systematically explore the nature, magnitude and consequences of the specification errors that can arise through the inappropriate or incomplete specification of inter-alternative correlation and inter-agent taste heterogeneity. We report the results of a comprehensive set of numerical experiments using quasi-simulated data in which we test a wide range of mis-specified models with respect to alternative data generating processes which include both independent and joint inter-alternative correlation and inter-agent heterogeneity. The results of this analysis indicate that there is substantial scope for confounding - inter-alternative correlation can be mis-indentified as inter-agent taste heterogeneity and vice versa - and that when such confounding occurs it leads to serious bias in parameter estimates and elasticities. We recommend that, whenever possible, models should be specified to accomodate the possibility of both phenomena simultaneously.
Michel Bierlaire, Timothy Michael Hillel, Janody Pougala
Mario Paolone, Vladimir Sovljanski