Felipe González – Identifiability of discrete choice models considering heterogeneous heuristics

  • Date:

Felipe González

Identifiability of discrete choice models considering heterogeneous heuristics

Wednesday, 28th June

Abstract:

Random utility maximization -RUM- (McFadden, 1973) is established as the most popular theory of discrete choice, having as the most iconic models the multinomial logit, hierarchical logit, and mixed logit models. Because these models are based on an additive measure of utility, they are compensatory across attributes and can represent a wide variety of behaviours. However, for the same reason, they cannot represent adequately some specific phenomena such as the decoy effect (Guevara and Fukushi, 2016). To address this among other issues, behavioural theories have been developed such as prospect theory (Kahneman and Tversky, 1979), regret theory (Zeelenbeg and Pieters, 2007), and satisficing theory (Simon, 1955). These theories have been materialized in the random regret minimization –RRM– (Chorus, 2008), elimination by aspects –EBA– (Tversky, 1972), and the stochastic satisficing –SS– models (Gonzalez-Valdes and Ortúzar, 2017), among others.

To exploit the advantages of both RUM and non-RUM heuristics, models that incorporate multiple heuristics have been used (Balbontin et al., 2017; Hess et al., 2012). However, combined models of this kind have been found frequently to lead to undistinguishable effects (unidentifiability), which is analysed in this presentation. First, this phenomenon is analysed theoretically to understand how behavioural differences lead to model identifiability. Then, RUM is compared with three choice heuristics (RRM, EBA, and SS) to explain its implications in practice by using a synthetic population together with a real choice set. Finally, these models are analysed to identify the choice mechanism that each individual used using mixed multiple heuristic models. The results of this show that RUM together with EBA leads to an identifiable model, RUM with RRM does not, whilst RUM with satisficing leads to a weakly identified model.

About Felipe: Felipe Gonzalez is an Industrial Engineer from Pontificia Universidad Católica de Chile. He is now a PhD student from that university working with emeritus professor Juan de Dios Ortúzar. He works on discrete choice models analysing new modelling formulations and has developed a model applying the psychological theory of satisficing. Felipe arrived as a visiting postgraduate research student at UCL in September 2017, to work with Benjamin Heydecker to untangle mathematical issues behind models with multiple heuristics.  Some of these results are presented in this seminar.