International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Making choice set formation practicable through direct elicitation of availability and choice
Habtamu Tilahun Kassahun, Joffre Swait

Last modified: 28 March 2017


Making choice set formation practicable through direct elicitation of availability and choice

Habtamu Tilahun Kassahun and Joffre Swait

Institute for Choice, University of South Australia, Sydney, Australia

Email: ;

Extended abstract

Misspecification of choice sets is a known major issue in choice model development. Both ignoring and misspecifying choice sets lead to biased parameter and welfare estimates. Modeling latent choice set formation is challenging due to the combinational complexity. The objective for this paper is therefore to propose a tractable generalized choice model (GCM) that on account for choice set formation practicable, even in situations with many alternatives. We circumvent this complexity by considering all alternatives with some probability, which is consistent with standard microeconomics theory. However, we show how to avoid enumeration of latent choice sets to eliminate complexity.

Data source and rationality of probabilistic choice set consideration

To illustrate our tractable GCM, data is obtained from a Retrospective Mode Choice Survey from 889 individuals living across the Sydney metropolitan area, Australia. We observe an individual’s work trips from home to work over a maximum period of 35 days. The total number of alternatives used in the analyses is 10 as shown in Figure 1, which depicts a mode choice transition matrix and complexity of mode choice switching over the observation period.  Similarly, mode consideration patterns change over time due to several factors, so that rigid mode consideration over time may not be a realistic representation. For example, a person who always considers and uses car must consider other alternatives if the car is not available due to breakdown or other reason. The representation of these factors in a choice model is the focus of this paper by advocating probabilistic consideration as a means of incorporating choice set formation   in discrete choice models of transport behavior (this is, of course, extensible to other domains).


Modeling approach and results

The GCM is based on the framework of earlier research that introduces utility constraints in discrete choice models. Different from these papers which focus on introducing attribute constraints, our approach focuses on the joint specification of probabilistic alternative consideration and choice as a structural relationship of choice of alternative and availability constraint. These include a joint unobserved factor correlation between choice of alternative and availability, and introduction of probabilistic availability constraints on the utility of alternatives; and explicit modeling of each alternatives’ consideration and availability interdependence in the choice set. Moreover, our approach evaluates sources of information jointly that lead to the inclusion and the exclusion of a particular alternative in a choice set and eventually leads to choice decision.  Thus, our approach is theoretically more sound and provides a wider perspective for policy implication and recommendation. It can be used for both stated and revealed preference applications. Our empirical findings support our proposed extensions, which is shown superior in terms of prediction and model fit.


Figure 1: Mode choice transition matrix pattern and complexity of mode switch over a month in 2015.


Conference registration is required in order to view papers.