International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Is there a systematic relationship between random parameters and process heuristics?
Camila Balbontin, David A. Hensher, Andrew Collins

Last modified: 28 March 2017


The manner in which individual’s process information in arriving at a choice is being increasingly recognised as an important feature of discrete choice modelling. The popular assumption that all attributes are traded in a form specified as linear in the parameters and additive in the attributes (LPAA) is increasingly shown to be out of line with ways in which many individuals typically make choice-related decisions. There exist a plethora of other ways in which attributes are processed in choice settings (be they real or hypothetical settings) such as attribute non-attendance, extreme aversion, elimination by aspects, and value learning. As we delve deeper into the potential role of various process heuristics, it has become apparent that there may exist a relationship between the information that is associated with a non-systematic (i.e., random) representation of preference heterogeneity in the LPAA form and the information offered through one or more process heuristics. If there is a link, then it is useful to explore what this might be, and to see to what extent the interaction between the standard preference heterogeneity assumption within an LPAA setting might be related (confounded or conditioned) to process heterogeneity as represented by one or more process heuristics. This paper investigates this possibility with a focus on value learning that accounts for how individual preferences associated with each attribute might change from one choice scenario to the next in a choice experiment. We use a mode choice data set to explore the conditioning of the mean and standard deviation of the standard random parameter specification by the value learning heuristic. Of special interest is the extent to which this conditioning, it if exists, modifies the two key moments of willingness to pay distributions (i.e., the mean and standard deviation), including whether the standard deviation increases or decreases, as a way of establishing the extent to which there is redundant or missing preference heterogeneity under the traditional random parameter assumption.

Keywords: Multiple heuristics, process rules, fully compensatory choices, value learning, Mode choice data


Conference registration is required in order to view papers.