Non-trading, Lexicographic and Inconsistent Behaviour in SP Choice Data

Non-trading, Lexicographic and Inconsistent Behaviour in SP Choice Data


S Hess, ITS, University of Leeds, UK; J Rose, ITLS, University of Sydney, AU; J Polak, CTS, Imperial College London, UK


This paper looks at issues with non-trading, lexicographic and inconsistent behaviour in stated choice data


The vast majority of choice modelling applications reported in the literature are now estimated on Stated choice (SC) data, including but not limited to the field of transport research. Although concerns regarding the internal and external validity of SC data are routinely acknowledged, the practical advantages of using SC data, when compared to Revealed Preference data, are generally overwhelming. These advantages include the elimination of measurement error and of ambiguities regarding the composition of choice sets, the availability of multiple observations per individual and the ability to design-in optimally informative trade off situations. In recent years a considerable amount of work has been undertaken to develop SC design procedures that fully exploit these advantages.

However, fundamental concerns regarding the validity of SC data remain. These concerns focus principally on two inter-related issues; firstly, the extent to which respondents in SC exercises may use ad hoc heuristics to filter or pre-process the information presented to them prior to considering their choice, and second, the extent to which when making a choice, respondents do so in a manner that is consistent with assumptions underlying the models used to analyses the choice data (typically, assumptions of some version of rational behaviour). In this paper, we focus on three particular issues of this sort, which relate to how respondents trade between different alternatives and between different attributes. These issues are; the treatment of non-traders, the presence of apparently lexicographic choice processes and the incidence of apparently irrational behaviour.

Almost without exception, SC datasets include a proportion of respondents who always choose the same alternative, often, but not always, an alternative that corresponds closely to their current behaviour (something that arises especially in the context of labelled choice experiments). A related phenomenon occurs when choices appear to be entirely lexicographically determined by a sub-set of attributes, a limiting case of which is when a single attribute, such as cost, appears to dominate all others. The third issue we look at in this paper is respondents who appear to behave in a manner that is inconsistent with the axioms of rational choice behaviour upon which most analysis models are based. In the most simple context, this would refer to a respondent who, when faced twice with the same choice, does not behave in the same way.

To a large extent, the above issues are recognised in the existing literature but are largely ignored in actual applications. In this paper, we present results from four case studies using very different SC designs, and show the rather worrying extent of non-trading, lexicographic and inconsistent behaviour in rather standard SC surveys. We then illustrate the potential effects of such patterns of response on model results by showing how their presence can potentially lead to significant bias in important indicators such as the valuation of travel time savings.


Association for European Transport