Detecting, Non-Transitive, Inconsistent Responses in Discrete Choice Experiments

Surveys focusing on choice behaviour, and in particular, Discrete Choice Experiments (DCEs) are widely used in studies across many disciplines, including transportation, marketing research, health economics, labour economics and environmental studies. Investigation of the 'rationality' of responses in choice experiments has received a fair bit of attention from researchers. Most of this research has focused on the identification of irrational behaviour as it relates to non-satiation or lexicographic behaviour. At the same time, irrational behaviour indicated by non-transitive choices (often referred to as inconsistent behaviour) has received less attention by researchers. Until now the identification of non-transitive inconsistent behaviour has concentrated on relatively simple choice experiments. This research aims to extend previous work by developing a method to identify non-transitive inconsistent behaviour in more complex experiments. In particular, a systematic test procedure to detect inconsistent behaviour is developed and applied to three DCEs. The consistency test is implying that each respondent has a given preference structure and that her/his choices should be consistent with this structure across their choices, and therefore satisfy the axiom of transitivity. As such, choices that are not consistent with an individual's observed preference structure are identified as inconsistent with his/her own choices. Our analysis shows that inconsistent choices commonly occur in DCEs with multiple tasks and attributes. Moreover, more inconsistent behaviour is detected in more complex experiments. Also, such behaviour has a significant impact on the valuation of respondent sensitivity to attributes in models estimated from DCE data. Finally, excluding inconsistent responses results in significant improvements in models fit.