Hypothetical bias in Stated Preference Experiments: Is it a Problem? And if so, how do we deal with it?

Abstract

The extent to which Stated Preference (SP) experiments suffer from hypothetical bias continues to be a controversial topic in the SP literature. This thesis provides further evidence in this debate by examining the existence of hypothetical bias in a transport-related SP experiment. Data for this thesis were sourced from a University of Sydney study exploring the effect of variable rate charging on motorist behaviour. The sample included 148 Sydney motorists who were recruited to take part in a 10-week GPS driving field study (Revealed Preference / RP data). In addition, participants were also required to complete an SP survey. The SP survey consisted of a Contingent Valuation (CV) and Choice Experiment (CE) task designed to mimic the RP decision context in order to capture what participants indicated they would do as opposed to what participants actually did in reaction to the charging regime. Hypothetical bias was established by examining important differences between what people said they would do in the SP experiment and what they actually did in the RP field study. The current state of practice for measuring hypothetical bias in the literature is to compare aggregate differences in model outcomes using SP and RP data sources. Aggregate analysis is limited in its scope and does not allow for the calculation of the prevalence of hypothetical bias (i.e., how many participants are affected by hypothetical bias) or give any insight into why hypothetical bias occurs (i.e., correlates of hypothetical bias). This research is uniquely structured to allow for individual categorisation of hypothetical bias by comparing SP and RP data from the same sample for the direct purpose of investigating the prevalence of hypothetical bias. Furthermore, the extent to which elicitation procedures (CV and CE), mitigation techniques (cheap talk and certainty scales), demographics (gender and age) and the level of experience influence hypothetical bias is also explored. The findings from this research show that the SP model estimates are prone to hypothetical bias and that the mitigation techniques have potential to compensate for this inherent bias.