Polarization of Tastes: Stated Preference Stability in Sequential Discrete Choices

Purpose: The aim of this study is to assess stated preference stability in long-format discrete choice experiments. As the number of choice situations increases, data reveal more precise information regarding preferences. However, there are many doubts concerning the incentives compatibility of long designs. Psychological effects such as respondents’ learning, fatigue and decreasing concentration in successive choice situations can result in biased estimators of parameters of utility functions. It is, not clear which group of successive choices reveal the most trustworthy information: the initial choices (undistorted but potentially not robust) or a later set (consciously formed preferences but potentially under conditions of fatigue). Design/Methodology/Approach: With the long-format (144 choice tasks) data concerning employment options, we estimated utility function parameters were estimated using MNL and MMNL models.To conduct inter- intra- respondent analysis we used imputed individual-level parameters of utility function. Findings: We show that preferences are formulated at the intra-respondent level according to a specific pattern and, at the same time, the preferences of single respondents show lower variance across choice tasks than across populations. An increase in the standard deviation of parameters across the sample does not necessarily mean an inconsistency of preferences in this type of study. This can result from polarization of preferences in the population with simultaneous intra-respondent preference consistency. Practical Implications: Long-format DCEs can reveal some of the behavioural mechanisms behind the decision-making process. We show that, using this kind of study, it is possible to observe preference formulation. In some specific cases obtaining accurate information, or even teaching respondents their preferences, can be of a substantial significance. Originality/Value: An increase in the standard deviation of parameters across the population does not necessarily mean inconsistency resulting from the 'negative' consequences of ordering effects, contrary to the findings of Swait and Adamowicz (2001). Instead, this can result from the polarization of preferences in the population alongside the intra-respondent consistency.