Consistently inconsistent: The role of certainty, acceptability and scale in automobile choice

Abstract

The way in which respondents behave in stated preference experiments is of interest to many practitioners in the field of choice modelling. This paper draws together three increasingly prevalent concepts in the literature; the role of scale; choice certainty calibration as a method for reducing hypothetical bias; and the acceptability of alternatives as a method for better representing respondent choice behaviour. Using a scaled multinomial logit model to focus on the role of scale, it is found that the amount of idiosyncratic error in the context of automobile is significant. Choice task certainty is found to be a function of several respondent characteristics and can be used to decompose scale. In doing so, it is found that for choice tasks where there is less certainty about the choices made, the scale parameter is lower and hence these choices are more stochastic, particularly in the case of reduced alternatives. In comparing different approaches to incorporating certainty, it is found that parameter estimates differ substantially depending on what method is employed. The implications of a lack of a theoretical framework with respect to choice certainty are discussed, as are the implications of removing unacceptable alternatives from the modelling process.