Joint estimation of process and outcome in choice experiments involving attribute framing

Abstract

There is a growing interest and recognition that the study of discrete choice outcomes should take into account the process rules that are used to establish eligibility of each attribute. This applies to both revealed preference and stated choice data but is especially relevant in the context of choice experiments where the analyst traditionally assumes the relevancy of all attributes imposed on the respondent through a series of choice sets. This paper proposes a joint processoutcome model in which the choices made are conditioned on the rules adopted by each respondent in assessing the attributes packaged in the definition of each alternative. We set out a joint model for four attribute processing rules and three alternatives (including a reference alternative), and estimate two sets of panel-based mixed logit models – one set in which we ignore the attribute processing rules and one set in which we explicitly account for the rules. We integrate the inclusion/exclusion rules and ‘code’ the outcomes of various prospects (i.e., alternatives) as either gains or losses relative to a reference point. Using data from a commuter car trip study of unlabelled packages of times and cost attributes (including a toll), we identify willingness to pay distributions for travel time savings under the various process rules. The main finding is that failing to account for the process rules tends to result in statistically higher mean estimates of values of travel time savings.