Accounting for both consideration set screening and attribute non-attendance when modelling stated choices

Standard choice models are based on the assumption that respondents examine all attributes and all alternatives across all choice tasks in the same fully compensatory manner. However, recent research has indicated that respondents often resort to a two-stage ‘consider-then-choose’ decision process if the choice task is considered too complicated (e.g., because they are not familiar with the context or because the number of alternatives and/or the number of attributes is large): In a first (screening-)stage respondents may identify a consideration-set of alternatives that needs further evaluation (Hauser, 2014) or eliminate attributes they find irrelevant (Hensher et al., 2005). In a second stage they make a choice using a standard compensatory model on the outcome of the screening-stage. Common consideration-set screening (CSS) rules assume that respondents first evaluate whether the attribute levels of an alternative are acceptable and next use combination rules to combine the attribute level evaluations across attributes. In particular, the acceptability of attribute levels has been modeled using probabilistic approaches including binary latent variables with possibly segment-specific distributions (Jedidi & Kohli, 2005) or using subject-specific randomly distributed thresholds (Gilbride & Allenby, 2004). Furthermore, many different types of combination rules have been proposed to combine attribute level evaluations across attributes (e.g., disjunctive or conjunctive rules, disjunctions of conjunctions (Hauser et al., 2010); subset-conjunctive rules (Jedidi & Kohli, 2005); elimination by aspects (Gilbride & Allenby, 2006)). For instance, using a conjunctive rule, an alternative is included in the consideration-set if it is acceptable on all attributes. The screening rule which assumes that respondents ignore part of the attributes, also called attribute non-attendance (ANA), has received a lot of attention in the literature on discrete choice modeling. Accounting for ANA is important as failure to account for such attribute processing heterogeneity may lead to biased marginal WTP estimates (Hensher & Rose, 2009). Although there is compelling evidence that consumers resort to screening processes as ANA or CSS when choosing between the alternatives of choice set, and that accounting for such screening processes is useful, screening rule models have not yet been fully developed. Existing screening rule models have focused on accounting for either CSS or ANA but not on the combination of both processes. A first objective of this paper is to extend existing CSS models to account for ANA. This extension is important because failure to account for ANA may lead to biased willingness-to-pay estimates. Moreover, by accounting for ANA in CSS models the fit and predictive accuracy of such models may be improved. Second, existing CSS models do not yet allow to account for heterogeneity in the taste parameters of the choice model. However, using a latent class or a mixed logit approach to account for the fact the consumers assign different utilities to attribute levels has become the standard when modeling preferences with a compensatory choice model. Therefore, a second objective of our paper is to extend existing CSS models to account for heterogeneity in the taste parameters of the choice model. This extension is important as willingness-to-pay estimates may be severely biased if heterogeneity in the taste parameters is erroneously ignored. Furthermore, appropriately modeling the heterogeneity in the taste parameters and relating this heterogeneity to relevant respondent characteristics is important as it yields key substantive insights about the respondents. For instance, in a marketing context, it is crucial to identify and describe segments of consumers with homogeneous preference. The goal of this paper is to build new CSS models that account for both ANA and heterogeneity in taste parameters. In particular, we will focus on incorporating these sources of heterogeneity in probabilistic subset-conjunctive models (Jedidi & Kohli, 2005). Probabilistic t subset-conjunctive models assume that respondents will include an alternative as part of the consideration set if it is acceptable on at least t attribute levels. The involved screening process is probabilistic as respondents are assumed to classify the attribute levels of the alternatives in each choice set as acceptable/unacceptable with a certain (possibly class-specific) probability. To account for ANA in subset-conjunctive models, we will implement an extended three-fold screening process in which respondents first determine which subset of attributes will be accounted for, next classify attribute levels of the presented alternatives for attended attributes as acceptable/unacceptable, and finally apply a specific combination rule to combine the attribute level evaluations across attributes. In the subsequent choice-stage, choices will be modelled including the outcome of the screening-stage. Furthermore, in addition to accounting for ANA, we will use a latent class approach to model heterogeneity in the taste parameters of probabilistic subset-conjunctive models. The models will be illustrated using stated choice data on housing preferences of students. References Gilbride, T.J., Allenby, G. M. (2006). Estimating Heterogeneous Eba and Economic Screening Rule Choice Models. Marketing Science, 25 (5), 494-509. Gilbride, T.J., Allenby, G.M. (2004). A Choice Model with Conjunctive, Disjunctive and Compensatory Screening Rules. Marketing Science, 23(3), 391-406. Hauser, J.R. (2014). Consideration-Set Heuristics. Journal of Business Research, 67, 1688-1699. Hensher, D., Rose, J., Greene, W., (2005). The Implications on Willingness to Pay of Respondents Ignoring Specific Attributes. Transportation, 32(3), 203-222. Hensher, D., Rose, J. (2009). Simplifying Choice Through Attribute Preservation Or Non-Attendance: Implications for Willingness to Pay. Transportation Research Part E, 45(4), 583-590. Jedidi, K., Kohli, R. (2005). Probabilistic Subset Conjunctive Models for Heterogeneous Consumers. Journal of Marketing Research, 17, 483-494.status: publishe