Search

A latent class model for identifying classes of subjects in paired comparison choice experiments is developed. The model simultaneously estimates a probabilistic classification of subjects and the logit models' coefficients relating characteristics of objects to choices for each respective group among two alternatives in paired comparison experiments. A modest Monte Carlo analysis of algorithm performance is presented. The proposed model is illustrated with empirical data from a consumer psychology experiment that examines the determinants of perceived consumer risk. The predictive validity of the method is assessed and compared to that of several other procedures. The sensitivity of the method to (randomly) eliminate comparisons, which is important in view of reducing respondent fatigue in the task, is investigated.

The role of perceived risk in consumer behavior has been studied extensively by academic researchers. This paper introduces a methodology for the measurement of the effects of product features, marketing mix components, and individual differences on perceived consumer risk based on theoretical foundations in the literature. A conjoint-type model based on paired comparison judgments is estimated to provide attribute weights. A modification of a stochastic multidimensional scaling-based vector model is then used to measure and summarize individual consumer differences with respect to the impact of brand attributes and marketing mix components on latent levels of perceived consumer risk. An illustration is provided using students’ risk perceptions of sports cars.

Conjoint analysis studies typically utilize orthogonal fractional factorial experimental designs to construct a set of hypothetical stimuli. Occasionally, these designs include environmentally correlated attributes that can lead to stimulus profiles that are not representative of the subject's environment. To date, no one has proposed a remedy well-grounded in statistical theory. This note presents a new methodology utilizing combinatorial optimization procedures for creating modified fractional factorial designs that are as “orthogonal” as possible, which do not contain nonrepresentative stimulus profiles.

Two recently developed probabilistic multidimensional models for analyzing pairwise choice data are introduced, discussed in terms of their differential properties, and extended in several ways. The first one, the wandering vector model, was originally suggested by Carroll [12] and extended by De Soete and Carroll [30]. The second model, called the wandering ideal point model, is a more recently proposed [32] unfolding analog of the wandering vector model. A general maximum likelihood estimation method for fitting the various models described is mentioned, as well as a statistical test for assessing the goodness of fit. Finally, an application of the models is provided concerning consumer choice for some 14 brands of over-the-counter analgesics to illustrate how such models can be gainfully utilized for marketing decision making concerning product positioning.

Product-concept testing is a popular activity in marketing research. Often the number of new product/service concepts under study far exceeds the time available for any single respondent. Respondents therefore may receive only a subset of the concepts comprising the total design. Researchers are interested in making plausible imputations for the missing evaluations of any given respondent. This paper proposes a model and an iterative estimation procedure to impute missing entries for each evaluator. The model and the procedure incorporate (1) the internal structure of the response matrix and (2) an ancillary matrix of (nonmissing) respondent background data; they also (3) allow for individual differences in respondents' uses of the numerical rating scale. The model is applied to both real and synthetic data. Suggestions also are given on how the data imputations may be used in market segmentation and product-line decisions.

Our reply to Curry, Louviere, and Augustine's critique of our earlier paper focuses on differences in motivation between our research and theirs. Our interest in the problem relates to the possible incorporation of self-explicated evaluations in conjoint data collection methods; subsequent to the appearance of our original paper, we have developed hybrid models that combine elements of self-explicated (compositional) and conjoint (decompositional) data collection procedures. As far as we can surmise from their critique, Curry, Louviere, and Augustine are concerned with much broader strategic issues relating share of choices in the consumer population to changes in the shape of attribute weight distributions, shape of the Pareto tradeoff boundary, and so on.

Choice-constrained conjoint analysis (CCCA) is a new method for metric conjoint analysis studies. It computes part-worth utility functions that account for “revealed preference”—those products a respondent actually selects in an independent choice situation. CCCA uses an iterative penalty function estimation procedure that successively modifies initial regressionderived part worths so that respondent choices (either actual or intended) of real brands are predicted as accurately as possible. The paper first describes the motivation and rationale for CCCA and presents the mathematics of the algorithm. As an illustration, it applies the CCCA model and penalty function estimation procedure to a limited set of synthetic data. A second application of the technique is presented that uses data obtained by a major telecommunications firm that used conjoint analysis to examine the importance of several features of residential communication devices. The paper also discusses potential extensions of the CCCA model and the kinds of marketing applications for which it might be useful.

This study compares, via brand-choice simulation, the effect of various distortions in attribute importance weights on share of choices at the aggregate-sample level. The results indicate that severe distortions in importance weights can occur with little change in aggregate brand share. The implications of this result for utility estimation in main-effects conjoint analysis suggest that considerably simpler ways can be adopted to estimate part worths.