This article presents a way to model individual preferences and purchase intentions under the assumption that the pattern of relative similarity or substitutability among brands can be represented factorially. Factorial preference structure represent a compromise between the Luce model, which assumes that none of the alternatives share any common features, and EBA, which assumes that each possible subset of a choice set has a unique set of common features. Factorial preference structures are compared to several hierarchical structures and to one that assumes that preferences are simply scalable.

Unacceptable levels in conjoint analysis are studied using both a full profile method and Sawtooth Software's Adaptive Conjoint Analysis. The utility of an unacceptable level was set equal to a value that would have been assigned if the unacceptable levels question was not asked, but the respondent treated that level as unacceptable. With ACA, this resulted in a model with (a) superior predictive power relative to a model that eliminates alternatives with unacceptable levels and (b) equivalent predictive power to a compensatory model that does not consider unacceptable levels. Furthermore, in this application, this was accomplished even though the respondent was presented with approximately 30 percent fewer paired comparison observations.

When studying objects with environmentally correlated attributes, the use of orthogonal arrays in conjoint analysis can produce some highly unrealistic stimuli that might threaten a study's ecological validity. The effect of such departures from realism induced by environmental correlations is examined. A series of three experiments compares a stimulus set that has non-zero, but lower than environmental, correlations to one using an orthogonal array in terms of both perceived realism and predictive power. The results indicate that environmentally correlated attributes may pose fewer problems in practice than in theory.

This paper describes a straightforward procedure for assessing the predictive validity of joint-space models applied to the generation and evaluation of new concepts. The results indicate a marked deterioration in predictive power when using joint spaces derived from data on real objects to predict evaluations of new concepts. Conjoint analysis appears to offer a sounder approach to this aspect of product development.

In consumer judgments of product designs, the numbers of additive and configural feature effects are apt to differ between verbal and pictorial stimulus presentations. A study of judgmental responses to sweater designs suggests the significant occurrence of feature interactions, a tendency for pictorial (versus verbal) displays to evoke more main effects, and a tendency for pictorial displays to promote cue configurality when one controls for type of information-processing strategy.

In an earlier Journal of Consumer Research article, Kahn, Moore, and Glazer examined the phenomenon of constrained choice in experimental settings and found empirical results that ran counter to generally accepted models of hierarchical choice (e.g., preference trees). In this article, we reexamine those findings. Specifically, our main purposes are (1) to provide an explanation for the deviations from accepted theoretical models, (2) to suggest operational measures for the new constructs identified by Kahn, Moore, and Glazer, and (3) to propose a modification of hierarchical models that improves choice prediction by adjusting for bias against lone alternatives.

This article examines the relative importance of a variety of factors in influencing hierarchical choice. In our first experiment, we test some implications of Tversky and Sattath's (1979) Hierarchical Elimination Model (HEM) relating to a choice set in which an external constraint has been imposed. (An external constraint changes the decision process by partitioning the brands in a different way than the consumer naturally would.) Our experimental results and the theoretical predictions do not converge. While they agree that external partitions do affect choice probabilities, they differ on the nature of the effect. Next, we run a second experiment to test alternative explanations of our empirical results. Using these results, we propose managerial implications for positioning a #1 and #2 brand.

The recently developed “pick-any” approach to data collection and analysis is described and illustrated by examples that support its face validity, reliability, and convergent validity with other multidimensional scaling techniques. Some solutions to problems that arise in applying the pick-any procedure are suggested, and potential extensions are proposed for use of the procedure in perceptual mapping applications.