135  2001 by JOURNAL OF CONSUMER RESEARCH, Inc. ● Vol. 28 ● June 2001 All rights reserved. 0093-5301/2002/2801-0009$03.00 The Influence of Task Complexity on Consumer Choice: A Latent Class Model of Decision Strategy Switching JOFFRE SWAIT WIKTOR ADAMOWICZ* The literature indicating that person-, context-, and task-specific factors cause consumers to utilize different decision strategies has generally failed to affect the specification of choice models used by practitioners and academics alike, who still tend to assume an utility maximizing, omniscient, indefatigable consumer. This article (1) introduces decision strategy selection, within a maintained compensatory framework, into aggregate choice models via latent classes, which arise because of task complexity; (2) it demonstrates that within an experimental choice task, the model reflects changing aggregate preferences as choice complexity changes and as the task progresses. The import of these findings for current practice, model interpretation, and future research needs is examined. T he judgment and decision-making (JDM) literature has devoted considerable attention to identifying and char- acterizing the strategies used by human beings and organ- izations to make decisions (Bettman, Johnson, and Payne 1991; Payne, Bettman, and Johnson 1993). Some research- ers have concerned themselves with formulating descriptive and mathematical models of different decision strategies (for conjunctive and disjunctive decision rules, see Dawes [1964]; for satisficing, see Simon [1955]; for elimination- by-aspects, see Tversky [1972]); another stream of the JDM literature has concerned itself with finding evidence of the utilization of compensatory and noncompensatory decision strategies as task complexity and context change (Ball 1997; Payne et al. 1993; Russo and Dosher 1983). While this literature has established that people utilize multiple choice strategies depending on a number of factors (product, occasion, information presentation format, time pressure, alternative similarity, etc.), there has been little linkage of these findings to the literature on multiattribute, multialternative experimental choice tasks (Louviere and *Joffre Swait is a partner of Advanis, 12 West University Avenue, #205, Gainesville, FL 32601 (Joffre_Swait@Advanis.ca), and an associated fac- ulty member of the Department of Marketing, Warrington College of Busi- ness, University of Florida. Wiktor Adamowicz is Professor, Department of Rural Economy, University of Alberta, Edmonton, Alberta T6G 2H1, Canada (vic.adamowicz@ualberta.ca). Send correspondence to Joffre Swait. The authors acknowledge the essential contributions of the editor, associate editor, and reviewers to the improvement of the research. The first author acknowledges the financial support of Advanis. The data used in this study originate from an independent research effort by Joffre Swait and Doug Olsen (University of Alberta). Woodworth 1983). Choice experiments, while taking on many different forms, commonly present respondents with the task of choosing one alternative among multiple product profiles, each described in terms of a generally common attribute set. In addition, respondents are usually presented with multiple decision scenarios in a short time span. Tra- ditionally, these data are modeled through specifications such as the Multinomial Logit (MNL), Nested MNL, and Probit models. The models in general use are almost ex- clusively compensatory and single decision rule in nature (with some notable exceptions, including Andrews and Sri- nivasan 1995; Roberts and Lattin 1991; Swait 2001; Swait and Ben-Akiva 1987a, 1987b). In this article we build on the established knowledge that decision strategies can change because of context and pro- pose a modeling framework that allows for decision strategy and/or preference structure changes. We construct a model that can use choice data, as typically collected in commercial and academic applications, and that accounts for the types of decision strategy changes that have been found using eye-tracking, verbal protocols, or other laboratory tech- niques. The proposed framework allows factors like task complexity to affect inferences about preferences, with the goal of improving the modeling of experimental or revealed preference choice data. We also propose and employ a new summary measure of task complexity, based on the infor- mation theoretic concept of entropy. Rather than using verbal protocols (Bettman 1970) or eye- tracking techniques (Russo and Dosher 1983), or registering information search patterns (Ball 1997) to develop measures of how individuals respond to complexity and other task