International Journal of Market Research Vol. 49 No. 3, 2007 p.341–364 Predicting purchase decisions with different conjoint analysis methods: a Monte Carlo simulation Klaus Backhaus and Robert Wilken Westphalian Wilhelms-University of Muenster Thomas Hillig Alstom Schweiz AG INTRODUCTION Predicting purchase decisions is highly relevant for marketers who want to estimate the potential success of their products, or market shares. In marketing research and practice, conjoint analyses (CAs) are widely used in this context. During the last 20 years, not only in marketing, but also in other disciplines, numerous variants of conjoint models and parameter estimation methods have been developed (Green & Srinivasan 1978; Green & Srinivasan 1990; Moore et al. 1998), but only some of them have gained broad acceptance in practice (Carroll & Green 1995) and not all conjoint variants seem to be appropriate in order to predict purchase decisions. The goal of the present paper is to substantially contribute to the literature that compares conjoint models with respect to forecasting consumer choices. Via the analysis to be carried out, we try to systematically answer the research question as to which CA variant performs best in terms of purchase predictions. In order to be able to vary systematically several effects that may have an influence on the appropriateness of one method or another, our analysis is based on a simulation study. Furthermore, in simulation analyses, undesirable effects can be held constant, allowing for non- biased estimates of the relevant model parameters. The remainder of the paper is structured as follows. First, we describe the two main groups of conjoint models, with specific relevance to the context of consumer choice predictions. These two groups are choice-based models and the rating-based limit conjoint analysis, the latter being a simple but theoretically substantial modification of traditional conjoint analysis. Both groups are described with respect to their theoretical foundations. Then we review the literature on empirical comparisons of different CA models, using both real and synthetic data. After that, we turn to the simulation study that has been performed in order to answer the aforementioned research 1 of 15 17 Oct 2017 11:08:37