An empirical appraisal of conjoint choice simulators

Estimating market share for alternative concepts is an important task in the new product development process. One way of predicting (or \u27simulating\u27) shares at the concept stage is by using conjoint choice simulators which use data and estimates from conjoint analysis experiments as input. These simulators, in addition to being used as predictors of share for alternative concepts, are also an integral component in several product and product line optimization models. The primary objective of this thesis is to examine the nature of conditions where each of six alternative conjoint simulators may provide accurate and reliable predictions of market share. Also, a new simulator, motivated by results of the primary analysis, is presented and its advantages are illustrated. The six simulators considered in the first part are based on six different choice rules. These are (i) The Maximum Utility Rule, (ii) The Bradley-Terry-Luce-1 Rule, (iii) The Multinomial Logit Rule, (iv) The Bradley-Terry-Luce-2 Rule, (v) The Multinomial Probit Rule and (vi) The Multivariate Generalization of the Bivariate Logistic Rule, due to Bock and Jones. Simulators based on the first three rules are widespread in their use currently, while those based on the others are promising alternatives in specific situations. The results indicate that a deterministic choice rule based simulator like the MUR will provide less accurate estimates of share than the other stochastic rules based simulators when the product category is of the non-durable/low involvement type. Also, the MUR is found to provide inconsistent estimates of share in competitive (or \u27efficient\u27) sets of profiles with a good deal of homogeneity in preferences across respondents. Additionally, evidence is found to question the robustness of share estimates of the MUR simulator. Some limited, inconclusive evidence is presented about the applicability of the BTL rule based simulators in large choice sets. The MNL simulator is shown to predict shares which are considerably robust. The MNP simulator is shown to work better than any other simulator in sets characterized by correlations among respondent utilities; also, its predictions are fairly robust. The Bock-Jones simulator is found to be very useful for sets with a specific type of correlation structure. The new simulator, which is proposed in the latter part of this thesis, seeks to model decision making in the spirit of two-stage information processing strategies suggested in the literature on protocols of consumer decision processes. Also, the choice rule underlying the simulator allows for stochasticities and correlations among individual-level preferences. At various points in the thesis, limitations inherent in the research are described; and also provided are suggestions for future research in the area