Kurz Kommentiert

Ohne Zusammenfassung     

Challenges and opportunities in high-dimensional choice data analyses

Modern businesses routinely capture data on millions of observations across subjects, brand SKUs, time periods, predictor variables, and store locations, thereby generating massive high-dimensional datasets. For example, Netflix has choice data on billions of movies selected, user ratings, and geodemographic characteristics. Similar datasets emerge in retailing with potential use of RFIDs, online auctions (e.g., eBay), social networking sites (e.g., mySpace), product reviews (e.g., ePinion), customer relationship marketing, internet commerce, and mobile marketing. We envision massive databases as four-way VAST matrix arrays of Variables?×?Alternatives?×?Subjects?×?Time where at least one dimension is very large. Predictive choice modeling of such massive databases poses novel computational and modeling issues, and the negligence of academic research to address them will result in a disconnect from the marketing practice and an impoverishment of marketing theory. To address these issues, we discuss and identify the challenges and opportunities for both practicing and academic marketers. Thus, we offer an impetus for advancing research in this nascent area and fostering collaboration across scientific disciplines to improve the practice of marketing in information-rich environment.     

Adjusting Choice Models to Better Predict Market Behavior

The emergence of Bayesian methodology has facilitated respondent-level conjoint models, and deriving utilities from choice experiments has become very popular among those modeling product line decisions or new product introductions. This review begins with a paradox of why experimental choices should mirror market behavior despite clear differences in content, structure and motivation. It then addresses ways to design the choice tasks so that they are more likely to reflect market choices. Finally, it examines ways to model the results of the choice experiments to better mirror both underlying decision processes and potential market choices. Co-chairs. Author order is alphabetical.     

Augmenting discrete-choice data to identify common preference scales for inter-subject analyses

Discrete-choice experiments are commonly used to measure subjects?? preference structures and are often preferred to other measurement methods because they better align with actual choice behavior and avoid some of the well-documented biases inherent in alternative elicitation methods. A limitation of discrete-choice methods is the loss of inter-subject comparability because preference estimates are invariant to linear transformations necessitating indentifying constraints that remove a common, between-subjects utility scale. This constraint limits the application of discrete-choice results to situations where within-subject comparisons are meaningful. They enable one to sort options for each subject but not to sort subjects according to the relative intensity of their preferences. This paper uses auxiliary data to recover a common preference scale for between-subject comparisons. The model combines discrete-choice data with ratings data while adjusting for response biases due to method effects. The joint model moves the identification constraints from the sub-model for the discrete-choice data to the sub-model for the ratings data. The proposed methodology is complementary to willingness-to-pay computations when studies lack price or its economic foundation is untenable.     

Bayesian estimation of multivariate-normal models when dimensions are absent

Multivariate economic and business data frequently suffer from a missing data phenomenon that has not been sufficiently explored in the literature: both the independent and dependent variables for one or more dimensions are absent for some of the observational units. For example, in choice based conjoint studies, not all brands are available for consideration on every choice task. In this case, the analyst lacks information on both the response and predictor variables because the underlying stimuli, the excluded brands, are absent. This situation differs from the usual missing data problem where some of the independent variables or dependent variables are missing at random or by a known mechanism, and the “holes” in the data-set can be imputed from the joint distribution of the data. When dimensions are absent, data imputation may not be a well-poised question, especially in designed experiments. One consequence of absent dimensions is that the standard Bayesian analysis of the multi-dimensional covariances structure becomes difficult because of the absent dimensions. This paper proposes a simple error augmentation scheme that simplifies the analysis and facilitates the estimation of the full covariance structure. An application to a choice-based conjoint experiment illustrates the methodology and demonstrates that naive approaches to circumvent absent dimensions lead to substantially distorted and misleading inferences.