Market segmentation with choice-based conjoint analysis

Choice-based conjoint analysis has increased in popularity in recent years among marketing practitioners. The typical practice is to estimate choice-based conjoint models at the aggregate level, given insufficient data for individual-level estimation of part-worths. We discuss a method for market segmentation with choice-based conjoint models. This method determines the number of market segments, the size of each market segment, and the values of segment-level conjoint part-worths using commonly collected conjoint choice data. A major advantage of the proposed method is that current (incomplete) data collection approaches for choice-based conjoint analysis can still be used for market segmentation without having to collect additional data. We illustrate the proposed method using commercial conjoint choice data gathered in a new concept test for a major consumer packaged goods company. We also compare the proposed method with ana priori segmentation approach based on individual choice frequencies.     

A Low-Effort Recommendation System with High Accuracy

In recent studies on recommendation systems, the choice-based conjoint analysis has been suggested as a method for measuring consumer preferences. This approach achieves high recommendation accuracy and does not suffer from the start-up problem because it is also applicable for recommendations for new consumers or of new products. However, this method requires massive consumer input, which causes consumer reluctance. In a simulation study, we demonstrate the high accuracy, but also the high user’s effort for using a utility-based recommendation system using a choice-based conjoint analysis with hierarchical Bayes estimation. In order to reduce the conflict between consumer effort and recommendation accuracy, we develop a novel approach that only shows Pareto-efficient alternatives and ranks them according to the number of dominated attributes. We demonstrate that, in terms of the decision accuracy of the recommended products, the ranked Pareto-front approach performs better than a recommendation system that employs choice-based conjoint analysis. Furthermore, the consumer’s effort is kept low and comparable to that of simple systems that require little consumer input.     

A Comparison of Conjoint Methods When There Are Many Attributes

This paper compares several methods of performing conjoint analysis when there is a large number of attributes. National parks were described in terms of 17 attributes and 56 levels. Subjects were randomly assigned to one of four groups and each person responded to a calibration questionnaire that allowed the estimation of one of the following conjoint analysis models: full profile, ACA, individual-level hybrid, or full profile on the person's eight stated most important attributes. Validations were performed in terms of individual choices and aggregate choice shares. Reliabilities were assessed on both ratings and choices.Surprisingly even with a large number of attributes, the full profile method consistently validated best. Second was a full profile model estimated on the respondent's stated eight most important attributes. ACA and individual hybrid conjoint analysis performed similarly, but worse than these two methods on most measures. Validation differences were more strongly related to differences in attribute importances than desirabilities for levels within an attribute. It appears that these respondents were not able to accurately report self-explicated importances with a large number of attributes.     

On the Similarity of Classical and Bayesian Estimates of Individual Mean Partworths

An exciting development in modeling has been the ability to estimate reliable individual-level parameters for choice models. Individual partworths derived from these parameters have been very useful in segmentation, identifying extreme individuals, and in creating appropriate choice simulators. In marketing, hierarchical Bayes models have taken the lead in combining information about the aggregate distribution of tastes with the individual's choices to arrive at a conditional estimate of the individual's parameters. In economics, the same behavioral model has been derived from a classical rather than a Bayesian perspective. That is, instead of Gibbs sampling, the method of maximum simulated likelihood provides estimates of both the aggregate and the individual parameters. This paper explores the similarities and differences between classical and Bayesian methods and shows that they result in virtually equivalent conditional estimates of partworths for customers. Thus, the choice between Bayesian and classical estimation becomes one of implementation convenience and philosophical orientation, rather than pragmatic usefulness.     

Integrated mixed logit and latent variable models

Extending the traditional discrete choice model by incorporating latent psychological factors can help to better understand the individual’s decision-making process and therefore to yield more reliable part-worth estimates and market share predictions. Several integrated choice and latent variable (ICLV) models which merge the conditional logit model with a structural equation model exist in the literature. They assume homogeneity in the part-worths and use latent variables to model the heterogeneity among the respondents. This paper starts from the mixed logit model that describes the heterogeneity in the part-worths and uses the latent variables to decrease the unexplained part of the heterogeneity. The empirical study presented here shows these ICLV models perform very well with respect to model fit and prediction.     

A simple method for estimating preference parameters for individuals

This paper demonstrates a method for estimating logit choice models for small sample data, including single individuals, that is computationally simpler and relies on weaker prior distributional assumptions compared to hierarchical Bayes estimation. Using Monte Carlo simulations and online discrete choice experiments, we show how this method is particularly well suited to estimating values of choice model parameters from small sample choice data, thus opening this area to the application of choice modeling. For larger sample sizes of approximately 100–200 respondents, preference distribution recovery is similar to hierarchical Bayes estimation of mixed logit models for the examples we demonstrate. We discuss three approaches for specifying the conjugate priors required for the method: specifying priors based on existing or projected market shares of products, specifying a flat prior on the choice alternatives in a discrete choice experiment, or adopting an empirical Bayes approach where the prior choice probabilities are taken to be the average choice probabilities observed in a discrete choice experiment. We show that for small sample data, the relative weighting of the prior during estimation is an important consideration, and we present an automated method for selecting the weight based on a predictive scoring rule.     

Setting prices in mixed logit model designs

We investigate different procedures to set prices in designs for choice-based conjoint analysis using the mixed logit model which captures latent consumer heterogeneity. Besides discrete attributes, we include a linear price term in the deterministic utility function thereby treating price as continuous variable. We consider two different price intervals and several price sets which contain either two or three prices. We compare these alternatives to set prices by simulating choices for different constellations on the basis of the mixed logit model. Furthermore, we generate ten designs simultaneously instead of just one. Using these simulated choices, we estimate the parameters of the mixed logit model in the next step. To reduce the needed sample size and computation time caused by accounting for latent consumer heterogeneity, we apply Halton draws and set a minimum potential design for prior draws. ANOVA with root mean squared error between estimated and true price coefficient values of individual consumers as dependent variable shows that using more extreme prices as interval bounds and one intermediate price positioned to the right of the interval performs best.     

The no-choice option and dual response choice designs

Choice set designs that include a constant or no-choice option have increased efficiency, better mimic consumer choices, and allow one to model changes in market size. However, when the no-choice option is selected no information is obtained on the relative attractiveness of the available alternatives. One potential solution to this problem is to use a dual response format in which respondents first choose among a set of available alternatives in a forced-choice task and then choose among the available alternatives and a no-choice option. This paper uses a simulation to demonstrate and confirm the possible gains in efficiency of dual response over traditional choice-based conjoint tasks when there are different proportions choosing the no-choice option. Next, two choice-based conjoint analysis studies find little systematic violation of IIA with the addition/deletion of a no-choice option. Further analysis supports the hypothesis that selection of the no-choice option is more closely related to choice set attractiveness than to decision difficulty. Finally, validation evidence is presented. Our findings show that researchers can employ the dual response approach, taking advantages of the increased power of estimation, without concern for systematically biasing the resulting parameter estimates. Hence, we argue this is a valuable approach when there is the possibility of a large number of no-choices and preference heterogeneity.     

Attribute Range Effects in Binary Response Tasks

This paper investigates attribute range effects in binary response conjoint analysis tasks. We investigate a long-standing conjecture that the regression estimates of attributes in choice tasks are influenced by a researcher's selected range of attribute levels across choice sets. Specifically, we examine the effect(s) of varying attribute ranges systematically over two ranges of levels (1=wide range, 2=levels in the middle of the wide range) in a public bus choice context. A master 2³ design is used to vary the range (i.e., wide, middle) of three numerical attributes (fare, service and time). In each of the eight master range conditions a 2³ factorial creates bus profiles, and a ninth condition is added to test for response non-linearity. Our results suggest that attribute range impacts attribute main effects to a small degree, yet exhibit substantial and systematic effects on attribute interactions and model goodness-of-fit. Implications of these results for practical design of academic and commercial choice-based conjoint analysis tasks are discussed.     

Hierarchical representations of market structures and choice processes through preference trees

Two ways of representing market structures in hierarchical forms are discussed and compared. The better-known method, hierarchical clustering of similarity measures derived from aggregate brand-switching probabilities, has three drawbacks. First, the aggregate-level switching probabilities may not be representative of the majority of the individuals; therefore, inferences made about individual or segmental behavior can be misleading. Second, regardless of the level of aggregation, it is not possible to infer the choice process from typical purchase data. Finally, in this example, this method produces a tree that is driven largely by the brands with small market shares. The other method, preference-tree analysis, models brand-choice behavior as a hierarchical elimination process. The parameters of the model are estimated from the way that relative brand-choice probabilities change as the choice set is varied. Preference-tree analysis, which requires paired-comparison preference data, is shown to eliminate the drawbacks observed with hierarchical clustering.     

The Energy Paradox Revisited: Analyzing the Role of Individual Differences and Framing Effects in Information Perception

In the ongoing debate about the “energy paradox”, a recent stream of literature highlights the importance of behavioural anomalies such as bounded rationality and self-control problems. However, the role of individual-level factors in explaining the energy paradox is still not fully understood. Combining literature on behavioural anomalies and consumer heterogeneity, the current paper analyses how individual differences influence the perception of energy-related information and susceptibility to choice-framing effects. A choice-based conjoint experiment about energy-saving home improvements was conducted with 363 homeowners in Switzerland. Results show that numeracy and energy literacy have no influence on how much attention individuals pay to energy cost savings. However, impulsivity and risk aversion are found to significantly impact homeowners’ weighting of future energy cost savings. Further, it is found that impulsive homeowners are significantly more susceptible to energy cost-framing effects. A key implication for consumer policy is that general educational programs targeted at enhancing citizens’ knowledge and cognitive abilities are unlikely to increase energy conservation investments. The findings further suggest that consumer policies and business models aimed at reducing impulsiveness and influencing risk perception might foster the uptake of energy-saving measures in the residential housing sector.     

A cross validation test of four models for quantifying multiattribute preferences

This paper examines the internal predictive validity of four multiattribute preference models: (a) a self-explicated model with equal importance weights; (b) a self-explicated model with unequal weights; (c) Sawtooth Software's Adaptive Conjoint Analysis (ACA); and (d) full profile conjoint analysis. We also discuss the problem of choosing criterion measures for comparing cross validations across models.The authors would like to acknowledge the support of the Huntsman Center for Global Competition and Innovation and the Crosby/Foggitt Fellowship from the Sol C. Snider Entrepreneurial Center, both at the Wharton School.     

Studying the level-effect in conjoint analysis: An application of efficient experimental designs for hyper-parameter estimation

Research in marketing, and business in general, involves understanding when effect-sizes are expected to be large and when they are expected to be small. An example is the understanding of the level-effect in marketing, where the effect of product attributes on utility is positively related to the number of levels present among choice alternatives. Knowing when consumers are sensitive to the competing levels of attributes is an important aspect of merchandising, selling and promotion. In this paper, we propose a model and a method for studying the level-effect in conjoint analysis. The model combines perceptual theories in psychology to arrive at a non-linear specification of hyper-parameters in a hierarchical model. The method applies an experimental design criterion for efficient estimation of hyper-parameters. The proposed model and method are validated using a national sample of respondents.      

Retention of latent segments in regression-based marketing models

Product design and marketing mix decisions for segmented markets depend crucially on the correct specification of marketing models used as input to these decisions. With real-world data, the true number of segments in a market is unknown. Current evidence from simulation studies suggests that the accuracy of commonly used criteria for determining the number of segments in a market depends on the usage context, including the type of distribution being used to describe the data, the model specification, and the characteristics of the market. This study investigates via simulation the performance of seven segment retention criteria used with finite mixture regression models for normal data. This is one of the most important analysis contexts in marketing research since regression models are used, for example, in conjoint analysis and market response analysis, yet no previous study in either the marketing or statistics literatures explores the segment retention problem for mixture regression models. The study shows that one criterion, Akaike's Information Criterion (AIC) with a per-parameter penalty factor of 3 (AIC3), is clearly the best criterion to use across a wide variety of model specifications and data configurations, having the highest success rate and producing very low parameter bias. Currently, this criterion is rarely, if ever, used in the marketing literature.     

Towards a brain-to-society systems model of individual choice

Canonical models of rational choice fail to account for many forms of motivated adaptive behaviors, specifically in domains such as food selections. To describe behavior in such emotion- and reward-laden scenarios, researchers have proposed dual-process models that posit competition between a slower, analytic faculty and a fast, impulsive, emotional faculty. In this paper, we examine the assumptions and limitations of these approaches to modeling motivated choice. We argue that models of this form, though intuitively attractive, are biologically implausible. We describe an approach to motivated choice based on sequential sampling process models that can form a solid theoretical bridge between what is known about brain function and environmental influences upon choice. We further suggest that the complex and dynamic relationships between biology, behavior, and environment affecting choice at the individual level must inform aggregate models of consumer choice. Models using agent-based complex systems may further provide a principled way to relate individual and aggregate consumer choices to the aggregate choices made by businesses and social institutions. We coin the term “brain-to-society systems” choice model for this broad integrative approach.     

Models and optimal designs for conjoint choice experiments including a no-choice option

In a classical conjoint choice experiment, respondents choose one profile from each choice set that has to be evaluated. However, in real life, the respondent does not always make a choice: often he/she does not prefer any of the options offered. Therefore, including a no-choice option in a choice set makes a conjoint choice experiment more realistic. In the literature, three different models are used to analyze the results of a conjoint choice experiment with a no-choice option: the no-choice multinomial logit model, the extended no-choice multinomial logit model, and the nested no-choice multinomial logit model. We develop optimal designs for the two most appealing of these models using the D-optimality criterion and the modified Fedorov algorithm and compare these optimal designs with a reference design, which is constructed while ignoring the no-choice option, in terms of estimation and prediction accuracy. We conclude that taking into account the no-choice option when designing a no-choice experiment only has a marginal effect on the estimation and prediction accuracy as long as the model used for estimation matches the data-generating model.     

A latent structure factor analytic approach for customer satisfaction measurement

The linkage of customer satisfaction, customer retention, and firm profitability has been well established in the marketing literature, and provides ample justification as to why customer satisfaction measurement (CSM) has been a focal point in marketing decision making. Although aggregate market level research on understanding the determinants of customer satisfaction is abundant, CSM decisions at segment level are possible only if the individual or market segment differences in the formation of overall satisfaction judgments and subsequent heterogeneity in the role these various determinants play are understood. Based on expectancy-disconfirmation theory in customer satisfaction, we propose a maximum likelihood based latent structure factor analytic methodology which visually depicts customer heterogeneity regarding the various major determinants of customer satisfaction judgments involving multiple attributes, and provides directions for segment-specific CSM decisions. We first describe the proposed model framework including the technical aspects of the model structure and subsequent maximum likelihood estimation. In an application to a consumer trade show, we then demonstrate how our proposed methodology can be gainfully employed to uncover the nature of such heterogeneity. We also empirically demonstrate the superiority of the proposed model over a number of different model specifications in this application. Finally, limitations and directions for future research are discussed.     

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.     

Customer satisfaction study via a latent segment model

The aim of this study is to apply a new conceptual model, and a new technique as an approach to the modelling of customers’ satisfaction, and to develop an overall satisfaction index (OSI). This study evaluates customers’ satisfaction of a certain public organization service, and argues that in order to estimate the global customers’ satisfaction measure we must appeal to methodologies recognizing that satisfaction must be understood as a latent variable, quantified through multiple indicators. Thus, it is natural that we consider the latent segment models (LSM) approach to proceed to the evaluation of customer's service satisfaction. As a result of these models estimation, we selected a three latent segment model, that is to say, the latent variable customer satisfaction has three classes: segment 1, with 50.4 percent of the customers, that represents “The Very Satisfied”, for those to whom everything is very well in the organization service; a segment 2, with 33.4 percent of the customers, representative of the “The Well Satisfied”, not totally satisfied with the quality of the organization, and a segment 3, with 16.2 percent of the customers, “Satisfaction Demanders”, thinking that organizational quality can be improved. Finally, we developed an overall satisfaction index which is important to show how the company as a whole is performing.