Rick L. Andrews

A Comparison of Segment Retention Criteria for Finite Mixture Logit Models

Despite the widespread application of finite mixture models in marketing research, the decision of how many segments to retain in the models is an important unresolved issue. Almost all applications of the models in marketing rely on segment retention criteria such as Akaikes information criterion, Bayesian information criterion, consistent Akaikes information criterion, and information complexity to determine the number of latent segments to retain. Because these applications employ real-world data in which the true number of segments is unknown, it is not clear whether these criteria are effective. Retaining the true number of segments is crucial because many product design and marketing decisions depend on it. The purpose of this extensive simulation study is to determine how well commonly used segment retention criteria perform in the context of simulated multinomial choice data, as obtained from supermarket scanner panels, in which the true number of segments is known. The authors find that an Akaikes information criterion with a penalty factor of three rather than the traditional value of two has the highest segment retention success rate across nearly all experimental conditions. Currently, this criterion is rarely, if ever, applied in the marketing literature. Experimental factors of particular interest in marketing contexts, such as the number of choices per household, the number of choice alternatives, the error variance of the choices, and the minimum segment size, have not been considered in the statistics literature. The authors show that they, among other factors, affect the performance of segment retention criteria.

An Empirical Comparison of Logit Choice Models with Discrete Versus Continuous Representations of Heterogeneity

Currently, there is an important debate about the relative merits of models with discrete and continuous representations of consumer heterogeneity. In a recent JMR study, Andrews, Ansari, and Currim (2002; hereafter AAC) compared metric conjoint analysis models with discrete and continuous representations of heterogeneity and found no differences between the two models with respect to parameter recovery and prediction of ratings for holdout profiles. Models with continuous representations of heterogeneity fit the data better than models with discrete representations of heterogeneity. The goal of the current study is to compare the relative performance of logit choice models with discrete versus continuous representations of heterogeneity in terms of the accuracy of household-level parameters, fit, and forecasting accuracy. To accomplish this goal, the authors conduct an extensive simulation experiment with logit models in a scanner data context, using an experimental design based on AAC and other recent simulation studies. One of the main findings is that models with continuous and discrete representations of heterogeneity recover household-level parameter estimates and predict holdout choices about equally well except when the number of purchases per household is small, in which case the models with continuous representations perform very poorly. As in the AAC study, models with continuous representations of heterogeneity fit the data better.

Hierarchical Bayes Versus Finite Mixture Conjoint Analysis Models: A Comparison of Fit, Prediction, and Partworth Recovery

A study conducted by Vriens, Wedel, and Wilms (1996) and published in Journal of Marketing Research found that finite mixture (FM) conjoint models had the best overall performance of nine conjoint segmentation methods in terms of fit, prediction, and parameter recovery. Since that study, hierarchical Bayes (HB) conjoint analysis methods have been proposed to estimate individual-level partworths and have received much attention in the marketing research literature. However, no study has compared the relative effectiveness of FM and HB conjoint analysis models in terms of fit, prediction, and parameter recovery. To conduct such a comparison, the authors employ the simulation methodology proposed by Vriens, Wedel, and Wilms with some modification. The authors estimate traditional individual-level conjoint models as well. The authors show that FM and HB models are equally effective in recovering individual-level parameters and predicting ratings of holdout profiles. Two surprising findings are that (1) HB performs well even when partworths come from a mixture of distributions and (2) FM produces good parameter estimates, even at the individual level. The authors show that both models are quite robust to violations of underlying assumptions and that traditional individual-level models overfit the data.

Parameter Bias from Unobserved Effects in the Multinomial Logit Model of Consumer Choice

Over the past two decades, validation of choice models has focused on predictive validity rather than parameter bias. In real-world validation of choice models, true parameter values are unknown, so examination of parameter bias is not possible. In contrast, the main focus of this study is parameter bias in simulated scanner-panel choice data with known parameter values. Study of parameter bias enables the assessment of a fundamental issue not addressed in the choice modeling literature—the extent to which the logit choice model is capable of distinguishing unobserved effects that give rise to persistence in observed choices (e.g., heterogeneity and state dependence). Although econometric theory provides some information about the causes of bias, the extent of such bias in typical scanner data applications remains unclear. The authors present an extensive simulation study that provides information on the extent of bias resulting from the misspecification of four unobserved effects that receive frequent attention in the literature—choice set effects, heterogeneity in preferences and market response, state dependence, and serial correlation. The authors outline implications for model builders and managers. In general, the potential for parameter bias in choice model applications appears to be high. Overall, a logit model with choice set effects and the Guadagni–Little loyalty variable produces the most valid parameter estimates.

Inferring Market Structure from Customer Response to Competing and Complementary Products

We consider customer influences on market structure, arguing that market structure should explain the extent to which any given set of market offerings are substitutes or complements. We describe recent additions to the market structure analysis literature and identify promising directions for new research in market structure analysis. Impressive advances in data collection, statistical methodology and information technology provide unique opportunities for researchers to build market structure tools that can assist “real-time” marketing decision-making.

Recovering and profiling the true segmentation structure in markets: an empirical investigation

Abstract Although a variety of approaches for inferring market segments exist, little, if any, effort has been devoted to comparing the relative validity of these approaches. This study conducts two extensive simulation experiments in a scanner data setting to empirically compare and validate alternative mixture model-based procedures for segmenting households using choice behaviors and household characteristics. Compared to existing two-stage approaches, a new method known as the joint approach produced 23–27% less error in estimates of characteristics and 30–38% less error in estimates of choice model parameters. Contrary to conventional wisdom, the joint approach, which simultaneously uses household choice and characteristic data, is shown to be superior even when one is interested in recovering only characteristic-based segments.

Two-stage discrete choice models for scanner panel data: An assessment of process and assumptions

Discrete choice models such as the multinomial logit assume that consumers choose from the full set of alternatives available to them. However, because (i) consumers may not be able to recall or recognize available brands, (ii) consumers may not have the cognitive capacity or mental energy to process information pertaining to all available brands, or (iii) careful consideration of all available brands might be suboptimal from an economic standpoint given the cost of information search, consumers tend to make choices from a relatively small subset of the available brands. This study assesses the process assumptions of existing two-stage models for scanner panel data and their consistency with the actual processes believed to be used by consumers in forming choice sets. After reviewing what is known from two-stage models in scanner data applications, we highlight issues in need of research.

Behavioural differences between consumers attracted to shopping online versus traditional supermarkets: implications for enterprise design and marketing strategy

Despite the dot.com shakeout, online revenues continue to increase and are projected to impose greater pressure on traditional distribution channels. However, there is a striking absence of published empirical work on how consumers attracted to shopping online behave relative to consumers shopping in a traditional store. Such behavioural differences, if they exist, could guide online enterprise design and marketing strategy. This study uses data from both traditional supermarket scanners and an online supermarket to test expected differences in choice behaviours of such consumers. For two product categories, statistically significant differences are found between consumers attracted to shopping online versus traditional supermarkets with regard to the parameters describing the choice process. Compared to traditional supermarket consumers, online consumers are less price sensitive, prefer larger sizes to smaller sizes (or at least have weaker preferences for small sizes), have stronger size loyalty, do more screening on the basis of brand names but less screening on the basis of sizes, and have stronger choice set effects. Many of these differences are found to be prevalent among the majority of online consumers rather than due to the substantially unique behaviour of a minority. Indeed, 11 to 39% of traditional supermarket consumers (depending on the product category) are found to behave like the majority of online consumers whilst 0 to 31% of online consumers are found to behave like the majority of traditional supermarket consumers. Implications of both sets of results for online enterprise design, marketing, and evolution are outlined.

Identifying segments with identical choice behaviors across product categories: An Intercategory Logit Mixture model

Abstract Because consumers are limited information processors seeking to conserve cognitive energy, it is likely that at least some use identical decision heuristics across product categories. This study develops a finite mixture logit model that identifies segments of households with identical behaviors across product categories. The proposed model is shown to fit in-sample choices and forecast out-of-sample choices at least as well as an unrestricted model in which all choice behaviors are independent across product categories. The results show that about 32% of the sample households have choice behaviors that are identical across at least two of the three product categories studied, while the remaining households have choice behaviors that are independent across all three categories. The empirical results show that the segment with identical behaviors is quite price sensitive, not at all sensitive to store feature advertising, and not very brand- or size-loyal. These households are more likely to have larger families and marginally lower incomes and to shop less frequently and spend less per shopping trip. They are also lighter users in two of the three product categories investigated. Implications of the model for manufacturers and retailers are discussed.

Forecasting Performance of Structural Time Series Models

Although theoretical research on the properties of structural time series models has regularly appeared in the literature, there is as yet scant evidence on the forecasting performance of structural models relative to more traditional methods. This study compares the empirical performance of structural time series models to four methods that are similar in complexity, using 111 business and economic time series. The structural approach appears to perform quite well on annual, quarterly, and monthly data, especially for long forecasting horizons and seasonal data. Of the more complex forecasting methods, structural models appear to be among the most accurate.