Observed and Unobserved Preference Heterogeneity in Brand-Choice Models (original) (raw)
Related papers
2006): “Observed and Unobserved Preference Heterogeneity
2015
In deciding what brand to buy consumers trade off between how valuable each brand is to them and its price. Scanner data based brand choice models that evaluate this trade off and allow for unobserved heterogeneity are a very popular topic. We extend this line of research by explicitly incorporating brand preferences rather than just utilizing brand specific constants and past purchase behavior to proxy for them. We illustrate our model using a unique dataset that combines survey and scanner data collected from the same individuals. The addition of individual specific brand preference information to a standard scanner choice model that incorporates brand specific constants, loyalty, promotion and price significantly improves how well choices are explained and predicted. Moreover, this “observed ” heterogeneity via inclusion of individual specific brand preferences better explains choice than does “unobserved” heterogeneity in the standard scanner model parameters. This is largely du...
Disentangling Preferences and Learning in Brand Choice Models
Marketing Science, 2012
I n recent years there has been a growing stream of literature in marketing and economics that models consumers as Bayesian learners. Such learning behavior is often embedded within a discrete choice framework that is then calibrated on scanner panel data. At the same time, it is now accepted wisdom that disentangling preference heterogeneity and state dependence is critical in any attempt to understand either construct. We posit that this confounding between state dependence and heterogeneity often carries through to Bayesian learning models. That is, the failure to adequately account for preference heterogeneity may result in over-or underestimation of the learning process because this heterogeneity is also reflected in the initial conditions. Using a unique data set that contains stated preferences (survey) and actual purchase data (scanner panel) for the same group of consumers, we attempt to untangle the effects of preference heterogeneity and state dependence, where the latter arises from Bayesian learning. Our results are striking and suggest that measured brand beliefs can predict choices quite well and, moreover, that in the absence of such measured preference information, the Bayesian learning behavior for consumer packaged goods is vastly overstated. The inclusion of preference information significantly reduces evidence for aggregate-level learning and substantially changes the nature of individual-level learning. Using individual-level outcomes, we illustrate why the lack of preference information leads to faulty inferences.
Management Science, 2005
We investigate the role of potential weekly brand-specific characteristics that influence consumer choices, but are unobserved or unmeasurable by the researcher. We use an empirical approach, based on the estimation methods used for standard random coefficients logit models, to account for the presence of such unobserved attributes. Using household scanner panel data, we find evidence that ignoring such time-varying latent (to the researcher) characteristics can lead to two types of problems. First, consistent with previous literature, we find that these unobserved characteristics may lead to biased estimates of the mean price response parameters. This argument is based on a form of price endogeneity. If marketing managers set prices based on consumer willingness to pay, then the observed prices will likely be correlated with the latent (to the researcher) brand characteristics. We resolve this problem by using an instrumental variables procedure. Our findings suggest that simply ig...
An experimental investigation of scanner data preparation strategies for consumer choice models
International Journal of Research in Marketing, 2005
Over the past two decades, marketing scientists in academia and industry have employed consumer choice models calibrated using supermarket scanner data to assess the impact of price and promotion on consumer choice, and they continue to do so today. Despite the extensive usage of scanner panel data for choice modeling, very little is known about the impact of data preparation strategies on the results of modeling efforts. In most cases, scanner panel data is pruned prior to model estimation to eliminate less significant brands, sizes, product forms, etc., as well as households with purchase histories not long enough to provide information on key consumer behavior concepts such as loyalty, variety seeking, and brand consideration. Further, product entity aggregation is usually part of data preparation also since hundreds of SKUs are available as choice alternatives in many product categories. This study conducts an extensive simulation experiment to investigate the effects of data pruning and entity aggregation strategies on estimated price and promotion sensitivities. Characteristics of the data that may moderate the effects of data preparation strategies are also manipulated. The results show that data preparation strategies can result in significant bias in estimated parameters. Based on the results, we make recommendations on how the model builder can prepare scanner panel data so as to avoid significant biases in estimated price and promotion responses. D
Purchase-Frequency Bias in Random-Coefficients Brand-Choice Models
2005
Conventional random-coefficients models of conditional brand choice using panel data ignore the dependence of the random-coefficients distribution on the purchase frequencies. We show that this leads to biased estimates and propose a conditional likelihood approach to obtain unbiased estimates. Unlike alternative approaches that require observation of “no-purchase” occasions, our proposed method relies only on purchase data.
Incorporating State Dependence in Aggregate Brand-level Demand Models
2012
Empirical investigations of household-level data show that state dependence is a signicant determinant of consumers’ choices in frequently purchased product categories. In this paper we examine how one can incorporate state dependence in an aggregate demand model for brandlevel data which are more often available and frequently used in applied research in marketing and economics. We dene state dependence as consumers receiving higher utility when buying the same brand in two consecutive periods. The main challenge we face is that we do not observe what portion of consumers who chose a given brand had chosen the same brand in the previous period. We overcome this challenge by constructing the brand choice probability using observed market shares from the previous period and the law of total probability. Through Monte Carlo simulations we show that our model is a good approximation of household-level models. We apply our method to estimating demand for salty snacks using a multi-marke...
Another look at loss aversion in brand choice data: Can we characterize the loss averse consumer?
International Journal of Research in Marketing, 2005
Much research has focused on the effects of reference prices on brand choice decisions using scanner panel data. The theory and application are well-documented and accepted. However, researchers have found contrary results on the existence of loss aversion in consumer goods markets. Loss aversion is a phenomenon based on the reference dependent theory that consumers respond more to losses (reference price b price) than to gains (reference price N price). The mixed results on the existence of loss aversion can be a result of not adequately accounting for consumer heterogeneity in response to marketing effects. Therefore, we focus our analysis on loss aversion and adequately accounting for consumer heterogeneity. We estimate a reference dependent model with a mixed logit specification that allows for a continuous distribution of response heterogeneity in the population. We use Gibbs Sampling to obtain individual estimates. Our estimation results from two different consumer goods categories, which show that the degree of loss aversion is small after properly accounting for heterogeneity. Further, we accomplish a posterior analysis and investigate whether the individual response to gains and losses can be attributed to consumer specific characteristics. The relation of the estimated individual specific variables to households' sociodemographic and psychographic variables as well as to observed purchase behavior reveal interesting insights into which consumers respond more or less to price deviations from their reference point. Hence, our results are important for the development of effective pricing strategies and the timing of price promotions. D
Incorporating Responsiveness to Marketing Efforts in Brand Choice Modeling
Econometrics, 2014
We put forward a brand choice model with unobserved heterogeneity that concerns responsiveness to marketing efforts. We introduce two latent segments of households. The first segment is assumed to respond to marketing efforts while households in the second segment do not do so. Whether a specific household is a member of the first or the second segment at a specific purchase occasion is described by household-specific characteristics and characteristics concerning buying behavior. Households may switch between the two responsiveness states over time.
The price consideration model of brand choice
Journal of Applied Econometrics, 2009
The workhorse brand choice models in marketing are the multinomial logit (MNL) and nested multinomial logit (NMNL). These models place strong restrictions on how brand share and purchase incidence price elasticities are related. In this paper, we propose a new model of brand choice, the "price consideration" (PC) model, that allows more flexibility in this relationship. In the PC model, consumers do not observe prices in each period. Every week, a consumer decides whether to consider a category. Only then does he/she look at prices and decide whether and what to buy. Using scanner data, we show the PC model fits much better than MNL or NMNL. Simulations reveal the reason: the PC model provides a vastly superior fit to inter-purchase spells.