Data pruning in consumer choice models

Abstract

Common, if not ubiquitous, Marketing practice when estimating models for scanner panel data is to: (a) observe the data, (b) prune the data to a “manageable” number of brands or SKUs, and (c) fit models to the remaining data. We demonstrate that such pruning practice can lead to significantly different (and potentially biased) elasticities, and hence different managerial/practical outcomes, especially in the context of model misspecification. We first justify our claims theoretically by writing the general problem in a classic missing-data framework and demonstrate that commonly used pruning mechanisms (gleaned from current academic Marketing literature) can lead to a nonignorable missing data mechanism. Secondly, we summarize an extensive set of simulations that were run to understand the driving factors of that bias. The results indicate much greater pruning bias in those cases where model fit is poor (small $$R^{2}$$ ), random utility errors are correlated with the covariates, or the model is misspecified (e.g., a homogeneous logit is specified when a mixed-logit is true). Empirically, we also demonstrate our findings on the well-cited and highly utilized fabric softener data of Fader and Hardie ( 1996 ). Our empirical findings suggest a number of estimates that vary according to the way in which the data is pruned including the magnitude of market mix and attribute elasticities, and purchase probabilities, but that the pruning effect is smaller for better fitting models. Copyright Springer Science + Business Media, LLC 2006Nonignorable missing data mechanism, Probabilistic sampling, Scanner panel data, Multinomial logit models, Mixed-logit models,

Modeling Marketing Dynamics by Time Series Econometrics

“…Even though each individual study is careful in stating its own pruning rule, the danger is imminent that these intricacies get lost when citing the results (e.g., the resulting average response or reactivity elasticity), or when meta-analytically combining results from multiple studies into updated empirical generalizations. Finally, Zanutto and Bradlow (2001) demonstrate, in the context of consumer choice models, how data pruning might lead to significantly biased parameter estimates (and hence, potentially different managerial implications) compared to models that are estimated on the full data set. While the Zanutto and Bradlow (2001) study is based on real data wherein true parameter estimates are unknown, Andrews and Currim (2004) conduct a simulation wherein true parameter estimates are known.…”

“…Finally, Zanutto and Bradlow (2001) demonstrate, in the context of consumer choice models, how data pruning might lead to significantly biased parameter estimates (and hence, potentially different managerial implications) compared to models that are estimated on the full data set. While the Zanutto and Bradlow (2001) study is based on real data wherein true parameter estimates are unknown, Andrews and Currim (2004) conduct a simulation wherein true parameter estimates are known. They investigate entity aggregation (analyses at brand vs. brand-size, vs. SKU levels) as well as data pruning decisions in the context of consumer choice models, and find that such data preparation decisions can have significant implications for the assessment of consumer response to price and promotion.…”

“…In response to the increasing number of brands, stores and/or SKUs, it has become common practice to prune the available data to a more manageable number Zanutto and Bradlow, 2001). For example, Nijs et al (2001) consider the competitive reactivity among the top-five brands in the market, Pauwels et al (2002) decompose the long-term effects of price promotions using data from the largest store in four different chains, while Srinivasan et al (2004) look at the profitability of price promotions for the three best-selling brands in 21 product categories.…”

“…Even though more research is needed on this issue, one might expect a similar bias in the context of dynamic (time-series) models. Andrews and Currim (2004) and Zanutto and Bradlow (2001) describe several bias reducing (or even eliminating) strategies that are potentially useful in the context of time-series models.…”

An experimental investigation of scanner data preparation strategies for consumer choice models

“…Some studies sample households, including the entire purchase history of each selected household purchasing only from the selected brands, while others sample purchases of the selected brands and omit purchases of other brands, possibly resulting in incomplete household purchase histories (see Gupta, Chintagunta, Kaul, & Wittink, 1996). Eliminating choice alternatives and/ or households from the data so that it is more amenable to statistical analysis is called data pruning (Zanutto & Bradlow, 2003).…”