Knowledge that Transforms

prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at

Two simulations assessed the statistical bias, consistency, and efficiency of 4 different signal detection theory (SDT) sensitivity measures; a corrected-hit probability, the traditional d′ statistic, and 2 nonparametric measures collected from a collapsed-data procedure. Overall, results reinforce evidence that collapsed procedures produce relatively unbiased and efficient estimators. Recommendations for the best approach to using SDT for advertisement recognition testing are offered.

In assessing the performance of a choice model, we have to answer the question, “Compared with what?” Analyses of consumer brand choice data historically have measured fit by comparing a model's performance with that of a naive model that assumes a household's choice probability on each occasion equals the aggregate market share of each brand. The authors suggest that this benchmark could form an overly naive point of reference in assessing the fit of a choice model calibrated on scanner-panel data, or any repeated-measures analysis of choice. They propose that fairer benchmarks for discrete choice models in marketing should incorporate heterogeneity in consumer choice probabilities, evidence for which is by now well documented in the marketing literature. They use simulated data to compare the performance of parametric and nonparametric benchmark models, which allow for heterogeneity in consumer choice probabilities, with the performance of the aggregate share-based benchmark model, which assumes consumers are homogeneous in their choice probabilities. They also assess the performance of two previously published consumer behavior models against the proposed fairer benchmark models that allow for heterogeneity in consumer choice probabilities. They find that one provides a significantly better fit than their more conservative benchmark models and the other performs less favorably.

The authors develop a two-stage forecasting methodology for estimating the sales responses to marketing and environmental variables when it is likely that their impacts will change unpredictably over time. The methodology is based on an integrated least squares procedure that uses regression analysis in Stage 1 to estimate the coefficients of the controllable and environmental variables in combination with a Stage 2 discounted least squares smoothing procedure that updates key parameters in response to changing market conditions. The effectiveness of the methodology is demonstrated by applying it to weekly sales data from a major retail chain.