Overview
Greg M. Allenby is a professor at the Max M. Fisher College of Business, Ohio State University. He is a leading figure in Bayesian marketing econometrics, known for developing hierarchical Bayes methods for modeling consumer heterogeneity in discrete choice panel data. Together with Peter Rossi and Robert McCulloch, he pioneered the hierarchical Bayes multinomial probit (HB-MNP) framework that replaced finite mixture models as the standard approach to household-level demand estimation in quantitative marketing.
Key Contributions
- Allenby and Rossi (1999): Pedagogical synthesis of the HB-MNP framework; advocates continuous normal mixing over finite mixture models for consumer heterogeneity; demonstrates decisively (Δlnp=2070) that the hierarchical Bayes continuous mixture fits the ketchup scanner panel data far better; derives optimal targeted price reductions using full posterior uncertainty propagation. See Multinomial Probit and Bayesian Hierarchical Model.
- Allenby-Rossi (1991): "Quality Perceptions and Asymmetric Switching between Brands" — nonhomothetic logit model; marginal utility ϕi(u)=exp(ai−kiu) where ki is the data-revealed quality rank; asymmetric switching: premium-brand promotions attract more switchers; BIC=−4,944.5 (17 params) beats nested logit −5,131.7 (58 params); correlated errors in standard logit explained as nonhomotheticity misspecification; margarine scanner panel (517 households). See Nonhomothetic Preferences.
- Allenby-Lenk (1994): Logistic normal regression model for scanner panel data — an early alternative formulation using logistic normal heterogeneity rather than normal random coefficients.
- Rossi-McCulloch-Allenby (1996): "On the value of household purchase history information in target marketing" — comprehensive treatment of targeted couponing strategies using the HB-MNP estimates.
- Allenby-Ginter (1995b): "Using Extremes to Design Products and Segment Markets" — two-level HB conjoint with covariance matrix D; the "extremes" argument: low annual fee generates 24.2% tail mass above utility threshold vs. 19.2% for low interest despite equal means; 111 vs. 58 respondents in segment (>70% membership); cov(low fee, out-of-state) = 5.34; HB MAD=0.235 vs. latent class 0.282 vs. aggregate 0.311; 6-block Gibbs sampler. See Allenby-Ginter (1995) and Conjoint Analysis.
- Allenby-Ginter (1995a): "The Effects of In-Store Displays and Feature Advertising on Consideration Sets" — companion study on retail display effects.
- Allenby-Arora-Ginter (1998): "On the heterogeneity of demand" — further extensions of the continuous heterogeneity framework.
- Allenby, Rossi, and McCulloch (2005): Practitioners guide to HB models in marketing; develops four-block Gibbs for random-effects logit conjoint; credit card study (N=946, 14,799 comparisons); demonstrates that Vβ off-diagonal structure determines tail-area market potential (low fee 7.5% vs. low interest 4.5% above profitability threshold). See Conjoint Analysis.
- Edwards and Allenby (2003): High-dimensional multivariate binomial probit for pick any/J survey data; postprocessing-based identification strategy (fit unconstrained Ω, standardize draws to recover R); PCA/canonical-correlation/conditional-probability applications; superiority over tetrachoric correlations in small samples and zero-cell cases. See Multivariate Probit and Edwards-Allenby (2003).
- Rossi and Allenby (2003): Programmatic synthesis of Bayesian statistics for marketing (Marketing Science 22(3): 304–328); three-component framework (within-unit likelihood, cross-unit heterogeneity, decision action); latent-variable unification of discrete-outcome models via data augmentation; extends heterogeneity to mixture-of-normals, structural (observation-level) heterogeneity, and spatial/network models; critiques BIC and plug-in decision-making. See Bayesian Hierarchical Model.
Related