Comparer des méthodes

	NBD-Dirichlet Model	BG/NBD Model
Domaine	Marketing	Marketing
Famille	Regression model	Regression model
Année d'origine	1984	2005
Auteur d'origine	Gerald J. Goodhardt, Andrew S. C. Ehrenberg & Christopher Chatfield	Peter S. Fader, Bruce G. S. Hardie & Ka Lok Lee
Type	Stochastic model of category purchase incidence and brand choice	Probabilistic buy-till-you-die model of repeat transactions
Source fondatrice	Goodhardt, G. J., Ehrenberg, A. S. C., & Chatfield, C. (1984). The Dirichlet: A Comprehensive Model of Buying Behaviour. Journal of the Royal Statistical Society: Series A (General), 147(5), 621-655.	Fader, P. S., Hardie, B. G. S., & Lee, K. L. (2005). "Counting Your Customers" the Easy Way: An Alternative to the Pareto/NBD Model. Marketing Science, 24(2), 275-284.
Alias	Dirichlet Model, NBD-Dirichlet, Goodhardt-Ehrenberg-Chatfield Model, Dirichlet Model of Buying Behaviour	Beta-Geometric/NBD Model, BG/NBD, Buy-Till-You-Die Model, Fader-Hardie-Lee Model
Apparentées	4	4
Résumé	The NBD-Dirichlet model is the canonical stochastic model of repeat buying and brand choice in stationary, competitive consumer-goods markets. Introduced by Gerald Goodhardt, Andrew Ehrenberg and Christopher Chatfield in their 1984 Journal of the Royal Statistical Society paper "The Dirichlet," it integrates two processes: how often households buy in a product category, modeled by the negative binomial distribution (NBD), and how those purchases are split across competing brands, modeled by a multinomial-Dirichlet process. From just a few parameters, the model reproduces a remarkably wide set of empirical regularities, including each brand's penetration (how many people buy it), its buyers' purchase frequency, repeat-purchase rates, the share of category requirements each brand earns, and the duplication of purchase between brands. The model encodes Ehrenberg's classic 'laws' of buying behavior, most famously double jeopardy, whereby small brands suffer twice over by having both fewer buyers and slightly less loyal buyers. It assumes a stationary, non-partitioned market with brand choice that looks like sampling 'as if from an urn,' and it serves as a benchmark of what normal, no-loyalty-segmentation buying looks like, against which deviations such as genuine partitioning or excess loyalty can be detected.	The BG/NBD (Beta-Geometric/Negative Binomial Distribution) model is a probabilistic buy-till-you-die model that predicts how many times a customer will transact in the future and whether that customer is still active, using only their past purchase recency and frequency. Introduced by Peter Fader, Bruce Hardie and Ka Lok Lee in their 2005 Marketing Science paper "Counting Your Customers the Easy Way," it was designed as a far simpler alternative to the Pareto/NBD model of Schmittlein, Morrison and Colombo while delivering comparable forecasts. The model couples a Poisson purchasing process, whose rate varies across customers by a gamma distribution, with a geometric dropout process governed by a beta-distributed dropout probability. The key behavioral story is that customers buy at a steady individual rate while alive and become permanently inactive with some probability immediately after any purchase. Because the latent attrition is unobserved, the model infers each customer's probability of still being alive from how recently and how often they bought. Its estimation requires only the (x, t_x, T) summary per customer and can even be fit in a spreadsheet, which made customer-base analysis practical for ordinary analysts.