Australasian Marketing Journal 13 (2), 2005 27 Repeated Binary Logit: Analysing Variation in Behavioural Loyalty 1. Introduction Brand and store managers are anxious to obtain high “loyalty” from their consumers. And, very often, they equate “loyalty” and repeat purchase behaviour. Assessing the level of repeat purchase, in a category at large or for a specific brand, is a key step in diagnosing a market, and in defining a marketing strategy. We argue that market analysts must go beyond the sheer measurement of “loyalty” through repeat purchase. Rather, they should identify the causes of loyalty, and measure their impact, through a statistical analysis. In this paper, we introduce the Repeated Binary Logit (RBL) model that analyses directly the impact of covariates on loyalty. RBL can be described as an extension of traditional logistic regression. We present several empirical applications of RBL. Finally, we discuss its relationships to several classical models. 2. The Repeated Binary Logit (RBL) Model 2.1 Functional Form We first describe the simple case where there are no covariates impacting loyalty. RBL is constructed from the beta binomial distribution. Let a randomly selected decision maker make k selections from a binary choice set containing alternatives A and B (A could be a specific brand A, and B all the other brands; or A could be a specific store A, and B all the other stores). Let the number of times alternatives A and B are selected be r A and r B where r A + r B = k . Over the population of decision makers, these selection rates are random variables, R A and R B . They have a beta binomial distribution (BBD) conditional on k and on the parameters α A and α B of the beta distribution. The probability density function for the BBD is: Repeated Binary Logit:Analysing Variation in Behavioural Loyalty Cam Rungie & Gilles Laurent Abstract Brand and store managers are anxious to obtain high “loyalty,” as operationalized by repeat purchase behaviour. In this paper, we introduce the Repeated Binary Logit (RBL) model, which analyses directly the impact of covariates on repeat purchase, and which can be described as an extension of traditional logistic regression. We present empirical applications of RBL, and we discuss its relationships to several classical models. Keywords: Brand choice, Buyer behaviour, Choice models, Data mining, Marketing research, Segmentation Equation 1