Revealed Preference Attribute Modelling using Repeated Purchases

Multinomial logit (MNL) models are routinely used in marketing to estimate the impact of attribute levels on the potential success of new alternatives in a market. In the paper the MNL model is extended, using the Dirichlet Multinomial Distribution (DMD), to analyse the attributes involved in the patterns in repeated purchase revealed preference data. The result is a model for which MNL is a special case and which estimates the impact of attribute levels on market share and loyalty. The success of a new alternative can now be estimated in terms of market share and loyalty simultaneously. The paper uses a basic model which demonstrates the process but is too simple to generate accurate or practical estimates. This is a limitation of the specific model and not the process.