SAS Macro for fitting Statistical Beta Binomial Models
A simple binomial may not be a good model for data from replicated paired preference tests, as it fails to account for the extra variation, or overdispersion, that can be introduced by differences in the ability of the assessors to discriminate. The Betabin macro fits a model that accounts for this extra variation. There are two equivalent parameterisations of the beta-binomial model that are in common use, in the sensory literature it is usual to determine the parameters gamma & theta, where gamma significantly greater than zero can be used as a direct test for overdispersion. In other fields the alpha-beta parameterisation is more common. The Betabin macro determines all four parameters simultaneously.
The macro is referenced in the RAND Health publication.