A Log-Linear Model for Predicting Magazine Audiences

Abstract

A log-linear model for predicting magazine exposure distributions is developed and its parameters are estimated by the maximum likelihood technique. The log-linear model is compared empirically with the best-found model for equal-insertion schedules, one of Leckenby and Kishi's Dirichlet multinomial models. For unequal-insertion schedules the log-linear model is compared with the popular Metheringham beta-binomial model. The results show that the log-linear model has significantly smaller prediction errors than either of the other models.