EVALUATING THE MULTIVARIATE BETA BINOMIAL DISTRIBUTION FOR ESTIMATING MAGAZINE AND INTERNET EXPOSURE FREQUENCY DISTRIBUTIONS

This study introduces a newly developed media exposure model, called the Multivariate Beta Binomial Distribution (MBD) Model. The MBD Model is developed based on three probability theories-Hyett's Beta Distribution (1958), Waring's Theorem (1792), and Greene's Personal Media Probability Method (1970)-to overcome the major limitations of the previous exposure models such as generation of negative probabilities of exposure (meaning that the proportion of people exposed to an ad is less than zero, which is illogical) and the models' incapability of handling asymmetrical media schedules (which is impractical). The MBD Model is the most advanced multivariate media exposure model, designed to work for not only the traditional media classes, but also the Internet. An illustrative sample of numerical examples demonstrates the accuracy of the MBD Model, based on magazine audience data. Then, the performance of the MBD Model is compared to that of eight existing media exposure models, tested on 440 randomly selected media schedules for a Web audience comScore data set. The findings indicate that the MBD Model performs well in both prediction and in overcoming the limitations of the prior models.