| Estimating the number of individuals who might see or hear an advertisement, and the number of times these individuals will be exposed, is a starting point of advertising management for effective media decision-making. A reach/frequency estimation model is one of methods providing this information.;The purpose of this dissertation is to explore a new multivariate media reach/frequency estimation model which improves the limitations and accuracy of the best currently-available, non-proprietary models. In addition, to test the robustness of the models' performances and their applicability to different nations, three different data sets, two U.S. data sets and one New Zealand data set were used; 1979 and 1984 Simmons Market Research Bureau (SMRB) data and AGB:McNair New Zealand Ltd. National Media Survey.;A new model, the Conditional Beta Distribution model, is developed and tested. For direct comparisons, seven existing models and one modified version of an existing model are also included in this study--the Beta Binomial Distribution model, the Canonical Expansion model, the Dirichlet Multinomial Distribution model, the Hofmans Beta Binomial Distribution model, the Hofmans Geometric Distribution, the Morgenzstern Sequential Aggregation model, the Poisson Binomial Distribution model, and the Modified Dirichlet Multinomial Distribution model. A thorough empirical test using three data sets and 1,726 sample schedules proves the superiority in accuracy of estimating reach and exposure probabilities of each frequency level across all data sets.;In addition to the model evaluations, a survey of media directors at the 200 largest U.S. advertising agencies is included to update practitioners' perspectives on media exposure models and media planning. It shows increased awareness of effective reach concept, importance of frequency distribution, and increased concern about the accuracy of media performance. |
