Models of Competitive Television Ratings

Abstract

In a time series analysis of prime time network television viewing data a weekly trigonometric regression model is used to predict the total network viewing for specific times on specific days of the year. These predictions are very accurate. Four network share models, each based on different assumptions about how the activity of television viewing interacts with nonviewing activities and the strength of program loyalties, are used to allocate the total network prediction to specific program predictions. The predictions of the network share model are compared with the highly accurate published predictions of an expert who explained his predictions on the basis of demographic and program content considerations. The predictions of the network share models, which do not utilize demographic or program content information, compare reasonably well with the expert judgment predictions. The authors suggest how demographic, program content, and competitive bridging information would be incorporated within the framework provided by the time series models to produce advanced network scheduling models.