Rating forecasts for television programs

This paper investigates the effect of aggregation and non-linearity in relation to television rating forecasts. Several linear models for aggregated and disaggregated television viewing have appeared in the literature. The current analysis extends this work using an empirical approach. We compare the accuracy of population rating models, segment rating models and individual viewing behaviour models. Linear and non-linear models are fitted using regression, decision trees and neural networks, with a two-stage procedure being used to model network choice and viewing time for the individual viewing behaviour model. The most accurate forecast results are obtained from the non-linear segment rating models.