Optimizing Television Program Schedules Using Choice Modeling

Abstract

The authors examine the rescheduling of television programs to maximize the total ratings for one network across a week. The key idea is to design a choice experiment in which television programs are rescheduled and presented to respondents. Respondents read these program schedules (much like the regular TV Guide listings) and give their preferences, including not watching any of the listed programs. Because there are potentially billions of possible schedules, the authors give a procedure for designing a fractional factorial experiment that can accommodate both programs of varying length and constraints on eligible program times. The authors also develop a latent class multinomial logit model for modeling program preferences and present a validation of our experimental procedure and the model. They also present an empirical test of the procedure in which they use the model to predict ratings for all the possible program schedules, not just those constituting the choice sets. In this example, the optimum schedule increases the predicted total weekly ratings during prime time by 18% for a network. The projected increase in total weekly ratings is achieved without the network needing to purchase any new programs; all it needs to do is reschedule eight programs in its existing prime-time lineup.