Modelling Viewing Behavior Towards Television Programs And Advertisements

With the fast development of electronic technologies, the television industry has greatly changed the living patterns of all walks of life. Television advertising, which is the main benefit resource of television stations, has become a popular issue discussed by many researchers. This thesis investigates television audience viewing behavior towards programs and advertisements respectively.This dissertation focuses on two areas:(1) modeling the dynamic multiple choice behavior of television audience to predict and improve program audience ratings; and (2) modeling commercial viewing patterns to convert minute information into second information to evaluate advertisement (ad) effectiveness.In the first area, we propose a new dynamic multiple programs adoption (DMPA) model to investigate audience multiple-category choice behavior during a peak-hour period. The simultaneous demand theory of Kim et al.(2002) is adopted to mimic the real situation. A state-space model that makes use of both the quantitative and qualitative dynamics of viewing is proposed to solve the dynamic problem. Here, quantitative dynamic captures the dynamic of variety choice by exogenously imposing dependence between past and current choices. Qualitative dynamic provides a behavioral explanation for why current choice should depend on past choices according to the Bayesian rules.In addition, most studies using the Bayesian theorem assume that the memory of past usage remains stable. However, the capability of recalling prior information may diminish over time. Forgetting limits the full use of prior information:people rely on partial prior information. Thus, an exponential decay function is used to modify the original Bayesian learning rules.The results show that the popular multiple-choice model in economics also performs well for studying audience viewing behavior. Including Bayesian learning in the multiple-choice model significantly improves model performance and prediction accuracy. Moreover, taking the forgetting effect into account when studying audience learning renders the Bayesian learning model much more accurate in real situations. Managerial implications are developed to provide useful insights for guiding advertisers in advertising placement decisions, and also to help the television industry to attract more viewers (and thus advertisers). The other area involved in this study is to model viewing patterns when television commercials are shown and then use them to convert the minute-by-minute people-meter data into second-by-second information to evaluate ad effectiveness. This information includes the second-by-second commercial audience ratings and individual ad viewing duration in the unit of seconds, which are not available from people-meters directly. The different viewing patterns between programs and commercials cause advertisers to pay more attention to advertisement ratings than program ratings. However, people-meters record viewing data in the unit of minutes, which is not accurate enough for advertisement ratings because segments of television advertisements are very short, usually of lengths of less than30seconds. We are endeavoring to develop a methodology of modeling viewing behavior to convert the minute-by-minute people-meter data into second-by-second information.