Modeling of television audience viewing behavior


Student thesis: Doctoral Thesis

View graph of relations


  • Xiaoling LU

Related Research Unit(s)


Awarding Institution
  • Hing Po LO (Supervisor)
Award date2 Oct 2007


Throughout the last half-century, television has established itself as the world’s most important mass medium dominating our leisure hours and our modes of family life. Nowadays television is diversifying its forms, extending its scope, penetrating further into public and private life, consequently becoming an object of social and political concern. Therefore, the activity of viewing television, to which we devote so much of our time, deserves study. The results of audience research have the potential to influence the efforts of policymakers to justify policy about mass media and society. Most importantly, the television industry is a fascinating and challenging business in which billions of dollars is invested annually in programs, commercials, equipment and people. Television audience research can provide much useful information to broadcasters and advertisers to better understand viewer behavior and improve the effectiveness of television broadcasting and advertising. In Hong Kong today, there are two categories of television broadcasting services targeting Hong Kong residents. The first one is the domestic free television program service which is free of charge. The second one is the domestic pay television program service which is available on subscription. The domestic free television program service is very pervasive and influential; its penetration rate is close to 100%. This means a potential market of the whole Hong Kong population, which is over seven million. In Hong Kong, the service provider of television audience measurement is Nielsen Media Research Company. They select a sample of households to represent the Hong Kong population and have the people-meter system connected to the participants’ television sets. The system provides an accurate minute-by-minute record of the viewing data of each individual in the sample. Recognizing the importance of audience research and the huge marketing opportunity for the TV industry, this thesis aims to study in depth the television audience viewing behavior, using both people-meter data in Hong Kong and survey data collected by the researcher. To study the audience viewing behavior of a TV program, the first question one may want to ask is “When and how does a viewer begin to watch it?” This is to study the first-time reach of the program. Cumulative reach is a measure of the overall popularity of a program and is used as a criterion to sell television time, so it is an important topic in television audience research. This thesis studies the first-time reach of TV programs from a diffusion point of view. It proposes a new diffusion model based on the hazard rate analysis, and adopts maximum likelihood estimation method to estimate the parameters. The model fits the people-meter data very well and outperforms the existing models. The use of a diffusion model to study the first-time reach of a TV program is believed to be the first of its kind in television audience research. This basic model is then extended to a proportional hazard model to study the covariate effects on the hazard rate of first-time reach, which is further extended to a proportional hazard mixture model to study the covariate effects on the likelihood of an individual becoming an eventual viewer. The EM algorithm is adopted to estimate the parameters. Empirical results show that both demographic and behavioral variables have significant effects on individuals’ TV viewing behavior in terms of the first-time reach of a program. Advertisers can benefit from these results in targeting their desired customers. As said previously, cumulative reach is used as a criterion to sell television time. It will be very useful for broadcasters and advertisers if one can forecast the initial cumulative reach before a program is aired and update the figures as data become available. A new parameter estimation method and Bayesian forecasting procedure are introduced for this purpose in this study. Empirical results show that the new estimation method is better than the maximum likelihood estimation method and the Bayesian forecasting procedure has a very high accuracy. The study of the first-time reach of TV programs is only the first step of examining viewer behavior. For the success of a program, it is also very important to know the viewers’ repeat viewing patterns and their viewing intensity. In this thesis, a random-coefficient Beta-binomial distribution regression model is developed for modeling repeat viewing patterns and a random-coefficient Beta distribution model is developed for modeling viewing intensity. These models provide very good fit to the people-meter data, and based on the results, viewers can be grouped into different segments. So far the models developed for audience viewing behavior are all based on people-meter data. Though people-meter system accurately records individuals’ viewing information at minute level, one of its limitations is that it cannot tell whether or not viewers appreciate the programs they watch. Not all viewers watch TV in the same manner. The individual-program relationship can be characterized on a continuum from no involvement to fanaticism. Understanding and evaluating these differences may be crucial in assessing the true impact of advertising and other media-based efforts. This study looks at audience satisfaction of TV programs by investigating its antecedents and consequences. Besides two constructs (“cognitive expectation” and “audience involvement”) proposed by previous researchers, “connectedness” and “TV programs’ performance at attribute level” are hypothesized as two additional antecedents of audience satisfaction. Also, people’s judgement on programs’ attributes display loss aversion according to the well-known prospect theory. Consistent with satisfaction research in marketing, “repeat watching intention” and “audience voice” are proposed as two consequences of audience satisfaction. Moreover, viewers’ “likelihood to watch the embedded ads in order not to miss the remaining part of the program” is considered as another consequence of audience satisfaction. Empirical data collected from a telephone survey support all the proposed hypotheses. Finally, the implications for management are discussed in the thesis.

    Research areas

  • Research, Television viewers