Skip to main navigation Skip to search Skip to main content

Empirically modeling of audience behavior in the television industry

  • Xiaohong LIU

    Student thesis: Doctoral Thesis

    Abstract

    Empirical modeling in marketing has been widely developed in recent decades. Researchers use quantitative techniques to investigate marketing issues by analyzing industries' real data. In particular, many studies are generated in the television industry, including viewer rating and choice, repeat viewing, joint decisions by household members. This thesis also focuses on the television industry, and uses television viewing data to model viewer behavior. The data in this thesis were provided by Television Broadcasts Ltd. (TVB), which is the dominant producer of local Cantonese television programs in Hong Kong; its Chinese JADE channel (TVBJ) maintains a remarkable average of 80 percent of Hong Kong television audience share. The data are collected by Nielsen Media Research. Nielsen Media Research selects a sample of about 600 households, which can be viewed as a good representative sample of the Hong Kong population. The dataset consists of different types of households and individuals from all over Hong Kong. According to the people-meter system, each individual’s viewing information is recorded at very accurate minute level. The data also consist of detailed demographic information about each person, such as gender, age, work status, education and income. For this thesis, one year's data, from July 19, 2004 to July 24, 2005, were made available. The study concentrated mainly on two areas: 1) the impact of viewer learning on viewer behavior in watching a drama series; 2) household members' interaction in group-viewing decision-making. The study also made an empirical comparison between Hong Kong and United States households' viewing behavior. A two-stage process is proposed to investigate the audience's viewing behavior during a television program series. Using this approach, the study models empirically viewer choice and watching duration behavior. According to the particularity of television programming, viewers' decisions can be summarized as follows: 1) whether or not to watch the program; and 2) if viewers choose to watch it, how long will they keep on watching. A probit model, which is a typical discrete choice model, is used to capture a viewer's decision whether or not to watch an episode of a drama. Should the viewer choose to watch, a proportional hazard model is applied to capture how long he/she keeps on watching the episode. To demonstrate the connection between the two decisions by the viewer, namely whether to watch and for how long to watch, the utility function of the probit model is assumed to be correlated with the covariate function in the proportional hazard model by partly sharing independent variables or having a correlation link. To capture the evolution of viewers' preference for a certain program, a learning structure in the two-stage model is proposed. Viewers are assumed to be uncertain about the quality of programs aired by television stations because of imperfect information. Past experience can affect viewers' information set and also their choice and watching duration decisions. Different from existing learning models, the learning structure in this study includes two unique characteristics due to the uniqueness of television watching behavior. Firstly, existing learning models assumed that all signals perceived by different consumers about product quality followed an independent and identical distribution with the same true mean quality. This implied that when consumers had enough information and their uncertainties about product quality had almost disappeared, their preferences would stay the same and would equal true mean quality. However, this may be not true in reality. In the learning model proposed here, viewer heterogeneity rather than homogeneity regarding the true mean quality of a television program is allowed. The empirical results show that considering heterogeneity in true mean quality is very important. Secondly, existing learning models assumed that perceptions about product quality can be either fully unobserved or fully observed. In this study, signals of viewers' perceptions are partially observed: viewers' watching duration can be regarded as an indicator of their perceptions because of the high correlation between them. The model is able to cope with this kind of situation through its complete stochastic framework. Regarding viewers' decisions on how long to watch for, the study found that viewers are more sensitive to current drama quality than to previous evaluations of the drama. The second part of the study concerned group decision-making by household members. The "household" has been recognized as an important decision-making unit. Because of strong ties among household members, intra-household interactions are likely to be significant in family decision making. This study focuses on family members' television viewing behavior and proposes a structural model of household resource allocation incorporating group decisionmaking. The household utility function can be assumed to combine individual utilities two ways: first, individual members obtain a weighted average of their separate utility functions and then maximize the resulting mixture function, an "average then maximize" process or the Harsanyi solution. This process leads to an additive way of combining individual members' utility functions to form the household utility function. Second, each individual member locates his/her most preferred outcome and then compromises by averaging along the negotiation frontier, a "maximize then average" process or the Nash solution. This process leads to a multiplicative way of combining individual members' utility functions to form the household utility function. The Kuhn-Tucker theory is then applied to solve the optimal resource allocations on goods consumed by household members individually and household members jointly, including the interior solution and the corner solution. The empirical results of the present study indicate that the wife plays a dominant role under the assumption of the Nash group-decision strategy but the relative influence of the husband and wife is more balanced under the assumption of the Harsanyi group-decision strategy. In addition, the Harsanyi strategy is more suitable for modeling household members' joint television-viewing behavior. Based on the additive way of combining individual utilities, the viewing behavior of Hong Kong and United States households is also examined. Empirical results indicate that, on average, wives from Hong Kong households have a higher influencing power in group-viewing decisions than their husbands, while husbands and wives in United States' households have almost equal power in group-viewing decision making. Finally, the previous viewing choice of an individual has a positive effect on the marginal utility of that person, but has no cross effect on the marginal utility of his/her spouse.
    Date of Award4 Oct 2010
    Original languageEnglish
    Awarding Institution
    • City University of Hong Kong
    SupervisorHing Po LO (Supervisor)

    Keywords

    • Research
    • Hong Kong
    • Television viewers
    • China

    Cite this

    '