Statistical analysis of television audience measurement systems and their implications
This thesis focuses on the identification of measurement errors in television audience measurement systems and their implications to the television industry. BARB in the UK and Mediamat in France are the case studies selected. Measurement errors are identified by assessing the sampling and non sampling operations implemented in these sample surveys but also by integrating the uses that are made of the estimates yielded and the meaning of the concept measured. The thesis argues that the commodities traded on the television market are not audiences but statistics: namely, television ratings, and that this commodity production function is a distinctive feature of these sample surveys. The means by which these commodities are produced by broadcasters, and priced and bought by agencies on behalf of advertisers are examined. It is shown that prices attached by buyers to these commodities rely on factors that are grounded in economic rather than statistical considerations, and that the commodities are not known at the moment they are priced but need to be projected. It is argued that television audience measurement systems are based on a construct of the audience that relies entirely on assumptions and operational definitions. The consequences of this approach to measurement on the uses to which the data are currently put and on the capacity of the industry for making predictions are examined. It is suggested that the data collection technique implemented in these measurement systems is of decisive importance to the sampling design and the economy of the medium. The validity of the people-metering data collection technique currently in use is assessed. It is argued that this technique imposes the use of samples that are not valid from a statistical viewpoint. Components of variance for audience estimates are identified and a method for searching for patterns in standard errors for audience estimates is proposed. The implications of the findings in the new television environment are developed.