Building and Interpreting Audience Networks: A Response to Mukerjee, Majo-Vazquez & Gonzalez-Bailon

Abstract

Building audience networks offers a way to apply the tools of network analysis to traditional audience data. Mukerjee, Majó-Vázquez, and González-Bailón (2018) have done this in their article on news consumption. They situate their analysis by making extensive reference to studies that we, and others, have published since 2011. Our colleagues at Annenberg and Oxford claim to have identified and corrected “flaws” in our work that have “started to percolate in the literature” (p. 9). Unfortunately, their approach contains serious conceptual and methodological errors, which Harsh Taneja discussed at length with the corresponding author in an person meeting in May 2017. Surprised at them not being addressed in the published manuscript, we offer a brief response here in the hope that these errors do not themselves begin to percolate in the literature. Our approach to building audience networks conceives of media outlets (e.g., websites) as nodes in a network. The audience shared by any pair of outlets determines if they are linked. The question we faced was how much audience overlap was needed to signal a link. Our answer came from marketing research and its long tradition of analyzing audience duplication. Specifically, the “duplication of viewing law” stipulates that the proportion of the total audience that sees any pair of programs is a function of the percent of the population that sees each program in the pair (i.e., each program’s ratings) (Goodhardt, Ehrenberg, & Collins, 1975, p. 20). Following the law of joint probability, multiplying these ratings determines the level of duplication that would be expected just by chance. Any audience overlap above that level is evidence of audience loyalty. By extension, when using dichotomous ties, we have argued that duplication above randomness signals a link. Ksiazek (2011) first described how to apply these principles to network analysis. What he refers to as “absolute duplication” is directly analogous to the criterion variable in the duplication of viewing law. Likewise, the expected (random) duplication is a probability of a person visiting both outlets in a pair if visiting each is independent of the other.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 13 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.