Matthew Gentzkow has made fundamental contributions to our understanding of the economic forces driving the creation of media products, the changing nature and role of media in the digital environment, and the effect of media on education and civic engagement. He has thus emerged as a leader in a new generation of microeconomists applying economic methods to analyze questions that were historically analyzed by non-economists. His empirical work combines novel data, innovative identification strategies and careful empirical methods to answer questions at the interface of economics, political science, and sociology. This work is complemented by significant theoretical work on information, communication, and persuasion. Gentzkow, both on his own and in collaboration with his frequent co-author, Jesse Shapiro, has played a primary role in establishing a new and extremely promising empirical literature on the economics of the news media.
I think of Matt’s work as straddling political economy and industrial organization. The latter area is extremely mature and yet Matt, often with Jesse Shapiro, has made extremely original contributions with a focus on the media. (I note that he has produced interesting papers on brand preferences more recently, closer to traditional IO concerns.)
I am most familiar with his work on media bias and reputation (JPE, 2006). Consumers want to take an action that fits the state of the world. They have some prior distribution over the states. The media can produce a signal that potentially informs this decision. The consumers are also trying to assess the quality of the media as well as take optimal actions. That is, there is a second dimension of uncertainty over the “type” of the media. If the media’s signal has low probability given the consumers’ prior, they will infer the media is bad quality. Hence, there is an incentive for the media to lie or “spin” to fit the prior of the consumers. There are many other results but this key logic underlies all of them. The model uses differing priors which is a non-standard assumption. It is non-standard as it may not yield a tractable model or crisp, falsifiable results (among other more philosophical reasons). But Gentzkow and Shapiro show that it is possible to handle such a setting.
I am less qualified to judge empirical work but I did see Matt and Jesse’s paper on ideological separation on the internet. Cass Sunnstein has claimed that the internet has created polarization of opinions as people just go to websites that fit their ideological preferences. Using data on internet use, reporting of ideological preferences, and measures of media bias of a site based on fraction of conservatives and liberals visiting the site, Gentzkow and Shapiro calculate conservative exposure of conservatives and liberal. The difference is a measure of isolation. This number is surprisingly low.
Finally, Kamenica and Gentzkow have introduced a new kind of principal-agent/receiver-sender into economic theory. The receiver’s (e.g. a judge or jury) action depends on the state of the world. The sender (e.g. a prosecutor) has his own preferences and chooses an information structure, i.e. a distribution of signals as a function of the state, to influence the receiver. They show that the sender will deliberately choose a noisy signal structure to persuade the receiver. For example, the prosecutor may choose an investigation that deliberately delivers guilty signals for innocent defendants. If this probability is low compared to the probability of guilty signals for guilty defendants, the judge will convict any defendant for whom there is a guilty signal, even though the defendant might be innocent. If the prosecutor wants to maximize the rate of conviction, by using noisy signals, we can achieve a higher conviction rate than with perfect signals. This paper has stimulated research on “Bayesian persuasion”.
All this work is quite original and quite sophisticated. The AEA is right in concluding:
His work is creative without sacrificing quality. He has established himself as a role model in both substance and execution.