Public Media Do Serve the State: Evidence from a Randomized Experiment

東京経済大学 経済学部

with 北村周平

This study examines the demand and supply of information in public media within democratic settings using natural language processing and a randomized experiment. By applying unsupervised machine learning to a corpus of approximately 200,000 TV programs, we find that public media tend to favor the government’s views over those of foreign governments. However, this bias does not extend to ideological divisions along partisan lines. To further explore the effects of public media exposure, we randomly increased access to public media programs. Participants in the treatment group evaluated government policies more favorably than those in the control group, suggesting that public media influences public opinion. Building on these findings, we develop a structural model that accounts for strategic interactions between broadcasters and conduct counterfactual simulations. In a counterfactual situation where all consumers were subjected to the same environment as the treatment group, equilibrium results indicate increased exposure to public media content, bringing it even closer to the government’s position. In contrast, reducing overall TV demand by half leads all broadcasters to produce more balanced information.