When Citizens Demand More Censorship: Evidence from the U.S. and China

• In Event: Understanding and Addressing Online Political Incivility, Hate, and Polarization

Abstract

Existing research on information control and content moderation on social media generally focuses on the roles of governments (in authoritarian contexts) and platforms reacting to false information (in democracies). However, observational evidence suggests that social media users themselves may sometimes prefer greater levels of censorship. For example, in China, despite the stringent information control, many internet users flag and report (jubao) others’ postings for political reasons, causing the additional removal of hundreds of millions of posts by the government or the social media company. In the United States, while polls show that most Americans think social media platforms censor their political viewpoints, approximately 30 percent of the respondents from a recent YouGov sample believe that there is too little censorship. This phenomenon is puzzling, as some citizens do not simply dismiss or refute a message that they do not approve but instead prefer a more costly approach of seeking its disappearance, which apparently does not increase their utility and may backfire and eventually lead to the censorship of information of interest to them.

We propose a theory to account for this phenomenon. Our theory characterizes demand for more censorship as a general equilibrium effect of information control: When censorship has become an established part of their information environment, people will utilize it in ways that may not be in accordance with the interests or intention of the regime or the social media platform that prescribes censorship in the first place. This includes facilitating the deletion of posts that they believe “ought” to disappear from social media platforms — although they themselves do not believe in (and will not be affected by) their content, the existence of these posts will influence other actors with prior beliefs and/or incentives different from theirs, and this community-level effect will jeopardize their utility more than the discounted risk of censorship being strengthened in the future at their expense.

We relate this behavior to the prediction of the third-person effect hypothesis in sociology and communication, i.e., people tend to believe that information from mass media will affect others more than themselves and may therefore support stricter information control. Therefore, people may seek more information control if it can prevent the information they regard as inaccurate, inappropriate, or offensive from being seen by others, including those with uninformative priors (the masses), those in a politically opposing out-group who may exploit the information to the detriment of their in-group interests, or those in the in-group who may be swayed by the information and reduce their support. As such, when the perceived community-level risks of certain content on social media increase, so will the demand for more information control. This is consistent with our empirical analysis based on Google Trends, which demonstrates a substantial 90 percent increase in search volume of queries on how to report on social media in the aftermath of the January 6, 2021 attack on U.S. Capitol.

We conduct at least two sets of survey experiments that empirically examine the two major propositions of our theory in the United States and China. In the experiments, we randomize background information that includes different framing of the extent of social media censorship, so that the respondents in the treatment group will tend to perceive censorship as significantly more prevalent than those in the control group. Similarly, we also randomize the respondents’ perception of external threats (party polarization and foreign interference in the US. version; foreign government’s attitudes toward China in the Chinese version). Then we ask respondents to react to various political postings, including both true and false information on politics (targeting Democratic and Republican politicians respectively in the U.S. version; criticism and praises of the Chinese government in the Chinese version) and COVID-19. For each post, we randomize, in a conjoint fashion, the reception of social media postings, including the numbers of “like”s and the comments’ attitudes. We then elicit the respondents’ censorship desire and actions from surveys. Finally, we inquire about their thought on the current censorship policies — whether they would ideally desire more, less, or equal amounts of censorship as practiced in the status quo. (We may also conduct a field experiment in China among social media users.)

Our research contributes to the literature by laying out an important downstream effect of authoritarian information control that alters citizen behavior as they adapt to the equilibrium of a regulated information environment. Our research also has policy implications relevant to social media companies and regulators in democracies, where content moderation has become an increasingly pressing issue. 

Authors

• Qitong Cao, Stanford Graduate School of Business

• Yunpeng Bai, Yale University