The Impact of Media Censorship: Evidence from a Field Experiment in China

The Impact of Media Censorship: Evidence from a Field Experiment in China Yuyu Chen David Y. Yang January 22, 2018 Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 1 / 39

Motivation Media censorship is a hallmark of authoritarian regimes. Scholars have long suggested that censorship is key to popular support and stability of these regimes. Nonetheless, direct empirical evidence about the effect of removing censorship is limited. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 2 / 39

This paper In this paper, the authors try to answer empirically two questions: Does providing access to an uncensored Internet lead citizens to acquire politically sensitive information? Does the acquisition of politically sensitive information change citizens beliefs, attitudes, and behaviors? The authors conduct a field experiment in China in order to answer these two questions. They randomly assign 1,800 students in Beijing to: (i) a tool to bypass censorship, and (ii) a temporary encouragement to visit international news outlets otherwise blocked. During the period, they repeatedly measure a wide range of outcomes. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 3 / 39

Literature This paper relates with two literatures: Political economy of mass media (in particular, the literature of mass media that emphasizes the importance of demand-side factors). The empirical literature on the endogenous formation of beliefs and preferences when authoritarian regimes have a direct incentive to intervene. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 4 / 39

Context The Great Firewall The Chinese government has imposed several administrative and legal regulations in order to impose severe business and political cost from publishing content that the state deems threatening and objectionable. As a result, domestic media outlet content is either routinely self-censored during the editorial process, or censored and filtered according to orders from the Propaganda Department of the Communist Party of China. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 5 / 39

Randomization Access and encouragement treatments 1,800 undergraduates students from two universities in Beijing. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 6 / 39

Randomization Encourage The authors design two phases of the encouragement treatment. a. Purely informational: It consist of 4 newsletters introducing students to a variety of foreign websites that are blocked by the Great Firewall. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 7 / 39

Randomization Encourage Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 8 / 39

Randomization Encourage The authors design two phases of the encouragement treatment. a. Purely informational: It consist of 4 newsletters introducing students to a variety of foreign websites that are blocked by the Great Firewall. b. Quizzes with monetary awards: These quizzes aim to encourage students to actually visit the Chinese edition of The New York Times. Each phase of the encourage last two months. They also randomize free Youku VIP (Chinese Netflix) in order to do the experiment less obvious. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 9 / 39

Randomization Timming Baseline Survey (11-2015): Potential participants are informed that this is an academic research project that involves repeated surveys over the course of 18 months, and that the study aims to to understand Chinese college students beliefs, attitudes, and behavior during the age of globalization. The provision of censorship circumvention tool is not mentioned during this stage. The students are paid US$15 for participating, and additional US$10 bonus payment, on average, depending on their surveys. Treatment (12-2015): They distribute access treatment and encourage newsletters simultaneously. This treatment assignment stage excludes the students using any censorship circumvention tools. Midline survey (05-2016): It was taken six months after the treatment. Students were rewarded with US$20 for participating and US$ 10 (on average) depending on their answers. 89.5% of retention. Endline survey (04-2017): Similar to midline survey. Students were rewarded with US$20 for participating and US$20 (on average) depending on their answers. 75.9% of retention (no selective attrition by treatment status). Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 10 / 39

Data Media Consumption Browsing foreign websites: Decisions to purchase access after experiment ends: Beliefs and attitudes regarding media: Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 11 / 39

Data Media Consumption Browsing foreign websites: For students in the treatment group who have activated their censorship circumvention tool accounts, they directly observe all of their online activities that route towards websites hosted outside China. They create four outcome variables: (i) whether a student activates the tool, (ii) whether the student actively uses the service after activation, (iii) total time spent on browsing foreign websites each day after activation, and (iv) total time spent on each foreign website category. Limitations? 1. They only observe the activities of the treated group 2. They do not observe the visits to local websites. Decisions to purchase access after experiment ends: Beliefs and attitudes regarding media: Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 12 / 39

Data Media Consumption Browsing foreign websites: Decisions to purchase access after experiment ends: During the end line survey, the authors offer students who receive the access treatment an opportunity to renew their subscription at a discounted price. The price was set as the median level of treated students willingness to pay for the censorship tool. In addition, they offered the control group the same opportunity to buy the circumvention tool. Beliefs and attitudes regarding media: Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 13 / 39

Data Media Consumption Browsing foreign websites: Decisions to purchase access after experiment ends: Beliefs and attitudes regarding media: To assess the mechanisms behind, the authors asked about (i) valuation of the access to uncensored Internet and foreign news outlets, (ii) trust in domestic and foreign news outlets, (iii) belief in the actual media censorship level and its drivers, and (iv) justification for media censorship. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 14 / 39

Data Beliefs, knowledge, attitudes, and behaviors In order to measure the effects once students are exposed to politically sensitive information, the authors measure 4 broad groups of interest in the survey: a. They assess a range of knowledge from contemporary to historical and from politically sensitive to non-sensitive. b. Economic beliefs. c. Attitudes respect to politics, broadly defined (e.g., trust in institutions) d. Self-report on a range of past and future behaviors. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 15 / 39

Does access increase acquisition of sensitive information? Free provision of access alone does not increase acquisition. Only 53% of students treated actually activate the tool. The Group-A students who actively use the tool are positively selected in terms of their English ability, oversea experience, and household income. This group s media consumption is uncorrelated with the occurrences of politically sensitive events. This suggest that students are unlikely to treat these foreign websites as primary sources of political sensitive information. The Group-AE students are 15 percentage points more likely to activate the censorship circumvention tool. The students does not seem to increase their consumption with pure informational encouragement. The monetary incentives effectively boost the demand to visit foreign news websites. Taken together, these results demonstrate that the free access to uncensored Internet alone has little effect on students acquisition of politically sensitive information from foreign news outlets. This suggest that student s demand for sensitive information may be low. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 16 / 39

Does access increase acquisition of sensitive information? Free provision of access alone does not increase acquisition. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 17 / 39

Does access increase acquisition of sensitive information? Increase in information acquisition persists after encouragement ends The Group-AE students who activate the censorship tool spend, on average, 4.4 more minutes on the New York Times per week. This increase is not driven by the Group-AE changing the underlying selection of who browse foreign websites, since essentially no students spend time on the foreign news websites without the encouragement. Neither is by a a big increase in media consumption of a small group, since the difference of the median student who activate the tool is even sharper. The Group-AE also spend more time on Wikipedia. The authors speculate that the sensitive information prompt students to explore similarly sensitive censored events. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 18 / 39

Does access increase acquisition of sensitive information? Increase in information acquisition persists after encouragement ends The time Group-AE students spend on the New York Times closely tracks the percentage of politically sensitive articles. However, the extensive margin does not vary with the variation in sensitive articles. This suggests that Group-AE students visit the website at a stable frequency, but spend additional time browsing during weeks when there are more news articles they have not yet seen on domestic news websites. Encouragement raises willingness to pay for uncensored Internet access, reflecting an increase in their demand for sensitive information. The average willingness to pay in the middle and end line survey differ between groups. In addition, the Group-AE s rate of renew the tool is larger that Group-A s rate. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 19 / 39

Does access increase acquisition of sensitive information? Why is the demand for uncensored information low? One hypothesis is that students belief is that uncensored information is not valuable. In the baseline survey, students hold diverse beliefs regarding the extent to which content on domestic media is censored. In addition, the unwillingness to pay for access may reflect a belief that such information is not valuable. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 20 / 39

Does access increase acquisition of sensitive information? Why is the demand for uncensored information low? Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 21 / 39

Does access increase acquisition of sensitive information? Why is the demand for uncensored information low? If we assume that exposure to foreign news outlets make beliefs about their quality and value more accurate (as we would expect if the students are Bayesian), the patterns above imply that students belief about the quality and value of foreign news outlets are biased downward in the baseline survey. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 22 / 39

What is the impact when students acquire sensitive information? Impact on students directly exposed The authors exploit the variation in their acquisition of information. They compare, across students in different groups during the endline survey, outcomes that are elicited in a private manner, removing various social incentives (such as signaling, coordinating, conforming) that may affect students answers. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 23 / 39

What is the impact when students acquire sensitive information? Impact on students directly exposed Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 24 / 39

What is the impact when students acquire sensitive information? Impact on students directly exposed The authors administer 7 quizzes on politically sensitive events that occurred within 3 months of the end-line survey. The Group-AE students have 0.902 (13%) more quizzes correctly than those in C or A groups. This quizzes are design to capture the knowledge stock (they cannot browse the information since the subscription is over). In contrast, there is no difference between group A, C or AE in their quizzes of local news. The students in the Group-AE are more pessimistic about China s economic performance. Also, they are more optimistic about US economic performance. The Group-AE students report lower trust in China s government and in politicians. The Group-AE students changed its behavior. They are more likely to report that they discuss political topics with other students and are more likely to apply to overseas graduate school and, hence, leave China. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 28 / 39

What is the impact when students acquire sensitive information? Magnitude of the effects The magnitude of the treatment effect is specific to the study sample, the time, and the foreign news outlet. With this in mind, the local effects of acquiring politically sensitive information that we identify are substancial in magnitude. The median of Group-AE is ranked at the 47th percentile of the distribution of all study participants at the baseline survey. After the treatment, the median has moved to the 55th percentile of the distribution. They estimate the persuasion rate and find that the median persuasion rate across outcomes of interest is 44.8%, similar to the rate in authoritarian regimes. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 29 / 39

What is the impact when students acquire sensitive information? Heterogeneity in the treatment effect on knowledge The authors re-estimate the treatment effect on subsamples split according to all demographic characteristics and fundamental preferences, as well as student s knowledge prior to the experiment. The treatment effect is larger among students from relatively more disadvantaged backgrounds. This result suggest that one should be cautious when generalizing the results to other demographic groups in China. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 30 / 39

What is the impact when students acquire sensitive information? Social spillover of politically sensitive knowledge Does the acquisition of uncensored information affect students beyond those who are directly exposed? The authors focus on college dorm roommates to answer this question. Ovarall, 57% of students who do not receive access and encouragement treatments reside with at least one roommate who is treated; and 42% of treated students reside with at least one roommate who is also treated. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 31 / 39

What is the impact when students acquire sensitive information? Social spillover of politically sensitive knowledge They estimate: Correct ij (I i (own), N i (roommate)) = + α j + I i p j ( ) 1 ((α j + p j )(1 q j.ii ) + (1 α j p j )) N i I i {0, 1} indicates if the students belongs to the E or AE group, N i {0, 1, 2} denotes the number of roommates who have access to uncensored internet. α j is a base-rate learning probability, I i p j indicates the marginal increase in the probability of correctly answering the quiz questions if a student has access to uncensored Internet herself and actively browses foreign news outlets. The third term describes social learning, where q i.ii The parameters of interest are: α j, p j, q j,ii=0 and q j,ii=1. is the social learning term. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 32 / 39

What is the impact when students acquire sensitive information? Social spillover of politically sensitive knowledge Three assumptions of the social learning structure: 1. Becoming informed of a censored news event is an absorbing state, such that learning from one roommate versus multiple roommates exhibits the same empirical outcome because the quiz outcomes are binary. 2. They assume that there is no information transmission from students who do not have direct access to uncensored Internet since it is not separately identifiable. 3. There exists only one degree of social transmission. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 33 / 39

What is the impact when students acquire sensitive information? Social spillover of politically sensitive knowledge Three patterns emerge with respect to the social transmission of information. 1. Knowledge spillover is significantly different from zero. 2. The social transmission rate monotonically increase with the direct learning rates. 3. Social learning double if the student has a uncensored roommate Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 34 / 39

What is the impact when students acquire sensitive information? Social spillover of politically sensitive knowledge Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 35 / 39

What makes censorship effective? Would all the students become informed of sensitive events if censorship were to be removed among a large share of them? Would students low demand for uncensored information be partially offset by the social transmission of information? The authors use the estimated parameters from previous section to simulate the response to these situations. It is important to notice that these estimated parameters capture partial equilibrium effects, since the experiment covers a little percentage of the student population. There are two benchmark: (i) status quo, and (ii) full saturation (when all students would be able to correctly answer the quiz question 80% of students actively browse foreign news sites.) Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 36 / 39

What makes censorship effective? The authors simulated three scenarios: 1. If all students have free access to uncensored Internet, the share of students who answer the quiz question correctly increase by 3 percentage points. This suggest that the current censorship apparatus is fairly robust in maintaining its effectiveness. 2. If all students have free access to uncensored Internet + encouragement, the share of students who answer the quiz correctly raises to 98% (close to the full saturation benchmark). This suggest that, although the current censorship apparatus is robust due to lack of demand for uncensored information, its effectiveness could be substantially and persistent diminished if demand were raised through encouragement and exposure. 3. Partial subsidization of the circumvention tool. If they receive encouragement, 72%of the newly exposed students would pay to continue accessing uncensored Internet. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 37 / 39

What makes censorship effective? Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 38 / 39

Conclusion The authors find the low demand for uncensored information is a crucial reason why students don t acquire such information. Nevertheless, the low demand is not permanent. Encouraging the demand of social media raise permanently the demand of censored information, changing the beliefs and perception of the users. Yuyu Chen David Y. Yang The Impact of Media Censorship: Evidence from a Field Experiment in ChinaJanuary 22, 2018 39 / 39