When “Fake News” Becomes Real: The Consequences of False Government Denials in an Authoritarian Country

Authored by journals.sagepub.com and submitted by smurfyjenkins

Governments around the world, particularly authoritarian ones, often deny inconvenient or unfavorable information, calling it fake news or false rumor, and yet what was denied often turns out to be true eventually. How will citizens react when the initial “fake news” is verified to be real? What are the consequences of false government denials on government credibility and citizen satisfaction? Using a survey experiment in China and a follow-up survey, we find that citizens can be persuaded by the authorities’ denials and reduce their belief in a piece of news that has been declared “fake.” But when the denied news turns out to be real, citizens will reduce their belief not only in the denial at hand but also in a similar denial in the future and reduce their satisfaction with the government. Thus, false denials have both immediate and lasting effects on government credibility and can erode citizen satisfaction with the government.

Research Design and Recruitment The ever-present information battle between the state and society in China presents an excellent opportunity for this research by providing an environment rich with rumors and denials, with some of the latter eventually turning out to be false. We use data from a two-wave online survey conducted in fall 2017 about Chinese internet users’ trust in different information sources and public policies. The first wave was a survey experiment, and the respondents were randomly assigned to one of three groups. The first group was exposed to a rumor regarding the government’s automobile registration policy (rumor group), the second group was exposed to the government’s denial of the rumor (denial group), and the third group was exposed to an event in which the rumor initially denied by the government was verified as true (verification group). Because our focus is how denial and verification of a rumor affect people’s attitudes toward the government, not the influence of rumor itself, we did not have a pure control group that was not exposed to any of the above messages, including the rumor; instead, the rumor or denial group served as the baseline in the analysis below. In the second wave follow-up survey, which occurred about 3 weeks after the first wave, respondents from the verification group of the first wave, who witnessed the government’s denial of the rumor and the eventual revelation of the falsehood of the denial, were re-enrolled and exposed to another rumor about automobile policy and the government’s denial. For the purpose of the second wave, this group will be called the experience group. For comparison, a new group of respondents, who did not participate in the first-wave experiment, was also recruited in the second wave and exposed to the same rumor and denial as the experience group. This new group will be called the no experience group. As should be clear, this second-wave survey was designed to explore whether witnessing a government denial proven false would have lasting effects on respondents who had that experience. While more details of the survey and experiment will be provided in the following sections, here we provide the outline. In each wave, all respondents were first asked about their (pre-treatment) political interest,3 trust in the government,4 life satisfaction,5 and sociodemographic information. Then they were given a brief message describing the worsening traffic congestion in a Chinese city. Afterwards respondents in the different groups were exposed to the rumor, denial, and/or verification messages, as described above, with all messages referring to the aforementioned city experiencing worsening traffic congestion (see the following sections for the content of each specific message). Finally, all respondents were asked to rate their belief in the various messages they had just been exposed to and their satisfaction with the government related to the issue at hand, namely the government’s automobile and traffic management, all on a scale from 1 to 100.6 In the surveys we referred to the unconfirmed rumor as “unverified information” rather than “rumor” to ensure the neutrality and naturalness of the information provision. The rumors and denials in our study were about local automobile registration policies. We chose these relatively low-stakes rumors rather than “big rumors” about issues like an epidemic, high-level corruption, or elite politics for a number of reasons. First, traffic congestion is a common issue in Chinese urban areas, so automobile management is a policy area that most people care about. Second, the government’s about-faces concerning this issue reflect a common issue of policy unpredictability and instability in authoritarian countries, due to the public’s lack of sufficient say in the policy formulation process.7 Therefore these seemingly low-stake rumors have significant political implications. Last, but not the least, such a choice could reduce the political sensitivity of the survey and increase respondents’ willingness to provide genuine answers. The rumors and denials in the study were based on two real events occurring, respectively, in southern and northern China, and real government and media messages. For ethical reasons, it is important to use information that is actually available on the Chinese media and internet rather than making up rumors or denials.8 To increase the external validity and not to tie the treatment to a specific real event, however, we replaced the names of the real places with generic English letters (e.g., City “H”). A pilot survey conducted prior to our formal surveys shows that there is a large variation in Chinese online respondents’ exposure to socio-political news, with some respondents following news very closely. Therefore, to minimize the possibility that some respondents may have heard the rumors and/or denials prior to the study, and their opinions may have been polluted by their attitudes toward the specific locations and events, it was reasonable to make the locations of the events vague. Using generic letters to make the specific location vague is also a common practice in Chinese research studies as well as in online news. Therefore, participants in our study understood that what they saw in the survey was referring to actual events in non-specified locations. Respondents in the two-wave study were recruited from Baidu Zhongce (https://test.baidu.com/), a Chinese crowd-sourcing website for recruiting and compensating agents to perform tasks. Baidu Zhongce has a larger user pool than many other crowd-sourcing websites in China: over 17 million registered users of age 18 or above, covering 300 cities in China. In addition, its closed user system and bonus structure is particularly useful for re-enrolling the same participants in follow-up studies, which is important for our purpose. To prevent repetitive participation, each Baidu ID (registered user) and each unique IP address could participate only once in the experiments (with the exception of participants in the verification group of the first wave, who were re-enrolled to participate in the follow-up survey, as explained above). Because we study internet rumors and denials, recruiting subjects online and having them take the surveys on their own computers in their own places increased the mundane realism of the experiment and the generalizability of the results. The study’s participants had a broad geographic representation: they came from all province-level administrative regions of mainland China except Tibet. They also represented diverse walks of life, including students, government employees, corporate employees, workers and professionals, agricultural labor, and other occupations. In addition, they were also from many different age groups and educational backgrounds. Although the respondents were somewhat younger and better educated than the general internet population in China (a feature common to internet surveys), they nevertheless achieved a broad demographic representation. Moreover, the younger and better educated generation is also more politically active and, therefore, worth particular attention. See Supplemental Appendix 1 for the demographic breakdowns of the study participants and the general Chinese internet population. Because a potential challenge to online surveys is to ensure that the respondents pay proper attention to the questions, we recorded the time each participant spent in completing the survey and dropped from the analysis a small number of unusual speeders who finished the procedure faster than a predetermined generous time threshold.9 This yielded 2905 effective participants in the first wave experiment and 1948 effective participants in the second wave survey.

The Immediate Effects of False Government Denials Our wave 1 experiment was designed to investigate respondents’ immediate reaction to a government denial and the revelation that the denial was false. The respondents were randomly assigned to one of three approximately equal-sized groups discussed above: the rumor group, the denial group, and the verification group. After a set of general questions on the respondents’ pre-treatment attitudes and sociodemographic information, and a reminder of the traffic congestion issue in City H, all respondents were presented with the following rumor about automobile registration policy: Some online sources said automobile registration in City H of Zhejiang Province would soon be allocated by a license-plate lottery system starting March 25, 2017. As a result, all automobile 4S dealers in City H became crowded with customers and stayed open all night. At least 200 cars lined up outside the Office of Automobile Administration, waiting to register for licenses. Some new car owners simply slept in their cars overnight. The denial group and the verification group were then presented with the government’s denial of the rumor. In addition, the verification group, after seeing the denial, was shown a news event in which the initial rumor was proven to be true. The rumor group did not see the denial or verification. The government’s denial read as follow: The traffic management authorities of City H issued a rebuttal via official media, stating that online speculation about the city’s car registration system moving to a license-plate lottery system on March 25th was pure fake news. The government reminded citizens not to believe the online rumors and exercise judgment in making car purchase decisions. The news item that verified the veracity of the initial rumor was as follows: On March 25th, 2017, the municipal government of City H abruptly announced that the city would limit the number of automobile plates by adopting a license-plate lottery system. This is despite the fact that, from 2016 until the announcement of the new policy, relevant government agencies of City H had altogether issued 11 denials regarding online speculation about the lottery system. After these information treatments, all respondents were asked about their belief in the initial rumor, and those in the denial group and verification group were asked about their belief in the official denial. Finally, all respondents were asked about their satisfaction with the government’s work in automobile and traffic management. Figure 1 details the differences in the groups’ information exposure (dashed boxes indicate outcome questions). Download Open in new tab Download in PowerPoint Figure 2 displays the means comparisons of belief in information and satisfaction with government across the three experimental conditions in wave 1.10 As expected, the left panel of Figure 2 indicates that, relative to the rumor condition, the denial significantly reduced the respondents’ belief in the rumor, with the effect size being 11.8 points on a 100-point scale. The verification, on the other hand, increased belief in the rumor relative to the denial condition by 8.6 points, although the level of belief was still about 3.1 points lower than the rumor group. That people believe more in the rumor when it was just a rumor than when it has been verified is likely because the verification group was also exposed to the denial message. This two-sided information flow thus cancelled out some effect of the one-sided information flow of the rumor group (Zaller, 1992, 1996). Download Open in new tab Download in PowerPoint The middle panel of Figure 2 compares the verification group and the denial group’s belief in the denial. As expected, verification of the rumor increased the respondents’ belief in the denial. The effect size was 5.8 points on the 100-point scale. The right panel of Figure 2 shows that the denial group and the rumor group had similar levels of satisfaction with government, while the verification group had a slightly lower level of satisfaction (2.7 points lower than the rumor group). Given that the government’s denial in the experiment was very simple and did not come with a great deal of explanation or evidence, this result is consistent with the findings in the existing literature (Huang, 2017). One potential concern with the results is that the rumor referred to an event in Zhejiang Province, so respondents from that province might have seen something similar before, even though the specific city was made vague in the experiment. To mitigate this concern, in Supplemental Appendix 3 we provided the means comparison of the three groups both with and without respondents from Zhejiang. The results are broadly similar, except the small differences between the rumor and verification group are no longer statistically significant at conventional levels when Zhejiang respondents are excluded, meaning the verification largely restored non-Zhejiang respondents’ belief in the rumor and satisfaction with the government to the rumor condition. For further robustness, we also added the speeders back to the sample. Doing so did not qualitatively change the main results either: In both cases the denial reduced belief in the rumor and verification increased belief in the rumor (see Supplemental Appendix 4). Most, but not all, demographic and attitudinal covariates across the three groups were well-balanced. Therefore we ran ordinary least square regressions on the respondents’ belief in the rumor, belief in the denial, and satisfaction with the government, controlling for a set of covariates. In all regressions, we include the usual demographic variables, political affiliation (membership in the Chinese Communist Party), and pre-treatment attitudinal questions including political interest, political trust, and life satisfaction. Demographic variables include gender, age group, education, income level, and rural/urban residence. Table 1 shows the regressions results. The baseline in regressions (1) and (2) is the rumor group, where we compare the denial group and the verification group with the rumor group. The baseline in regressions (3), (4), and (5) is the denial group, where we compare the verification group directly with the denial group. The results are consistent with the means comparisons: Denying a rumor reduced the respondents’ belief in that rumor; verification of the rumor increased belief in the rumor and reduced belief in the denial. Table 1 also shows that, after controlling for the covariates, the small difference in satisfaction with government between the rumor group and the verification group is no longer statistically significant. In other words, neither denial nor verification had immediate effects on satisfaction with the government. Table 1. Belief in Information and Satisfaction with Government (Wave 1). View larger version Among demographic variables, education increased belief in the government’s denial, income increased belief in the rumor, and females were more satisfied with the government. Members of the Chinese Communist Party did not have a significantly different level of belief in the rumor (or the denial) than non-members; nor did they have significantly higher satisfaction with the government. This suggests that in contemporary China party membership is to a large extent motivated by instrumental considerations rather than political or ideological reasons. Interestingly, with regards to attitudinal covariates, people with higher political trust and political interest were more likely to believe both the rumor and the denial (when they saw the denial); in other words, they were more responsive to whatever information they were exposed to. They were also more satisfied with the government, similar to those with higher life satisfaction. In sum, in the first-wave experiment official denial reduced the respondents’ belief in the rumor, while verification of the rumor decreased belief in the denial and increased belief in the original rumor. Aside from the verification of rumor, which is new in the literature, the key results from our first wave experiment are consistent with previous research on rumor and denial in China. This indicates the validity of our experiment and increases our confidence in the results on rumor verification in both waves.

Discussion and Conclusion We conducted a survey experiment and a follow-up survey to examine the consequences of government denials of initially unconfirmed rumors that later turn out to be true. The study’s main findings are that while an official denial can initially reduce citizens’ belief in the unconfirmed information, when the denial is revealed to be false citizens will have lower levels of belief not just in the current denial but also the government’s future denials of similar rumors. Moreover, the negative lasting effects will carry over to satisfaction with the authorities in the related policy area. Thus, while government denials of a piece of real news may temporarily reduce citizen belief in the news, in the long run they will reduce government credibility and citizens’ political trust and satisfaction. Note that the rumors in our study were about local automobile registration policies and thus relatively low stakes, and yet the result of false government denials was negative. One can imagine that if the government denial was about more significant news such as a public health crisis or intra-elite political conflicts, the effects of the revelation of false government denials might be even stronger. For example, after Li Wenliang’s warning turned out to be true, his death almost provoked an online revolt and the comment section of his last social media post became China’s virtual Wailing Wall (Buckley & Mozur, 2020; Li, 2020). Future research can look into the consequences of false government denials in these higher-stakes areas. Readers may be concerned that a recency effect drove the effect in our second-wave survey, that is, the experience group was simply reminded that the government had recently told lies and, therefore, reported a lower level of satisfaction. While 3 weeks are not a terribly short a time in our era of information overload, it is possible that a longer lapse of time would see the effect of a single false government denial diminished. However, in real life people come across false government denials from time to time, and thus their memories would be regularly reinforced. The potential recency effect of any particular denial is therefore inconsequential. One potential limitation of our research design is that the second wave survey did not use randomization. We have used matching strategies and sensitivity analysis to show that our results are robust with regards to functional form assumptions and potential unobserved confounders. An alternative and potentially cleaner design would be to have a pure control group in the first wave that was not exposed to any of the rumor, denial, or verification messages, and then re-enroll both this pure control group and the verification group in the second wave. This would assure randomization in our comparison of the experience group with the no-experience group. We could also re-enroll the rumor and denials groups in the first-wave experiment for the second wave, although then we would be comparing the full experience group with subjects who have had a rumor and/or denial experience, a comparison that is different from our full experience versus clean slate comparison. These alternative designs can be explored in the future. In addition, the denial in the second-wave survey of this study was similar to that in the first wave. Future research can examine if the loss of government credibility extends to government denials in general rather than just similar denials. Conversely, future research can also examine the consequences of truthful government denials of critical rumor and speculations. For example, Guo Wengui, an exiled Chinese billionaire and real-estate tycoon, has released on social media explosive claims against the Chinese government, including corruption at its highest levels. Naturally, the Chinese government has denied the claims. Among other things, it accuses Mr. Guo of forging “top secretive” government documents, the possession of which had been used by Guo as evidence that he had high-level access within the Chinese government (Xinhua News Agency, 2018). But while some of Guo’s revelations may be accurate, independent and Western sources have reported that many others are outlandishly aggrandized or provably false (Forsythe & Stevenson, 2017; Hilgers, 2018). As a result, Guo’s following among overseas Chinese dissidents has significantly waned, with some even accusing him of being a spy for the Chinese Communist Party and splitting the dissident community (Hilgers, 2018). What are the political consequences when fierce critics’ accusations against the authorities turn out to lack evidence and the latter’s denial was corroborated by independent and credible sources? This will be an interesting research question about opinion dynamics in authoritarian countries that would complement our current study.

Acknowledgements We would like to thank Jia Chen, Yue Hou, Tianguang Meng, Robert Ostergard, Xiaoyu Pu, Lai Wei, Yang Zhong, several anonymous reviewers, and participants of the 2019 Workshop on Chinese Politics and Society in Shanghai for helpful comments.

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article. ORCID iD

Supplemental material for this article is available online.

Rumor, fake news, and misinformation are not identical concepts. Rumors are claims “without secure standards of evidence being present” (Allport & Postman, 1947, p. ix), but they might sometimes turn out to be correct. Fake news and misinformation, by their very labels, refer to false claims and information. But due to the close connections among these concepts, the literature has treated them in similar ways. 2

We recognize that denials by governments are not always acts of deliberate dishonesty but may sometimes reflect different policy intentions or policy instability. But from the society’s point of view, revealed falsehood of official denials nevertheless represents the lack of credibility of official statements. 3

Political interest was measured by asking respondents’ interest in reading social-political news. The choices were “not interested at all,” “not very interested,” “neutral,” “somewhat interested,” and “interested.” 4

Political trust was measured by asking respondents the extent to which they agree with the following statement: “The local government’s policies are designed to serve the interest of the general public.” The choices were “disagree,” “somewhat disagree,” “neither agree nor disagree,” “somewhat agree,” and “agree.” Because transportation and automobile policy are local issues, we asked about trust in the local government rather than the central government. 5

Question wording: “All things considered, how do you feel about your life, work, and study these days?” The choices were “dissatisfied,” “somewhat dissatisfied,” “neither satisfied nor dissatisfied,” “somewhat satisfied,” and “satisfied.” 6

Question wording on satisfaction with the government: “On a scale from 1 to 100, with 1 being ‘completely dissatisfied’ and 100 being ‘completely satisfied,’ indicate the extent to which you are satisfied with the government’s urban traffic control and management.” Because automobile and traffic management is local government responsibility, the question should be understood as asking about satisfaction with local governments. The questions on belief in the rumor and denial used similar wording, with unbelievable/believable replacing dissatisfied/satisfied. 7

For example, in an effort to revive the economy and boost employment after the Covid-19 pandemic, Chinese premier Li Keqiang made a high-profile pitch for the street-stall economy at the 2020 National People’s Congress. Less than 2 weeks later, the Chinese government backpedaled on the push with a series of state media editorials cautioning or opposing it, perhaps because the idea conflicts with General Secretary Xi Jinping’s vision of China as an advanced and high-tech superpower (Chen, 2020). 8

The study received Institutional Review Board approval from the first author’s institution at the time of the study. 9

The pre-determined time threshold, set based on results from a pilot survey, was 150 seconds. The average time to finish the survey in wave 1’s rumor group, which was shorter than all other groups since this group did not see the denial or verification message, was 283 seconds. Therefore the threshold of 150 seconds was quite generous. 10

See Supplemental Appendix 2 for the balance of covariates in the first-wave experiment. 11

Respondents in the experience group did not know that they would be recruited again for a follow-up survey at the time of the first-wave survey experiment. 12

Note that belief in a rumor and belief in its denial are not mutually exclusive. For example, (some degree of) disbelief in a rumor does not necessarily mean (a corresponding degree of) belief in the denial. 13

See Supplemental Appendix 6 for the balance of covariates in the second-wave survey.

This_isR2Me on September 20th, 2020 at 23:17 UTC »

So you're telling me the people want a reliable government?

MoldyPlatypus666 on September 20th, 2020 at 21:12 UTC »

Which is why tech may prove to be our downfall. With misinformation and things like Deep Fake becoming more sophisticated and prevalent, the concern isn't so much that we won't believe information that isn't true. It's that we'll doubt information that is true and will be too caught up in analysis paralysis to tell the difference and trust it.

Boris740 on September 20th, 2020 at 18:49 UTC »

So it's like "cry wolf"?