Sometimes Less Is More: Censorship, News Falsiﬁcation, and Disapproval in 1989 East Germany

: Does more media censorship imply more regime stability? We argue that censorship may cause mass disapproval for censoring regimes. In particular, we expect that censorship backfires when citizens can falsify media content through alternative sources of information. We empirically test our theoretical argument in an autocratic regime—the German Democratic Republic (GDR). Results demonstrate how exposed state censorship on the country’s emigration crisis fueled outrage in the weeks before the 1989 revolution. Combining original weekly approval surveys on GDR state television and daily content data of West German news programs with a quasi-experimental research design, we show that recipients disapproved of censorship if they were able to detect misinformation through conflicting reports on Western television. Our findings have important implications for the study of censoring systems in contemporary autocracies, external democracy promotion, and campaigns aimed at undermining trust in traditional journalism. VerificationMaterials: The data, code, and materials required to verify the computational reproducibility of the results, procedures, and analyses in this article are available on the American Journal of Political Science Dataverse within the Harvard Dataverse Network, at: https://doi.org/10.7910/DVN/AZFHYN.

The German censors idiots .-Heinrich Heine, Reisebilder, Chapter XII I n 2003, Barbra Streisand filed a lawsuit against amateur photographer Kenneth Adelman for violating her privacy rights.To document accelerated erosion, Adelman had uploaded aerial images of the entire California coastline to his website, with one of them showing the singer's mansion in Malibu.Beyond the fact that the courts dismissed Streisand's claim for damages and injunction, only six people had downloaded the image before the lawsuit-"including twice by her own lawyers" (Cacciottolo 2012)-whereas the website attracted over 400,000 visitors in the following month.Named after this incident, the "Streisand effect" describes the unintended consequences counter to the censor's initial motivation for withholding information (Martin 2007).
To cement their power, both historical and contemporary political regimes suppress information (Roberts 2018).Especially autocratic governments often systematically control the media and communication flows to prevent the diffusion of information that could damage their reputation or spark public resistance against the state (Little 2016).Research generally agrees that censorship and propaganda can be powerful tools to boost regime legitimacy and maintain political stability (e.g., Cho, Lee, and Song 2017;Stockmann and Gallagher 2011).
In view of the asserted effectiveness of information manipulation, it is puzzling that contemporary autocracies do not fully exhaust their capacity to implement all-encompassing censorship (Gunitsky 2015;King, Pan, andRoberts 2013, 2017).Existing explanations focus on a regime's benefits from granting at least some freedom of expression.For example, autocracies without revelatory elections might shy away from total censorship to gain information on people's opinions and grievances (e.g., Dimitrov 2014).Governments might also use free speech to monitor local officials, release political pressure, and signal responsiveness (e.g., Huang, Boranbay-Akan, and Huang 2019;Lorentzen 2014). 1 Finally, regimes may grant freedom of expression to reduce the threat from revolution by revealing disunity within opposition movements (Chen and Xu 2017).
We suggest another reason why modern autocracies use censorship in a restrained way: More censorship does not imply more stability.Recent theoretical work holds that while information manipulation may improve citizens' beliefs about the regime, it can also diminish the perceived reliability of state media (Little 2017;Shadmehr and Bernhardt 2015).In effect, leaders lose the key tool to influence the population in their favor.We argue that blatant state censorship is likely to backfire, particularly when people can draw on trusted alternative sources of information to identify discrepancies.Such alternative sources can be traditional media and, increasingly, web-based communication technologies.Systematically conflicting narratives between state media and other sources increase the likelihood that recipients believe to be censored if they sufficiently trust the alternative source.Those who suspect that the state is withholding important information should become more interested in the censored issue and increasingly critical of the censor.Yet, individuals might react to censorship differently, depending on their personal regime ties.People with an ideological predisposition to the government may be less willing to trust alternative sources, whereas those without such ties should be more upset.
To empirically test our argument, we study the German Democratic Republic (GDR) in 1989.We show how excessive media censorship on the GDR refugee crisis fueled major dissatisfaction, which eventually contributed to the breakdown of one of the most stable Warsaw Pact regimes.Our findings demonstrate that people were able to identify their regime's bleak strategy of censorship and defamation based on the extensive reports on West German television (WGTV).Before the mass public uprising in late 1989, thousands of GDR citizens had left the country and occupied West German embassies to force their emigration to the West.State-controlled East German media largely denied these developments, construed a legend about a massive hoax by Western intelligence services, and insulted refugees as infamous traitors.Those who stayed and watched reports about refugees on WGTV could unmask domestic news reports as clumsy propaganda.
Using over 17,000 individual-level approval ratings of GDR state television programs and original content data based on West German news transcripts, we quantitatively show that censoring practices by East German state media resonated poorly with the audience-irrespective of the respondents' ideological ties to the regime.Like other viewers, members of the ruling Socialist Unity Party (SED) markedly rejected censored news content.To gauge causal effects, we leverage temporal variation in television news reporting and spatial variation of WGTV coverage (Crabtree, Darmofal, and Kern 2015).In addition, we run a placebo analysis of approval ratings for the national lottery broadcasting to rule out that our findings simply reflect a general trend in popular dissatisfaction.
This article contributes to our knowledge about backfiring dynamics of censorship (Roberts 2018;Shadmehr and Bernhardt 2015).Based on unique observational data on media approval in a highly authoritarian environment, we provide the first systematic test of how excessive information manipulation can damage the reliability of state media, which deprives the regime of its capacity to influence the population (Gehlbach and Sonin 2014;Little 2017).We hereby contribute to an emerging research strand on propaganda effects (Adena et al. 2015;DellaVigna et al. 2014;Peisakhin and Rozenas 2018;Yanagizawa-Drott 2014).Finally, our findings augment research on the nexus between media, framing, and social movements (e.g., Andrews and Biggs 2006;Myers 2000;Soule and Roggeband 2019), as well as the literature on political backlash in reaction to physical repression (e.g., Daxecker and Hess 2013;Moore 1998;Sutton, Butcher, and Svensson 2014).

Research on Political Backlash
Studies on political backlash largely focus on adverse consequences from physical repression.2Research shows that violence against protesters often not only fails to end public upheaval but even intensifies domestic resistance (e.g., Francisco 2004;Lichbach 1987;Moore 1998), while at the same time motivating international sanctions and defections in the security apparatus (e.g., Albrecht and Ohl 2016;Chenoweth and Stephan 2011;Nepstad 2013;Pion-Berlin, Esparza, and Grisham 2014;Sutton, Butcher, and Svensson 2014).The likelihood of backfiring depends on the severity of repressive measures (e.g., Albrecht and Ohl 2016;Moore 1998;Rasler 1996), initial opposition tactics, and political institutions (e.g., Carey 2006;Gupta, Singh, and Sprague 1993;Nepstad 2011).
Research on backlash from physical repression views the media as a tool for activists to communicate state atrocities, which helps mobilize resistance (Sutton, Butcher, and Svensson 2014).In contrast, states with control over the media can withhold information and contain the risk of protest escalation (Francisco 2004).This perspective corresponds with studies that consider media censorship as an effective strategy to generate political stability (Cho, Lee, and Song 2017;Stockmann and Gallagher 2011).By manipulating publicly available information, leaders can inflate people's opinion about the regime's performance and foster conformist behavior in support of the government (Hollyer, Rosendorff, and Vreeland 2015;Little 2017).
However, censorship and misinformation can entail an independent risk of backlash.Formal models of censorship (Shadmehr and Bernhardt 2015) and propaganda (Gehlbach and Sonin 2014;Horz 2018;Little 2017) hold that information manipulation carries a trade-off: Although biased reports may lead people to see the regime's performance in a more favorable light, they can also undermine the source's reliability.In other words, "excessive media bias works against the government's propaganda interest, as citizens who ignore the news cannot be influenced by it" (Gehlbach and Sonin 2014, 163).Censorship and misinformation may therefore backfire by depriving the regime of its key instrument for influencing the masses.Building upon this trade-off, we explain when and why recipients are likely to turn against their censor.
Most related empirical work examines how beliefs about the government's performance can be manipulated by propaganda (Adena et al. 2015;Enikolopov, Petrova, and Zhuravskaya 2011;Peisakhin and Rozenas 2018).Yet, little is known about adverse effects from censorshipparticularly regarding its impact on people's trust in state media.Participants in laboratory experiments have been found to disapprove of any evident withholding of information, while at the same time developing a heightened interest in the censored topic (Hayes and Reineke 2007;Worchel, Arnold, and Baker 1975).Recent studies on Chinese online censoring offer initial evidence for behavioral changes among affected internet users, showing that censoring measures are associated with more comments on dissident blogs and downloads of anonymity technology (Hobbs and Roberts 2018;Roberts 2018).Moreover, people who personally experience censorship on social media use more critical phrases within the first 10 days after having been censored (Roberts 2018, 127-46).We examine the conditions under which people can identify censorship that is not directed against individuals but society as a whole.We thereby systematically analyze the effect of censorship on the perceived reliability of authoritarian regime media.
With respect to differences between individuals' reactions to censorship, laboratory experiments suggest that people with an ideological connection to the censor are less likely to become upset (Worchel and Arnold 1973;Worchel, Arnold, and Baker 1975).Similarly, research on radio propaganda suggests that prior attitudes cause heterogeneous reactions.Peisakhin and Rozenas (2018, 535) find that Russian propaganda generally helped pro-Russian parties during the 2014 Ukrainian elections, but it was "counter-effective in persuading those with strong pro-Western priors."Furthermore, Adena, Enikolopov, Petrova, Santarosa, andZhuravskaya (2015, 1885) conclude that in Germany, "Nazi radio was [...] effective in places where anti-Semitism was historically high [but] had a negative effect in places with historically low anti-Semitism."Whereas these studies rely on aggregate outcome measures such as vote shares, we offer individual-level evidence on potential differences between people with and without ideological linkages to an autocratic regime.

When and How State Censorship Backfires
Does media censorship cause political backlash?In this section, we develop our theoretical argument, which is summarized in Figure 1: We argue that people disapprove of blatant media censorship when they are able to identify misinformation via an alternative source that provides them with credible information on an important

FIGURE 1 Argument
issue over a sustained period.People perceive censorship as a restriction of their personal freedom, upon which they develop an increased interest in the censored issue and adopt a critical opinion toward the censor.Yet, we do not expect censorship to affect recipients uniformly.Individuals with an ideological disposition toward the state should be less likely to acknowledge and disapprove of state censorship.Next, we outline the psychological foundations of our argument.Why should individuals reject censorship in the first place?Psychologists argue that withholding information restricts an individual's personal freedom, which induces psychological costs (Brehm 1966;Brehm and Brehm 1981;Wicklund 1974).People perceive censorship as an external and illegitimate attempt to curtail their behavioral choice.The greater these restrictions, the more severe the grievances felt by an individual.In what psychologists term reactance, people who suffer from being censored seek to restore their original level of freedom and develop an "enhanced interest in the object of the freedom lost" (Hayes and Reineke 2007, 425).If the external limitation persists, recipients adopt views that are contrary to the one intended by the censor.Laboratory experiments confirm that participants exposed to censored information develop a heightened interest in the censored issue and change their attitudes in support of the object (Worchel, Arnold, and Baker 1975).
These findings are also consistent with the propositions of dual process theories (e.g., De Neys 2018; Kahneman 2011), according to which the human brain generally operates in two states of mind-System I (autonomous, fast, effortless) and System II (logic-based, slow, memory intensive).While System I processes provide for most cognitive tasks, System II takes over only if the individual is willing to spend his or her cognitive resources to resolve conflicting impulses.The realization of censorship may therefore motivate individuals to switch toward a deliberate state of mind and critically engage with reporting practices (Horz 2018).Insights from reactance and dual process theories therefore suggest that blatant censorship constitutes an exogenous limitation of personal freedom and induces individuals to question the censor.
The precondition for such a reaction, however, is that recipients have to realize that someone is withholding information from them.Unlike in laboratory settings, in which researchers reveal censorship by design, authoritarian regimes typically deny the suppression of information (Roberts 2018).People therefore require additional sources, which enable them to cross-check reports, identify discrepancies, and infer the deliberate spread of misinformation.These sources can be personal experiences as well as alternative domestic or foreign media (Huntington 1991;Sutton, Butcher, and Svensson 2014).Having available such alternative sources of information should enable people to recognize incompatibilities in reporting.
Nevertheless, incompatible reports alone are presumably insufficient to raise the suspicion of state censorship.As alternative media might also exaggerate or even fabricate stories, people are faced with an attribution problem.In this situation, individuals' ideological predisposition likely influences which narrative they consider more plausible, and whom they blame for misinformation (Allcott and Gentzkow 2017).Research on the "hostile media" phenomenon suggests that partisans tend to perceive even objectively balanced reports as hostile and biased (e.g., Arceneaux, Johnson, and Murphy 2012;Baum and Gussin 2008), whereas they want to believe in the narrative of like-minded sources to keep their worldview coherent (Little 2019).When confronted with conflicting reports on alternative media, regime supporters are therefore likely to preclude the possibility of state censorship, discounting alternative media as fake news.In contrast, recipients without strong ideological regime ties should be more receptive to alternative media content, which makes them more likely to identify and disapprove of censorship. 3We therefore expect the magnitude of backfiring to be moderated by individuals' ideological predispositions.
Irrespective of recipient-related differences, the risk of political backlash should further rest on two scope conditions.First, recipients need to consider the issue unequivocally newsworthy, as well as personally and socially relevant.Second, the topic must be persistent.Censorship becomes especially manifest when alternative sources cover an important issue over a sustained period, whereas state media neglect it.In this case, chances are minimal that the discrepancies are due to different assessments of a topic's newsworthiness, and people are less likely to believe that they had simply missed out on reports discussing the issue.The repeated perception of systematic differences between reports on a focal issue should reinforce people's conclusion that the state is withholding information. 4n sum, we expect that censorship triggers disapproval when recipients may falsify information based on conflicting, more trustworthy accounts in alternative media.Individuals with ideological ties to the regime are less likely to show disapproval upon biased reports even if they can draw on alternative sources of information.
H1: Recipients who can draw on alternative sources of information disapprove of media censorship on a focal issue.H2: While individuals with pro-regime disposition are generally more approving of state news than other recipients, they update less negatively when censorship is identifiable.5

Empirical Case
Sparse information has largely impeded research on backfiring processes in reaction to censorship within autocracies.We test our hypotheses with the case of East Germany.In the summer of 1989, a sustained wave of illegal emigration constituted a focal issue, which was heavily censored by the regime.Yet, the majority of GDR citizens could draw on West German television as an alternative source of information.The case offers an exceptional opportunity to study individual-level reactions to censorship within a highly authoritarian environment.It allows us to examine not only the influence of alternative media and recipients' ideological predispositions but also people's reaction to the cessation of censorship.Next, we describe the conditions and developments that provided the breeding ground for political backlash in response to East German media censorship.

Emigration Wave
Throughout its existence, the German Democratic Republic was one of the most repressive Eastern Bloc states.
The ruthless surveillance apparatus together with the poorly planned economy caused substantial political and economic grievances, while the Iron Curtain largely prevented people from moving to the West.Over 100,000 formal applications for emigration had been pending for years when the Hungarian government facilitated the border crossing to Austria in May 1989.Although the easing was limited to domestic merchants, the symbolic effect on GDR citizens was enormous.Many decided not to return from their summer holidays in the southern Bloc states and to cross the Austro-Hungarian border instead (Hirschman 1993).
Since most escape attempts had failed, the number of East Germans waiting behind the border noticeably increased in August (Timmer 2000).Others sought asylum at the West German embassy in Budapest.In response to growing pressure from the Western community, the Hungarian government finally allowed foreigners to leave for Austria (Richter 2009).As soon as East German officials ceased travel permits to Hungary, refugees went to occupy the West German embassies at Prague and Warsaw, where they set up camps and held out for weeks to enforce their emigration.The GDR leadership remained uncompromising until the large-scale celebrations for the fortieth regime anniversary came closer.To avoid any disruption to the festivities, the East German government agreed that the embassy occupants were allowed to emigrate to West Germany-provided that the evacuation trains would pass through the territory of the GDR (Richter 2009).The regime's goal was to signal sovereignty, but the result was mass commemorations all along the train tracks (Bahr 1990;Hofmann 2014).

Censorship by East German State Media
Since the founding of the GDR, all media had been subject to firm regime control and surveillance (Opp, Voss, and Gern 1993;Timmer 2000).Editors of Aktuelle Kamera, the national news program, had to submit all broadcasting sequences to the Secretariat for Agitation, which would check them and adjust the program.To ensure the implementation of socialist propaganda, the regime hired media workers based on their ideological conviction and allocated secret police agents to all major editorial departments (Holzweißig 1989, 90).
With regard to the 1989 emigration crisis, the regime imposed a strict censorship policy.This was a direct decision by Erich Honecker, the head of state, who sometimes personally edited media content or demanded changes during ongoing programs (Holzweißig 1989, 60-63;Schabowski 1991, 235).At an August 1989 Politburo meeting, Honecker ordered the national media to refrain from covering the emigration crisis (Hesse 1990, 337).Our systematic review of East German news transcripts shows that East German media followed this instruction. 6f at all, the news referred to the crisis by accusing West German media of a smear campaign against socialism.When the regime granted the refugees their emigration to the West, state news heavily disparaged them as "traitors, over whom it was not worth to shed a single tear" (Ludes 1990, 217).7

West German Television as an Alternative Source of Information
Given the regime's tight grip on domestic media, people informed themselves about the emigration crisis through reports on West German television (WGTV).West German public media and in particular the main evening newscast (Tagesschau) had the mission to inform people in both German states.Broadcasting statutes governed that the media should aim at the political reunification of Germany (Hesse 1990, 334).About 90% of the East German population received and regularly watched WGTV, and many of them considered WGTV as trustworthy (Kern and Hainmueller 2009;Meyen 2010, 29).
In contrast to GDR media, the West German Tagesschau extensively covered the 1989 emigration wave, providing updates about the number of refugees as well as information on their motivation, needs, and desires.

Liberalization of State Media
From early on, parts of the East German ruling party (SED) had been worried about adverse consequences of the blatant censoring strategy, fearing that the population would increasingly rely on Western media (Holzweißig 1997, 107).However, only after the resignation of Erich Honecker on October 18, 1989, did the Politburo under Egon Krenz order the national media to fundamentally change reporting practices.The editors welcomed this decision and quickly delivered much more objective coverage (Hesse 1990, 339;Ludes 1990, 290). Figure 2 shows the change in reporting.After the liberalization of state media, East German news reported on emigration even more extensively than their Western counterpart.

Data and Method
To assess whether media censorship induced political backlash in 1989 East Germany, we compile an original data set from various archival sources.To measure people's assessment of state media, we draw on the GDR's national media survey-a series of weekly representative surveys on people's consumption and evaluation of individual TV programs (GESIS 2011).In face-to-face interviews, respondents were asked to rate individual broadcasts that had been aired in the preceding days on a 5-point scale.We use respondents' approval ratings for the East German newscast (Aktuelle Kamera) as our dependent variable, ranging from 1 (lowest approval) to 5 (highest approval).Drawing on all surveys conducted in 1989, 9 the final data set includes 17,551 ratings of individual Aktuelle Kamera broadcasts from repeated cross-sections. 10 The East German national media survey is exceptionally well suited to study the effects of censorship in an authoritarian state.In the absence of free and fair elections, the survey served the regime as an important indicator of political attitudes and public opinion (Hausstein 2008, 68).Political leaders therefore were highly interested in accurate, representative, and unbiased answers.The survey results were top secret and for regime-internal use only.The National Committee for Television in charge of the survey put great emphasis on a sophisticated scientific sampling procedure and survey design.Due to the high scientific standards and the practical efforts toward valid and reliable results, the national media survey is considered the only East German social research institution with the capacity, resources, and authority to produce reliable and representative data on people's attitudes (Braumann 1994, 527).
Each week, interviewers polled around 1,000 respondents.To avoid social desirability bias, interviewers received extensive training and had to guarantee to the respondents full anonymity. 11A card index of all televiewers from the Ministry for Postal and Telecommunication Services served as an up-to-date sampling frame, which allowed for representative sampling, stratification, and clustering.The sampling procedure was standardized and did not change during the period under observation (Hausstein 2008). 129 Data for calendar week 22 are unavailable.
10 Repeated cross-sections allow us to study the variation in ratings by different individuals across time and space, but not withinperson variation.
12 Appendix SI.2 describes the sampling process and measures to maximize the validity of the surveys.

Research Design and Explanatory Variables
We test our hypotheses in two ways.In a first step, we exclusively leverage the onset of the emigration issue and the change in censoring practices to estimate the effects of censorship on people's evaluation of the East German state news.Informed by historical accounts, we divide the year into three successive periods.The Baseline period runs from January until July 1989.The following Censorship period from August to October 18 comprises the highly censored first phase of the emigration crisis until the resignation of Honecker, which marked the end of blatant censorship.The last period, which we call Liberalization period, covers the time of liberalized reporting practices from October 19 to New Year's Eve.In our analyses, we include the two mutually exclusive dummy variables Censorship period and Liberalization period, with the Baseline period serving as the reference category.Based on Hypothesis 1, we expect the coefficient for Censorship period to be negative.Although the Baseline period was not free from misinformation, our argument suggests that only the sustained and identifiable discrepancies between East and West German media coverage on the emigration crisis caused mass disapproval of the regime's apparent censorship practices.
In the second, more refined step, we specifically scrutinize the impact of alternative sources of information, which allow individuals to identify censorship.In order to do so, we no longer impose the simplified distinction between Baseline and Censorship period.Instead, we test whether actual West German reports on the emigration crisis were associated with disapproval of East German state media.We compile an original data set, providing detailed information on the reported content in each edition of the daily Tagesschau newscast from July to December 1989. 13The variable Tagesschau indicates whether West German news reported on the emigration crisis the day before the Aktuelle Kamera edition to be rated by the respondent. 14To gauge the effects of West German news broadcasts before and after the liberalization of reporting practices, we include an interaction term between the variables Tagesschau and Liberalization period.In line with Hypothesis 1, we expect the constituent term of Tagesschau to have a negative effect on respondents' evaluation of Aktuelle Kamera.This would suggest that people had identified the censoring practices via contrasting reports on refugees on WGTV.
To test for heterogeneous effects, as proposed in Hypothesis 2, we include the variable SED.It indicates whether a respondent reported being a member of the socialist ruling party, serving as a proxy for the individual's ideological linkage with the regime.The proportion of self-reported SED members in our data remains fairly stable at around 20-30%, approximating the estimated SED membership rate among the general population at that time (Jurich 2006).We interact the variable SED with the period dummies to check whether SED members disapproved of censorship less than nonmembers.

Control Variables
We control for variables that might confound our results.To account for general time trends, we include the variable Average rating last week, which gives the average rating of East German news in the previous week.We further include the variable Week trains coded 1 for calendar week 40, when the evacuation trains were passing through GDR territory.The binary variable District trains indicates whether a respondent (from calendar week 40 onwards) was living in the districts of Karl-Marx-Stadt or Dresden, as the evacuation trains crossed these districts and caused substantial attention.
We also control for sociodemographic respondent characteristics based on a standardized set of questions in the questionnaires of the national media survey.The variable Rural indicates whether a respondent lived in a rural area.To account for gender differences, we include the variable Male, whereas Age and its squared term account for generational effects.Finally, the variable Education gives people's educational attainment on a 4-point scale. 15

Estimation Strategy
We treat the dependent variable as metric, assuming that the rating is cardinal and categories are equidistant.We therefore model the approval of Aktuelle Kamera using linear regression models with standard errors clustered over households to account for within-household correlation.The unit of analysis is the broadcast-interviewee.

Results
Before turning to the statistical results, we present descriptive evidence for the approval of East German news 15  in the course of 1989.Figure 3 shows the mean approval ratings for Aktuelle Kamera by SED members and nonmembers.Descriptive results tentatively support Hypothesis 1.When West German television started reporting on the emigration attempts of GDR citizens, viewers identified the regime's censorship strategy and approval rates dropped.
Contrary to Hypothesis 2, differences between the approval ratings by SED members and nonmembers appear marginal.Regime party members similarly disapproved of the state media's misinformation campaign.Figure 3 also shows a swift recovery of ratings following the liberalization of reporting practices, indicating that the reactance effect of censorship was short-lived and reversible.

Basic Analyses
Next, we discuss the evidence from our statistical analyses.Table 1 shows the results of the basic regression models.Models 1-3 lend strong support for Hypothesis 1.The negative and highly significant coefficient for Censorship period confirms that approval ratings for the GDR state news substantially decreased in view of the stark discrepancies between the censored state media and the extensive reports about the asylum seekers on WGTV.In line with the descriptive evidence, Liberalization period has a positive and statistically significant effect on people's approval ratings, even exceeding the approval rates before the emigration crisis.
To test the moderating effect of a respondent's proregime disposition, as suggested by Hypothesis 2, we interact SED with the period variables in Model 3. Results show that SED members were generally more satisfied with GDR state news.However, the coefficient for the interaction between Censorship period and SED is indistinguishable from zero.SED members seem to have shown the same level of discontent for the censored news program as nonmembers.Figure 4 depicts the substantive effects of censorship and liberalization for both SED members and nonmembers.Whereas the positive effect of liberalization is significantly stronger for SED members, we find no statistically significant difference between the reactions to censorship (p = .598).Personal regime linkages did not tame disapproval of censorship but facilitated recovery upon reformation.SED members may have disapproved of the regime's misinformation strategy, as they antici-pated that people with access to WGTV could easily identify the state's crude attempt to hush up problems, further diminishing the regime's legitimacy.This might also explain why party members welcomed the liberalization of state media even more than other viewers.Regime affiliates, whose private privileges depended on the party's rule, likely considered the change in reporting as a necessary step to win back trust and to save the regime from collapse.16Regarding the control variables, people seem to have been less satisfied with East German news when they were better educated, lived in urban areas, or lived in districts through which the evacuation trains passed.Approval ratings additionally decreased in the week in which the evacuation trains crossed GDR territory.All models show a substantial fit to the data explaining up to 27% of the variance in the dependent variable.

Refined Analyses
Table 2 shows the refined analyses of the relationship between domestic censorship, alternative information, and news approval.Here, we no longer impose the starting date of the censorship period.Instead, Models 4 and 5 include the variable Tagesschau, which indicates whether West German news reported on the emigration crisis on the day before.We interact the variable Tagesschau with Liberalization period, such that the coefficient for Tagesschau gives the effect of West German news reports before the liberalization.
Again, results strongly support Hypothesis 1.The coefficient for Tagesschau is negative and highly statistically significant.This suggests that Western reports about the emigration wave helped to showcase the regime's censoring activities, inducing viewers to disapprove of the subsequent edition of East German news.Figure 5 shows the substantive effect of Tagesschau reports on the approval of Aktuelle Kamera.Before the liberalization, viewers significantly decreased their approval of Aktuelle Kamera when Tagesschau reported on the emigration wave.After the regime had abolished censoring, West German news coverage no longer affected the ratings.
Model 5 includes a three-way interaction between SED, Tagesschau, and Liberalization period to test for heterogeneous effects.Again, the coefficient of the interaction term between Tagesschau and SED, which gives the effect of West German news reports among party members compared to nonmembers before the liberalization, is statistically insignificant. 17Respondents with links to the regime reacted to emigration reports on WGTV in the same way as individuals without SED membership.Both groups equally downgraded the subsequent edition of Aktuelle Kamera.SED members and nonmembers did not significantly differ in their reaction to Tagesschau reports before or after the liberalization of GDR media. 18 Consistent with the results of our basic analyses, we do not find a moderating effect of regime linkages.

Robustness Checks
Our findings remain stable for a wide range of robustness checks.First, we rerun our main Models 3 and 4 using ordered probit regression models (Table SI.5.1).Second, we include either random intercepts or fixed effects on the district level, and random intercepts on the household and respondent levels to control for 17 Figure SI.4.1 visualizes the substantive effect. 18As can be seen from Table SI.5.3 the results are robust when we substitute the dummy variable Tagesschau with a continuous measure of the duration Tagesschau reported on emigration each day.
unobserved heterogeneity (see Table SI.5.2).Third, we control for the weekday of the GDR news broadcast to account for potential biases from systematically different approval ratings on specific days of the week (Table SI.5.4).Fourth, we exclude and use alternative measures for the control variable Average rating last week to rule out that our results are sensitive to the modeling of potential time trends (Table SI.5.5).Finally, we rerun our analyses without observations from November and December, as the proportion of self-reported SED membership slightly decreased during these months (Figure SI.3.1 and Table SI.5.6).

Observable Implications and Alternative Explanations
Next, we empirically test observable implications and discuss potential alternative explanations for our findings.First, we test the observable implication of our claim that Tagesschau served as an alternative source of information and presented the decisive tool to falsify domestic media content by drawing on spatial variation in access to West German television.Second, we assess whether our results are merely picking up general time trends in people's sentiments that are independent of the regime's censoring practices.Third, we check whether the series of anti-regime demonstrations and the noncoverage by GDR state news drive our results.All tests provide corroborative evidence for our findings.Alternative explanations are unlikely to account for our results, whereas observable implications support our suggested mechanism.

Spatial Variation in Access to West German Television
First, we exploit the spatial variation in the access to West German television.Due to limited signal strength and topography, people in some parts of the GDR had little or no access to WGTV. 19 (Bursztyn, and Cantoni 2016, 30). 20To test for different reactions in regions with and without WGTV, we perform our main analysis for the time before the liberalization, now including an interaction term between the variables Tagesschau and No WGTV.
In line with our suggested mechanism, we find that evaluations by respondents without WGTV were indeed unaffected by Tagesschau reports.In contrast, people with WGTV access significantly disapproved of the censored state news when the West German Tagesschau had covered 19 This exogenous variation has been used to study protest (Crabtree, Darmofal, and Kern 2015;Kern 2011), regime legitimacy (Kern andHainmueller 2009), and consumer behavior (Bursztyn, and Cantoni 2016). 20Applying the coding rule by Crabtree, Darmofal, and Kern (2015), this classification corresponds to any minimum signal strength threshold between -84.7 and -77.8 dBm.Appendix SI.6 offers additional information on operationalization (including an explanation for the use of a binary measure), identification, and robustness checks.Table SI.6.1 provides the regression results using more extreme cut-off values and alternative coding decisions.Figure SI.6.3 shows that the substantive interpretation of our results remains unchanged.We thank Leonardo Bursztyn and Davide Cantoni for kindly sharing their data.the emigration crisis before.Figure 6 visualizes the corresponding first-difference effects from Tagesschau reports among respondents from areas with and without WGTV access.The result strongly corroborates our suggested mechanism in that domestic censorship triggers backlash if people can draw on alternative sources of information.

Placebo Test
Next, we scrutinize alternative explanations for our findings.Changes in the approval of Aktuelle Kamera might simply reflect a general time trend in the rating of the national broadcast program.People might use the survey to voice their general current dissatisfaction with the regime.
To check whether this is the case, we run a placebo analysis regressing recipients' approval of the GDR Tele-Lotto broadcasts (live drawing of the lottery numbers) on the same set of explanatory variables. 21Analogous to the ratings of East German news, people rated Tele-Lotto on the  Table 3 shows that this is not the case.Neither the variable Censorship period (Model 6) nor the variable Tagesschau (Model 7) is significantly correlated with evaluations of Tele-Lotto.The negative coefficient for Liberalization period in Model 7 suggests that respondents' approval of Tele-Lotto even decreased after the liberalization of reporting practices, which clearly contradicts a general time trend behind our main findings.Furthermore, the models only account for 3% of the variance in the Tele-Lotto ratings as compared to 27% in Aktuelle Kamera ratings. 23The placebo test corroborates our main findings.The changing approval rates for East German news are unlikely to reflect a time trend in people's general sentiments.

Demonstrations in the Fall of 1989
Finally, our models might not reflect the effects of state media's censorship of refugees, but that of the subsequent anti-regime protests.From a theoretical perspective, the demonstrations between September and December 1989 were also important and persistent events censored by state media for a substantial period of time, whereas WGTV provided some coverage.The censorship of nationwide protests might have unfolded a similar backlash effect on people's assessment of the East German news.To test whether this is the case, we code all Tagesschau reports on anti-regime demonstrations and add the variable Tagesschau demonstrations as a control.Table 4 shows our analyses of the full year (Model 8) and the time before the liberalization only (Model 9).The results demonstrate that the substantive interpretation of our main finding remains unchanged when we control for reports on demonstrations. 24While the variable Tagesschau demonstrations also has a negative effect, WGTV reports on the emigration crisis still significantly reduce the approval of Aktuelle Kamera.
The results suggest that Tagesschau reports on emigration were the main driver for the decreasing approval of East German news.Both the emigration crisis and the decrease in approval of Aktuelle Kamera had started already one month before the first protest events.

Conclusion
This article studies whether and under which conditions censorship backfires against authoritarian regimes.We find that alternative sources of information enable people to identify misinformation on censored state media.
Our statistical analysis of East German media surveys throughout the revolutionary year of 1989 clearly shows that the socialist regime paid a high price for censoring the country's emigration crisis.In view of the regime's blatant censorship strategy, the population increasingly disapproved of the national news program.Consistent with our theoretical argument, we empirically show that the disapproval of the regime's main newscast was closely associated with foreign reports on the emigration crisis.
West German news exposed censorship on state television as part of the regime's larger misinformation campaign.
Our results present the first empirical evidence that excessive information manipulation may work against the regime's interests by diminishing the reliability of state media (Little 2017;Shadmehr and Bernhardt 2015).
In contrast to laboratory research, we do not find heterogeneous effects.Pro-regime dispositions did not moderate people's rejection of censorship.Members of the ruling party opposed state media's censoring activities as much as nonmembers.Future research might examine under which conditions regime supporters endorse or reject censorship.In addition, our analysis questions the persistence of backfiring effects.We find that respondents' approval ratings for state media recovered soon after the regime had ceased its censoring practice and adopted a more objective style of reporting.
Our article has four important implications.First, more censorship does not imply more stability.Whereas research on social movements and state repression commonly conceives of censorship as an effective government tool to contain resistance, we show that media censorship entails a backlash potential on its own.Blatant censoring of state media reveals the regime's deceptive intentions and its vulnerability to a large audience.Instead of terminating public resistance, censorship may thus help social movements to gain momentum and mobilize a critical mass.Concerning the German Democratic Republic, Hirschman (1993) and Richter (2009) claim that the emigration wave represented the country's first opposition movement that successfully changed the power relationship between the regime and its citizens, setting the stage for the Peaceful Revolution.Future research might study when outrage over misinformation translates into resistance in the streets and when censorship successfully prevents organized opposition.
Second, alternative sources of information may provoke distrust in traditional media and political elites.Our findings demonstrate that foreign efforts to provide people with alternative information can influence public opinion, putting pressure on a censoring regime.In some respects, East and West Germany may be interpreted as a most likely case to observe such effects: People in both states shared a historical and cultural background, spoke the same language, and showed great mutual interest.On other dimensions, however, the German Democratic Republic appears as a hard case.The socialist regime had been extensively indoctrinating its citizens for decades to render them immune against foreign subversion and exerting strong social pressures for conformist behavior.This suggests that external information campaigns can foster liberalization processes even in highly ideological regimes.Recent such efforts include the "Flash Drives for Freedom" campaign and the BBC's Korean-language radio service for North Korea (Ryall 2017).The harsh reaction by the North Korean government against these initiatives suggests that also the regime perceives the influx of external information as distinctly threatening.
Third, past failures of all-encompassing media censorship might explain why current authoritarian regimes appear to use it in a more restrained and nuanced way.Authoritarian regimes may have learned from historic failures.Especially the breakdown of the German Democratic Republic might serve as a cautionary tale, as the East German regime's inability to seal off society from alternative media resembles the challenges faced by modern dictatorships in times of the Internet.In order not to suffer the same fate, current autocracies might have adjusted their repressive tool kit accordingly and refined their misinformation strategies.
Finally, our findings suggest that campaigns aimed at undermining trust in traditional journalism may even cause political backlash against righteous media and regimes that grant full freedom of expression-including democracies.If such campaigns manage to build confidence, they can spread doubts about the integrity of established news sources and fuel people's perception of being systematically misinformed.Quality journalism appears to be particularly vulnerable to accusations of censorship since it is based on the obligation to check facts carefully and refrain from reporting when a story's authenticity remains in doubt.Given the backlash potential from perceived censorship, states should therefore have an interest in actively protecting quality journalism from such illusive campaigns.

American
Journal of Political Science, Vol.00, No. 00, xxxx 2019, Pp. 1-17 C 2019 The Authors.American Journal of Political Science published by Wiley Periodicals, Inc. on behalf of Midwest Political Science Association DOI: 10.1111/ajps.12501This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

FIGURE 4
FIGURE 4 Effects of Censorship and Liberalization on the Approval of GDR State News

FIGURE
FIGURE 5 Effect of Tagesschau Reports onApproval Rates for GDR State News

FIGURE
FIGURE 6 Effect of Tagesschau Reports amongPeople with and without WGTV

FIGURE 2 East and West German News Coverage from July to December 1989
8Many reports included interviews with emigrants and video footage, while news anchors repeatedly referred to Note:The plot shows the differences between Tagesschau (West German news) and Aktuelle Kamera (East German news) coverage of the emigration wave in days per week.The dashed line marks the liberalization of GDR state media.Positive y-values indicate that Tagesschau reported on the emigration crisis more often than Aktuelle Kamera, and vice versa.thecensoring practices by the East German state media.Tagesschau started reporting on the emigration wave in July 1989.Between the second week of August and the end of September, it covered the emigration wave on a daily basis.

FIGURE 3 Average Approval of GDR State News throughout the Year 1989
Table SI.1.1 provides coding details and summary statistics.

TABLE 1 Approval of GDR State News: Basic Analyses Using Time Periods
Note:The table reports coefficients with standard errors clustered on households.* p < .05,* * p < .01,* * * p < .001 a Reference category: Baseline period.

TABLE 2 Effect of Tagesschau Reports on the Approval of GDR State News
Note:The table reports coefficients with standard errors clustered on households.* p < .05,* * p < .01,* * * p < .001 Our theory predicts that Tagesschau reports should not be correlated with approval ratings of Aktuelle Kamera in areas without WGTV access since respondents were not able to identify censorship via alternative sources.The variable No WGTV indicates whether the respondent lived in a district with limited or no WGTV access.Consistent with prior research on 1989 East Germany, we employ a dichotomous measure of WGTV access based on signal strength (e.g., Crabtree, Darmofal, and Kern 2015; Kern 2011; Kern and Hainmueller 2009) and code the variable No WGTV to 1 if a respondent lived in the districts of Dresden, Neubrandenburg, or Rostock, and 0 otherwise

TABLE 3 Placebo Test: Approval of Tele-Lotto
Note:The table reports coefficients with standard errors clustered on households.*p< 0.05, * * p < 0.01, * * * p < 0.001 same 5-point scale each week.22Ifpeople's disapproval of Aktuelle Kamera was unrelated to news censorship and if our estimates picked up a general trend in public sentiment, the time periods should have similar effects on the ratings of Tele-Lotto.

TABLE 4 Approval of GDR State News Controlling for Reports on Demonstrations (8) (9)
Tagesschau reported on demonstrations in the GDR for the first time.Another reason why reports on emigration exhibit more explanatory power than those on protest is that West German journalists were better able to provide detailed footage and firsthand testimonies of the emigration crisis because these developments took place outside GDR territory.Conversely, East German security forces closely observed foreign journalists and largely prevented Western camera teams from filming protests in the early days of the uprising.Based on the results of our content analysis of Tagesschau, the emigration wave received substantially more attention during the censorship period than the demonstrations in East German cities.