The World Bank Economic Review, 36(2), 2022, 455–487 https://doi.org10.1093/wber/lhab018 Article Opposition Media, State Censorship, and Political Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Accountability: Evidence from Chavez’s Venezuela Brian Knight and Ana Tribin Abstract This study investigates the effects of state censorship in the context of the 2007 government closing of RCTV, a popular opposition television channel in Venezuela. Some parts of the country had access to a second opposition channel, Globovision, while other parts completely lost access to opposition television. The first finding, based upon ratings data, is that viewership fell on the progovernment replacement, following the closing of RCTV, but rose on Globovision in areas with access to the signal. Based upon this switching, the paper investigates whether support for Chavez fell in areas that retained access to opposition television, relative to those that completely lost access. Using three measures, Latinbarometer survey data, electoral returns, and data on protest activity, the second finding is that support for Chavez fell in municipalities that retained access to opposition television, relative to municipalities that lost access to opposition television. Taken together, these two findings suggest that voters switching from censored outlets to uncensored outlets can limit the effectiveness of state censorship. JEL classification: D70, D80 Keywords: censorship, media bias, voting 1. Introduction Opposition media has the potential to help voters hold incumbent politicians and parties account- able. By reporting on the performance of government, including coverage of scandals and corruption, the media provides information to voters that is useful when deciding whether or not to reelect in- cumbent politicians and parties.1 In this way, the media also provides incentives for officials to act in the best interests of their constituents.2 Despite these potential benefits of opposition media to society, Brian Knight (corresponding author) is a professor at Brown University, Providence, Rhode Island, USA; his email address is brian_knight@brown.edu. Ana Tribin is a Public Policy Specialist at the Regional Bureau of Latin America and the Caribbean of the United Nations Development Programme (UNDP); her email address is ana.tribin@undp.org. The authors thank au- diences at the California Institute of Technology, Columbia University, Harvard University, London School of Economics, Nottingham University, Stanford Graduate School of Business, the University of Warwick, the Paris Empirical Political Econ- omy Seminar, the New York City Media Seminar, and the RIDGE/LACEA-PEG Workshop on Political Economy. Julia Cage, Greg Martin, Adam Szeidl, Ferenc Szucs, and Maria Petrova provided helpful comments. Thanks to Juan Pereira for careful research assistance. A supplementary online appendix is available with this article at The World Bank Economic Review website. 1 On media coverage of government performance, see, among others, Ferraz and Finan (2008). 2 On these incentives provided by media coverage, see, among others, Besley and Burgess (2002) and Ferraz and Finan (2011). © The Author(s) 2021. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 456 Knight and Tribin incumbent governments, given the threats to their power, face incentives to limit press scrutiny of their office via censorship.3 When the government censors opposition media, what are the effects on support for incumbent politicians? Is censorship effective? And how does the answer to this question depend upon market structure and voter responses to censorship? Importantly, opposition voters might respond to censorship and the associated change in ideological content by switching away from censored out- lets and towards other opposition outlets (Durante and Knight 2012).4 Thus, by adjusting their media Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 consumption patterns, voters can limit the effectiveness of government censorship of opposition media outlets.5 This paper investigates these issues in the setting of Venezuela during the presidency of Hugo Chavez and the closing of RCTV, a popular opposition television channel with national coverage, by the Chavez government. The change occurred overnight on May 27, 2007, following the expiration of RCTV’s 20- year broadcast license and the government’s decision to not renew the license. The government replaced RCTV with a progovernment public channel TVES, which inherited both the channel number and the broadcast infrastructure of RCTV. Two large metro areas, Caracas and Valencia, retained access to oppo- sition television via Globovision, the remaining source of opposition news available via broadcast signal following the closing of RCTV, while the remainder of the country no longer had access to opposition television via broadcast signal. Incorporating information on the Globovision broadcast towers and, us- ing the irregular terrain model, we first predict the strength of the Globovision broadcast signal across municipalities. While the Globovision signal is not randomly assigned and the decision to close RCTV likely reflected political considerations, RCTV was closed following the expiration of a 20-year broad- cast license. Given this, the key identifying assumption is that trends in support for Chavez in areas with the Globovision signal would have followed trends in support in areas without the signal following the closing of RCTV. Using Nielsen ratings data on viewership of each show for four metro areas, the first finding is that viewers responded, in terms of adjusting their media consumption patterns, to the change in ideological content. In particular, there is a reduction in viewership of TVES, relative to RCTV, following the change. Likewise, there is a corresponding increase in viewership of Globovision, the remaining opposition televi- sion channel, in the two metro areas with access to the broadcast signal. Taken together, these two findings suggest that voters respond to censorship of opposition media outlets by switching to other opposition channels when available. Using geographic variation in the Globovision signal across municipalities, smaller units than the metro areas in the viewership data, our analysis considers annual survey data (Latinbarometer) conducted both before and after the closing of RCTV. In particular, there is a reduction in approval ratings for Chavez in those municipalities with access to the Globovision signal, relative to municipalities without such access, following the closing of RCTV. Addressing possible threats to our identifying assumption, these results are robust to observable controls and an analysis that focuses on the narrow geographic region surrounding the Caracas and Valencia metro areas. Our results also cannot be explained by trends in internet use, a potential alternative explanation. Finally, in a further investigation of dynamics, the effects tend to fade 3 Limitations on opposition media can take a variety of forms, from licensing requirements, to restricting access to gov- ernment officials, to outright censorship. Censorship itself can be manifested in a variety of ways, from the closing of opposition media outlets, to content requirements, to the outright blocking of specific media reports. 4 This mechanism also highlights the importance of market structure since it relies on the existence of at least one uncen- sored outlet. Of course, if other uncensored opposition outlets are not available, viewers can respond to censorship by simply tuning out, reducing their overall media consumption (Knight and Tribin 2019). 5 Another possible response involves discounting, by which rational voters may simply account for censorship when processing content, understanding that censorship implies a loss of access to information that might reflect negatively on the government. This type of discounting should reduce the impact of censorship (Chiang and Knight 2011). The World Bank Economic Review 457 over time, as access to the Globovision signal matters less towards the end of our sample period when compared to the period just following the closing of RCTV. There are similar findings when using two other measures of support for Chavez: electoral returns data and protests data, both of which vary across space and time. Taken together, these findings document that viewers with access to the broadcast signal increased their viewership of Globovision, leading to a reduction in support for Chavez, relative to viewers in areas that Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 no longer had access to opposition television via broadcast signal, following the closing of RCTV. Thus, censorship appears to have increased support for the Chavez government in areas without access to other opposition television channels, relative to areas that retained access to opposition television. This suggests that behavioral changes in media consumption patterns are an effective response to censorship. Yet these differences across municipalities fade over time, suggesting that viewers in areas without other opposition channels might adjust to their new media environment and discount censored material accordingly. These ideas are formalized in the context of a model of an informative media and government censorship of negative media reports. In the model, voters account for censorship when processing of media reports and have a preference for opposition media. While voters might be manipulated by censorship in the short run, they discount appropriately in the long run after learning about the censorship environment from the lack of negative media reports about the government. 2. Related Literature This paper comes at the intersection of two literatures on the political economy of the media: one involv- ing media consumption patterns according to ideology and another involving the persuasive effects of the media. Studies on media consumption patterns include Durante and Knight (2012), who show that right- leaning viewers were more likely to watch public television during periods in which it was controlled by Silvio Berlusconi, the leader of the center-right coalition in Italy during the 1990s and early 2000s. Like- wise, Knight and Tribin (2019), also using data from Venezuela, examine high frequency changes in ratings for news programs that are interrupted by cadenas, government propaganda that is not announced in ad- vance to stations and viewers. The key finding is that the drop-off in ratings is particularly pronounced on opposition-aligned channels when they are airing news programming, consistent with a preference for like-minded information (Gentzkow and Shapiro 2010; Mullainathan and Shleifer 2005). In the US context, Martin and McCrain (2019) document small decreases in viewership following the acquisition of local television stations by a national conglomerate (Sinclair). In the Russian context, there was a drop in viewership of NTV, a key opposition channel, following its takeover by the state-controlled outlet Gazprom in 2001 (Gehlbach and Sonin 2014). This is consistent with a lack of opposition television in Russia, as the other two key channels were both state controlled (Enikolopov, Petrova, and Zhuravskaya 2011). Taken together, these studies document that viewers are not passive receivers of information and instead respond to changes in ideological content. On the persuasive effects of opposition media and state censorship, Enikolopov, Petrova, and Zhu- ravskaya (2011) find that access to NTV, the key opposition channel in Russia, reduced support for the incumbent government in the 1999 elections, and there is some evidence that this effect persisted even after the 2001 takeover of NTV by Gazprom. Adena et al. (2015) document that government-controlled radio facilitated the rise of the Nazi party in Germany during the 1930s. Chen and Yang (2019) conduct a field experiment in which students in China are provided free access to the uncensored internet. When combined with encouragement, acquisition of information increases, leading to changes in knowledge, beliefs, and attitudes.6 Roberts (2018) covers censorship in China, arguing that inconveniencing users 6 Regarding the supply side, King, Pan, and Roberts (2013) show that censors in China allow a significant amount of controversial material on social media to go uncensored but are particularly likely to block posts facilitating collective 458 Knight and Tribin for accessing information is an important form of censorship. She also draws an important distinction between observed and unobserved censorship. In our setting, the closing of RCTV was widely observed, making it easier for voters to adjust to the new media environment.7 While there are only a few empirical studies on the persuasive effects of censorship, there is of course a large and active literature, more generally, on the persuasive effects of media bias. Evidence that biased outlets may influence voting outcomes include George and Waldfogel (2003), DellaVigna and Kaplan Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 (2007), Gerber, Karlan, and Bergan (2009), Snyder and Stromberg (2010), Gentzkow, Shapiro, and Sink- inson (2011), Martin and Yurukoglu (2017), Prat (2018), and Durante, Pinotti, and Tesei (2019). Chiang and Knight (2011) argue that any persuasive effects of bias might be limited by voter discounting. In particular, based upon responses to newspaper endorsements in US presidential elections, voters appear to discount biased information. That is, crossover endorsements, those for Republican candidates from left-leaning papers and for Democratic candidates from right-leaning papers, are more influential than less surprising endorsements, those for Republican candidates from right-leaning papers and for Demo- cratic candidates from left-leaning papers. This limits the influence of media bias in the sense that shifting a newspaper’s ideology towards the right reduces the electoral impact of an endorsement of a Republican candidate by that newspaper. Relative to these two literatures, our paper is one of the first to simultaneously consider media con- sumption and persuasion. The key advantage of this approach is that we can quantify the degree to which behavioral responses by consumers, switching to noncensored outlets when available, limit the persuasive effects of censorship. That is, our paper conducts counterfactuals in which we measure the persuasive effects of censorship under alternative media consumption environments, including one in which viewers do not switch to noncensored outlets and another in which both Globovision and RCTV are closed. These approaches are only possible via our simultaneous consideration of media consumption and persuasion in the same setting. The only other paper of which we are aware to simultaneously consider media consumption and per- suasion is Martin and Yurukoglu (2017), who study these issues in the context of cable TV news in the United States. One finding is that the media may lead to increased polarization via a feedback loop, with Fox News, for example, shifting viewer ideology towards the right, further increasing their preference for right-leaning content. Our results, by contrast, document that the electoral effects of censorship, if anything, seem to fade over time. We argue that this is consistent with a model in which viewers learn about both government performance and the censorship environment from news reports. While cen- sorship might be effective in the short run, viewers learn about the degree of censorship over time and discount accordingly. That is, voters recognize that censorship entails a loss of information that might reflect negatively upon the government. Due to this learning and discounting, the effects of censorship may tend to fade over time. Our study is also closely related to independent work by Kronick and Marshall (2018), who, like us, study the electoral effects of the closing of RCTV in Venezuela. Their comparison is between areas with significant cable penetration (RCTV reemerged on cable later in 2007) and areas without significant cable penetration, the idea being that areas with access to cable are less influenced by the closing of RCTV, when compared to areas with less access to cable. Relative to our study, their geographic variation is more fine grained. In particular, while our analysis uses municipalities as the unit of observation, they use polling stations as the unit of observation and exploit variation across these smaller geographic units and within action. Consistent with this idea, Enikolopov et al. (2020) document that social media availability plays a key role in protest activity against the government in Russia. 7 Hobbs and Roberts (2018) investigate how social media users responded to the blocking of Instagram by the Chinese government and document increased adoption of virtual private networks (VPNs), leading an unexpected increase in access to censored information. The World Bank Economic Review 459 parishes (geographic units that are larger than polling stations and smaller than municipalities). They study electoral returns and find that retaining access to RCTV via cable access led to a relative increase in support for Chavez. They argue that this finding is consistent with viewers who lost access to RCTV, punishing Chavez for closing RCTV at the ballot box. Our study, by contrast, does not use variation in cable penetration but instead exploits geographic variation in access to the Globovision signal. We argue that our finding, losing access to opposition TV leading to an increase in support for Chavez, can be Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 explained by a different mechanism : switching to another opposition channel, Globovision. Given these two different mechanisms and the different sources of geographic variation, the two sets of findings are not inconsistent with one another, and we thus view the two studies as complementary in nature. Finally, this paper is also related to a literature on media capture. Besley and Prat (2006) develop a model in which, even in the absence of state censorship, private media outlets might self-censor negative information about the government in exchange for favorable policies. Media capture benefits governments by concealing potentially damaging information from the electorate, thereby increasing reelection rates. Such media capture is less likely under media pluralism, providing an additional motive for competition. Di Tella and Franceschelli (2011) and Szeidl and Szucs (2021) empirically examine media capture, doc- umenting that media outlets provide favorable coverage of the government in exchange for government advertising. In terms of incentives for favorable coverage in our context, the government provided sticks, revoking the broadcast license of RCTV, rather than carrots, in response to unfavorable coverage. 3. Institutional Context This section provides background on the political situation in Venezuela, with a focus on the role of television in politics. It begins by describing Chavez’s rise to power and the key elections during our sample period, followed by a discussion of the main television channels, including RCTV and Globovision, and the events surrounding the closing of RCTV in 2007. 3.1. Political Situation Chavez, first elected in 1998, promised a “Bolivarian revolution,” designed to eliminate exclusion, poverty, and government corruption. Since the beginning of Chavez’s term, the right-wing opposition was com- mitted to removing him from office.8 In April 2002 the opposition led a coup, which failed after some initial successes. Later that year, in December 2002, the opposition organized a strike in the oil industry, aimed at toppling Chavez, but, despite lasting three months, this too was ultimately unsuccessful. Two years later, in 2004, the opposition organized a recall referendum against Chavez. During our sample period, there were a number of elections, and our empirical analysis of electoral returns focuses on five key elections during Chavez’s term. In 2004, the right-wing coalition tried to re- move Chavez from power via a presidential recall referendum, as mentioned above, but Chavez won, with 59 percent voting against the recall.9 Two years later, in the 2006 presidential election, Chavez was reelected to a second term, with 63 percent of the vote, against the right-wing candidate Manuel Ros- ales. The following year, in December 2007, Chavez proposed to amend the Constitution via a referen- dum. These amendments included empowering the president to declare states of emergency for unlimited periods, extending presidential terms from six years to seven years, and, perhaps most importantly, re- moving presidential term limits, which restricted presidents to just two terms in office.10 Chavez lost this 8 The description of attempts to revoke President Chavez’s mandate are extensively documented in Wilpert (2007), Corrales and Penfold (2011), and Nelson (2009). 9 Hsieh et al. (2011) analyze the labor-market implications of publicly supporting the recall campaign. 10 See https://elpais.com/internacional/2007/12/03/actualidad/1196636401_850215.html. 460 Knight and Tribin Table 1. Key Television Channels Channels Ownership/ideology Metro areas Dates Daily news hours Globovision Private/opposition Caracas and Valencia All 19.57 RCTV Private/opposition National Until May 27, 2007 4.10 TVES Public National Starting May 28, 2007 3.01 Televen Private/moderate National All 3.83 Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Venevision Private/moderate National All 3.65 VTV Public National All 14.62 Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Note: Ideology determined by political content. Daily news hours represent averages for each station during the sample period. referendum, with only 49 percent of voters supporting these amendments designed to increase his powers. Another Constitutional referendum was held less than two years later, in February 2009, and Venezue- lans did vote this time, with 54 percent supporting, to remove terms limits for the president and other key offices. This opened the door for Chavez to seek reelection for a third term in October 2012, and he won, with 54 percent of the votes, against the right-leaning candidate Henrique Capriles. This election occurred during Chavez’s illness, and he died just a few months later, in March 2013. He was replaced on an interim basis by his vice president, Nicolas Maduro, who then won a disputed election in April 2013 and has served as president thereafter. 3.2. Television in Venezuela Television is an important source of news for Venezuelans during our sample period. According to Lat- inbarometer survey data, the focus of our later analysis, 35 percent of respondents reported watching news on television all seven days during the prior week, with only 14 percent reporting that they watched news none of the days, during our baseline periods 2005 and 2006.11 News viewership on television is higher than corresponding news consumption from both newspapers (23 percent reported reading news all seven days and 23 percent also reported none of the days) and radio (27 percent reported listening to news all seven days and 29 percent reported listening none of the days). Table 1 provides an overview of television channels in Venezuela. While the key public channel (VTV) tends to align itself with Chavez, the private television channels have tended to align their ideology with the opposition, especially at the start of Chavez’s term, supporting the coup in 2002 and providing extensive coverage of antigovernment protests during the oil strike later that year.12 After these events, the tension between the private media and government was at its peak, and Chavez’s stance became very aggressive, referring to the major private television channels (Venevision, RCTV, Globovision, and Televen) as the “four horsemen of the apocalypse.”13 In 2004, before the recall referendum, Chavez met with the owner of Venevision, which thereafter moderated their tone towards Chavez.14 Then, Televen followed the initiative to moderate the anti-Chavez speech around the same period.15 Following this, only two opposition channels remained. RCTV (Radio Caracas Television) is the old- est station, had the highest ratings, and had national broadcast coverage. In addition to its well-known antigovernment tone, RCTV was a major producer of entertainment programming, including soap operas 11 These survey questions change over time and thus comparable statistics cannot be developed during the period after the closing of RCTV. 12 See Nelson (2009) and Dinneen (2012). 13 Reporters without Borders (2003) documents that Chavez accused the private channels of “inciting rebellion and dis- respect for legitimate institutions and authorities,” “broadcasting false, misleading or biased news reports,” “harming the reputation and good name of persons or institutions,” and promoting “subversion of public and social order.” 14 http://www.nytimes.com/2007/07/05/world/americas/05venez.html?_r=1&. 15 Wilpert (2007). The World Bank Economic Review 461 Figure 1. EU Media Monitoring of 2006 Presidential Election. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: EU Report (http://www.eods.eu/library/FR% 20VENEZUELA%202006_en.pdf). Note: Coverage of Chavez (solid area) and opposition (dashed area) during the 2006 presidential election campaign, separately by station. and talk shows.16 Globovision is a news-only channel, providing nearly 20 hours of news per day (table 1) but its broadcast signal was available in only two large metro areas, Caracas and Valencia. While Globo- vision content is available nationally via cable, penetration is relatively low during our sample period, and RCTV was the only opposition channel available via broadcast signal in the remainder of the country, outside of these two metro areas. To summarize, there was a partitioning of television channels into opposition (RCTV and Globovision), moderate (Televen and Venevision), and progovernment (VTV). This is consistent with media monitoring during the 2006 presidential elections. In particular, fig. 1, based upon data reported in EU-EOM (2006), documents that RCTV and Globovision devoted roughly two-thirds of their coverage to the opposition party and only one-third to Chavez. The tone of this coverage, not reported here, followed similar patterns, with decidedly positive coverage of the opposition and decidedly negative coverage of Chavez. The main public channel, VTV, by contrast, provided roughly 85 percent of their coverage to Chavez and only 15 percent to the opposition. This coverage of the opposition was negative, with coverage of Chavez disproportionately positive. The two moderate channels, Venevision and Televen, were also biased in favor of the government in terms of the share devoted to each party. In terms of tone, moderate channels tended to provide positive coverage of both parties. Thus, media monitoring data support the notion that there were only two opposition channels in Venezuela in 2006 and thus only one following the closing of RCTV in 2007. 16 While RCTV provided only 4 hours of news per day (table 1), these programs had high ratings, as will be shown below. 462 Knight and Tribin 3.3. The Closing of RCTV The 20-year broadcast license for RCTV, which was granted on May 27, 1987, expired on May 27, 2007 and was not renewed by the government. In terms of the process leading up to the closing of RCTV, the nonrenewal was first announced five months prior, on December 28, 2006, and the decision was then upheld by the Supreme Court of Justice on April 17, 2007.17 The government’s rationale for not renewing the license of RCTV had two key components: the alleged violations of broadcast laws and the Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 participation in the coup and strike in the oil sector.18 RCTV was forced to stop broadcasting at midnight on May 27, 2007, and the channel was replaced overnight by TVES, a public channel. TVES inherited both the channel number of RCTV and its broadcast infrastructure. The government appointed a new president for TVES, Lil Rodríguez, with much of the staff of RCTV losing their positions overnight.19 RCTV did reemerge on cable later that summer, under the name RCTV International, but was later (in early 2010) shut down again.20 The decision to take RCTV off the air was marked by significant public protests, and 70 percent of Venezuelans disapproved of the decision to not renew the license.21 As discussed above, Globovision was the only remaining broadcast channel aligned with the opposition but, unlike RCTV, which had a national broadcast infrastructure, the broadcast coverage of Globovision was limited to two metro areas. While Globovision was available nationwide via cable subscription, cable penetration, according to Conatel, was relatively low in Venezuela during our sample period, ranging from 17 percent in 2004, the start of our analysis, to 40 percent of households by 2011.22 Thus, broadcast was a more important delivery means than cable, although the gap between the two narrows towards the end of our sample period. A key question is whether Globovision responded to the closing of RCTV by changing their ideological placement, and an implicit assumption in our analysis is that Globovision remained a key source of infor- mation to opposition voters.23 According to media monitoring of the 2012 presidential election, Capriles, the opposition candidate, received 64 percent of the Globovision coverage, which is nearly identical to the 65 percent share of coverage received by Rosales in the 2006 election, as documented in fig. 1.24 While Globovision was purchased by a progovernment investor in 2013, it continued to provide coverage of the opposition during key events of 2014, such as the street demonstrations held early that year.25 Taken together, there is little evidence that Globovision changed their ideological placement in response to the closing of RCTV. To summarize, Venezuela has been characterized by a tense conflict between the government and the opposition, and this divide was also reflected in the media, with public channels aligned with the govern- ment and two key private channels, RCTV and Globovision, aligned with the opposition. Following the closing of RCTV and the opening of TVES, Globovision became the only source of opposition television, and this was available via broadcast signal in only two metro areas. This left the majority of the territory 17 See https://en.wikipedia.org/wiki/RCTV. 18 Dinneen (2012). 19 https://en.wikipedia.org/wiki/TVes (accessed July 11, 2019) and http://www.el-nacional.com/noticias/sociedad/rctv- anos-cierre-amigo-para-siempre_241770 (accessed July 11, 2019). 20 Wilpert (2007). 21 Wilpert (2007). 22 Similarly, Weisbrot and Ruttenberg (2010) show that the audience share for cable is small, relative to broadcast TV, increasing from roughly 10 percent in 2004 to around one-third in 2010. 23 It is natural that Globovision might have moderated their tone, viewing the closing of RCTV as a threat to their own existence. Yet it is also possible that there was a backlash effect, with Globovision further shifting towards the opposition. 24 The 2012 monitoring was conducted by the Carter Center and the report is available at https://www.cartercenter.org/resources/pdfs/news/peace_publications/election_reports/venezuela-2012-election-study- mission-final-rpt.pdf (accessed May 3, 2019). 25 For evidence on this point, see https://www.american.edu/centers/latin-american-latino-studies/upload/bias-or- neutrality-report.pdf (accessed April 2020). The World Bank Economic Review 463 without any opposition television and with only moderate and progovernment channels. The next sec- tion examines the implications of this closing of RCTV for media consumption patterns, Chavez approval ratings, voting returns, and collective action. 4. Censorship and News Consumption Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Using ratings data, this section examines viewership of news programming in Venezuela during the time period before and after the closing of RCTV. The data are provided by Nielsen and measure ratings show by show and day by day, covering the years 2006 and 2007. Ratings are also provided separately for four metro areas (Caracas, Valencia, Barquisimeto, and Maracaibo) in Venezuela, and our analysis compares two metro areas (Caracas and Valencia) with access to the Globovision broadcast signal to two metro areas without access to the signal (Barquisimeto and Maracaibo). These metro areas consist of multiple municipalities, and the data do not include separate ratings for each municipality. Nonetheless, for con- sistency with our later municipality-based analyses, these two metro areas with Globovision, Caracas and Valencia, are referred to as Globo cities. In these data, shows are categorized according to their content, and, in some of our analyses, ratings of news programming are compared to ratings of non-news programming. Changes in the ratings of non-news programming, following the closing of RCTV, should be less affected by access to Globovision, which, as noted above, only provides news content. Our analysis also exploits the fact that ratings are provided separately for households with and without cable subscriptions. In particular, since Globovision is available via cable on a national basis, access to the Globovision broadcast signal should not matter in terms of viewership patterns for those households with cable subscriptions. 4.1. Viewership of News Programming This section begins by examining how viewership of news programming on RCTV and then the replace- ment, TVES, evolves on a week-by-week basis during the period 2006–2007 and for households with- out cable subscriptions. Importantly, TVES took over the RCTV broadcast infrastructure, used the same channel number, and began broadcasting almost immediately following the closing of RCTV. Thus, to the extent that viewers are passive and do not respond to changes in ideological content, there might be no change in viewership patterns following the closing of RCTV. Our investigation begins with an analysis of viewership of the channel number associated with both RCTV and TVES.26 As shown in fig. 2, there is an immediate drop-off in viewership of news programming on TVES, relative to viewership of news programming on RCTV, following the closing of RCTV on May 27, 2007 and for households without cable subscriptions. In particular, viewership falls from roughly 5 to 10 percent prior to the closing of RCTV (i.e., to the left of the dashed line) to very low levels following the closing of RCTV (i.e., to the right of the dashed line) for these households without cable subscriptions. Viewership of RCTV news tended to be higher in the areas with access to the Globovision signal and this difference, if anything, seems to grow larger just prior to the closing of RCTV. The low viewership for TVES news, by contrast, is apparent in both types of areas following the closing of RCTV. This result is consistent with a preference for opposition news programming. Given this finding of very low ratings for TVES, relative to RCTV, our analysis next examines whether these viewers switched from RCTV/TVES to other channels aligned with the opposition. For households without cable, this involved Globovision and, as noted above, this type of switching was only available in areas with access to the Globovision signal. As shown in fig. 3, there was indeed an increase in viewership of Globovision following the closing of RCTV for these households without cable subscriptions and with 26 Given our focus on this change in content, our analysis does not incorporate ratings of RCTV International, the cable channel that emerged during the summer of 2007. 464 Knight and Tribin Figure 2. RCTV/TVES Viewership for Households without Cable. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. Figure 3. Globo Viewership for Households without Cable. Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. The World Bank Economic Review 465 Table 2. Summary: Consumption of Opposition Television Before After Change Panel A: Globo cities ratings (AMR)a RCTV 8.91 0.00 −8.91 Globovision 1.30 2.25 0.95 Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Panel B: Globo cities average weekly minutes RCTV 153.43 0.00 −153.43 Globovision 106.85 184.94 78.08 Total 260.28 184.94 −75.35 Implied switching rate 50.89a Panel C: Non-Globo cities ratings (AMR)a RCTV 5.91 0.00 −5.91 Globovision 0.00 0.00 0.00 Panel D: Non-Globo cities average weekly minutes RCTV 101.77 0.00 −101.77 Globovision 0.00 0.00 0.00 Total 101.77 0.00 −101.77 Implied switching rate 0.00a Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Note: AMR represents average minute rating at the household level. a Percentages. access to the Globovision signal.27 Some of this increase is temporary, with a large increase in viewership during June 2007, immediately following the closing of RCTV. Note that this temporary increase may reflect viewer interest in Globovision coverage of protests associated with the closing of RCTV. Even following these protests, however, there is an increase in viewership of Globovision, relative to the period prior to the closing of RCTV. That is, ratings average 1.3 percent during the period prior to closing of RCTV but increase to an average of approximately 2 percent during the months July to December of 2007, representing an increase of over 50 percent. 4.2. Magnitudes To quantify these results, our analysis first calculates switching rates. That is, our analysis measures the fraction of the reduction in the consumption of opposition news on RCTV that was offset by the increased consumption of opposition news on Globovision in areas with access to the broadcast signal. An exclusive focus on ratings suggests relatively low switching rates: as shown in panel A of table 2, the viewership of opposition news on RCTV in areas with access to the Globovision fell from 8.91 percent of households to 0 percent following the closing of RCTV, and viewership of Globovision news rose from 1.30 percent to only 2.25 percent.28 But, as noted above, Globovision offers 19.57 hours of news per day, compared to 4.10 hours on RCTV. Thus, weekly viewership of news on RCTV fell by 153 minutes per household, while weekly viewership of news on Globovision rose by 78 minutes. Thus, overall consumption of opposition news fell by 75 minutes per week. As summarized in panel B of table 2, this implies an economically significant switching rate of 51 percent, meaning that roughly one-half of the reduction in viewership of news on RCTV was offset by increased viewership of Globovision where available. While these measures of switching rates focus on Globovision areas, the subsequent analysis of support for Chavez is based upon comparisons of areas with and without the Globovision signal. Indeed, key to 27 Consistent with this finding, fig. 4 documents that overall viewership of news programming, averaged across all channels, increased in areas with access to the Globovision signal, following the closing of RCTV, relative to areas without access to the signal. 28 These are average ratings in Globovision areas before and after the closing of RCTV in figs 2 and 3. 466 Knight and Tribin Figure 4. News Viewership for Households without Cable. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. our empirical strategy is the idea that, as a result of switching, consumption of opposition news fell less in areas with the Globovision signal, relative to areas without the signal. To address this issue, our analysis next measures the drop in consumption of opposition news in non-Globovision areas. As shown in panels C and D of table 2, viewership of opposition news in non-Globovision areas fell by 102 minutes per week. Thus, viewership of opposition news in areas with access to the Globovision signal (75 minutes per week) fell less, representing a relative increase in viewership of opposition news of 27 minutes per week. 4.3. Viewership of Non-News Programming For comparison purposes, our analysis next examines changes in the ratings of non-news programming. These ratings, following the closing of RCTV, should be less affected by access to Globovision, which is news only. As shown in fig. 5, there was a reduction in viewership of non-news programming across all areas following the closing of RCTV. This reduction in viewership of non-news is likely driven by the loss of popular entertainment programming on RCTV and the associated lack of interest in the entertainment programming offered by TVES. More importantly, the viewership patterns for non-news, both before and after the closing of RCTV, are similar for both areas with and without access to the Globovision signal. Thus, the relative decrease in consumption of news in areas without access to the Globovision signal cannot be solely explained by general trends in television viewership across both news and non-news programming in these areas. 4.4. The Role of Cable Our analysis next compares these trends in viewership for households with and without cable subscrip- tions. Since Globovision is available in all areas via cable subscriptions, one should expect a similar pat- tern of switching regardless of whether or not the area has access to the Globovision broadcast signal. As shown in fig. 6, there is again an immediate drop-off in viewership on RCTV/TVES following the closing The World Bank Economic Review 467 Figure 5. Non-News Viewership for Households without Cable. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. Figure 6. RCTV/TVES Viewership for Households with Cable. Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. 468 Knight and Tribin Figure 7. Globo Viewership for Households with Cable. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of viewership data from Nielsen Venezuela 2006 and 2007. Dashed line = cities with Globovision signal; solid line = cities without Globo- vision signal. Note: Weekly viewership averages based upon average minute rating at the household level. Vertical line represents the week in which the license for RCTV expired. of RCTV and the opening of TVES on May 27. Moreover, the patterns, for areas with and without access to the Globovision signal, are similar to those for households without cable (fig. 2). Our analysis next compares areas with and without access to the Globovision signal. As shown in fig. 7, there is again an in- crease in viewership of Globovision following the closing of RCTV. Importantly, these patterns are present in areas both with and without access to the Globovision signal, suggesting that the underlying demand for Globovision content is similar across these two areas. Taken together, the fact that viewership trends for these households with cable access is similar across areas with and without access to the Globovision signal suggests that our baseline results are not driven by other factors that may have changed in these areas. Instead, it is consistent with our interpretation that access to the Globovision signal led to changes in viewership patterns for households without cable following the closing of RCTV. 4.5. Summary Our analysis documents changes in viewership patterns following the government closing of RCTV and the corresponding change in station ideology. In particular, viewers quickly moved away from TVES, and there is also a corresponding increase in viewership of Globovision for households with access to the station either via broadcast signal or via a cable subscription. This leads to differences in both the ideological bundle consumed by viewers and in exposure to news overall. In the remainder of the paper, our analysis examines whether these changes in news consumption patterns affect approval for Chavez. 5. Censorship and Persuasion This section examines whether this change in news consumption shifted approval of Chavez. After de- veloping three measures of support for Chavez and explaining measurement of access to the signal, the empirical result are presented. The World Bank Economic Review 469 5.1. Measures of Chavez Approval The first measure is based upon annual survey data (2005–2011 from Latinbarometer) and examines trends in evaluations of Chavez, before and after the closing of RCTV in May 2007 and accounting for the Globovision broadcast signal. These surveys occur towards the end of the calendar year, and data thus include two years (2005 and 2006) of surveys prior to the closing of RCTV in May 2007 and five years of surveys after the closing of RCTV (2007–2011). Given that RCTV closed in May 2007, the Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 2007 survey, which occurred later that year, is considered posttreatment.29 In these data, respondents evaluate Chavez on a 1 to 10 scale (with 1 representing very bad and 10 representing very good).30 These surveys unfortunately do not include any information on cable subscriptions and thus our estimates can be interpreted as averaged over cable and noncable households. The second measure is based upon data on electoral returns from municipalities for key elections in Venezuela during the period before and after the closing of RCTV in May 2007. In particular, these data are based upon the five key elections discussed in more detail above. Two elections, the 2004 presiden- tial recall referendum and the 2006 presidential election, occurred prior to the closing of RCTV, and three elections, the 2007 and 2009 constitutional referenda and the 2012 presidential election, occurred after the closing of RCTV. For each election, voting returns are translated into a measure of the frac- tion of the electorate supporting Chavez, relative to the opposition. For the presidential elections, this is done in the natural way, by measuring the vote share for Chavez relative to the opposition candidate (Manuel Rosales in 2006 and Henrique Capriles in 2012). For the 2004 recall referendum, support for Chavez is measured as the fraction of the electorate voting against the effort to recall Chavez. For the two constitutional referenda, support for Chavez is based upon the fraction of the electorate in each munic- ipality that supported the referendum. This measure is natural given that both referenda were proposed by Chavez and were broadly interpreted as an attempt to consolidate political power in the office of the presidency. The third measure involves protest activity, a measure of political expression against the government and thus a lack of support for Chavez. These data are taken from the GDELT event database, which uses internet data in order to record physical activities around the world.31 For our purposes, there are four important pieces of information collected for each event. First, using information on landmarks, GDELT records the latitude and longitude of the event, and, for consistency with our previous analysis, our analysis focuses on events in Venezuela and then maps these events into municipalities. Second, GDELT records the purpose of the event in the form of an event code, and our analysis focuses on protest activities. Third, GDELT records the date of the event. Given that there are many days without any protest events in Venezuela, these daily level data are aggregated to the monthly level. Fourth, GDELT records the number of mentions of each event on the internet, and this is used as a proxy for the size of the event. As a robustness check, results are also presented based upon the number of events, rather than the number of mentions of events. This analysis focuses on the time period 2000–2014, covering roughly seven years before and seven years after the closing of RCTV. In total, our analysis focuses on 148 municipalities with at least one protest event recorded during our sample period, yielding a sample size of 26,640 month- municipality pairs. Despite aggregation to the municipality and monthly level, there are still a very large number of observations, roughly 96 percent of cases, without any protest events during that month in 29 Note also that the 2006 Latinbarometer survey was conducted between October 3, 2006 and November 3, 2006 and thus prior to even the announcement of RCTV closing on December 28, 2006. 30 We also use these survey data to examine other outcomes that might be affected by access to the Globovision signal, such as general ideology, overall support for the Venezuelan government, and support for key allies and enemies of Chavez. 31 We also investigated two other databases on protests but found that they lacked the necessary geographic variation. The PRIO Dataset on Urban Unrest only includes information for one city in Venezuela (Caracas). We also investigated using search phrases (e.g., protesta) in the Google trends database but found that the geographic variation within Venezuela is typically suppressed due to low search activity on a municipality-by-municipality and month-by-month basis. 470 Knight and Tribin that municipality. Given this, our analysis uses an inverse hyperbolic sine transformation that is defined at zero but approximates the logarithmic function. To address the large number of zeros in these data, our analysis also presents results in which the data are aggregated in two ways, one involving geographic aggregation from the municipal level to the state level and another involving temporal aggregation from the monthly level to the annual level. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 5.2. Globovision Signal Access to the Globovision signal is measured separately for each municipality. Note that the geography here is more detailed than in the previous section, which was based upon ratings data for four metro areas, which are collections of municipalities. Our measure of access to the Globovision signal for each municipality is based upon information provided directly by Globovision on the location (latitude and longitude) of their four towers, three of which are located in the Caracas metro area and one of which is located in the Valencia metro area. In addition, Globovision provided information on the power and height of each tower. To measure access to the Globovision signal, our analysis then uses the irregular terrain model, which uses information on the source of the signal and the terrain to predict the strength of the signal at destination locations.32 Using information on these four towers, the signal is measured (categorized as no signal, a weak signal, or a strong signal) for geographic grid cells.33 These measures are then aggregated from grid cells to the municipality level, and our analysis uses two such aggregates. First, a dummy variable, referred to as Globovision city, indicates whether or not a majority of the grid cells in the municipality receive the signal. Second, a continuous measure is based upon the fraction of grid cells in the municipality that receive the signal. Figure 8, which focuses on the heavily populated northern part of the country, depicts the results from this model. As shown, the signals are indeed concentrated in the Valencia and Caracas metro areas. Moreover, their reach depends upon the terrain, with the signal traveling long distances over, for example, the water but much shorter distances over the hilly terrain in this part of Venezuela. In total, the presence of a signal is detected in at least one grid cell for 68 out of 335 municipalities, with only 16 meeting our definition (majority of the municipality receiving the signal) as a Globovision city. Our analysis next uses this variation across municipalities to examine changes in support for Chavez following the closing of RCTV. To be clear, we do not claim that signal access is randomly assigned, and, as noted above, the signal is concentrated in the densely populated part of the country. Instead, our analysis also uses variation induced by the timing of the closing of RCTV following the expiration of the 20-year broadcast license. Given this, the key identifying assumption is that trends in support for Chavez in Globovision municipalities would have followed trends in support for Chavez in non-Globovision municipalities in the absence of the closing of RCTV. This identifying assumption and possible threats to this assumption are discussed after presenting our baseline results. 5.3. Hypotheses Our key hypothesis is that retaining access to opposition television via Globovision should be associated with a reduction in support for Chavez following the closing of RCTV. To test this hypothesis, measures of support for Chavez described above are related to access to the signal and the timing of the closing of RCTV in the following regression: Support = β1 Aftert + β2 Signalm × Aftert + αm . In this equation, the municipality fixed effects (α m ) capture time-invariant differences between places with and without access to the Globovision signal, and the parameter β 1 captures changes in approval for 32 Other studies in media economics using the irregular terrain method include Olken (2009), Enikolopov, Petrova, and Zhuravskaya (2011), Yanagizawa-Drott (2014), and Durante, Pinotti, and Tesei (2019). 33 In particular, our analysis uses the radio coverage tool on the website http://www.nautel.com/. The World Bank Economic Review 471 Figure 8. Globovision Coverage Map. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis based upon location, height, and power of Globovision towers according to information provided by Globovision and the irregular terrain method, which predicts signal strength in each municipality. Note: Dark gray represents high signal and light gray represents low signal. Chavez following the closing of RCTV in May 2007. After accounting for these time-invariant differences across cities and any changes in nationwide support for Chavez following the closing of RCTV, the key parameter of interest, β 2 , captures trends in the municipalities that receive the Globovision signal, relative to trends in those municipalities that do not receive the Globovision signal. Thus, our key hypothesis is that β 2 < 0 and involves a relative shift in support across municipalities over time. To account for the panel structure of these data, standard errors are clustered at the municipality level in all specifications. 5.4. Baseline Results This section begins with an analysis of Chavez approval ratings and examines trends according to access to the Globovision signal. In particular, based upon the research design described above, fig. 9 provides information on the distribution of Chavez approval ratings, separately for municipalities with and with- out access to the Globovision signal and then also separately before and after the closing of RCTV. As shown, respondents in both types of municipalities expressed considerable support for Chavez in 2005 and 2006, with over 40 percent giving Chavez the highest possible score of 10 in municipalities without the Globovision signal and a higher proportion, over 50 percent, in the corresponding set of municipalities with access to the signal. These both drop considerably, to just over 20 percent in municipalities with the Globovision signal and roughly 25 percent in municipalities without the Globovision signal, following the closing of RCTV in 2007. Thus, the decline in support for Chavez was sharper in areas with access to the Globovision broadcast signal. There appears to be a corresponding increase in the lowest possible score (very bad) in municipalities with access to the Globovision signal, relative to those municipalities without access to the signal. But, in general, the relative reduction in top scores in municipalities with access to the Globovision signal is offset by increases throughout the distribution of approval ratings. 472 Knight and Tribin Figure 9. Support for Chavez in Survey Data. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of survey data from Latinbarometer 2005–2011. Note: The left panel represents the time period prior to the closing of RCTV and the right panel represents the time period after the closing of RCTV. The upper panel represents respondents from municipalities that do not receive the Globovision signal, and the lower panel represents respondents from municipalities that receive the Globovision signal. To summarize, the drop in support for Chavez during the sample period was strongest in the areas with access to opposition television via the Globovision signal. To investigate these patterns more rigorously, our analysis next considers an ordered logit model, which accounts for the ordinal nature of approval ratings. Results from this regression are reported in column 1 of table 3. As shown, there is a drop in support for Chavez in both types of municipalities following the closing of RCTV, as reflected by the negative coefficient in the first row.34 But the decline is sharper in those municipalities that retained access to opposition television via the Globovision signal, as reflected in the negative coefficient on the interaction in the second row, and this effect is statisti- cally significant at conventional levels. Thus, in municipalities with access to the Globovision signal, the switching from RCTV to Globovision, as documented above, appears to be associated with a reduction in support for Chavez following the closing of RCTV in May 2007.35 Results are similar when considering a 34 The results are similar in nature when including a set of year fixed effects. 35 To provide readers with a sense of the broader impact, above and beyond support for Chavez, of access to opposition television, our analysis also examines responses to other questions related to Venezuelan politics in these Latinbarometer survey data. In particular, it examines whether access to opposition television via the Globovision signal following the closing of RCTV shifts ideology to the right, reduces approval of the Venezuelan government, reduces support for allies of Chavez, and increases support for his enemies. As shown in table 4, there is general support for these hypotheses, although three out of the five results are not statistically significant at conventional levels. In particular, access to the Globovision signal is associated with a movement to the right in the ideological spectrum (column 1) following the The World Bank Economic Review 473 Table 3. Globovision Signal and Support for Chavez (1) (2) (3) (4) (5) (6) (7) Controls and Continuous flexible Narrow control Internet Variables Approval signal High signal Controls coefficients group controls After −0.4865*** −0.4663*** −0.4973*** −0.2954 −1.0163*** −0.5338*** −0.2665 Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 (0.1291) (0.1324) (0.1268) (0.1929) (0.3447) (0.1936) (0.1871) Globo city × After −0.6796*** −0.7293*** −0.6635** −0.6133** −0.7273*** (0.1975) (0.2488) (0.3307) (0.2488) (0.2464) Signal × After −1.0109*** (0.2757) Globo city (high) × After −0.6880*** (0.1974) Observations 7,976 7,976 7,976 7,976 7,976 4,119 7,907 Source: Authors’ analysis of survey data from Latinbarometer 2005–2011. Note: Standard errors clustered at the city level. Controls include gender, SES, age, and urban. *** p < 0.01, ** p < 0.05, ** p < 0.10. Table 4. Globovision Signal and Other Outcomes (1) (2) (3) (4) (5) Variables Ideology L to R Approve government Approve Castro Approve Uribe/Santos Approve US president After −0.2209 −0.4860*** −0.6042*** −0.6661*** 0.9184*** (0.2909) (0.1522) (0.1735) (0.2231) (0.2063) Globo city × After 0.1754 −0.2247 −0.6057* 0.4528* 0.2681 (0.3014) (0.1691) (0.3661) (0.2431) (0.2368) Observations 6,890 7,666 7,727 4,582 6,934 Source: Authors’ analysis of survey data from Latinbarometer 2005–2011. Note: Standard errors clustered at the city level. *** p < 0.01, ** p < 0.05, ** p < 0.10. continuous measure of the signal strength (column 2) or a higher threshold for being classified as a Globo city (column 3). Our analysis next considers electoral data, where our measure of support is based upon the log odds measure (i.e., the natural log of the ratio of the Chavez vote share to the opposition vote share).36 As shown in column 1 of table 5, there is again a general decline in support for Chavez following the closing of RCTV in 2007. But, more importantly, this effect is particularly pronounced in municipalities that re- tained access to opposition television via the Globovision signal, and both of these differences are again statistically significant at conventional levels. These findings suggest that access to Globovision following the closing of RCTV, and viewers switching towards this channel, contributed towards a decline in sup- port for Chavez at the voting booth. To measure the magnitude of the electoral effects, and to show that it does not depend upon the log-odds formulation, our analysis next considers a similar specification but with the Chavez vote share, rather than the log odds formulation, as the dependent variable. As shown in closing of RCTV (although this result is not statistically significant). Likewise, there appears to be reduction in support for the government (column 2) in municipalities with access to the Globovision signal and following the closing of RCTV (although this result is also not statistically significant). Turning to other leaders, we document a statistically significant reduction in support for Fidel Castro, a key foreign ally of Chavez, following the closing of RCTV in municipalities with access to the Globovision signal. Likewise, there is a corresponding increase in support for key enemies, including Uribe and Santos, presidents of Colombia (column 4), and Bush and Obama, presidents of the USA (column 5), in municipalities with access to the Globovision signal following the closing of RCTV. 36 This log odds specification can be motivated by assuming that voters receive utility from supporting Chavez, relative to the opposition, equal to β 1 Aftert + β 2 Signalm × Aftert + α m + im , with im a distributed type-1 extreme value. 474 Knight and Tribin Table 5. Globovision Signal and Electoral Support for Chavez (1) (2) (3) (4) (5) (6) Variables Baseline Vote share High baseline support Low baseline support Log turnout Narrow control group After −0.2256*** −0.0503*** −0.2817*** −0.1695*** 0.0377*** −0.2476*** (0.0168) (0.0036) (0.0249) (0.0216) (0.0082) (0.0253) Globo city × After −0.1117*** −0.0183** −0.0972 −0.1261*** 0.0426*** −0.0897** Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 (0.0391) (0.0091) (0.0682) (0.0301) (0.0147) (0.0436) Observations 1,668 1,668 833 835 1,668 688 R-squared 0.8370 0.8474 0.6438 0.8095 0.9558 0.8815 Source: Authors’ analysis of official voting returns from Venezuela 2004–2012. Note: Standard errors clustered at the city level. Controls include gender, SES, age, and urban. *** p < 0.01, ** p < 0.05, ** p < 0.10. column 2 of table 5, there is a roughly 5 percentage point decline in the vote share for Chavez in mu- nicipalities without access to the Globovision signal following the closing of RCTV but a 7 percentage point decline in the Chavez vote share in municipalities with access to the Globovision signal. Thus, the differential decline equals 2 percentage points, and this decline is statistically significant at conventional levels. One advantage of the voting data is that our analysis can consider heterogeneous effects according to preexisting support for Chavez. Accordingly, the analysis next splits the sample according to baseline (averaged over 2004 and 2006) support for Chavez by municipality. As shown in columns 3 and 4, there are slightly stronger results for municipalities with low levels of support for Chavez at baseline. But the coefficients are similar in magnitude and the results imprecise in general, meaning that there is a sub- stantial degree of overlap in the confidence intervals of the two coefficients. An additional advantage of the electoral data involves measurement of turnout, another possible response to exposure to opposition television.37 As shown in column 5, turnout rose roughly 4 percent following the closing of RCTV in municipalities without access to the Globovision signal but 8 percent in municipalities with access to the Globovision signal, a statistically significant differential increase in turnout of 4 percentage points.38 Our third analysis involves the number of mentions of protest events, using the inverse hyperbolic sine transformation.39 Note that one should expect opposite coefficients in this case since protests reflect collective action against Chavez. As shown in the first row of column 1 of table 6, protest activities increased following the closing of RCTV in 2007, and some of this increase is likely driven by an increase in the size of the GDELT database over time. More importantly, as shown by the estimate of β 2 , the increase is particularly pronounced in municipalities that retained access to opposition television via the Globovision signal, and this difference is statistically significant at the 90 percent level. As a robustness check, our analysis next measures protest activity based upon the number of events rather than the 37 Given that we do not have annual data on the number of eligible voters and thus cannot measure turnout rates directly, our analysis instead measures participation via the log of the number of votes cast. Thus, our key coefficient on the inter- action between the post-RCTV closing indicator and an indicator for access to the Globovision signal can be interpreted as the percent change in turnout following the closing of RCTV in municipalities with access to the Globovision signal, relative to the change in turnout in municipalities without access to the Globovision signal. In particular, note that the log turnout rate can be written as ln (votes/eligible) = ln (votes) − ln (eligible). Since we do not have information on the number of eligible voters on a year-by-year basis, our analysis instead incorporates ln (eligible) into the municipality fixed effects. 38 Figure 10 considers the timing of these effects with respect to turnout in an event study, documenting an immediate and large increase (15 percent) in turnout during 2007 in municipalities with access to Globovision, relative to municipal- ities without Globovision. The effects diminish over time, and there are essentially no differences in turnout between municipalities with and without access to the Globovision signal by 2012, the end of the sample period. That is, our dependent variable is ln[ymt + (y2 0.5 39 mt + 1) ], where ymt represents the number of mentions of protests in municipality m in year t. The World Bank Economic Review 475 Table 6. Globovision Signal and Protest Activity (1) (2) (3) (4) (5) (6) (7) (8) Narrow Variables Mentions Events Opposition Progovernment Other State-level Annual control After 0.0534*** 0.0261*** 0.0205*** 0.0122*** 0.0089*** 0.2572*** 0.3597*** 0.0489*** (0.0085) (0.0043) (0.0045) (0.0034) (0.0023) (0.0496) (0.0532) (0.0110) Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Globo city × After 0.4938* 0.2827* 0.3019* 0.2620 0.1882 0.5352 0.4982* (0.2818) (0.1666) (0.1759) (0.1616) (0.1296) (0.4426) (0.2829) Globo state × After 1.1495*** (0.3904) Observations 26,640 26,640 26,640 26,640 26,640 4,320 2,220 12,780 R-squared 0.4230 0.4727 0.3287 0.3549 0.2856 0.4677 0.4419 0.5570 Source: Authors’ analysis of viewership data from the GDELT Events Database 2000–2014. Note: Standard errors clustered at the city level. Controls include gender, SES, age, and urban. *** p < 0.01, ** p < 0.05, ** p < 0.10. Figure 10. Change in Turnout in Globovision Cities. Source: Authors’ analysis of official voting returns from Venezuela 2004–2012. Note: Coefficients on access to the Globovision signal, separately by year, from a regression with dependent variable equal to the natural log of turnout. Coefficient for 2006 normalized to zero. number of mentions of these events. As shown in column 2 of table 6, the key coefficient is again positive and statistically significant, indicating an increase in the number of protest events following the closing of RCTV in cities that retained access to opposition television via Globovision. So far, our analysis has implicitly assumed that protests involve collective action against the government. Yet some protests might be progovernment and others might be less political in nature. Given this, protest events are next split into three separate categories using the type of actor initiating the protest, as coded in the GDELT 476 Knight and Tribin database: opposition protests, progovernment protests, and other protests.40 As shown in columns 3, 4, and 5 of table 6, the results are only statistically significant for protests associated with the political opposition. While the other coefficients are positive and the standard errors are large, the coefficients are smaller in magnitude and statistically insignificant. Thus, the increase in protests documented in our baseline regressions appear to be driven, at least in part, by protests associated with the opposition rather than progovernment protests or other types of protests. This is consistent with Globovision providing Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 information for viewers who are likely to express their opposition to the government via collective action. As noted above, these data on protests are sparse, with 96 percent of municipal-month observations having no protest activity. To address this issue, our analysis next considers two specifications in which the sample is aggregated to higher levels, first aggregating protests from the municipal level to the state level. Thus, the unit of observation in this case is the state-month. Relative to our baseline strategy, this Globo state measure is based upon having at least one Globo city in the state and thus does not exploit within-state variation. As shown in column 6 of table 6, these differences are positive and statistically significant at the 99 percent level, with a significant increase in protests in Globo states, relative to states without Globovision, following the closing of RCTV in May 2007. Our second measure aggregates the baseline measure from the monthly level to annual level, and the unit of observation is thus the city-year. Relative to our baseline measure, this measure does not exploit variation within calendar year 2007 and considers this year as after the closing of RCTV. As shown in column 7 of table 6, the results are similar in magnitude to our baseline results in column 1 but are statistically insignificant in this case, presumably due to the reduced sample size and the loss of identifying variation within calendar year 2007. In summary, cities with the Globovision signal experienced a significant reduction in support for and an increase in collective action against Chavez following the closing of RCTV in May 2007. These findings are consistent with a link between censorship of opposition television and support for government. 5.5. Threats to Identification As noted above, our key identifying assumption is that trends in support for Chavez in Globovision municipalities would have followed trends in support for Chavez in non-Globovision municipalities in the absence of the closing of RCTV. If support for Chavez was already falling in Globovision municipalities in a differential manner for other reasons, then our interaction coefficient will overstate the role of access to Globovision in support for Chavez. This concern is addressed in four ways. The first involves accounting for observable differences between areas that receive the Globovision signal and other areas. For example, the Caracas and Valencia metro areas, where the signal is concentrated, are clustered in the more densely populated northern part of the country. If any negative perceptions of the government spread more rapidly in urban areas, then one might expect a differential drop in support for Chavez in these Globovision municipalities during our sample period. To address this point, our analysis next includes observable controls in the survey data (urban, gender, age, and socioeconomic status) and, as shown in column 4 of table 3, the differential decline in approval of Chavez in municipalities with access to the Globovision signal is robust to these controls.41 More importantly, in column 5, the coefficients on these controls are allowed to change following the closing of RCTV, relative to the period prior to the closing of RCTV. This allows, for example, support for Chavez to fall more quickly in urban areas and thus compares the differential decline in support between urban areas with and without access to Globovision. Again, our results are robust to these more general controls. 40 In particular, our analysis uses the GDELT variable Actor1Type1Code and code opposition protests as those associated with the political opposition, students, civilian groups, and labor. Progovernment protests are coded as those associated with the government, police, or military. 41 This analysis is not feasible for the analyses of electoral returns and protests as we are not aware of any annual demo- graphic measures at the municipal level. The World Bank Economic Review 477 Table 7. Structural Break Analysis: Support for Chavez (1) (2) (3) (4) (5) (6) Shift treatment Shift treatment Shift treatment Shift treatment Shift treatment Variables 1 year before Baseline 1 year after 2 years after 3 years after 4 years after After −0.3732** −0.4865*** −0.3137** −0.1283 −0.0926 0.1185 (0.1791) (0.1291) (0.1548) (0.0985) (0.0989) (0.1101) Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Globo city × After −0.4591 −0.6796*** −0.5838*** −0.0644 0.1945 0.3095 (0.3887) (0.1975) (0.1795) (0.1877) (0.2335) (0.2366) Observations 7,976 7,976 7,976 7,976 7,976 7,976 Log-likelihood −16,259 −16,221 −16,244 −16,277 −16,279 −16,275 Source: Authors’ analysis of survey data from Latinbarometer 2005–2011. Note: Standard errors clustered at the city level. Controls include gender, SES, age, and urban. *** p < 0.01, ** p < 0.05, ** p < 0.10. Our second strategy towards addressing these threats to identification involves developing a more narrow comparison group, which, in our baseline specification, includes all municipalities in Venezuela that do not receive the Globovision signal. In particular, our analysis focuses the comparison group on those states that either include Globovision municipalities or border these Globovision states.42 Taken together, these 10 states account for less than 20 percent of the area of the country and are comparable in nature, located in the densely populated northern central region of the country. As shown in column 6 of table 3, the results based upon this narrow geographic control group are similar in nature, with a statistically significant decline in support for Chavez in Globovision municipalities, relative to this narrow comparison group, following the closing of RCTV. Likewise, the results are also robust to focusing on the narrow comparison group in the electoral returns and protests data (column 6 of table 5 and column 8 of table 6). Thus, our baseline results are robust to a more narrow comparison group. Our third strategy towards addressing these threats to identification involves structural break analyses. This allows for examination of whether the changes in support for Chavez in Globo municipalities, relative to control municipalities, occurred in the year that RCTV closed or in other years surrounding this event. More specifically, Aftert is measured using alternative definitions before and after the actual closing of RCTV. That is, our analysis estimates specifications in which RCTV is assumed to have closed in other time periods, shifting the treatment backwards and forwards. For the survey data, as shown in table 7, our baseline specification (column 2) has the largest coefficient on the key interaction term, when compared to specifications that allow for changes in Globovision one year before the actual closing and one to four years after the closing. Moreover, only the specification that uses the following year as the change has an interaction coefficient that is statistically significant. Related to this, following a literature on structural breaks, our analysis locates the single break that best fits the data. As shown in the final row, the baseline specification has the best fit, in terms of maximizing the log likelihood.43 As shown in table 8, which considers structural break analyses for electoral data, the results for our baseline specification (column 2) are very similar to those using 2009 as the event date. Using 2006 or 2012 as the event date, however, yields interaction coefficients that are close to zero and statistically insignificant. Moreover, as shown in the final row, the baseline specification has the best fit, as measured by the R-squared in this case. For the protests data, regressions are run separately for each possible monthly closing in each year 2000–2014 and report the fit of each regression, as reported in the R-squared. Given the high-frequency nature of these data, there are now 180 separate regressions, one for each month in the 2000–2014 period. As 42 These states are Yaracuy, Carabobo, Cojedes, Guarico, Anzoategui, Aragua, Vargas, Distrito Federal, Miranda, and Falcon. 43 While national reductions in support for Chavez following 2006 may have contributed to this improved fit, the t-statistic on the interaction coefficient is also largest in the baseline specification. 478 Knight and Tribin Table 8. Structural Break Analysis: Vote Shares (1) (2) (3) (4) Variables Shift treatment 1 year before Baseline Shift treatment 2 years after Shift treatment 5 years after After −0.0497*** −0.2256*** −0.0245 −0.0731*** (0.0183) (0.0168) (0.0151) (0.0196) Globo city × After −0.0337 −0.1117*** −0.1155*** 0.0037 Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 (0.0393) (0.0391) (0.0279) (0.0330) Observations 1,668 1,668 1,668 1,668 R-squared 0.8001 0.8370 0.7999 0.8014 Source: Authors’ analysis of official voting returns from Venezuela 2004–2012. Note: Standard errors clustered at the city level. Controls include gender, SES, age, and urban. *** p < 0.01, ** p < 0.05, ** p < 0.10. Figure 11. Structural Break Analysis: Protests. Source: Authors’ analysis of viewership data from the GDELT Events Database 2000–2014. Note: R-squared statistic from regressions in which the closing of RCTV is shifted before and after the actual closing of May 2007. Dependent variable is the inverse hyperbolic sine transformation of the number of mentions of protests in a given municipality-month. shown in fig. 11, there is a clear break in the series during May 2007, the actual month of the closing, with the R-squared generally increasing over time until this point. After May 2007, the fit declines for a few months and then increases again, with a very similar R-squared five months after the closing. The fit then generally declines again as treatment dates are shifted forwards in time, moving beyond the actual closing year of 2007, and there is no evidence of an improved fit after 2007. To summarize, while we cannot definitely rule out the role of changing unobserved factors in Globovision municipalities following the closing of RCTV, these results suggest that any such changes would likely have occurred in exactly the same year as the closing of RCTV. This allows us to rule out changes in other years surrounding the closing of RCTV as drivers of our baseline results and thus yields credence to our interpretation related to the closing of RCTV in 2007. Fourth, our analysis addresses a specific concern related to internet access and internet use. If the in- ternet expanded more rapidly in more urban areas, those with the Globovision signal, during our sample The World Bank Economic Review 479 period, this would be a key threat to identification, with increased internet use following the closing of RCTV, rather than access to Globovision, explaining our key results. Based upon a survey question in the survey data, there is not substantially higher internet use in the years 2007–2011, those following the closing of RCTV, relative to the survey years 2005 and 2006, those prior to the closing of RCTV. That is, across all municipalities, internet use rose from 34.4 percent to 53.2 percent. Yet this increase occurred in both Globovision and non-Globovision municipalities, and there are not statistically significant dif- Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 ferences in the increase between Globovision and non-Globovision municipalities. In particular, internet use increased by 22 percentage points, from 41.0 percent to 63.0 percent, in Globovision municipalities and by 17.4 percentage points, from 32.9 percent to 50.3 percent, in non-Globovision municipalities, and these differences in the growth of internet use between Globovision municipalities and other mu- nicipalities are not statistically significant. When considering the only non-Globovision municipalities in the narrow geographic control group, internet use rose from 35.1 percent to 52.0 percent, an increase of 16.9 percent, and differences relative to Globovision municipalities are again statistically insignifi- cant. Our analysis also considers specifications with controls for internet use. As shown in column 7 of table 3, our results are similar when including individual-level controls for internet use in our base- line specification.44 Finally, note that these results are specific to our time period studied, 2005–2011. Following our sample period, internet penetration continued to climb, reaching 72 percent by 2019, a bit higher than that of their neighbors Colombia (65 percent) and Brazil (67 percent), and Twitter has become a key source of information and mobilization for the opposition.45 Thus, had RCTV closed more recently, the internet would have been a more crucial alternative source of information for the opposition. 5.6. Possible Backlash Effects Despite addressing threats to our key identifying assumption in four ways, our analysis could still be confounded by other responses to the closing of RCTV. More specifically, our interpretation of this relative effect involves voters in Globo cities being exposed to negative information about Chavez via opposition television. This interpretation could be confounded by other effects associated with the closing of RCTV. Most importantly, it is possible that the closing of RCTV could create a backlash effect, meaning that voters are unhappy with Chavez for closing RCTV. Given that RCTV had a national reach, it is most natural that this type of backlash would be reflected in a national drop in support, and this is consistent with our negative coefficient on After in the specifications above. Yet After is a just a year effect, and there are many possible alternative interpretations of this negative coefficient. While the economic depression in Venezuela does not begin until after our sample period, the worldwide financial crisis occurs during our sample period, following the closing of RCTV, and was associated with a collapse in the price of oil, the key source of government revenue in Venezuela. This makes it difficult to separate a national backlash effect during our sample period from other aggregate shocks. It is also possible that any backlash effects could have played out differently in different municipalities. If voters were unhappy with the loss of opposition television, then a backlash effect should have been more significant in places without Globovision, leading us to understate our results. On the other hand, viewership of RCTV was higher in the two metro areas with the Globovision signal, relative to the two metro areas without the signal, as shown in fig. 2. Then, if a backlash effect was purely driven by the loss of RCTV, the backlash effect could have been more significant in places with Globovision, leading us to overstate our results. Thus, the direction of any relative backlash effects could go in either direction. 44 The results are also robust to a specification allowing for flexible coefficients by allowing for the effects of internet use and other controls to vary before and after the closing of RCTV. 45 https://data.worldbank.org/indicator/IT.NET.USER.ZS?view=map (accessed December 23, 2020). 480 Knight and Tribin Figure 12. Event Study for Support for Chavez in Survey Data. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of survey data from Latinbarometer 2005–2011. Note: Coefficient of the interaction between Globo city and the number of months until the closing of RCTV in May 2007 from an ordered logit specification. The coefficient for one year prior to the closing is normalized to zero. The dependent variable is approval of Chavez on a 1–10 scale. 5.7. Event Studies To further investigate the timing of our results, our analysis next estimates event-study-ordered logit model specifications, in which the difference between signal and no-signal municipalities is allowed to vary over time in a flexible way: Support = βt Signalm + αm + μt , where μt is now a series of year fixed effects and β 2006 is normalized to zero. When considering the survey data, there are not significant differences in support for Chavez between municipalities with and without access to the Globovision signal in 2005, relative to 2006, as shown in fig. 12. Following the closing of RCTV, by contrast, access to the Globovision signal is associated with a drop in support for Chavez in 2007, relative to 2006, and this drop becomes even sharper in 2008. Following this, there is a slow rebound starting in 2009 and continuing until the end of the sample period.46 All of these five post-RCTV closing coefficients on access to the Globovision signal are negative, and four out of five, all except the final year 2011, are statistically different from zero at conventional levels. Turning to voting data, as shown in fig. 13, there is some evidence of a preclosing difference between 2004 and 2006. But, more importantly, the effect of having access to the Globovision signal also appears in 2007 (held December 2, roughly five months following the closing of RCTV), and the effect becomes stronger in 2009, followed by a rebound back to 2007 levels during the 2012 elections. This post-closing pattern of effects is similar to that in fig. 12, which documented that the effect on approval for Chavez associated with access to the Globovision signal was strongest during the survey interviews held in 2008 and 2009 before rebounding back to baseline levels towards the end of the sample period. 46 Among other possible interpretations, the rebound could reflect the fact that cable penetration was increasing during our sample period, as noted above, providing access to Globovision in cities without the broadcast signal. The World Bank Economic Review 481 Figure 13. Change in Support for Chavez in Globovision Cities. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of official voting returns from Venezuela 2004–2012. Note: Coefficients of access to the Globovision signal, separately by year, from a regression with dependent variable equal to the natural log of support of Chavez relative to support for the opposition. Coefficient for 2006 normalized to zero. For protests data, the effect for the month before the closing of RCTV (April 2007) is normalized to zero. As shown in fig. 14, there is no pretrend in the differences in protest activity between Globo cities and other cities during the months leading up to the closing of RCTV in May 2007. There is an immediate increase in protest activity in Globo cities, relative to other cities, in the month of the closing of RCTV, relative to the period just prior to the closing of RCTV. While the results are imprecise in general, they suggest that, unlike the results for approval ratings and voting, this higher level of protest activity tends to persist throughout the sample period, all the way into 2014. To summarize, our analysis documents reductions in support for Chavez following the closing of RCTV in cities that retained access to opposition television via the Globovision signal. For the survey data and electoral data, the effects diminish over time. For the protests results, by contrast, the effects are more persistent, remaining until the end of the sample period. 6. Magnitudes and Counterfactuals Combining the results from the media consumption analysis with the results from the persuasion analysis, our analysis next attempts to quantify our results. To do so, and for comparison with the existing literature, this is based upon the voting results. Recall from above that consumption of opposition news rose by 27 minutes per week and per household in areas with access to the Globovision signal, relative to areas without access to the signal. Likewise, the vote share for Chavez fell by 1.87 percentage points in areas with access to the Globovision signal, relative to areas without access to the signal. This implies that an additional weekly minute of opposition news reduces support for Chavez by 0.07 percentage points. By comparison, Martin and Yurukoglu (2017) conclude that a 3-minute increase of weekly viewership of Fox News (MSNBC) increases voting for Republicans among centrists by 0.65 (−0.04) percentage points, 482 Knight and Tribin Figure 14. Event Study for Protests. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ analysis of viewership data from the GDELT Events Database 2000–2014. Note: Coefficient of the interaction between Globo city and the number of months until the closing of RCTV in May 2007. The coefficient for one month prior to the closing is normalized to zero. The dependent variable is the inverse hyperbolic sine transformation of the number of mentions of protests in a given municipality-month. when averaged across years. Thus, our corresponding 3-minute magnitude of 0.21 percentage points is stronger than the MSNBC effect but weaker than the Fox News effect. Combining the media consumption analysis with the Chavez approval analysis also allows us to con- duct counterfactuals. In particular, our analysis next conducts counterfactual exercises designed to inves- tigate the link between media consumption and persuasion. These include a no-switching counterfactual (i.e., a switching rate of zero) and a counterfactual with a switching rate of 100 percent, when compared to our measured switching rate of 51 percent. In addition, our analysis conducts a counterfactual in which Chavez closes both RCTV and Globovision. Figure 15 summarizes the results from these counterfactual experiments. Our baseline difference-in- difference estimate implies that the vote share for Chavez fell by 1.83 percentage points in areas with access to Globovision following the closing of RCTV. Under a counterfactual in which RCTV did not close, there is no differential change in the Chavez vote share, by assumption. Had RCTV closed but viewers did not switch to Globovision in areas with access (i.e., a switching rate of zero), viewership of opposition news would have fallen more, by 52 minutes per week per household, in areas with access to the Globovision signal, reflecting the fact that viewership of RCTV was higher in areas with the Globovision signal. Given this, the Chavez vote share would have actually risen by 3.58 percentage points. Comparing this to our baseline estimates, this implies that viewer responses, in the form of switching to noncensored outlets, reduced the Chavez vote share by a substantial 5.41 percentage points. At the other extreme, had RCTV closed and all RCTV viewers switched to Globovision in areas with access (i.e., a switching rate of 1), viewership of opposition television would have risen by 102 minutes per household per week in areas with access to the Globovision signal, relative to areas without. Given this, the Chavez vote share would have fallen by an economically significant 7.05 percentage points. Finally, had both RCTV and Globovision closed, viewership of opposition news in areas with the Globovision signal would have fallen by 159 minutes per week per household, relative to areas without the signal. Given this, the Chavez The World Bank Economic Review 483 Figure 15. Vote Change in Globo Cities: Counterfactuals. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Source: Authors’ calculations based upon official voting returns from Venezuela 2004–2012 and viewership data from Nielsen Venezuela 2006 and 2007. Note: All effects involve the predicted change in vote shares between Globovision and non-Globovision cities after the closing of RCTV, relative to before. Baseline is the estimated effect. No RCTV closing has no effect, by assumption. No switching is the effect in the absence of switching from RCTV to Globovision. A switching rate of 100 percent is the effect if lost RCTV minutes were fully offset by higher minutes on Globovision. Close both RCTV and Globovision is the effect had both channels been closed. vote share would have increased by 11 percentage points in areas with the signal, relative to areas without. This large increase, relative to our baseline decline of 1.83 percentage points, reflects three mechanisms: RCTV viewers can no longer switch to Globovision, existing Globovision viewers also lose access to opposition television, and viewership of RCTV was also higher in areas with the Globovision signal. Thus, censorship is less effective in an environment where opposition outlets remain available, and, conversely, censorship is more effective when implemented in a comprehensive manner, via targeting all opposition outlets. 7. Theoretical Explanation This section develops a simple economic model that is consistent with our key empirical results: (a) voters switch to Globovision, when available, following the closing of RCTV, (b) the closing of RCTV increased support for the government in areas that lost access to opposition television, relative to places that retained access to opposition television via Globovision, (c) similar results are found when examining protest activities, and (d) the differences in support for government between places with and without access to opposition television, via the Globovision signal, tend to fade over time, following the closing of RCTV. The baseline model is presented and then followed by five extensions. In the model, the media issues informative reports but the government might censor reports that reflect negatively upon their performance. There are two states of the world, good (s = g) and bad (s = b), reflecting the performance of the incumbent government.47 The probability of the bad state equals π b . A media outlet receives a signal (σ ) containing no information about the state of the world (σ = φ ) with probability p, and, with probability 1 − p, receives a signal that reveals the true state (σ = s). Based upon this information, the media then issues a report (r), which reflects either their signal (r = σ ) or no information (r = φ ). Thus, media outlets cannot issue a good report when the state is bad and vice versa. 47 These could be interpreted, for example, as the state of the economy or the degree of corruption in government. 484 Knight and Tribin While media outlets would like to issue a report when they have relevant information, the government may censor reports when the signal is bad.48 While the censorship environment is determined exogenously in the baseline model, an extension is developed in which the incumbent government chooses the degree of censorship. In particular, there are two media environments, uncensored (m = u) and censored (m = c). In the former case, there is no censoring of bad signals. In the latter environment, which occurs with probability π c , the government censors bad signals with probability q, and, in this case, the report contains Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 no information (r = φ ). Thus, when the media does not issue a report, this could reflect either censoring or a lack of information available to the media. To summarize, in the censored environment, the media issues a bad report with probability (1 − p)π b (1 − q), issues a good report with probability (1 − p)(1 − π b ), and issues no report with probability p + (1 − p)π b q. While voters are assumed to know all relevant probabilities (π b , π c , p, q), assume for now that they do not observe the media environment and can only learn about the environment from contemporaneous media reports. This assumption is later relaxed in two ways, one involving voters learning about the media environment from past media reports and another involving voters receiving independent and informative signals regarding the media environment. Given this setup, citizens observing an informative report (r = b or r = g) learn the true state. Upon receiving no report (r = φ ), by contrast, voters do not know whether the government censored a bad report or the media received no signal.49 In this case, voters follow Bayes rule. Since the government does not censor favorable reports, the probability that the media outlet issues no report when the state is good equals the probability that the outlet received no information. That is, Pr(r = φ |s = g) = p. When the state is bad, by contrast, it is possible that the report is censored and thus Pr(r = φ |s = b) = p + (1 − p)πc q. Finally, using the fact that Pr(s = b) = πb , voters update as follows upon receiving no report: [ p + (1 − p)πc q]πb Pr(s = b|r = φ ) = . [ p + (1 − p)πc q]πb + p(1 − πb ) In the absence of censoring (π c = 0 or q = 0), voters do not update in either direction in the absence of a media report, as shown by Pr(s = b|r = φ ) = πb . With censoring (π c > 0 and q > 0), by contrast, voters understand that it is possible that the government censored the report and, when faced with no media reports, thus infer that the bad state is now more likely. That is, Pr(s = b|r = φ ) > πb . More generally, one can show that the ex post belief of a bad state following no report is increasing in the degree of censoring (i.e., in both π c and q). Is censoring an effective government strategy? That is, does censorship systematically decrease the likelihood that voters believe the state is bad? To investigate this issue, consider the prob- ability that voters believe that the state is bad averaged across media reports. This can be written as: Pr(s = b|r ) = Pr(r = b) + Pr(r = φ ) Pr(s = b|r = φ ). In the censored environment, Pr(r = b) = (1 − p)πb (1 − q ) and Pr(r = φ ) = p + (1 − p)πb q. Since the environment is unknown to voters, and using the results from above, this can be written as follows for 48 We later consider an extension in which opposition media might also engage is censorship by suppressing signals revealing that the state is good. 49 This basic idea, that a lack of media reports might reflect either no information available to the media or government censoring of negative media reports, has played a key role in the theoretical literature on state censorship. Shadmehr and Bernhardt (2015) argue that governments might gain by committing to less censorship since this leads voters to update more favorably towards the government upon receiving no media reports. Guriev and Treisman (2015) develop an infor- mational model in which dictators can survive via manipulation of information, including censorship and propaganda, rather than via more traditional means, such as repression. The World Bank Economic Review 485 the censored environment: Pr(s = b|r ) = (1 − p)πb (1 − q ) [ p + (1 − p)πc q]πb + [ p + (1 − p)πb q] . [ p + (1 − p)πc q]πb + p(1 − πb ) One can show that this is less than π b so long as π c < 1.50 That is, censorship systematically decreases Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 voter perceptions of the bad state. If voters know that the media environment is censored (π c = 1), by contrast, then there is no uncertainty over the media environment, and Pr(s = b|r ) = πb . In this case, voters are sufficiently sophisticated to account for censorship when processing content, understanding that censorship implies a loss of access to information that might reflect negatively on the government. This result, that censorship systematically manipulates voter beliefs when π c < 1, highlights the role of voter uncertainty over the censorship environment in manipulating voters. The extensions below further explore how voters might learn about the censorship environment. The supplementary online appendix considers five extensions of the model. The first embeds our in- formational model of censorship into a simple model of collective action in which opposition to the government exhibits strategic complementarities, generating a demand for opposition media. Due to col- lective action, voters benefit from distinguishing between the media receiving negative information about the government (i.e., bad signals) and the media receiving no information about the government (i.e., no signal). In this sense, voters have a preference for opposition media.51 The second accounts for the fact that the closing of RCTV was observed by voters.52 In particular, voters now learn about the censorship environment via an informative signal of the media environment before receiving media reports. The key result here is that, unless this signal perfectly reveals the media environment, then voter beliefs are still systematically manipulated by the media environment. The third allows voters to learn about the censor- ship environment from both contemporaneous and past media reports. A numerical analysis documents that voters learn the true media environment over time and can account for this when processing media reports. Due to this, and given the results above, voter beliefs over the likelihood of the bad state in the censored environment following media reports converge to the ex ante probability, highlighting a crucial difference between the short run and the long run. The fourth extension allows for endogenous state censorship. In particular, the incumbent, knowing whether the state is good or bad, chooses the media environment. If the costs of censorship are sufficiently low, there is an equilibrium in which censorship is adopted in the bad state but not in the good state. Given this, voters update more negatively upon receiving no report since censorship is now more likely in the bad state. Despite this, incumbent politi- cians do benefit from censorship in the bad state since it allows them to pool with politicians in the good state. The fifth extension considers the case in which opposition media might also engage in a form of self-censorship. In particular, if opposition media is biased against the government, it might choose to suppress signals revealing that the state is good. In this case, state censorship is still effective in the sense that it lowers, on average, voter perceptions that the state is bad. In summary, the model and the associated extensions can explain our key empirical findings. First, as developed in the first extension, there is a preference for opposition media, meaning that voters might switch to Globovision, when available, following the closing of RCTV. Second, censorship systematically manipulates voter beliefs in favor of the incumbent, meaning that areas without access to Globovision might experience an increase in support for the government, relative to areas that retained access to 50 This follows from the fact that the expression equals π b when π c = 1 and that the right-hand side is increasing in π c , meaning that voter perceptions of the bad state decrease as voters uncertainty over the media environment increases (i.e., π c falls below 1). 51 In the context of this model, censorship reduces voter welfare and the government might benefit from increased censor- ship, via a corresponding reduction in collective action, when censorship levels are low. 52 In the Chinese setting, by contrast, citizens are not aware of many forms of censorship. See Roberts (2018). 486 Knight and Tribin opposition television, following the closing of RCTV. Third, as developed in the third extension, voters rationally adjust to the media environment in the long run, and beliefs are not systematically manipulated by censorship. This explains the fading of the gap in electoral support for Chavez between areas with and without access to Globovision following the closing of RCTV. 8. Conclusion Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 This paper investigates the role of government censorship of opposition television in the accountability of governments and incumbent politicians in the context of the expiration of the television license of RCTV, a key opposition channel in Venezuela, and the subsequent opening of TVES, a public channel. The closure of RCTV, along with the limited signal availability of Globovision, provides a natural experiment in which some areas lose access to opposition television overnight, while other areas retain access to opposition news. Using ratings data, our first finding is that viewers switched from TVES, the public progovernment channel, to Globovision, the only other opposition channel, in areas with access to the broadcast signal following the closing of RCTV. Building upon this result, the analysis documents that support for Chavez fell in these areas that retained access to opposition television via the Globovision broadcast signal, relative to areas without access to the signal, following the closing of RCTV. Counterfactuals document that changes in media consumption contribute towards an economically significant reduction in Chavez vote shares. Finally, a theoretical model helps to rationalize this set of findings.53 References Adena, M., R. Enikolopov, M. Petrova, V. Santarosa, and E. Zhuravskaya. 2015. “Radio and the Rise of the Nazis in Pre-War Germany.” Quarterly Journal of Economics 130(4): 1885–939. Besley, T., and R. Burgess. 2002. “The Political Economy of Government Responsiveness: Theory and Evidence from India.” Quarterly Journal of Economics 117(4): 1415–51. Besley, T., and A. Prat. 2006. “Handcuffs for the Grabbing Hand? Media Capture and Government Accountability.” American Economic Review 96(3): 720–36. Chen, Y., and D. Y. Yang. 2019. “The Impact of Media Censorship: 1984 or Brave New World?” American Economic Review 109(6): 2294–332. Chiang, C.-F., and B. Knight. 2011. “Media Bias and Influence: Evidence from Newspaper Endorsements.” Review of Economic Studies 78(3): 795–820. Corrales, J., and M. Penfold. 2011. Dragon in the Tropics: Hugo Chavez and the Political Economy of Revolution in Venezuela. Washington, DC: Brookings Latin America Initiative Books. DellaVigna, S., and E. Kaplan. 2007. “The Fox News Effect: Media Bias and Voting.” Quarterly Journal of Economics 122(3): 1187–234. Di Tella, R., and I. Franceschelli. 2011. “Government Advertising and Media Coverage of Corruption Scandals.” American Economic Journal: Applied Economics 3(4): 119–51. Dinneen, M. 2012. “The Chavez Government and the Battle over the Media in Venezuela.” Asian Journal of Latin American Studies 25(2): 27–53. Durante, R., and B. Knight. 2012. “Partisan Control, Media Bias, and Viewer Responses: Evidence from Berlusconi’s Italy.” Journal of the European Economic Association 10(3): 451–81. Durante, R., P. Pinotti, and A. Tesei. 2019. “The Political Legacy of Entertainment TV.” American Economic Review 109(7): 2497–530. Enikolopov, R., A. Makarin, and M. Petrova. 2020. “Social Media and Protest Participation: Evidence from Russia.” Econometrica 88(4): 1479–1514. Enikolopov, R., M. Petrova, and E. Zhuravskaya. 2011. “Media and Political Persuasion: Evidence from Russia.” American Economic Review 101(7): 3253–85. European Union Election Observation Mission. 2006. “Final Report: Presidential Elections Venezuela 2006.” Technical report, European Union Election Observation Mission. 53 The data underlying this article will be shared on reasonable request to the corresponding author. The World Bank Economic Review 487 Ferraz, C., and F. Finan. 2008. “Exposing Corrupt Politicians: The Effects of Brazil’s Publicly Released Audits on Electoral Outcomes.” Quarterly Journal of Economics 123(2): 703–45. Ferraz, C., and F. Finan. 2011. “Electoral Accountability and Corruption: Evidence from the Audits of Local Govern- ments.” American Economic Review 101(4): 1274–1311. Gehlbach, S., and K. Sonin. 2014. “Government Control of the Media.” Journal of Public Economics 118(C): 163–71. Gentzkow, M., and J. M. Shapiro. 2010. “What Drives Media Slant? Evidence from US Daily Newspapers.” Econo- metrica 78(1): 35–71. Downloaded from https://academic.oup.com/wber/article/36/2/455/6364498 by LEGVP Law Library user on 08 December 2023 Gentzkow, M., J. M. Shapiro, and M. Sinkinson. 2011. “The Effect of Newspaper Entry and Exit on Electoral Politics.” American Economic Review 101(7): 2980–3018. George, L., and J. Waldfogel. 2003. “Who Affects Whom in Daily Newspaper Markets?” Journal of Political Economy 111(4): 765–84. Gerber, A. S., D. Karlan, and D. Bergan. 2009. “Does the Media Matter? A Field Experiment Measuring the Effect of Newspapers on Voting Behavior and Political Opinions.” American Economic Journal: Applied Economics 1(2): 35–52. Guriev, S., and D. Treisman. 2015. “How Modern Dictators Survive: An Informational Theory of the New Authori- tarianism.” Working Paper 21136, Technical report March, National Bureau of Economic Research. Hobbs, W. R., and M. E. Roberts. 2018. “How Sudden Censorship Can Increase Access to Information.” American Political Science Review 112(3): 621–36. Hsieh, C.-T., E. Miguel, D. Ortega, and F. Rodriguez. 2011. “The Price of Political Opposition: Evidence from Venezuela’s Maisanta.” American Economic Journal: Applied Economics 3(2): 196–214. King, G., J. Pan, and M. E. Roberts. 2013. “How Censorship in China Allows Government Criticism but Silences Collective Expression.” American Political Science Review 107(2): 326–43. Knight, B., and A. Tribin. 2019. “The Limits of Propaganda: Evidence from Chavez’s Venezuela.” Journal of the European Economic Association 17(2): 567–605. Kronick, D., and J. Marshall. 2018. “Opposition Media and Political Accountability: Evidence from Venezuela.” Working Paper. Martin, G. J., and A. Yurukoglu. 2017. “Bias in Cable News: Persuasion and Polarization.” American Economic Review 107(9): 2565–99. Martin, G. J., and J. McCrain. 2019. “Local News and National Politics.” American Political Science Review 113(2): 372–84. Mullainathan, S., and A. Shleifer. 2005. “The Market for News.” American Economic Review 95(4): 1031–53. Nelson, B. A. 2009. The Silence and the Scorpion: The Coup against Chavez and the Making of Modern Venezuela. New York: Nation Books. Olken, B. 2009. “Do Television and Radio Destroy Social Capital? Evidence from Indonesian Villages.” American Economic Journal: Applied Economics 1(4): 1–33. Prat, A. 2018. “Media Power.” Journal of Political Economy 126(4): 1747–83. Reporters without Borders. 2003. “Venezuela: Caught between an Authoritarian President and an Intolerant Media.” Technical report, Reporters without Borders. Roberts, M. E. 2018. Censored: Distraction and Diversion inside China’s Great Firewall. Princeton, NJ: Princeton University Press. Shadmehr, M., and D. Bernhardt. 2015. “State Censorship.” American Economic Journal: Microeconomics 7(2): 280– 307. Snyder, J. M., and D. Stromberg. 2010. “Press Coverage and Political Accountability.” Journal of Political Economy 118(2): 355–408. Szeidl, A., and F. Szucs. 2021. “Media Capture through Favor Exchange.” Econometrica 89(1): 281–310. Weisbrot, M., and T. Ruttenberg. 2010. “Television in Venezuela: Who Dominates the Media?” Technical report, Center for Economic and Policy Research. Wilpert, G. 2007. Changing Venezuela by Taking Power: The History and Policies of the Chavez Government. London, England: Verso Books. Yanagizawa-Drott, D. 2014. “Propaganda and Conflict: Evidence from the Rwandan Genocide.” Quarterly Journal of Economics 129(4): 1947–94.