The Limits of Propaganda: Evidence from Chavez’s Venezuela

The Limits of Propaganda: Evidence from Chavez’s Venezuela Abstract We investigate viewer responses to ideological changes in television programming induced by cadenas, unannounced government propaganda in Venezuela. The drop-off in ratings during cadenas is concentrated among viewers of news programming on opposition channels, relative to progovernment channels. Also, the drop-off in ratings for moderate channels takes an intermediate value. The drop-off is stronger for viewers with access to cable channels, which do not air cadenas and experience an increase in viewership during cadenas. Structural estimation of our model allows us to quantify the degree to which opposition viewers limit their exposure to and ultimately the influence of propaganda via tuning out. 1. Introduction The media is often considered essential in the functioning of democracy via the provision of information to voters. At the same time, there is a temptation for incumbent governments to use media outlets to deliver political propaganda. This propaganda can be used by the government, among other ways, to promote its policies, increase its standing with the population in advance of elections, and to criticize opposition leaders and parties. If influential, propaganda may lead to moral hazard, via poor monitoring of incumbents by voters, and the re-election of low quality politicians and parties. Sophisticated consumers of information may respond to such propaganda in a variety of ways. One response involves consumers discounting biased information. For example, Chiang and Knight (2011) show that newspaper endorsements are more influential when they are less biased. That is, an endorsement of a Republican candidate by a liberal newspaper has more influence than an endorsement of a Republican by a conservative newspaper. Thus, discounting by readers limits the influence of biased information. Another response involves changing consumption patterns. For example, Durante and Knight (2012) document switching in viewership in response to a change in government in Italy and an associated change in the content of the main public television channel. With a preference for like-minded information, this switching is particularly relevant for consumers affiliated with the opposition. Note that switching will only be available in media sectors that are pluralistic, those offering a variety of ideological viewpoints. Given this, another possibility involves consumers simply “tuning out”, or consuming less information overall across all media outlets. Like switching, tuning out may be especially relevant for the opposition. Given this, tuning out may limit the ability of the government to convert the opposition to their cause via propaganda.1 In this paper, we investigate this second response, changing consumption patterns, using high-frequency television ratings data from the country of Venezuela. Hugo Chavez and his successor have routinely used cadenas, translated as “chains” and defined as government propaganda that is required to be aired live by all broadcast television channels. Thus, during a cadena, viewers watching television face the same programming on every broadcast channel. Importantly, these cadenas are not announced in advance to viewers, providing an experiment through which to examine short-run responses, in terms of changes in viewership, to government propaganda. In addition, cadenas were not required to be aired by cable channels during our sample period, allowing us to examine whether households with larger choice sets are more likely to switch to other outlets when faced with propaganda. Finally, broadcast channels in Venezuela during our sample period cover the political spectrum and can be naturally categorized as opposition, moderate, or progovernment. This allows us to examine whether switching and tuning out are more common among opposition viewers, who, as we document using survey data, are more likely to watch opposition news programming. To develop a set of testable hypotheses, we begin by building a simple model of consumer choice of television programming. In the model, there are two types of consumers, opposition and progovernment, both with a preference for like-minded information, two types of channels, opposition and government, and two types of programming, news and cadenas. We begin by assuming that both channels are required to air cadenas and thus initially focus on tuning out. The model predicts that, with positive switching costs and a preference for like-minded news, the drop-off in viewership when transitioning from news to cadena is more significant for the opposition channel than for the progovernment channel. This is due to the selection of opposition viewers into news programming on the opposition channel and the selection of progovernment viewers into news programming on the government channel. Introducing a third channel, which is moderate in nature, the model predicts that the drop-off in ratings when moving from news programming to cadenas should be most significant for the opposition channel, followed by the moderate channel, followed by the government channel. Finally, we consider an extension of the model to allow for switching via a cable channel, which is not required to air cadenas, and this extension provides two additional predictions. First, the model predicts that the drop-off in viewership on the private network, relative to the public network, should be more significant for households with access to cable, when compared to households without cable. Second, cable viewership, due to its role as an outside option, should be higher during the airing of cadenas on broadcast channels, relative to when cadenas are not aired on broadcast channels. We then test these predictions using data on television ratings from Venezuela. These data cover the years 2006 and 2007 and are high-frequency in nature (i.e., day-by-day and show-by-show). Consistent with the first prediction of the model, we find that the drop-off in viewership when transitioning from news programming to cadenas is more significant for the opposition channel than for the government channel, and these differences are both economically and statistically significant. Consistent with the second prediction of the model, we find that the drop-off in viewership for news programming on the moderate channel takes an intermediate value, between that of opposition channels and that of government channels. Next, focusing on the outside option, we find that, consistent with switching, cable viewership rises during cadenas and the drop-off in viewership is more significant for those with access to cable. Complementing this analysis, we also estimate the underlying structural parameters of the model; these include switching costs and the value of ideological information. This approach allows us to conduct counterfactual scenarios, which are not possible in the reduced form analysis. In particular, we consider a counterfactual in which cadenas are replaced with news programming, and we also consider a no switching counterfactual. One key finding is that opposition viewers significantly limit their exposure to propaganda by tuning out. In particular, although only 10% of opposition viewers are exposed to propaganda under the baseline, that figure rises to roughly 23% in absence of switching. Based upon this, we use persuasion rates from the literature to estimate the degree to which tuning out limits the persuasive impact of propaganda. Again, we find that switching matters for persuasion. In particular, although only 1% of opposition viewers are persuaded to support the government under the baseline, that figure rises to 3% in the absence of switching. Finally, we consider dynamic viewer responses and also provide a welfare comparison of government propaganda and media pluralism. This paper contributes to several literatures on media bias. A large literature measures the persuasive effects of the media, with DellaVigna and Gentzkow (2010) noting persuasion rates, defined as the probability that a viewer is persuaded by content, conditional on exposure, varying between 6% and 20%.2 Our study contributes to this literature by emphasizing the role of viewer responses in limiting exposure to biased information and thus ultimately limiting persuasion. On this issue, a theoretical literature, including Besley and Prat (2006) and Gehlbach and Sonin (2014), has emphasized the role of viewer responses in limiting the ability of the government to bias mass media in their favor. There is also a related empirical literature more focused on the role of government propaganda.3 Closest to our paper, several studies have documented a preference for like-minded news. These include Gentzkow and Shapiro (2010), Martin and Yurukoglu (2015), and Gentzkow, Shapiro, and Sinkinson (2014). The most closely related study in this literature is Durante and Knight (2012), who, as noted previously, investigate switching in viewership in Italy. There are two important differences between this study on Venezuela and Durante and Knight (2012). The first involves the frequency of responses. Although Durante and Knight (2012) study changes in the choice of media outlets over several years, this paper measures high-frequency, short-run changes in media consumption associated with a preference for like-minded news. Given inertia, it is possible that short-run responses are much smaller than long-run responses. This distinction is particularly relevant in countries with high frequency changes in government. The second important difference involves the structural analysis in this paper, allowing us to quantify the role of viewer responses in exposure to propaganda and in limiting ideological persuasion. More generally, our paper makes three contributions to the media economics literature. First, as noted previously, although most of the literature examines long-run responses, we focus on short-run, minute-by-minute responses. As a result, identification is cleaner than existing studies since viewer ideology is arguably fixed within a day. Second, this is one of the first studies in the literature with structural estimation. This analysis allows us to quantify the degree to which behavioral responses limit exposure to propaganda and ultimately limit political persuasion among the opposition. Third, studies in media economics have tended to focus on both developed countries and on democracies, with studies focused on developing countries and authoritarian regimes, such as Venezuela, more scarce. The paper proceeds as follows. Section 2 provides an overview of the key institutional details. Section 3 develops our key hypotheses in the context of a simple choice model. Section 4 describes the data, and Section 5 provides our results. Section 6 provides the structural estimates and counterfactual exercises. Finally, Section 7 offers a brief conclusion. 2. Institutional Context This section covers the political situation in Venezuela, with a focus on the role of television.4 In 1998, the leftist candidate Hugo Chavez won the presidential election in Venezuela, promising to lessen social exclusion, poverty and government corruption. Chavez was re-elected in 2000, 2006, and 2012 and served as President until his death in 2013. Since the beginning of Chavez’s time in office, the right-wing opposition was committed to removing him from power, with an attempted coup in April 2002, a national strike in December 2002, and recall referendum in 2004.5 During these confrontations, the private media sector tended to side with the opposition.6 Tensions between the private media and government were at their peak, with Chavez referring to the major private television channels (Venevision, Radio Caracas Television (RCTV), Globovision, and Televen) as the “four Horsemen of the apocalypse”, and, more generally, his language against the private media became very aggressive.7 In 2004, before the recall referendum, Chavez met with the owner of Venevision, leading to a warming in relations between the channel and President Chavez.8 Then, Televen followed the initiative to moderate their anti-Chavez tone around the same period.9 However, Globovision and RCTV, the oldest and largest television station, remained in opposition to the government. This partitioning of private channels into opposition (RCTV and Globovision) and moderate (Televen and Venevision) is consistent with media monitoring during the 2006 Presidential elections. In particular, EU-EOM (2006) document that RCTV and Globovision devoted a majority of their coverage to the opposition party, whereas Televen and Venevision devoted a majority of their coverage to Chavez’s party. Not surprisingly, the main public channel, Venezolana de Television (VTV), also devoted disproportionate coverage to Chavez’s party. Similar patterns were found with respect to the tone of the coverage, with positive coverage of the opposition and negative coverage of Chavez on RCTV and Globovision. Coverage of both Chavez and the opposition by Televen and Venevision, by contrast, was largely positive in nature. Finally, coverage of Chavez on the main public channel VTV was primarily positive, with decidedly negative coverage of the opposition. Given this evidence, we classify channels into three ideological categories: opposition (RCTV and Globovision), moderate (Televen and Venevision), and progovernment (VTV). In May 2007, the broadcasting license of RCTV expired and was not renewed by the government, and RCTV was replaced overnight by TVES, a government-run channel. The government’s rationale for closing RCTV had two key components: alleged violations of broadcast laws and their coverage of the coup and the strike in the oil sector. Later that year, during July 2007, RCTV re-emerged as a cable channel under the name RCTV International.10 In addition to not renewing the broadcast license of RCTV, Chavez attempted to influence the media via government channels and cadenas, government programming that must be aired live by noncable (i.e., broadcast) channels.11 Bisbal (2009) estimates that 1,731 cadenas were broadcast between 1999 and June 2008, totaling over 1,000 h. According to Kitzberger (2010) and Reporters Without Borders (2003), cadenas are used by Chavez to mobilize supporters, criticize and threaten adversaries, and more generally, for political campaigning. Two aspects of cadenas are particularly useful for our identification strategy. First, cadenas are not announced in advance to stations or viewers and, as we argue in what follows, are difficult to forecast in terms of either their starting time or duration. Given this, we assume that viewers do not anticipate cadenas when choosing whether or not to watch news programming. Second, all broadcast channels must air cadenas and thus, in the absence of a cable subscription, every available channel carries the cadena. Moreover, our understanding is that, due to the high volume of cadenas, viewers are aware of this and do not attempt to change channels when a cadena comes on the air. This is useful for our empirical strategy since we can infer the fraction of viewers of each news program who watch the cadena via the change in viewership on a given channel when transitioning from news to cadena. 3. Theoretical Model This section develops a simple theoretical model to provide a set of hypotheses for the empirical analysis of ratings data. In addition, the model provides a framework for the structural analysis to follow. We begin with the simple case of only two types of viewers (opposition and progovernment), two channels (opposition and government), and two types of programming (news and cadenas). In extensions of the model, we then introduce a third channel, which is moderate in nature, and then separately consider how the results differ with the presence of a cable channel that is not required to air cadenas. 3.1. Baseline Case Viewers, indexed by $$v$$, are of two types: progovernment (g) and opposition (o). Let the fraction of each type in the population be given by πg and πo = 1 − πg, respectively. News stations, indexed by i, are also of two types: government (g) or opposition (o). Each outlet offers news programming (p = n), and both outlets are also required to carry cadenas (p = c). Viewers differ in the degree to which they value news programming. For progovernment types, the value of government news is θs and the value of opposition news is θd, where we assume that viewers prefer same-ideology news over different-ideology news (i.e., θd < θs). For opposition types, by contrast, the value of government news is θd and the value of opposition news is θs. Cadenas are assumed to have progovernment content and thus provide payoffs of θd to opposition types and θs to progovernment types. Then, letting u$${vip}$$ ∈ {θd, θs} represent these systematic payoffs, viewer $$v$$ receives the following overall payoff from watching programming p on station i: \begin{equation*} U_{vip}=u_{vip}+\epsilon _{vip} \end{equation*} where ε$${vip}$$ represents unobserved preferences and is assumed to be distributed type-1 extreme value.12 These unobserved preferences are assumed to be independently distributed, both across viewers and across channels. We consider a scenario in which both stations are airing news, and viewers have three options: (1) watching the government station, (2) watching the opposition station, and (3) watching neither (which yields a systematic payoff of zero). Then, letting σ$${in}$$ be the market share on channel i when both channels are airing news programming, we have the following market shares on the government channel: \begin{equation*} \sigma _{gn}=\pi _{g}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}+\pi _{o}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}. \end{equation*} The first term is the product of the fraction of progovernment viewers and the market share within this group, and the second term is the product of the fraction of opposition viewers and the market share within this group. Thus, the overall market share for the government channel is a weighted average of market shares for progovernment and opposition viewers. Likewise, the market share on the opposition channel is given by \begin{equation*} \sigma _{on}=\pi _{g}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}+\pi _{o}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}. \end{equation*} Now, suppose that the government airs a cadena and that this is not anticipated by viewers. That is, as discussed previously, viewers do not account for the possibility of a cadena when choosing whether or not to watch news. Further, for simplicity, assume that viewers who are not watching news (the third option described previously) do not come back to watch the cadena on either of the two channels. Also, assume a switching cost of η > 0 so that viewers will not change the channel when the cadena comes on the air. As discussed previously, all broadcast channels must air cadenas and thus there is no incentive for viewers to change channels when a cadena comes on the air. Instead the only margin involves whether or not to watch the cadena. More formally, this simply requires positive switching costs (η > 0) and that unobserved preferences over programming (ε$${vi\!p}$$) are constant across channels when a cadena comes on the air.13 Then, let the fraction of progovernment viewers who choose to watch the cadena, conditional on watching the news on that channel, be given by \begin{equation*} p_{g}=\exp (\theta _{s})[1+\exp (\theta _{s})]^{-1}, \end{equation*} and the analogous fraction for opposition viewers is given by \begin{equation*} p_{o}=\exp (\theta _{d})[1+\exp (\theta _{d})]^{-1},\text{where}p_{o}<p_{d} \text{since} \theta _{d}<\theta _{s}. \end{equation*} Then, we have that market shares for cadenas on the two stations are given by \begin{equation*} \sigma _{gc}=\pi _{g}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{g}+\pi _{o}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{o}, \end{equation*} \begin{equation*} \sigma _{oc}=\pi _{g}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{g}+\pi _{o}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{o}. \end{equation*} Then, define the drop-off in viewership moving from news to cadena, for government and opposition channels, respectively, as \begin{equation*} \Delta ^{o}=\ln \left[\sigma _{oc}/\sigma _{on}\right] \quad \text{and} \quad \Delta ^{g}=\ln \left[\sigma _{gc}/\sigma _{gn}\right]. \end{equation*} Given the log transformation, these measures can be interpreted as the percentage reduction in viewership on a given channel when moving from news programming to cadenas. We first compare the drop-off in viewership on opposition and government channels in the following proposition. Proposition 1. With positive switching costs (η > 0) and a preference for like-minded news (θd < θs), the drop-off in viewership moving from news to cadena is more significant for the opposition channel than for the government channel. That is, Δo < Δg. We provide proofs of all propositions in the Online Appendix. The intuition for this proposition is simply that opposition viewers, relative to progovernment viewers, are more likely to watch opposition news, relative to government news. Moreover, these opposition viewers also have a distaste for the content of the cadena, relative to progovernment viewers. Given all of this, viewers of opposition news are more likely to tune out when a cadena comes on the air. 3.2. Moderate Channel Extension We next extend the model to allow for a third channel, which is assumed to air moderate news. For simplicity, assume that both opposition and progovernment viewers get a payoff of θm from watching news programming on this channel, with θd < θm < θs. Then, again comparing the drop-off in viewership across the channels, we have the following proposition. Proposition 2. With positive switching costs (η > 0) and a preference for like-minded news (θd < θm < θs), we have that the drop-off in viewership for the moderate channel lies in between the opposition and the government channel. That is, Δo < Δm < Δg. The intuition for this proposition is simply that the moderate channel attracts a less polarized audience for its news programming, whereas the opposition channel disproportionately attracts opposition viewers and the government channel disproportionately attracts progovernment viewers. Thus, the drop-off in viewership for the moderate channel takes an intermediate value, when compared to the government and opposition channels. 3.3. Cable Extension To investigate the possibility of switching to other outlets in a pluralistic media environment, we return to the baseline model of two broadcast channels but now allow for a cable channel, which is assumed to be linked to the opposition, and, as discussed previously, is not required to air cadenas. In the context of this extension, we investigate two questions. First, due to the presence of this new opposition channel, is the drop-off in viewership, when moving from opposition news to cadena, more significant for those viewers with cable than for those viewers without cable? Second, consistent with switching, does cable viewership increase during cadenas? Given the empirical application to the cable channel RCTV International, we assume here that cable also has opposition news, yielding a payoff of θd to progovernment viewers and θs to opposition types. Now, suppose that the government unexpectedly decides to air a cadena. As stated previously, assume that viewers who are not watching do not come back to watch the cadena. Also, as stated previously, assume a switching cost of η > 0 so that viewers will not change the channel when the cadena airs. Finally, for simplicity, we assume that viewers do not switch from cable to either the opposition or the government channel when the cadena comes on the air. They can switch from one of the broadcast stations to cable but must incur the switching cost. Then, we have the following result with respect to the drop-off measures considered previously. Proposition 3. With positive switching costs (η > 0) and a preference for like-minded news (θd < θs), the drop-off in viewership on the opposition channel, relative to the government channel, for viewers with cable is larger than for viewers without cable. That is, Δo − Δg falls when cable is introduced. The intuition for Proposition 3 is that, in addition to turning off the television, opposition viewers with access to cable now have another attractive outside option, switching to watch opposition news on cable during the cadena. Given this, even fewer viewers of opposition news will watch the cadena. Finally, we consider how viewership of cable changes when a cadena comes on broadcast television, and we have the following result. Proposition 4. With positive switching costs (η > 0), a preference for like-minded news (θd < θs), and a cable option, viewership of cable rises during the cadena. The logic behind Proposition 4 is straightforward. Since opposition viewers value cable as an outside option, viewership of cable programs rises during cadenas. To summarize, the theoretical model makes four predictions. First, the drop-off in viewership when moving from news to cadenas should be more significant on private channels, when compared to the government channel. Second, the drop-off in viewership on moderate channels should take an intermediate value, between the opposition channel and the government channel. Third, the drop-off in viewership for the opposition channel, relative to the government channel, should be more significant for those with access to cable. Fourth, cable viewership should rise during cadenas. 4. Data Our data on television ratings were purchased from AGB Nielsen Media Research Venezuela and include broadcast ratings of each television show aired on each channel, from January 1, 2006 to December 31, 2007. During the part of the analysis focused on ratings of broadcast channels, we focus on data from the period prior to the closing of RCTV in May 2007 in order to have a consistent set of channels. Ratings are provided separately for the four largest metropolitan areas (Caracas, Barquisimeto, Maracaibo, and Valencia). In constructing our measure of ratings for each show we use the average minute rating (AMR) measure, and, given their very low ratings, ignore shows aired between midnight and 6 a.m.14 In addition to analyzing aggregate ratings for each show, channel, and metropolitan area, we also test Proposition 3 by employing measures of ratings separately for those with and without cable subscriptions. Likewise, our structural analysis uses gender-specific ratings.15 Our analysis considers the most significant channels, those discussed in Section 2. In particular, and as shown in Table 1, we focus on four private broadcast channels, Globovision, Televen, RCTV, and Venevision, one public channel, VTV, and one cable channel, RCTV International. As described in Section 2, television in Venezuela during the sample period is considered to be highly polarized. This political polarization allows us to create three categories for the channels based upon their ideology, as discussed previously. Although the main public channel (VTV) is assumed to be progovernment, private channels are split into opposition (RCTV and Globovision) and moderate (Venevision and Televen). Table 1. Channels analyzed. Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 View Large Table 1. Channels analyzed. Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 View Large Of course, there may be other differences across channels, and Table 2 provides some evidence on the types of programming offered during our sample period.16 As shown, three of the private channels disproportionately air entertainment programming and one of the private channels, Globovision, primarily offers news programming. The public channel VTV also offers more news programming than entertainment programming.17 As noted previously, key to our identification strategy is the assumption that viewers are not aware of cadenas in advance. The law does not require the government to preannounce cadenas, and our understanding is that cadenas are not preannounced in practice. Nonetheless, it is still possible that viewers can predict the airing of cadenas to the extent that they follow regular patterns. We investigate this issue by analyzing the distribution of cadenas across days, their starting time, and their duration. As shown in Figure 1, although cadenas are most commonly aired on Wednesdays, followed by Tuesdays, Thursdays, and Fridays, cadenas may appear on any day of the week, and there is not a noticeable spike on any particular day. Likewise, as shown in Figure 2, although cadenas are most commonly aired during prime time (i.e., between 7 p.m. and 10 p.m.), cadenas can occur at nearly any hour. In addition, as shown in Figure 3, although many cadenas start at the top of the hour, they can also begin at any minute within the hour. Finally, the duration of cadenas is difficult to predict. As shown in Figure 4, cadenas can be either very short in duration, less than 30 min, or very long in duration, in excess of four or even 5 h. To summarize, there is not a specific pattern in terms of the timing of cadenas, and there is thus an important element of surprise for the viewer, who can be exposed to these interruptions by the government at any time, without anticipating the day, the hour, the minute, or the length of the interruption. Figure 1. View largeDownload slide Day of the week of cadenas. Figure 1. View largeDownload slide Day of the week of cadenas. Figure 2. View largeDownload slide Starting hour of cadenas. Figure 2. View largeDownload slide Starting hour of cadenas. Figure 3. View largeDownload slide Starting minute of cadenas. Figure 3. View largeDownload slide Starting minute of cadenas. Figure 4. View largeDownload slide Duration in minutes of cadenas. Figure 4. View largeDownload slide Duration in minutes of cadenas. A key mechanism in our model is a preference for like-minded news, implying that opposition viewers are more likely to watch opposition news and that progovernment viewers are more likely to watch news on public channels. Unfortunately, our ratings data do not have any measures of viewer ideology. Instead, to examine this issue, we have analyzed separate data from the Latin American Public Opinion Project (LAPOP) Survey, conducted during 2007 for Venezuela. The survey includes questions about political preferences and media consumption for a total of 1,510 Venezuelan citizens. In particular, LAPOP asks respondents which candidate they voted for in the last election and the channel they watch most often for news. For the purposes of this analysis, we group the channels into opposition (RCTV and Globovision), moderate (TVES and Venevision), and public (VTV). As shown in Figure 5, respondents who voted for Chavez are most likely to watch moderate channels, followed by the public channel. They are unlikely to watch opposition channels RCTV and Globovision. For respondents who voted for the opposition, by contrast, the patterns are reversed. In particular, and, as shown in Figure 6, these respondents have a very low propensity of watching the public channel, and a majority report watching news on either RCTV or Globovision. To summarize, Chavez supporters are roughly 10 times more likely than opposition supporters to watch VTV, and opposition supporters are roughly three times more likely than Chavez supporters to watch opposition channels. This provides support for our maintained assumption of a preference for like-minded news.18 Figure 5. View largeDownload slide Favorite news channels for Chavez supporters. Figure 5. View largeDownload slide Favorite news channels for Chavez supporters. Figure 6. View largeDownload slide Favorite news channels for the opposition. Figure 6. View largeDownload slide Favorite news channels for the opposition. 5. Analysis of Ratings data In this section, we test the key hypotheses of the theoretical model in an investigation of viewer responses to political propaganda via cadenas in Venezuela during 2006 and 2007, a key period during Chavez’s time in office. Before turning to the regression results, we present summary statistics. In particular, Table 3 provides, for each channel, the average percent change in rating for transitions between the three types of programming: news (N), cadenas (C), and entertainment (E). Standard deviations are provided in parentheses. There are several notable findings here. As shown in the first column, opposition channels exhibit a reduction of viewership when a cadena interrupts a news program, with no change for moderate channels and an increase on the government channel. As shown in the third column, reverse transitions, those from news to cadena exhibit opposite patterns in viewership, with an increase on opposition channels and a decrease in viewership on public channels. Transitions from entertainment to cadena (column (2)) lead to only small changes in viewership on private channels but significant increases in viewership on public channels. Finally, as shown in the fifth column, transitions from news to entertainment exhibit sizable increases in viewership on private channels and significant reductions in viewership on the public channel. We next examine these relationships more formally in a regression analysis. Table 2. Content by channel. Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 View Large Table 2. Content by channel. Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 View Large Table 3. Descriptive statistics: Log change in ratings. Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Notes: Measures represent the mean log change in rating for news (N), cadena (C), and entertainment (E). Standard deviation in parentheses. View Large Table 3. Descriptive statistics: Log change in ratings. Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Notes: Measures represent the mean log change in rating for news (N), cadena (C), and entertainment (E). Standard deviation in parentheses. View Large Table 4. Log change in ratings: News to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 4. Log change in ratings: News to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 5. Log change in ratings: News to cadena with fixed effects. Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 ***p < 0.01. View Large Table 5. Log change in ratings: News to cadena with fixed effects. Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 ***p < 0.01. View Large Table 6. Log change in ratings: Heterogeneity. Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *** p < 0.01. View Large Table 6. Log change in ratings: Heterogeneity. Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *** p < 0.01. View Large Table 7. Log change in ratings: Cadena to news. Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Notes: The dependent variable is the log change in ratings when transitioning from a cadena to a news program. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 7. Log change in ratings: Cadena to news. Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Notes: The dependent variable is the log change in ratings when transitioning from a cadena to a news program. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 8. Log change in ratings: Entertainment to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Notes: Dependent variable is the log change in ratings when transitioning from entertainment to cadena. Public channel VTV is the base outcome. Robust standard errors in brackets. **p < 0.05; ***p < 0.01. View Large Table 8. Log change in ratings: Entertainment to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Notes: Dependent variable is the log change in ratings when transitioning from entertainment to cadena. Public channel VTV is the base outcome. Robust standard errors in brackets. **p < 0.05; ***p < 0.01. View Large Table 9. Log change in ratings: News to entertainment. Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Notes: Dependent variable is the log change in ratings when transitioning from news to entertainment. Public channel VTV is the base outcome. Robust standard errors in brackets. ***p < 0.01. View Large Table 9. Log change in ratings: News to entertainment. Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Notes: Dependent variable is the log change in ratings when transitioning from news to entertainment. Public channel VTV is the base outcome. Robust standard errors in brackets. ***p < 0.01. View Large Table 10. Drop-off for cable versus no cable. Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Notes: Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *p < 0.1; **p < 0.05. View Large Table 10. Drop-off for cable versus no cable. Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Notes: Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *p < 0.1; **p < 0.05. View Large Table 11. Cable channel RCTV international. Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Notes: All results for the cable channel RCTV International when a cadena is aired on the broadcast channels. Robust standard errors in brackets. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 11. Cable channel RCTV international. Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Notes: All results for the cable channel RCTV International when a cadena is aired on the broadcast channels. Robust standard errors in brackets. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 12. Cadenas content. Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Notes: Robust standard errors in parentheses; **p < 0.05; ***p < 0.01. View Large Table 12. Cadenas content. Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Notes: Robust standard errors in parentheses; **p < 0.05; ***p < 0.01. View Large Table 13. Summary of payoff structure. $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ View Large Table 13. Summary of payoff structure. $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ View Large Table 14. Structural estimates. Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Notes: Standard errors in brackets. ***p < 0.01. View Large Table 14. Structural estimates. Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Notes: Standard errors in brackets. ***p < 0.01. View Large 5.1. Drop-Off: News to Cadena Our econometric analysis begins with an investigation of how ratings change when a cadena interrupts news programming depending upon the political orientation of the station, under the assumption that viewers prefer to watch like-minded news. Given, as shown previously, that opposition viewers have a higher probability of watching opposition news channels, and, under the assumption that opposition viewers dislike cadenas, we expect viewers of opposition news to be more likely to tune out when cadenas are aired on television, relative to viewers of progovernment news. As argued previously, we hypothesize that viewers watching opposition news will respond more strongly to cadenas when compared to viewers watching news programming on government channels. To test this hypothesis, we estimate the following econometric model of viewer responses to cadenas: \begin{equation} \Delta ^{ic}=\ln \left[\frac{s_{ic}}{s_{in}}\right]=\beta _{i}+\epsilon _{ic}, \end{equation} (1) where s$${ic}$$ represents the measured rating for a cadena aired on channel i and s$${in}$$ is the ratings for the news program that aired just before cadena c on channel i.19 That is, consistent with the theoretical predictions, the drop-off in viewership is measured as the log change in ratings between cadenas and the previous news program for each cadena aired between January 2006 and May 2007.20 On the right-hand side, βi is a channel-specific constant. To test Proposition 1, we use a dummy variable that takes the value of 1 for a private channel and the value of 0 for a public channel. To test the second proposition, we employ a set of dummy variables based on political ideology of the station (i.e., opposition, moderate, and public). Then, we estimate a more flexible specification that uses a separate dummy variable for each channel. Finally, ε$${ic}$$ represents the unobserved determinants of the drop-off in ratings on channel i during cadena c. We begin with a simple comparison of private and public channels, where public channels are the omitted category. Thus, the results are interpreted as reflecting drop-off for the private channel relative to the public channel. As shown in the first column of Table 4, the coefficient on private channels is negative and statistically significant. That is, airing cadenas after news programming on private channels, relative to the public channel, is associated with more viewers tuning out. This provides support for Proposition 1, which predicted that the drop-off in viewership should be more significant for private channels than for public channels. Moreover, the magnitudes of these effects are large, with the drop-off for private channels 45 percentage points larger than the drop-off for public channels. As a robustness check, we next include time fixed effects in order to control for the timing of cadenas. As shown in Table 5, when including fixed effects for the starting hour of cadenas, the results are somewhat stronger, with a 52% reduction in viewership on private channels, relative to public. We next include day of week fixed effects, and, as shown in column (2), the results are again similar to the baseline results. Finally, we include starting hour by day of the week fixed effects. As shown in the final column, there is a 54% reduction in viewership on private, relative to public, in this case. Thus, these baseline results are robust to the inclusion of time fixed effects. Returning to Table 4, we next allow for two separate categories of private channels (opposition and moderate), with public again the omitted category. The coefficients are also large in magnitude and statistically significant for the two categories, opposition and moderate, relative to the public channel. The coefficients in the second column demonstrate that viewers of news on the opposition and moderate channels, relative to viewers of the public channel, are more likely to turn off the television when a cadena is aired. That is, consistent with Proposition 2, which predicted that the drop-off for moderate channels should take an intermediate value, the change in viewership for moderate channels is 19 percentage points higher than the public channel but is 35 percentage points lower than the opposition channels.21 Finally, in the third column of Table 4, we present the results separately for each channel, where the effects should again be considered relative to the public channel VTV. As shown, and for all channels, we find a statistically significant reduction of viewership, relative to the change in viewership of VTV, when a cadena is aired. Consistent with the results in the second column, the effect of switching to an outside option is most significant for Globovision and RCTV, the most extreme channels in terms of opposition to the government. We next explore heterogeneity in these viewer responses along three dimensions. First, we examine whether responses differ during prime time hours, when viewership tends to be higher. As shown in the first column of Table 6, we do find that dropoff is larger during prime time but the interaction between prime time and private is statistically insignificant. Second, we explore whether responses differ during the weekends, when viewership tends to be lower. As shown in the second column, the interaction between weekend and private is again statistically insignificant. Finally, we examine whether responses differ between short and long cadenas, defined as those in excess of 2 h. As shown in the third column, responses are stronger on private channels during long cadenas but these differences are not statistically significant. Overall, these results are consistent with Propositions 1 and 2, which predict that viewers of news on private channels are more likely to turn off the television during cadenas and that the drop-off on the moderate channels during cadenas lies between the opposition channels and the public channel. This behavioral response of shifting to an outside option associated with unanticipated exposure to ideological content that is not like-minded in nature suggests that the impact of political propaganda may be limited. The results are in line with theories of television program choice, which predict that people select television content in order to satisfy their preferences (Youn 1994; Durante and Knight 2012; Yao, Wang, and Chen 2017), whereas, at the same time, suggesting that inertia in television viewership is incomplete (see Moshkin and Shachar 2002; Goettler and Shachar 2001; Perretti and Esteves-Sorenson 2012). 5.2. Other Transitions For comparison purposes, we next extend the analysis to investigate the effect of the reverse experiment: transitioning from cadenas to news programs. Although the formal model did not consider this possibility, it is natural to conjecture that the results should go in the opposite direction, with viewership of news rising on private, relative to public, following a cadena. As shown in Table 7, the coefficient in the first column is positive and statistically significant, documenting that private channels do experience an increase in viewership of 23%, relative to the public channel, when cadenas are followed by a news program. As shown in columns (2) and (3), the effect is driven by opposition channels, especially Globovision, which is the only channel that has a statistically significant coefficient, re-enforcing the idea that viewers of the opposition channel search for ideological content similar to their own ideology. Overall, these results are consistent with notion that viewers have preferences for watching like-minded political content. For comparison purposes, in Table 8, we examine the drop-off in rating when entertainment programs are interrupted by a cadena. We again find similar results to those in the analysis of a change in content from news to cadenas. Nevertheless, as shown in column (2), the results are similar for opposition and moderate channels, and, as shown in column (3), the results are economically significant for all four private channels. Finally, we analyze the change in viewership when moving from news to an entertainment program. Note that, although both involve transitions away from news, transitions from news to entertainment are not directly comparable to transitions from news to cadenas since the former are anticipated and the latter are not. As shown in Table 9, we find that private channels, relative to the public channel, generate a statistically significant 45% increase in ratings when moving from a news program to an entertainment program. This is similar in magnitude to the result for the drop-off when moving from news to cadenas, suggesting that our results may be about viewership of news on different channels per se rather than political ideology. On the other hand, it is not clear that entertainment programming on public channels is comparable to entertainment programming on private channels, which is very popular in Venezuela. Moreover, as shown in columns (2) and (3), the effects are again similar for opposition and moderate channels. The similarity of these results for entertainment across these private channels of differing ideology suggests that our baseline results are driven, at least in part, by channel ideology, rather than other characteristics of news programming on different channels. Taken together, the results for these other transitions suggest that our baseline results relating to channel ideology are not driven by other channel-specific characteristics. 5.3. Cable Television We next consider Propositions 3 and 4 in the context of cable channels, which were not required to broadcast cadenas. Given this, Proposition 3 predicts that the disproportionate drop-off in viewership on the private channel, relative to the public channel, should be more significant for households with cable subscriptions, relative to households without cable subscriptions. Likewise, Proposition 4 predicts that viewership of cable should rise during cadenas, and we test this prediction using data from RCTV International, which began as a cable channel during July 2007. In terms of Proposition 3, we begin by estimating the following regression: \begin{equation} \Delta ^{ic}(\mathit {cable})-\Delta ^{ic}(\mathit {nocable})=\beta _{i}+\epsilon _{ic}, \end{equation} (2) where the drop-off in viewership is now measured separately for cable and noncable households, and, according to Proposition 3, the coefficient for private channels, relative to public channels, should be negative. As shown in Table 10, and consistent with Proposition 3, the drop-off in ratings for those with cable, relative to households without cable, is indeed more significant for private channels, relative to public channels. In columns (2) and (3), we break out this effect by type of channel, finding that the effect is somewhat larger and only statistically significant for moderate channels and is driven in large part by Televen. Taken together, these results demonstrate that the opposition may be exposed to political propaganda to an even lesser degree when a source of opposition programming remains available during cadenas. This implies that viewers who are not able to afford cable, especially those already inclined to support the government, are disproportionately exposed to propaganda. Using ratings data from RCTV International, a cable channel created following the closing of RCTV on broadcast television, we next test Proposition 4, which predicts that RCTV cable ratings should rise during cadenas as viewers use this channel as a source of opposition programming. In particular, we estimate the following regression specification: \begin{equation} \Delta ^{p}(\mathit {RCTV}\,\mathit {cable)}=\beta _{1}\mathit {Change}\,\mathit {in}\,\mathit {cadena}\,\mathit {overlap}^{p}+\epsilon ^{p}, \end{equation} (3) where the left-hand side variable is the percentage change in ratings for program p airing on RCTV International, when compared to the previous program aired on RCTV International. To compute the key right-hand-side variable, we first compute cadena overlap for each RCTV cable show. Cadena overlap is defined as the fraction of minutes for which the RCTV cable show overlapped with a cadena. Thus, cadena overlap varies between zero and one, where the former value is attained if there is no cadena aired at any point of the show, and the latter value is attained if the show overlaps entirely with a cadena. Taking first differences of cadena overlap, we then compute the change in cadena overlap, which ranges in value from negative one to plus one. For this analysis, we use the sample from July 2007 to December 2007, the period in which RCTV is aired on cable. As shown in Table 11, and consistent with Proposition 4, we do find that RCTV cable ratings do rise during cadenas, and the effect is both economically and statistically significant. In particular, considering moving from no overlap to complete overlap (i.e. change in cadena overlap equal to one), we have that ratings on RCTV cable rise by an economically significant 69%. In the second column, we investigate whether these results differ according to the type of programming on RCTV cable. As shown, the results are larger for news programming on RCTV cable, when compared to other types of programming on RCTV cable. More concretely, viewership of RCTV cable news programming increases by 171% when a cadena comes on broadcast television, whereas viewership of non-news programming increases by only 61%. These results provide further support for our hypothesis of viewer choice of like-minded ideological content. 5.4. Content of Cadenas We next use more detailed information about individual cadenas, as provided by Nielsen in the form of short descriptions of the content of each cadena. Using this description and supplementing this with information found online, we create five categories of cadena content, and these are described in what follows. 5.4.1. Foreign Relations Coverage of foreign policy accomplishments, such as visits of presidents, multilateral agreements, and international travel by Chavez. 5.4.2. Delivery Coverage of events involving government promises of the provision of public goods, services, etc. 5.4.3. Elections Vroadcasts focusing on elections, especially coverage of the 2006 Presidential elections and the 2007 constitutional referendum. 5.4.4. Celebrations Coverage of public events, such as the birth of Simon Bolivar, marches, etc. 5.4.5. Information Summary of the progress of the country in several areas, such as economic and political. For cadenas that do not meet one of these definitions, we create a sixth category, other. Table 12 examines the drop-off in rating, separately, for each of these categories on the private channels, compared to the same categories in the public channels, when transitioning from news to cadena. This specification is consistent with the baseline analysis in column (1) of Table 4. The regression also controls for main effects of these categories, not reported in the table. Comparing the magnitude of the coefficients on the interactions, we have that the largest drop-off on private, relative to public, occurs for the categories delivery and elections. The large drop-off for delivery cadenas aired on private television may reflect the fact that many of these broadcasts involve Chavez himself delivering promises of public goods and services to his core voters. Given the targeting of these goods and services, there may be a particular distaste among opposition viewers for these cadenas. Likewise, cadenas about elections are, by their very nature, politically oriented and may have created polarized responses in terms of viewership. Finally, the smaller coefficient on the information category may reflect the fact that both opposition and progovernment viewers find these transmissions to be truly informative about the state of the economy or along other dimensions. 5.5. Summary To summarize, the results of the empirical analysis are consistent with the four key predictions of the model. First, the drop-off in ratings is more substantial for private channels, when compared to the public channel. Second, this effect is concentrated among opposition channels, and results for the moderate channels take an intermediate value. Third, the drop-off in viewership for the private channel is more significant for households with a cable subscription. Fourth, viewership of RCTV International, an opposition cable channel that opened during 2007, rises significantly during cadenas. Finally, we examine heterogeneity according to the content of cadenas, with the largest drop-off of viewership on private channels for cadenas associated with the delivery of public goods and services and for cadenas related to elections. 6. Structural Estimation Building upon this evidence, we next provide estimates of a structural version of the model. We begin by extending the model and the notation to allow for non-news programming and gender-specific preferences over this programming. We then detail several issues in the empirical implementation and describe identification. After presenting the parameter estimates, we use the model to conduct counterfactual experiments. These experiments allow us to quantify the degree to which switching limits exposure to propaganda and ultimately political persuasion among the opposition. 6.1. Approach As in the first extension of the model, we consider three types of stations: government (i = g), moderate (i = m), and opposition (i = o). As before, let $$v$$ ∈ {o, g} index viewer ideology, opposition, and progovernment. Then, viewers receive payoffs equal to θs from same-type programming (cadenas and government news for progovernment viewers and opposition news for opposition viewers) and payoffs equal to θd from different-type programming (cadenas and government news for opposition viewers and opposition news for progovernment viewers). Both opposition and progovernment viewers receive payoffs of θm from moderate news. To estimate switching costs, we also consider the following additional types of non-news programming: soap operas (telenovelas), sports, and other. Following Esteves-Sorenson and Perretti (2012), we measure switching costs via gender-specific preferences over soaps and sports.22 In particular, let k ∈ {m, f} index viewer gender, and let u$${vkip}$$ represent gender-specific systematic payoffs for a viewer with ideology $$v$$ watching programming p on station i. Table 13 summarizes these payoffs. As shown, payoffs from news programming and cadenas are assumed to differ across viewer ideology but not gender, and payoffs from non-news programming, such as sports and soaps, differ across gender but not ideology. Likewise, although preferences for news programming vary across stations, a simplifying assumption is that preferences for sports, soaps, and other programming vary across viewer types but not across stations. In the context of this model, we next derive market shares, separately for viewers of ideology $$v$$ and gender k. We sequence shows within a day according to the time aired (t = 1, 2, 3, …).23 Then, with positive switching costs (η > 0), market shares for a viewer with ideology $$v$$ and gender k watching programming p on station i at time t ($$\sigma _{vkip}^{t}$$), as a function of market shares during the previous time slot ($$\sigma _{vkip}^{t-1}$$), are given by \begin{eqnarray*} \sigma _{vkip}^{t}&=&\sigma _{vkip}^{t-1}\frac{\exp \big(u_{vkip}^{t}\big)}{\exp \big(u_{vkip}^{t}\big)+\sum _{j\ne i}\exp \big(u_{vkjp}^{t}-\eta \big)}\nonumber\\ && +\sum _{l\ne i}\sigma _{vklp}^{t-1}\frac{\exp \big(u_{vkip}^{t}-\eta \big)}{\exp \big(u_{vklp}^{t}\big)+\sum _{j\ne l}\exp \big(u_{vkjp}^{t}-\eta \big)} . \end{eqnarray*} The first term represents the likelihood that a viewer is both watching channel i during the previous time slot (t − 1) and does not switch to another channel at time t. The second term represents the likelihood that a viewer is both watching a different channel (l ≠ i) during the previous time slot and switches from channel l to channel i at time t, incurring switching costs equal to η. This is then summed across all other options. This includes the outside option of not watching television, which, as stated previously, is normalized to provide a systematic payoff of zero. To illustrate the intuition behind these market shares, consider two special cases. First, with high switching costs (η → ∞), market shares do not change between time t − 1 and time t; that is, $$\sigma _{vkip}^{t}=\sigma _{vkip}^{t-1}$$. In this case, inertia is complete, and viewership does not respond to the airing of cadenas. Second, in the absence of switching costs (η = 0), market shares at time t are independent of market shares at time t − 1 and collapse to the standard multinomial logit form: \begin{equation*} \sigma _{vkip}^{t}=\frac{\exp \big(u_{vkip}^{t}\big)}{\exp \big(u_{vkip}^{t}\big)+\sum _{j\ne i}\exp \big(u_{vkjp}^{t}\big)}. \end{equation*} In this case, there is no inertia. Although viewership does respond to the airing of cadenas, the impact lasts for only one period, with viewership during future periods unchanged. In intermediate cases, with moderate switching costs, inertia exists but is incomplete. In particular, a positive shock to viewership of channel i at time t − 1 leads to higher viewership of that channel at time t. For example, if females have a stronger preference for soaps than males, then a soap airing at time t − 1 will, all else equal, tend to increase viewership of that channel for females, relative to males, at time t. This is due to the presence of switching costs, resulting in inertia in viewership. Since our data distinguish between male and female viewers but not between progovernment and opposition viewers, we next aggregate market shares across opposition and progovernment. Recalling that πg represents the fraction of progovernment viewers, we have that market shares among gender k for station i airing programming p equal \begin{equation*} \sigma _{kip}^{t}=\pi _{g}\sigma _{gkip}^{t}+(1-\pi _{g})\sigma _{okip}^{t}. \end{equation*} For the purposes of estimation, these model-based market shares ($$\sigma _{kip}^{t}$$) are then linked to observed market $$(s_{kip}^{t})$$ shares via the following log-odds formulation: \begin{equation*} \ln \left(\frac{s_{kip}^{t}}{1-s_{kip}^{t}}\right)=\ln \left(\frac{\sigma _{kip}^{t}}{1-\sigma _{kip}^{t}}\right)+\epsilon _{kip}^{t}, \end{equation*} where $$\epsilon _{kip}^{t}$$ is assumed to be normally distributed. Then, the parameters of the model (e.g., θd, θm, θs, η) are estimated via maximum likelihood. 6.2. Empirical Implementation and Identification Before presenting estimates of the parameters of this model, we first address three issues regarding empirical implementation. We then provide an intuitive overview of identification. First, although the previous formulation assumed that the sequence of programming (t = 1, 2, 3…) was identical across channels, programming schedules differ across channels within a day. For example, RCTV may air a program from 6 p.m. to 6:30 p.m., whereas Globovision may air a program from 5:30 p.m. to 6:20 p.m. and then another show from 6:20 p.m. to 7 p.m. In this case, for a given program, it is unclear how to define the set of competing shows, those aired on other channels. To do so, we define, for each show, the set of competing shows on other channels as those with the maximal time overlap with the focal program. In the previous example, a show airing from 6 p.m. to 6:30 p.m. on RCTV would compete for viewership with the show airing from 5:30 p.m. to 6:20 p.m. on Globovision, which shares 20 min of programming, rather than the show airing from 6:20 p.m. to 7 p.m., which shares only 10 min of programming. Second, given the recursive formulation stated previously, in which viewership at time t depends upon viewership at time t − 1, one must define initial conditions for market shares. To do so, we assume zero viewership before 6 a.m., when most of the population is sleeping, and ratings are consequently extremely low. That is, we assume that the entire market is consuming the outside option of no television, which provides a systematic payoff of zero, prior to t = 1. This allows us to write viewership during the first time slot (t = 1) as follows: \begin{equation*} \sigma _{vkip}^{1}=\frac{\exp \big(u_{vkip}^{1}-\eta \big)}{1+\sum _{j\ne 0}\exp \big(u_{vkjp}^{1}-\eta \big)}, \end{equation*} where $$u_{vkip}^{1}-\eta$$ is the payoff from switching from the outside option to channel i airing programming p at t = 1 and j = 0 refers to the outside option. In addition to closing the model, this assumption implies no dynamic linkages in viewership between midnight and 6 a.m., allowing us to treat each day as an independent observation. Third, since we do not observe market shares separately for opposition and government viewers, we must aggregate across these groups, as outlined previously. Given this, one must thus measure the fraction of progovernment viewers (πg) in each municipality. To do so, we measure these via municipality-specific vote shares for the opposition party and Chavez, respectively, in the 2006 Presidential election. The intuition for identification is explained in several steps. First, gender-specific preferences over sports and soap operas are identified simply by comparing ratings for these types of programming across male and female viewers. Then, with these estimates of gender-specific programming, switching costs can be identified by examining gender-specific ratings for shows aired on the same channel but after these sports and soaps programs. Finally, with estimates of these switching costs, one can identify ideology-specific preferences over news and cadenas by examining, similarly to the reduced form evidence presented previously, changes in ratings during cadenas that interrupt news programming across different types of stations (opposition, moderate, and progovernment). This identifies preferences over ideological content, as given by θd, θm, and θs.24 6.3. Parameter Estimates Our parameter estimates are provided in Table 14. Note that these coefficients should be considered relative to programming other than news, cadenas, sports, and soaps. This includes categories such as movies and game shows, which receive payoffs equal to the constant term, and the payoff from not watching television is normalized to zero. Following the identification logic given previously, we begin by discussing gender-specific preferences over news programming. As seen, we find overall high viewership for soaps. This is true for men and, consistent with prior evidence (Esteves-Sorenson and Perretti 2012), especially so among female viewers. Likewise, we find slightly lower viewership for sports but especially so among female viewers. These two gender-specific coefficients are both economically and statistically significant, with females, relative to males, having 56% higher viewership for soaps and 18% lower viewership for sports. In addition, females have 18% higher viewership across all categories. As noted previously, by comparing gender-specific ratings on shows immediately following sports and soaps, we can identify switching costs. As shown, these estimated switching costs are also statistically significant, and evidence on their economic significance will be documented in a counterfactual analysis to follow, in which we trace out the dynamic response to cadenas for viewers and channels of differing ideology. Finally, using these estimates of switching costs to identify preferences over ideological content, we have that payoffs from information are associated with lower viewership overall. This is the payoff for both progovernment and opposition viewers from consuming moderate news. As shown, this negative effect is partially offset for same-type information, cadenas, and news on government channels for progovernment viewers and news on opposition channels for opposition viewers. Conversely, payoffs are substantially lower for different-type information, cadenas, and news on government channels for opposition viewers and news on opposition channels for progovernment viewers. Note also the asymmetry between same-type and different-type ideology, with the benefits associated with same-type information (0.2409) smaller than the costs associated with different-type information (1.6717). Taken together, these estimates provide additional support for the hypothesis of preferences for like-minded information. 6.4. Counterfactual Viewership Using these parameter estimates, we then summarize viewership patterns under three scenarios. First, we predict viewership of cadenas as predicted by the structural model separately by channel and separately for progovernment and opposition viewers. Second, we predict viewership patterns under a counterfactual scenario in which cadenas are replaced with news programming. That is, propaganda is replaced by opposition content on opposition channels and moderate content on moderate channels, with no change in content on government channels. Third, we predict viewership patterns under a counterfactual scenario in which cadenas are aired but under which viewers face infinite switching costs and thus cannot switch channels or tune out. This allows us to quantify the degree to which switching limits exposure to government propaganda. Analysis of these three scenarios proceeds in the following four steps. First, we focus on the set of days on which a single cadena was broadcast. This allows us to assume that viewership just before the cadena is identical under the three scenarios. Second, using this sample of days and normalizing the time slot of the cadena to equal zero, we use the estimated model to predict viewership of shows aired throughout the day. For simplicity, we focus on viewership among females and in cities with the full set of available channels (i.e., Caracas and Valencia). Third, we use the model to predict how viewership would have evolved were cadenas to be replaced by news programming, with ideological content depending upon the channel under consideration. Only programming in the focal time slot (t = 0) is altered, and programming during the other time slots is unchanged under the counterfactual. Similarly, and as noted previously, we also consider a counterfactual scenario in which viewers cannot tune out when cadenas come on the air. Table 15 compares viewership patterns under these three scenarios. For progovernment viewers, patterns of viewership are similar under the three scenarios. Viewership combined across the five channels totals 20.4% under our baseline scenario (propaganda), 22.1% under propaganda without switching, and 16.7% under media pluralism. The lower consumption under media pluralism reflects the fact that the media landscape is more opposition-oriented in this case. Table 15. Counterfactual viewership summary. Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 View Large Table 15. Counterfactual viewership summary. Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 View Large For opposition viewers, by contrast, consumption is substantially different under the three scenarios. Viewership combined across the five channels totals 9.8% under propaganda, 22.8% under propaganda without switching, and 19.7% under media pluralism. The difference between propaganda with and without switching is driven by significant dropoff on both opposition and moderate channels when a cadena comes on the air, and viewership of the public channel is close to zero under all three scenarios. Taken together, we find that switching significantly limits exposure to propaganda. That is, exposure to propaganda would be more than twice as high for opposition viewers in the absence of switching. 6.5. Implications for Persuasion Given that switching limits exposure to propaganda, it is natural to investigate how this impacts political persuasion. To do so, we next calculate the electoral effects of propaganda under the three scenarios described previously. We use estimates from the literature on persuasion rates, defined as the probability of converting a voter not already persuaded to support the favored candidate based upon media exposure.25 DellaVigna and Gentzkow (2010) summarize persuasion rates in the literature on switching votes as varying between 6% and 20%, and we use the midpoint of 13%. That is, we assume that opposition viewers exposed to propaganda are converted to the government’s side with 13% probability. Likewise, under the media pluralism scenario, we assume that exposure of progovernment viewers to opposition news are converted to the opposition side with 13% probability.26 Finally, we assume that moderate news converts progovernment viewers to the opposition with 6.5% probability (one-half of baseline persuasion) and converts opposition viewers to the progovernment side with 6.5% probability.27 Note that persuasion rates may vary across contexts, and the set of countries considered in DellaVigna and Gentzkow (2010) does not include Venezuela. In addition, persuasion rates should naturally depend upon the types of interventions, with long cadenas having more influence than short cadenas. Given these limitations, we also provide estimates using the lower end of this interval (6%) and the upper end of this interval (20%). As shown in Table 15, we find that having the government control the airways via propaganda leads to 1.3% of opposition viewers persuaded to support the government. Progovernment viewers already support the government and hence there is no persuasion for this group. Given that the opposition represents 45% of the electorate in Caracas, we have support for the government increases by 0.6 percentage points (net persuasion) and thus support for the opposition falls by 0.6 percentage points, a swing of 1.2 percentage points toward the government. Under media pluralism, by contrast, 1.1% of progovernment viewers are persuaded to support the opposition, with the patterns reversed for 0.7% of opposition viewers. These effects largely cancel out, and we have that support for the government falls by 0.3 percentage points. Under propaganda without switching, we have large effects for opposition viewers, with 3.0% switching their support to the government. Comparing these two propaganda scenarios, one with and one without switching, we again demonstrate that switching limits exposure to propaganda and thus also ultimately limits ideological persuasion among opposition viewers. Finally, not reported in this table, we consider the fraction persuaded under both lower and higher persuasion rates. When the persuasion rate is 6% instead of the baseline 13%, we have that 0.6% of opposition viewers are persuaded by cadenas and this increases to 1.4% in the absence of switching. When the persuasion rate is 20%, by contrast, we have that 2.0% of opposition viewers are persuaded by cadenas and this increases to a very large 4.6% in the absence of switching. As noted previously, these estimates are based on persuasion rates from other contexts. Nonetheless, these results demonstrate the potential for behavioral responses in limiting the influence of government propaganda. 6.6. Dynamic Responses Despite substantial behavioral responses, many opposition viewers are exposed to cadenas in our baseline propaganda scenario. Moreover, those who do tune out may consume less opposition news during subsequent time slots if they do not return to opposition channels. We next examine this issue by analyzing the dynamics of viewer responses under two scenarios: propaganda and media pluralism. The results from this exercise are provided in Figure 7, in which we plot viewership during the two shows aired before the cadena, the cadena, and the six shows aired after the cadena. The x-axis is time to cadena and is normalized to equal zero during the cadena time slot. The y-axis is the viewership market share, separately by channel and viewer ideology. The upper panel provides results for ratings on the opposition channel, with progovernment viewers on the left and opposition viewers on the right. The middle and bottom panels provide corresponding results for moderate and government channels. Finally, we consider both market shares under propaganda, as given the solid line, and market shares under the media pluralism counterfactual, as given by the dashed line. Figure 7. View largeDownload slide Predicted viewership under propaganda and counterfactual media pluralism. Figure 7. View largeDownload slide Predicted viewership under propaganda and counterfactual media pluralism. Consistent with the viewership summary described previously, we see a sharp dropoff in viewership among opposition viewers when a cadena comes on the air for both opposition and moderate channels. Moreover, due to the presence of switching costs, differences in viewership, depending upon whether the previous show is a cadena or opposition news, are also apparent during the subsequent time slot (i.e., t = 1), with the counterfactual path of viewership then converging back to the predicted path of viewership several time slots following the cadena. Thus, cadenas have a persistent effect on viewership of opposition channels, with a sustained decrease in viewership by opposition viewers. This can be interpreted as a multiplier effect, under which cadenas reduce viewership of private channels not only during the cadena but also during subsequent time slots. Due to the reduction in exposure to opposition news, opposition viewers may also be less informed in general, making it more difficult for them to hold incumbents accountable. 6.7. Welfare Analysis Finally, we consider consumer welfare under two scenarios: government propaganda and media pluralism. In particular, we measure the overall welfare of opposition and progovernment viewers, respectively, when cadenas come on the air, relative to the counterfactual in which each channel airs news programming. This allows us to measure the welfare gains from media pluralism, defined as moving from an environment with only government programming on all channels to a situation with two channels airing opposition news and two channels airing moderate news. Welfare is measured using the inclusive value, the standard measure in discrete choice models. This is calculated by taking the expected value of the maximal utility over the choice set. Abstracting from gender and taking viewership probabilities at time slot t − 1 as given, the welfare of a viewer with ideology $$v$$ at time t is given by, \begin{equation*} W_{v}^{t}=\sum _{l}\sigma _{vlp}^{t-1}\ln \left[\exp \big(u_{vlp}^{t}\big)+\sum _{j\ne l}\exp \left(u_{vjp}^{t}-\eta \right)\right]. \end{equation*} Within the summation, the term \begin{equation*} \ln \left[\exp \big(u_{vlp}^{t}\big)+\sum _{j\ne l}\exp \big(u_{vjp}^{t}-\eta \big)\right] \end{equation*} represents the value to viewers with ideology $$v$$ watching channel l at time t − 1, where $$u_{vlp}^{t}$$ is the payoff from continuing to watch channel l at time t and $$u_{vjp}^{t}-\eta$$ is the payoff associated with switching at time t from channel l to a different channel j ≠ l. These values associated with watching a given channel at time t − 1 are then aggregated across channels, weighting by viewership at time t − 1. The results from this welfare analysis are presented in Table 16. As shown, welfare for progovernment viewers falls when moving from an environment in which all channels air government programming to media pluralism, an environment with one channel airing government programming, two channels airing moderate programming, and two channels airing opposition programming. This simply reflects the fact that overall ideological content is more opposition-oriented under the counterfactual, relative to the scenario in which only government programming is aired on all five channels. For opposition viewers, by contrast, the pattern is reversed, with an increase in welfare under media pluralism, again reflecting the fact that overall ideological content is more opposition-oriented in this case. Table 16. Welfare analysis. Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 View Large Table 16. Welfare analysis. Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 View Large Using the 55% share of progovernment viewers, we have that aggregate welfare rises under media pluralism, despite the fact that opposition viewers comprise a minority. This simply reflects the fact that the welfare gains from media pluralism for the opposition exceeds the welfare losses to progovernment viewers.28 In closing, we note three limitations of this welfare analysis. First, we do not have monetary measures of welfare and thus cannot determine whether or not these welfare differences are economically significant.29 Second, this analysis assumes that viewer choices are personal and reflect their underlying preferences. This assumption might be violated if, for example, public employees are expected to watch cadenas and do so in order to enhance their job security. Third, our welfare measures do not account for the consequences of voters using information to hold incumbents accountable. 7. Conclusion In future work, we plan to pursue related topics involving media bias in Venezuela. First, we plan to study the government’s strategic decision regarding the timing of cadenas. For example, does the government air cadenas in order to censor opposition news during key events, such as protests? Second, we plan to examine the bundling of entertainment and news programming. For example, do viewers choose to watch news on channels with popular entertainment programming, and do channels with an ideological objective respond to this type of behavior when developing their programming? Third, although this study has relied on persuasion rates from the existing literature, we hope to measure the persuasive impact of propaganda in the Venezuelan context. Consistent with a preference for like-minded ideological content, we find that viewers respond to high frequency variation in the ideological slant of television programming. These responses are stronger for private channels, when compared to public channels, and for the most ideological channels. The responses are stronger for viewers with larger choice sets, as proxied via cable. Consistent with this result, we also show that viewership of cable increases during cadenas. The results are also stronger for the most polarizing cadenas, those involving the delivery of goods and services and those related to elections. Building upon this evidence, we structurally estimate the model. Based upon counterfactuals, we document that these behavioral responses significantly limit exposure to propaganda among opposition viewers. This may potentially limit the persuasive impact of government propaganda. There have been many changes in the media landscape in Venezuela since our sample period, with the rise of social media a key factor. The intensive use of political propaganda and the homogenization of ideologies in the television channels in Venezuela have made social media particularly popular. The opposition uses social media given the lack of opposition television channels at current. As an example, RCTV began online broadcasting through the Internet, and protesters often find social media an effective means for organizing demonstrations. Likewise, the government now uses social media to deliver propaganda. Nevertheless, television remains a key government vehicle for delivering propaganda since mainstream media continues to reach a large fraction of the population. Notes The editor in charge of this paper was Paola Giuliano. Acknowledgments: For helpful comments, we thank seminar participants at Brown University, Rice University, Carnegie Mellon University, Barcelona GSE Summer Forum, Stony Brook Political Economy Conference, CFPE Conference (Vancouver School of Economics), the New York City Media Seminar, USC Marshall, Workshop in Political Economy (Uppsala), CEMFI, Singapore Management University, University of Calgary, and Washington University. The opinions and statements are the sole responsibility of the authors and do not necessarily represent neither those of the Banco de la República de Colombia nor of its Board of Directors. Footnotes 1 This paper studies three types of behavioral responses by viewers: tuning out, switching, and staying tuned in. Hirschman (1970) identifies Exit, Voice, and Loyalty as three responses to propaganda. In this case, exit is the equivalent of tuning out, since viewers can avoid propaganda completely, but they are also consuming less information overall across all media. Thus, exit may lead viewers to become more disengaged in general with politics, thus becoming less informed and potentially weakening political accountability. Voice is the equivalent of switching since it is a form of protest and people can choose to consume like-minded information. Finally, loyalty is the equivalent of staying tuned in. 2 Recent studies on persuasion include DellaVigna and Kaplan (2007), Enikolopov, Petrova, and Zhuravskaya (2011), George and Waldfogel (2003), Chiang and Knight (2011), Gentzkow, Shapiro, and Sinkinson (2011), Gerber, Karlan, and Bergan (2009), Martin and Yurukoglu (2015), Prat (2014), and Snyder and Stromberg (2010). See also Prat and Stromberg (2013) for a comprehensive overview of this literature. 3 DiTella, Galiani, and Schargrodsky (2012) study the effects of government propaganda against privatization of water services after the 2006 nationalization in Argentina. Qian and Yanagizawa-Drott (2017) document an increase in U.S. news coverage of human rights abuses in countries not aligned with the United States when they rotated onto the United Nations. Security Council during the Cold War. They report similar patterns for reports produced by the U.S. State Department, suggesting an important role for government propaganda. Other literature focuses on the power of propaganda to mobilize the masses. Adena et al. (2015) document the importance of political propaganda to mobilize support for the Nazis. One interesting finding in this study is that a predisposition against propaganda can limit its effectiveness, and our study explores a potential mechanism underlying this finding. Yanagizawa-Drott (2014) provides evidence on the role of propaganda broadcast on radio by the Hutu government during the Rwandan genocide. DellaVigna et al. (2014) document an instance in which propaganda had negative consequences: cross-border exposure to Serbian radio among Croats is associated with anti-Serbian sentiment and anti-Serbian behavior. 4 This section draws upon Wilpert (2007), Corrales and Penfold (2011), Nelson (2009), Republica Bolivariana de Venezuela (2012), and Dinneen (2012). 5 Chang-Tai et al. (2011) document that voters who supported the Presidential recall referendum against Chavez experienced a significant reduction in earnings and employment following the public release of a list of voters who signed the recall petition. 6 For example, private television channels tended to cover only antigovernment protests during the coup and pointed to the government as the cause of violence in the struggle between Pro-Chavez and Anti-Chavez protesters. Once Chavez returned to power, private channels stopped broadcasting news, and a Chavez speech was aired in split-screen to broadcast anti-Chavez protests in parallel with the speech by Chavez. During the strike, the media gave priority to this issue for more than two months, often suspending regular programming for more extensive coverage of the crisis. Even when the protests were significantly weakened, some private media commentators continued to call for Chavez’s resignation in order to end the crisis. 7 Chavez accused the private channels publicly of “inciting rebellion and disrespect 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”. See Reporters Without Borders (2003). 8 New York Times (2007). 9 Besley and Prat (2006) analyze government capture of the media sector. 10 RCTV International was later shut down, closing in 2010. 11 In addition to cadenas, Chavez also hosts a public television program titled “Alo Presidente”, where he promoted the Bolivarian revolution. The show started at 11 a.m. every Sunday and lasted about 5 h (Kitzberger 2010). Frajman (2014) argues that Alo Presidente was a “grand stage for Chavez to promote his position as revolutionary leader and be cheered by crowds of loyal supporters”. 12 The type-1 extreme value distribution assumption is necessary and sufficient for generating logit probabilities. See McFadden (1973) for further details. 13 We do assume, for tractability reasons, that viewers receive a new ε$${vip}$$ draw for the outside option when a cadena comes on the air. This assumption is not crucial to our results given in what follows and can be relaxed. 14 Given our use of average minute rating, the measure of drop off combines viewers who switch off immediately and viewers who switch off part way through. 15 Each member of the household has a separate code, allowing Nielsen to separate viewership within a household according to gender. 16 We group them into three categories: news, entertainment, and cadenas and thus the numbers in the table do not add to 100%. News programs include the categories “Information/Opinion” and “Documentaries”. Entertainment includes “Sports”, “Entertainment”, “Children”, “Games”, “Microseries”, “Miniseries”, “ Movies”, “Series”, and “Soap Operas”. Finally, we leave the category “cadenas” as is. 17 This evidence is consistent with EU-EOM (2006), which shows that VTV and Globovision devoted greater time to political information during 2006 elections and the private channels RCTV, Venevision, and Televen devoted far less time to political information. 18 Likewise, using other measures of political preferences, not reported here, we find that people who watch news on public channels report higher levels of trust in Chavez than people who watch private channels. 19 Our measure of viewership is based upon the percentage of (potential) viewers. But our measures would be identical if we used the absolute number of viewers since drop-off is based upon the percent change when transitioning from news to cadenas. That is, drop-off for channel i is given by Δ$${ic}$$ = ln [s$${ic}$$/s$${in}$$]. Viewer shares for cadenas and news equal s$${ic}$$ = N$${ic}$$/N and s$${in}$$ = N$${in}$$/N, where N$${ic}$$ is the number of viewers of the cadena, N$${in}$$ is the number of viewers of the preceding news program, and N is the population (number of potential viewers). Thus, our dropoff measure would be identical if it were based upon the absolute number of viewers (i.e., s$${ic}$$ = N$${ic}$$ and s$${in}$$ = N$${in}$$) since population appears in both the numerator and the denominator of our measure of dropoff. 20 For this analysis we drop cases where the gap between the end time of the news and the start time of the cadena exceeds 10 min. 21 The difference between these two coefficients is statistically significant at conventional levels. 22 Note that our data also include ratings according to age and socioeconomic status. We choose to focus on gender given the evidence in Esteves-Sorenson and Perretti (2012). 23 As in the baseline theoretical model, viewers receive a new draw from the distribution for unobserved preferences (ε$${vkip}$$) for each new time period/show. 24 Note that this intuition is incomplete, as the model also implies linkages between switching costs and preferences over ideological content via the ideological mechanism. That is, in order to observe a disproportionate drop-off in viewership when a cadena is aired on private channels, it must be the case that switching costs are positive. Otherwise, as noted previously, viewership of cadenas is independent of previous programming, which includes the ideological content of news. Thus, switching costs are identified via both gender-specific preferences over programming and channel-specific changes in viewership when cadenas are aired. 25 In using this definition of persuasion, we abstract from an alternative goal of propaganda: maintaining a base of core supporters. 26 One could also consider asymmetric persuasion rates. If, for example, opposition news is more persuasive than propaganda, then our estimates may overstate the importance of switching among opposition viewers exposed to propaganda. 27 Net persuasion results are very similar when one ignores any persuasive effects of moderate news since the effects for progovernment and opposition viewers tend to cancel out. 28 To better understand why these gains to the opposition under media pluralism exceed the losses to progovernment viewers, consider the following simple example. Suppose there are only two stations and an equal number of opposition and progovernment viewers (i.e., πg = 0.5). With both channels airing government programming, welfare for opposition viewers equals ln [2 exp (θd)], and welfare for progovernment viewers is now ln [2exp (θs)]. Under media pluralism, by contrast, welfare for each group equals ln [exp (θs) + exp (θd)]. One can then show that aggregate welfare rises under media pluralism due to a preference for like-minded information (θd < θs). This simply reflects the fact that the gain to opposition viewers from the introduction of a like-minded station is less than the loss to progovernment viewers from losing one of two like-minded stations. 29 Moreover, the cross-group comparisons rely on the assumption that the variance of the unobservables is the same across groups. References Adena Maja , Enikolopov Ruben , Petrova Maria , Santarosa Veronica , Zhuravskaya Ekaterina ( 2015 ). “ Radio and the Rise of the Nazis in Pre-War Germany .” Quarterly Journal of Economics , 130 , 1885 – 1939 . Google Scholar CrossRef Search ADS Besley Timothy , Prat Andrea ( 2006 ). “ Handcuffs for the Grabbing Hand? Media Capture and Government Accountability .” American Economic Review , 96(3) , 720 – 736 . Google Scholar CrossRef Search ADS Bisbal Marcelino ( 2009 ). “ La Comunicacion Masiva Como Politica de Gobierno de Hugo Chavez Frias .” In Hegemonia y Control Comunicacional , edited by Bisbal Marcelino . Editorial Alfa, Caracas, Venezuela . Chang-Tai Hsieh , Miguel Edward , Ortega Daniel , Rodriguez Francisco ( 2011 ). “ The Price of Political Opposition: Evidence from Venezuela’s Maisanta .” American Economic Journal: Applied Economics , 3 , 196 – 214 . Google Scholar CrossRef Search ADS Chiang Chun-Fang , Knight Brian ( 2011 ). “ Media Bias and Influence: Evidence from Newspaper Endorsements .” Review of Economic Studies , 78 , 795 – 820 . Google Scholar CrossRef Search ADS Corrales Javier , Penfold Michael ( 2011 ). Dragon in the Tropics: Hugo Chavez and the Political Economy of Revolution in Venezuela . Brookings Latin America Initiative Books , Washington DC, USA . DellaVigna Stefano , Enikolopov Ruben , Mironova Vera , Petrova Maria , Zhuravskaya Ekaterina ( 2014 ). “ Cross-Border Media and Nationalism: Evidence from Serbian Radio in Croatia .” American Economic Journals: Applied Economics , 6 , 103 – 132 . Google Scholar CrossRef Search ADS DellaVigna Stefano , Gentzkow Matthew ( 2010 ). “ Persuasion: Empirical Evidence .” Annual Review of Economics , 2 , 643 – 669 . Google Scholar CrossRef Search ADS DellaVigna Stefano , Kaplan Ethan ( 2007 ). “ The Fox News Effect: Media Bias and Voting .” Quarterly Journal of Economics , 122 , 1187 – 1234 . Google Scholar CrossRef Search ADS Dinneen Mark ( 2012 ). “ The Chavez Government and the Battle over the Media in Venezuela .” Asian Journal of Latin American Studies , 25 , 27 – 53 . DiTella Rafael , Galiani Sebastian , Schargrodsky Ernesto ( 2012 ). “ Reality versus Propaganda in the Formation of Beliefs about Privatization .” Journal of Public Economics , 96 , 553 – 567 . Google Scholar CrossRef Search ADS Durante Ruben , Knight Brian ( 2012 ). “ Partisan Control, Media Bias, and Viewer Responses: Evidence from Berlusconi’s Italy .” Journal of the European Economic Association , 10 , 451 – 481 . Google Scholar CrossRef Search ADS Enikolopov Ruben , Petrova Maria , Zhuravskaya Ekaterina ( 2011 ). “ Media and Political Persuasion: Evidence from Russia .” American Economic Review , 101 ( 7 ), 3253 – 3285 . Google Scholar CrossRef Search ADS EU-EOM ( 2006 ). “ Final Report: Presidential Elections Venezuela 2006 .” Tech. rep., European Union Election Observation Mission . Frajman Eduardo ( 2014 ). “ Broadcasting Populist Leadership: Hugo Chavez and Alo Presidente .” Journal of Latin American Studies , 43 , 501 – 526 . Google Scholar CrossRef Search ADS Gehlbach Scott , Sonin Konstantin ( 2014 ). “ Government Control of the Media .” Journal of Public Economics , 118 , 163 – 171 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. ( 2010 ). “ What Drives Media Slant? Evidence from US Daily Newspapers .” Econometrica , 78 , 35 – 71 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. , Sinkinson Michael ( 2011 ). “ The Effect of Newspaper Entry and Exit on Electoral Politics .” American Economic Review , 101 ( 7 ), 2980 – 3018 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. , Sinkinson Michael ( 2014 ). “ Competition and Ideological Diversity: Historical Evidence from US Newspapers .” American Economic Review , 104 ( 10 ), 3073 – 3114 . Google Scholar CrossRef Search ADS George Lisa , Waldfogel Joel ( 2003 ). “ Who Affects Whom in Daily Newspaper Markets? ” Journal of Political Economy , 111 , 765 – 784 . Google Scholar CrossRef Search ADS Gerber Alan S. , Karlan Dean , Bergan Daniel ( 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 , 35 – 52 . Google Scholar CrossRef Search ADS Goettler Ronald L. , Shachar Roni ( 2001 ). “ Spatial Competition in the Network Television Industry .” RAND Journal of Economics , 32 , 624 – 656 . Google Scholar CrossRef Search ADS Hirschman Albert ( 1970 ). Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States . Harvard University Press , Cambridge . Kitzberger Philip ( 2010 ). “ The Media Activism of Latin America’s Leftist Governments: Does Ideology Matter? ” GIGA Working Paper 151, German Institute of Global and Area Studies, Hamburg, Germany . Google Scholar CrossRef Search ADS Martin Gregory , Yurukoglu Ali ( 2015 ). “ Bias in Cable News: Real Effects and Polarization .” NBER Working Paper No. 20798, National Bureau of Economic Research, Cambridge, MA, USA. McFadden Daniel ( 1973 ). “ Conditional Logit Analysis of Qualitative Choice Behavior .” In Frontiers in Econometrics , edited by Zarembka P. . Academic Press, New York, USA . Moshkin Nickolay , Shachar Ron ( 2002 ). “ The Asymmetric Information Model of State Dependence .” Marketing Science , 21 , 435 – 454 . Google Scholar CrossRef Search ADS Nelson Brian A. ( 2009 ). The Silence and the Scorpion: The Coup Against Chavez and the Making of Modern Venezuela . Nation Books , New York, USA . Perretti Fabrizio , Esteves-Sorenson Constanca ( 2012 ). “ In Micro-Costs: Inertia in Television Viewing .” The Economic Journal , 122 , 867 – 902 . Google Scholar CrossRef Search ADS Prat Andrea ( 2014 ). “ Media Power .” CEPR Discussion Paper No. 10094, Center for Economic and Policy Research, Washington, DC, USA. Prat Andrea , Stromberg David ( 2013 ). “ The Political Economy of Mass Media .” In Advances in Economics and Econometrics , edited by Daron Acemoglu, Manuel Arellano, and Eddie Dekel . Cambridge University Press, Cambridge , pp. 135 – 187 . Google Scholar CrossRef Search ADS Qian Nancy , Yanagizawa-Drott David ( 2017 ). “ Government Distortion in Independently Owned Media: Evidence from U.S. Cold War News Coverage of Human Rights .” Journal of the European Economic Association , 15 , 463 – 499 . Google Scholar CrossRef Search ADS Reporters Without Borders ( 2003 ). Venezuela: Caught between an authoritarian president and intolerant media . Available at: http://www.refworld.org/docid/47fcdc365.html [accessed 12 March 2018] . Republica Bolivariana de Venezuela ( 2012 ). Abril , Memorias de un golpe de estado . Republica Bolivariana de Venezuela . Snyder James M. , Stromberg David ( 2010 ). “ Press Coverage and Political Accountability .” Journal of Political Economy , 118 , 355 – 408 . Google Scholar CrossRef Search ADS Wilpert Gregory ( 2007 ). Changing Venezuela by Taking Power: The History and Policies of the Chavez Government . Verso Books , London, England . Yanagizawa-Drott David ( 2014 ). “ Propaganda and Conflict: Evidence from the Rwandan Genocide .” Quarterly Journal of Economics , 129 , 1947 – 1994 . Google Scholar CrossRef Search ADS Yao Song , Wang Wenbo , Chen Yuxin ( 2017 ). “ TV Channel Search and Commercial Breaks .” Journal of Marketing Research , 54 , 671 – 686 . Google Scholar CrossRef Search ADS Youn Sug-Min ( 1994 ). “ Program Type Preference and Program Choice in a Multichannel Situation .” Journal of Broadcasting & Electronic Media , 38 , 465 – 475 . Google Scholar CrossRef Search ADS Supplementary Data Supplementary data are available at JEEA online. © The Author(s) 2018. Published by Oxford University Press on behalf of European Economic Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of the European Economic Association Oxford University Press

The Limits of Propaganda: Evidence from Chavez’s Venezuela

Loading next page...
 
/lp/ou_press/the-limits-of-propaganda-evidence-from-chavez-s-venezuela-3bL7D7aVQy
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of European Economic Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1542-4766
eISSN
1542-4774
D.O.I.
10.1093/jeea/jvy012
Publisher site
See Article on Publisher Site

Abstract

Abstract We investigate viewer responses to ideological changes in television programming induced by cadenas, unannounced government propaganda in Venezuela. The drop-off in ratings during cadenas is concentrated among viewers of news programming on opposition channels, relative to progovernment channels. Also, the drop-off in ratings for moderate channels takes an intermediate value. The drop-off is stronger for viewers with access to cable channels, which do not air cadenas and experience an increase in viewership during cadenas. Structural estimation of our model allows us to quantify the degree to which opposition viewers limit their exposure to and ultimately the influence of propaganda via tuning out. 1. Introduction The media is often considered essential in the functioning of democracy via the provision of information to voters. At the same time, there is a temptation for incumbent governments to use media outlets to deliver political propaganda. This propaganda can be used by the government, among other ways, to promote its policies, increase its standing with the population in advance of elections, and to criticize opposition leaders and parties. If influential, propaganda may lead to moral hazard, via poor monitoring of incumbents by voters, and the re-election of low quality politicians and parties. Sophisticated consumers of information may respond to such propaganda in a variety of ways. One response involves consumers discounting biased information. For example, Chiang and Knight (2011) show that newspaper endorsements are more influential when they are less biased. That is, an endorsement of a Republican candidate by a liberal newspaper has more influence than an endorsement of a Republican by a conservative newspaper. Thus, discounting by readers limits the influence of biased information. Another response involves changing consumption patterns. For example, Durante and Knight (2012) document switching in viewership in response to a change in government in Italy and an associated change in the content of the main public television channel. With a preference for like-minded information, this switching is particularly relevant for consumers affiliated with the opposition. Note that switching will only be available in media sectors that are pluralistic, those offering a variety of ideological viewpoints. Given this, another possibility involves consumers simply “tuning out”, or consuming less information overall across all media outlets. Like switching, tuning out may be especially relevant for the opposition. Given this, tuning out may limit the ability of the government to convert the opposition to their cause via propaganda.1 In this paper, we investigate this second response, changing consumption patterns, using high-frequency television ratings data from the country of Venezuela. Hugo Chavez and his successor have routinely used cadenas, translated as “chains” and defined as government propaganda that is required to be aired live by all broadcast television channels. Thus, during a cadena, viewers watching television face the same programming on every broadcast channel. Importantly, these cadenas are not announced in advance to viewers, providing an experiment through which to examine short-run responses, in terms of changes in viewership, to government propaganda. In addition, cadenas were not required to be aired by cable channels during our sample period, allowing us to examine whether households with larger choice sets are more likely to switch to other outlets when faced with propaganda. Finally, broadcast channels in Venezuela during our sample period cover the political spectrum and can be naturally categorized as opposition, moderate, or progovernment. This allows us to examine whether switching and tuning out are more common among opposition viewers, who, as we document using survey data, are more likely to watch opposition news programming. To develop a set of testable hypotheses, we begin by building a simple model of consumer choice of television programming. In the model, there are two types of consumers, opposition and progovernment, both with a preference for like-minded information, two types of channels, opposition and government, and two types of programming, news and cadenas. We begin by assuming that both channels are required to air cadenas and thus initially focus on tuning out. The model predicts that, with positive switching costs and a preference for like-minded news, the drop-off in viewership when transitioning from news to cadena is more significant for the opposition channel than for the progovernment channel. This is due to the selection of opposition viewers into news programming on the opposition channel and the selection of progovernment viewers into news programming on the government channel. Introducing a third channel, which is moderate in nature, the model predicts that the drop-off in ratings when moving from news programming to cadenas should be most significant for the opposition channel, followed by the moderate channel, followed by the government channel. Finally, we consider an extension of the model to allow for switching via a cable channel, which is not required to air cadenas, and this extension provides two additional predictions. First, the model predicts that the drop-off in viewership on the private network, relative to the public network, should be more significant for households with access to cable, when compared to households without cable. Second, cable viewership, due to its role as an outside option, should be higher during the airing of cadenas on broadcast channels, relative to when cadenas are not aired on broadcast channels. We then test these predictions using data on television ratings from Venezuela. These data cover the years 2006 and 2007 and are high-frequency in nature (i.e., day-by-day and show-by-show). Consistent with the first prediction of the model, we find that the drop-off in viewership when transitioning from news programming to cadenas is more significant for the opposition channel than for the government channel, and these differences are both economically and statistically significant. Consistent with the second prediction of the model, we find that the drop-off in viewership for news programming on the moderate channel takes an intermediate value, between that of opposition channels and that of government channels. Next, focusing on the outside option, we find that, consistent with switching, cable viewership rises during cadenas and the drop-off in viewership is more significant for those with access to cable. Complementing this analysis, we also estimate the underlying structural parameters of the model; these include switching costs and the value of ideological information. This approach allows us to conduct counterfactual scenarios, which are not possible in the reduced form analysis. In particular, we consider a counterfactual in which cadenas are replaced with news programming, and we also consider a no switching counterfactual. One key finding is that opposition viewers significantly limit their exposure to propaganda by tuning out. In particular, although only 10% of opposition viewers are exposed to propaganda under the baseline, that figure rises to roughly 23% in absence of switching. Based upon this, we use persuasion rates from the literature to estimate the degree to which tuning out limits the persuasive impact of propaganda. Again, we find that switching matters for persuasion. In particular, although only 1% of opposition viewers are persuaded to support the government under the baseline, that figure rises to 3% in the absence of switching. Finally, we consider dynamic viewer responses and also provide a welfare comparison of government propaganda and media pluralism. This paper contributes to several literatures on media bias. A large literature measures the persuasive effects of the media, with DellaVigna and Gentzkow (2010) noting persuasion rates, defined as the probability that a viewer is persuaded by content, conditional on exposure, varying between 6% and 20%.2 Our study contributes to this literature by emphasizing the role of viewer responses in limiting exposure to biased information and thus ultimately limiting persuasion. On this issue, a theoretical literature, including Besley and Prat (2006) and Gehlbach and Sonin (2014), has emphasized the role of viewer responses in limiting the ability of the government to bias mass media in their favor. There is also a related empirical literature more focused on the role of government propaganda.3 Closest to our paper, several studies have documented a preference for like-minded news. These include Gentzkow and Shapiro (2010), Martin and Yurukoglu (2015), and Gentzkow, Shapiro, and Sinkinson (2014). The most closely related study in this literature is Durante and Knight (2012), who, as noted previously, investigate switching in viewership in Italy. There are two important differences between this study on Venezuela and Durante and Knight (2012). The first involves the frequency of responses. Although Durante and Knight (2012) study changes in the choice of media outlets over several years, this paper measures high-frequency, short-run changes in media consumption associated with a preference for like-minded news. Given inertia, it is possible that short-run responses are much smaller than long-run responses. This distinction is particularly relevant in countries with high frequency changes in government. The second important difference involves the structural analysis in this paper, allowing us to quantify the role of viewer responses in exposure to propaganda and in limiting ideological persuasion. More generally, our paper makes three contributions to the media economics literature. First, as noted previously, although most of the literature examines long-run responses, we focus on short-run, minute-by-minute responses. As a result, identification is cleaner than existing studies since viewer ideology is arguably fixed within a day. Second, this is one of the first studies in the literature with structural estimation. This analysis allows us to quantify the degree to which behavioral responses limit exposure to propaganda and ultimately limit political persuasion among the opposition. Third, studies in media economics have tended to focus on both developed countries and on democracies, with studies focused on developing countries and authoritarian regimes, such as Venezuela, more scarce. The paper proceeds as follows. Section 2 provides an overview of the key institutional details. Section 3 develops our key hypotheses in the context of a simple choice model. Section 4 describes the data, and Section 5 provides our results. Section 6 provides the structural estimates and counterfactual exercises. Finally, Section 7 offers a brief conclusion. 2. Institutional Context This section covers the political situation in Venezuela, with a focus on the role of television.4 In 1998, the leftist candidate Hugo Chavez won the presidential election in Venezuela, promising to lessen social exclusion, poverty and government corruption. Chavez was re-elected in 2000, 2006, and 2012 and served as President until his death in 2013. Since the beginning of Chavez’s time in office, the right-wing opposition was committed to removing him from power, with an attempted coup in April 2002, a national strike in December 2002, and recall referendum in 2004.5 During these confrontations, the private media sector tended to side with the opposition.6 Tensions between the private media and government were at their peak, with Chavez referring to the major private television channels (Venevision, Radio Caracas Television (RCTV), Globovision, and Televen) as the “four Horsemen of the apocalypse”, and, more generally, his language against the private media became very aggressive.7 In 2004, before the recall referendum, Chavez met with the owner of Venevision, leading to a warming in relations between the channel and President Chavez.8 Then, Televen followed the initiative to moderate their anti-Chavez tone around the same period.9 However, Globovision and RCTV, the oldest and largest television station, remained in opposition to the government. This partitioning of private channels into opposition (RCTV and Globovision) and moderate (Televen and Venevision) is consistent with media monitoring during the 2006 Presidential elections. In particular, EU-EOM (2006) document that RCTV and Globovision devoted a majority of their coverage to the opposition party, whereas Televen and Venevision devoted a majority of their coverage to Chavez’s party. Not surprisingly, the main public channel, Venezolana de Television (VTV), also devoted disproportionate coverage to Chavez’s party. Similar patterns were found with respect to the tone of the coverage, with positive coverage of the opposition and negative coverage of Chavez on RCTV and Globovision. Coverage of both Chavez and the opposition by Televen and Venevision, by contrast, was largely positive in nature. Finally, coverage of Chavez on the main public channel VTV was primarily positive, with decidedly negative coverage of the opposition. Given this evidence, we classify channels into three ideological categories: opposition (RCTV and Globovision), moderate (Televen and Venevision), and progovernment (VTV). In May 2007, the broadcasting license of RCTV expired and was not renewed by the government, and RCTV was replaced overnight by TVES, a government-run channel. The government’s rationale for closing RCTV had two key components: alleged violations of broadcast laws and their coverage of the coup and the strike in the oil sector. Later that year, during July 2007, RCTV re-emerged as a cable channel under the name RCTV International.10 In addition to not renewing the broadcast license of RCTV, Chavez attempted to influence the media via government channels and cadenas, government programming that must be aired live by noncable (i.e., broadcast) channels.11 Bisbal (2009) estimates that 1,731 cadenas were broadcast between 1999 and June 2008, totaling over 1,000 h. According to Kitzberger (2010) and Reporters Without Borders (2003), cadenas are used by Chavez to mobilize supporters, criticize and threaten adversaries, and more generally, for political campaigning. Two aspects of cadenas are particularly useful for our identification strategy. First, cadenas are not announced in advance to stations or viewers and, as we argue in what follows, are difficult to forecast in terms of either their starting time or duration. Given this, we assume that viewers do not anticipate cadenas when choosing whether or not to watch news programming. Second, all broadcast channels must air cadenas and thus, in the absence of a cable subscription, every available channel carries the cadena. Moreover, our understanding is that, due to the high volume of cadenas, viewers are aware of this and do not attempt to change channels when a cadena comes on the air. This is useful for our empirical strategy since we can infer the fraction of viewers of each news program who watch the cadena via the change in viewership on a given channel when transitioning from news to cadena. 3. Theoretical Model This section develops a simple theoretical model to provide a set of hypotheses for the empirical analysis of ratings data. In addition, the model provides a framework for the structural analysis to follow. We begin with the simple case of only two types of viewers (opposition and progovernment), two channels (opposition and government), and two types of programming (news and cadenas). In extensions of the model, we then introduce a third channel, which is moderate in nature, and then separately consider how the results differ with the presence of a cable channel that is not required to air cadenas. 3.1. Baseline Case Viewers, indexed by $$v$$, are of two types: progovernment (g) and opposition (o). Let the fraction of each type in the population be given by πg and πo = 1 − πg, respectively. News stations, indexed by i, are also of two types: government (g) or opposition (o). Each outlet offers news programming (p = n), and both outlets are also required to carry cadenas (p = c). Viewers differ in the degree to which they value news programming. For progovernment types, the value of government news is θs and the value of opposition news is θd, where we assume that viewers prefer same-ideology news over different-ideology news (i.e., θd < θs). For opposition types, by contrast, the value of government news is θd and the value of opposition news is θs. Cadenas are assumed to have progovernment content and thus provide payoffs of θd to opposition types and θs to progovernment types. Then, letting u$${vip}$$ ∈ {θd, θs} represent these systematic payoffs, viewer $$v$$ receives the following overall payoff from watching programming p on station i: \begin{equation*} U_{vip}=u_{vip}+\epsilon _{vip} \end{equation*} where ε$${vip}$$ represents unobserved preferences and is assumed to be distributed type-1 extreme value.12 These unobserved preferences are assumed to be independently distributed, both across viewers and across channels. We consider a scenario in which both stations are airing news, and viewers have three options: (1) watching the government station, (2) watching the opposition station, and (3) watching neither (which yields a systematic payoff of zero). Then, letting σ$${in}$$ be the market share on channel i when both channels are airing news programming, we have the following market shares on the government channel: \begin{equation*} \sigma _{gn}=\pi _{g}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}+\pi _{o}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}. \end{equation*} The first term is the product of the fraction of progovernment viewers and the market share within this group, and the second term is the product of the fraction of opposition viewers and the market share within this group. Thus, the overall market share for the government channel is a weighted average of market shares for progovernment and opposition viewers. Likewise, the market share on the opposition channel is given by \begin{equation*} \sigma _{on}=\pi _{g}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}+\pi _{o}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}. \end{equation*} Now, suppose that the government airs a cadena and that this is not anticipated by viewers. That is, as discussed previously, viewers do not account for the possibility of a cadena when choosing whether or not to watch news. Further, for simplicity, assume that viewers who are not watching news (the third option described previously) do not come back to watch the cadena on either of the two channels. Also, assume a switching cost of η > 0 so that viewers will not change the channel when the cadena comes on the air. As discussed previously, all broadcast channels must air cadenas and thus there is no incentive for viewers to change channels when a cadena comes on the air. Instead the only margin involves whether or not to watch the cadena. More formally, this simply requires positive switching costs (η > 0) and that unobserved preferences over programming (ε$${vi\!p}$$) are constant across channels when a cadena comes on the air.13 Then, let the fraction of progovernment viewers who choose to watch the cadena, conditional on watching the news on that channel, be given by \begin{equation*} p_{g}=\exp (\theta _{s})[1+\exp (\theta _{s})]^{-1}, \end{equation*} and the analogous fraction for opposition viewers is given by \begin{equation*} p_{o}=\exp (\theta _{d})[1+\exp (\theta _{d})]^{-1},\text{where}p_{o}<p_{d} \text{since} \theta _{d}<\theta _{s}. \end{equation*} Then, we have that market shares for cadenas on the two stations are given by \begin{equation*} \sigma _{gc}=\pi _{g}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{g}+\pi _{o}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{o}, \end{equation*} \begin{equation*} \sigma _{oc}=\pi _{g}\frac{\exp (\theta _{d})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{g}+\pi _{o}\frac{\exp (\theta _{s})}{1+\exp (\theta _{s})+\exp (\theta _{d})}p_{o}. \end{equation*} Then, define the drop-off in viewership moving from news to cadena, for government and opposition channels, respectively, as \begin{equation*} \Delta ^{o}=\ln \left[\sigma _{oc}/\sigma _{on}\right] \quad \text{and} \quad \Delta ^{g}=\ln \left[\sigma _{gc}/\sigma _{gn}\right]. \end{equation*} Given the log transformation, these measures can be interpreted as the percentage reduction in viewership on a given channel when moving from news programming to cadenas. We first compare the drop-off in viewership on opposition and government channels in the following proposition. Proposition 1. With positive switching costs (η > 0) and a preference for like-minded news (θd < θs), the drop-off in viewership moving from news to cadena is more significant for the opposition channel than for the government channel. That is, Δo < Δg. We provide proofs of all propositions in the Online Appendix. The intuition for this proposition is simply that opposition viewers, relative to progovernment viewers, are more likely to watch opposition news, relative to government news. Moreover, these opposition viewers also have a distaste for the content of the cadena, relative to progovernment viewers. Given all of this, viewers of opposition news are more likely to tune out when a cadena comes on the air. 3.2. Moderate Channel Extension We next extend the model to allow for a third channel, which is assumed to air moderate news. For simplicity, assume that both opposition and progovernment viewers get a payoff of θm from watching news programming on this channel, with θd < θm < θs. Then, again comparing the drop-off in viewership across the channels, we have the following proposition. Proposition 2. With positive switching costs (η > 0) and a preference for like-minded news (θd < θm < θs), we have that the drop-off in viewership for the moderate channel lies in between the opposition and the government channel. That is, Δo < Δm < Δg. The intuition for this proposition is simply that the moderate channel attracts a less polarized audience for its news programming, whereas the opposition channel disproportionately attracts opposition viewers and the government channel disproportionately attracts progovernment viewers. Thus, the drop-off in viewership for the moderate channel takes an intermediate value, when compared to the government and opposition channels. 3.3. Cable Extension To investigate the possibility of switching to other outlets in a pluralistic media environment, we return to the baseline model of two broadcast channels but now allow for a cable channel, which is assumed to be linked to the opposition, and, as discussed previously, is not required to air cadenas. In the context of this extension, we investigate two questions. First, due to the presence of this new opposition channel, is the drop-off in viewership, when moving from opposition news to cadena, more significant for those viewers with cable than for those viewers without cable? Second, consistent with switching, does cable viewership increase during cadenas? Given the empirical application to the cable channel RCTV International, we assume here that cable also has opposition news, yielding a payoff of θd to progovernment viewers and θs to opposition types. Now, suppose that the government unexpectedly decides to air a cadena. As stated previously, assume that viewers who are not watching do not come back to watch the cadena. Also, as stated previously, assume a switching cost of η > 0 so that viewers will not change the channel when the cadena airs. Finally, for simplicity, we assume that viewers do not switch from cable to either the opposition or the government channel when the cadena comes on the air. They can switch from one of the broadcast stations to cable but must incur the switching cost. Then, we have the following result with respect to the drop-off measures considered previously. Proposition 3. With positive switching costs (η > 0) and a preference for like-minded news (θd < θs), the drop-off in viewership on the opposition channel, relative to the government channel, for viewers with cable is larger than for viewers without cable. That is, Δo − Δg falls when cable is introduced. The intuition for Proposition 3 is that, in addition to turning off the television, opposition viewers with access to cable now have another attractive outside option, switching to watch opposition news on cable during the cadena. Given this, even fewer viewers of opposition news will watch the cadena. Finally, we consider how viewership of cable changes when a cadena comes on broadcast television, and we have the following result. Proposition 4. With positive switching costs (η > 0), a preference for like-minded news (θd < θs), and a cable option, viewership of cable rises during the cadena. The logic behind Proposition 4 is straightforward. Since opposition viewers value cable as an outside option, viewership of cable programs rises during cadenas. To summarize, the theoretical model makes four predictions. First, the drop-off in viewership when moving from news to cadenas should be more significant on private channels, when compared to the government channel. Second, the drop-off in viewership on moderate channels should take an intermediate value, between the opposition channel and the government channel. Third, the drop-off in viewership for the opposition channel, relative to the government channel, should be more significant for those with access to cable. Fourth, cable viewership should rise during cadenas. 4. Data Our data on television ratings were purchased from AGB Nielsen Media Research Venezuela and include broadcast ratings of each television show aired on each channel, from January 1, 2006 to December 31, 2007. During the part of the analysis focused on ratings of broadcast channels, we focus on data from the period prior to the closing of RCTV in May 2007 in order to have a consistent set of channels. Ratings are provided separately for the four largest metropolitan areas (Caracas, Barquisimeto, Maracaibo, and Valencia). In constructing our measure of ratings for each show we use the average minute rating (AMR) measure, and, given their very low ratings, ignore shows aired between midnight and 6 a.m.14 In addition to analyzing aggregate ratings for each show, channel, and metropolitan area, we also test Proposition 3 by employing measures of ratings separately for those with and without cable subscriptions. Likewise, our structural analysis uses gender-specific ratings.15 Our analysis considers the most significant channels, those discussed in Section 2. In particular, and as shown in Table 1, we focus on four private broadcast channels, Globovision, Televen, RCTV, and Venevision, one public channel, VTV, and one cable channel, RCTV International. As described in Section 2, television in Venezuela during the sample period is considered to be highly polarized. This political polarization allows us to create three categories for the channels based upon their ideology, as discussed previously. Although the main public channel (VTV) is assumed to be progovernment, private channels are split into opposition (RCTV and Globovision) and moderate (Venevision and Televen). Table 1. Channels analyzed. Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 View Large Table 1. Channels analyzed. Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 Name Ideology Coverage Period RCTV Opposition National Until May 27, 2007 Venevision Moderate National Whole period Televen Moderate National Whole period Globovision Opposition Caracas and Whole period Valencia VTV Government National Whole period RCTV International Opposition Cable Starting July 16, 2007 View Large Of course, there may be other differences across channels, and Table 2 provides some evidence on the types of programming offered during our sample period.16 As shown, three of the private channels disproportionately air entertainment programming and one of the private channels, Globovision, primarily offers news programming. The public channel VTV also offers more news programming than entertainment programming.17 As noted previously, key to our identification strategy is the assumption that viewers are not aware of cadenas in advance. The law does not require the government to preannounce cadenas, and our understanding is that cadenas are not preannounced in practice. Nonetheless, it is still possible that viewers can predict the airing of cadenas to the extent that they follow regular patterns. We investigate this issue by analyzing the distribution of cadenas across days, their starting time, and their duration. As shown in Figure 1, although cadenas are most commonly aired on Wednesdays, followed by Tuesdays, Thursdays, and Fridays, cadenas may appear on any day of the week, and there is not a noticeable spike on any particular day. Likewise, as shown in Figure 2, although cadenas are most commonly aired during prime time (i.e., between 7 p.m. and 10 p.m.), cadenas can occur at nearly any hour. In addition, as shown in Figure 3, although many cadenas start at the top of the hour, they can also begin at any minute within the hour. Finally, the duration of cadenas is difficult to predict. As shown in Figure 4, cadenas can be either very short in duration, less than 30 min, or very long in duration, in excess of four or even 5 h. To summarize, there is not a specific pattern in terms of the timing of cadenas, and there is thus an important element of surprise for the viewer, who can be exposed to these interruptions by the government at any time, without anticipating the day, the hour, the minute, or the length of the interruption. Figure 1. View largeDownload slide Day of the week of cadenas. Figure 1. View largeDownload slide Day of the week of cadenas. Figure 2. View largeDownload slide Starting hour of cadenas. Figure 2. View largeDownload slide Starting hour of cadenas. Figure 3. View largeDownload slide Starting minute of cadenas. Figure 3. View largeDownload slide Starting minute of cadenas. Figure 4. View largeDownload slide Duration in minutes of cadenas. Figure 4. View largeDownload slide Duration in minutes of cadenas. A key mechanism in our model is a preference for like-minded news, implying that opposition viewers are more likely to watch opposition news and that progovernment viewers are more likely to watch news on public channels. Unfortunately, our ratings data do not have any measures of viewer ideology. Instead, to examine this issue, we have analyzed separate data from the Latin American Public Opinion Project (LAPOP) Survey, conducted during 2007 for Venezuela. The survey includes questions about political preferences and media consumption for a total of 1,510 Venezuelan citizens. In particular, LAPOP asks respondents which candidate they voted for in the last election and the channel they watch most often for news. For the purposes of this analysis, we group the channels into opposition (RCTV and Globovision), moderate (TVES and Venevision), and public (VTV). As shown in Figure 5, respondents who voted for Chavez are most likely to watch moderate channels, followed by the public channel. They are unlikely to watch opposition channels RCTV and Globovision. For respondents who voted for the opposition, by contrast, the patterns are reversed. In particular, and, as shown in Figure 6, these respondents have a very low propensity of watching the public channel, and a majority report watching news on either RCTV or Globovision. To summarize, Chavez supporters are roughly 10 times more likely than opposition supporters to watch VTV, and opposition supporters are roughly three times more likely than Chavez supporters to watch opposition channels. This provides support for our maintained assumption of a preference for like-minded news.18 Figure 5. View largeDownload slide Favorite news channels for Chavez supporters. Figure 5. View largeDownload slide Favorite news channels for Chavez supporters. Figure 6. View largeDownload slide Favorite news channels for the opposition. Figure 6. View largeDownload slide Favorite news channels for the opposition. 5. Analysis of Ratings data In this section, we test the key hypotheses of the theoretical model in an investigation of viewer responses to political propaganda via cadenas in Venezuela during 2006 and 2007, a key period during Chavez’s time in office. Before turning to the regression results, we present summary statistics. In particular, Table 3 provides, for each channel, the average percent change in rating for transitions between the three types of programming: news (N), cadenas (C), and entertainment (E). Standard deviations are provided in parentheses. There are several notable findings here. As shown in the first column, opposition channels exhibit a reduction of viewership when a cadena interrupts a news program, with no change for moderate channels and an increase on the government channel. As shown in the third column, reverse transitions, those from news to cadena exhibit opposite patterns in viewership, with an increase on opposition channels and a decrease in viewership on public channels. Transitions from entertainment to cadena (column (2)) lead to only small changes in viewership on private channels but significant increases in viewership on public channels. Finally, as shown in the fifth column, transitions from news to entertainment exhibit sizable increases in viewership on private channels and significant reductions in viewership on the public channel. We next examine these relationships more formally in a regression analysis. Table 2. Content by channel. Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 View Large Table 2. Content by channel. Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 Name Percent entertainment Percent news RCTV 0.742 0.237 Venevision 0.731 0.252 Televen 0.769 0.217 Globovision 0.082 0.906 VTV 0.227 0.756 View Large Table 3. Descriptive statistics: Log change in ratings. Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Notes: Measures represent the mean log change in rating for news (N), cadena (C), and entertainment (E). Standard deviation in parentheses. View Large Table 3. Descriptive statistics: Log change in ratings. Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Transitions from N to C E to C C to N C to E N to E E to N 1. Private −0.2506 −0.0518 0.0788 0.1737 0.2130 0.0382 (0.8636) (0.5884) (1.0278) (0.6758) (0.9285) (1.0674) 1.1 Opposition −0.3454 −0.0307 0.1534 0.1472 0.1759 0.2049 (0.8841) (0.5357) (0.9763) (0.5382) (0.8203) (1.0410) Globovision −0.3953 −0.2099 0.2111 −0.5603 −0.1182 0.1793 (0.9275) (1.5479) (0.9637) (0.5345) (0.9759) (1.0959) RCTV −0.1648 −0.0278 −0.2186 0.1499 0.2848 0.2117 (0.6817) (0.5067) (0.9886) (0.5370) (0.7250) (1.0261) 1.2 Moderate 0.0091 −0.0625 −0.2384 0.1870 0.2324 −0.0563 (0.7493) (0.6136) (1.1797) (0.7354) (0.9797) (1.0706) Televen −0.0752 −0.0906 −0.1503 0.2438 0.1665 −0.1749 (0.7928) (0.7255) (1.4786 ) (0.8932) (1.2184) (1.2268) Venevision 0.0825 −0.0346 −0.2926 0.1310 0.2817 0.0214 (0.7087) (0.4765) (0.9692) (0.5311) (0.7489) (0.9467) 2. Public (VTV) 0.1949 0.3853 −0.1495 −0.2595 −0.2390 0.1991 (1.0430) (0.9837) (1.0568) (0.9479) (1.2929) (1.1732) Notes: Measures represent the mean log change in rating for news (N), cadena (C), and entertainment (E). Standard deviation in parentheses. View Large Table 4. Log change in ratings: News to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 4. Log change in ratings: News to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.4456*** (0.0672) Opposition −0.5403*** (0.0732) Moderate −0.1858*** (0.0898) Globovision −0.5902*** (0.0807) RCTV −0.3598*** (0.101) Televen −0.2701*** (0.1254) Venevision −0.1125*** (0.1083) Constant 0.1950*** 0.1950*** 0.1950*** 0.0504 (0.0504) (0.0505) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 5. Log change in ratings: News to cadena with fixed effects. Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 ***p < 0.01. View Large Table 5. Log change in ratings: News to cadena with fixed effects. Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.5189*** −0.4579*** −0.5400*** (0.0755) (0.0693) (0.0812) Fixed effects Hour Day of week Hour by day of week Observations 807 807 807 ***p < 0.01. View Large Table 6. Log change in ratings: Heterogeneity. Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *** p < 0.01. View Large Table 6. Log change in ratings: Heterogeneity. Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Variable Change in ratings Change in ratings Change in ratings Private −0.6952*** −0.4172*** −0.4400*** (0.1277) (0.0734) (0.0734) Prime −0.5359*** (0.1241) Weekend 0.1871 (0.1482) Long cadena 0.3012 (0.1711) Private × Prime 0.2325 (0.1522) Private × Weekend −0.1961 (0.1831) Private × Long Cadena −0.1638 (0.1991) Constant 0.5722*** 0.1679*** 0.1570*** (0.1133) (0.0540) (0.0524) Observations 807 807 807 Notes: The dependent variable is the log change in ratings when transitioning from a news program to a cadena. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *** p < 0.01. View Large Table 7. Log change in ratings: Cadena to news. Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Notes: The dependent variable is the log change in ratings when transitioning from a cadena to a news program. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 7. Log change in ratings: Cadena to news. Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Variable Change in ratings Change in ratings Change in ratings Private 0.2283*** (0.0695) Opposition 0.3029*** (0.0721) Moderate −0.0889 (0.1531) Globovision 0.3606*** (0.0751) RCTV −0.0691 (0.1678) Televen −0.0008 (0.2989) Venevision −0.1431 (0.1588) Constant −0.1495*** −0.1495*** −0.1495*** (0.0404) (0.0405) (0.0405) Observations 1,014 1,014 1,014 Notes: The dependent variable is the log change in ratings when transitioning from a cadena to a news program. Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. ***p < 0.01. View Large Table 8. Log change in ratings: Entertainment to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Notes: Dependent variable is the log change in ratings when transitioning from entertainment to cadena. Public channel VTV is the base outcome. Robust standard errors in brackets. **p < 0.05; ***p < 0.01. View Large Table 8. Log change in ratings: Entertainment to cadena. Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Variable Change in ratings Change in ratings Change in ratings Private −0.4371*** (0.1604) Opposition −0.4160*** (0.1615) Moderate −0.4479*** (0.1609) Globovision −0.5953 (0.5371) RCTV −0.4131** (0.1614) Televen −0.4760*** (0.1632) Venevision −0.4200*** (0.1612) Constant 0.3853** 0.3853** 0.3853** (0.1596) (0.1597) (0.1598) Observations 1,505 1,505 1,505 Notes: Dependent variable is the log change in ratings when transitioning from entertainment to cadena. Public channel VTV is the base outcome. Robust standard errors in brackets. **p < 0.05; ***p < 0.01. View Large Table 9. Log change in ratings: News to entertainment. Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Notes: Dependent variable is the log change in ratings when transitioning from news to entertainment. Public channel VTV is the base outcome. Robust standard errors in brackets. ***p < 0.01. View Large Table 9. Log change in ratings: News to entertainment. Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Variable Change in ratings Change in ratings Change in ratings Private 0.4519*** (0.0228) Opposition 0.4148*** (0.0244) Moderate 0.4713*** (0.0237) Globovision 0.1208*** (0.0345) RCTV 0.5238*** (0.0246) Televen 0.4055*** (0.0289) Venevision 0.5207*** (0.0238) Constant −0.2390*** −0.2390*** −0.2390*** (0.0214) (0.0214) (0.0214) Observations 17,721 17,721 17,721 Notes: Dependent variable is the log change in ratings when transitioning from news to entertainment. Public channel VTV is the base outcome. Robust standard errors in brackets. ***p < 0.01. View Large Table 10. Drop-off for cable versus no cable. Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Notes: Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *p < 0.1; **p < 0.05. View Large Table 10. Drop-off for cable versus no cable. Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Variable Difference between Difference between Difference between cable and no cable cable and no cable cable and no cable Private −0.2001* (0.1099) Opposition −0.1385 (0.1191) Moderate −0.3811** (0.1708) Globovision −0.1974 (0.1345) RCTV 0.0409 (0.1660) Televen −0.4493* (0.2357) Venevision −0.3280 (0.2221) Constant −0.1218 −0.1218 −0.1218 (0.0775) (0.0776) (0.0777) Observations 632 632 632 Notes: Public channel VTV is the base outcome for all columns. Robust standard errors in brackets. *p < 0.1; **p < 0.05. View Large Table 11. Cable channel RCTV international. Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Notes: All results for the cable channel RCTV International when a cadena is aired on the broadcast channels. Robust standard errors in brackets. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 11. Cable channel RCTV international. Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Variable Change in ratings Change in ratings Change in cadena overlap 0.6882*** 0.6087*** (0.0945) (0.0986) News −0.0126 (0.0211) News × Change in cadena overlap 1.1047*** (0.3188) Constant 0.0206** 0.0259* (0.0105) (0.0137) Observations 9,404 9,404 Notes: All results for the cable channel RCTV International when a cadena is aired on the broadcast channels. Robust standard errors in brackets. *p < 0.1; **p < 0.05; ***p < 0.01. View Large Table 12. Cadenas content. Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Notes: Robust standard errors in parentheses; **p < 0.05; ***p < 0.01. View Large Table 12. Cadenas content. Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Variables Change in ratings Foreign × Private −0.4652*** (0.1391) Delivery × Private −1.0560*** (0.2609) Elections × Private −0.7583*** (0.2647) Celebrations × Private −0.5436*** (0.1626) Information ×Private −0.2638** (0.1109) Others × Private −0.4817 (0.4725) Observations 807 Notes: Robust standard errors in parentheses; **p < 0.05; ***p < 0.01. View Large Table 13. Summary of payoff structure. $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ View Large Table 13. Summary of payoff structure. $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$v$$ = o, k = m $$v$$ = o, k = f $$v$$ = g, k = m $$v$$ = g, k = f p = n, i = g θd θd θs θs p = n, i = m θm θm θm θm p = n, i = o θs θs θd θd p = c θd θd θs θs $$p=\mathit {sports}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$\mathit {sports}_{m}$$ $$\mathit {sports}_{f}$$ $$p=\mathit {soaps}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$\mathit {soaps}_{m}$$ $$\mathit {soaps}_{f}$$ $$p=\mathit {other}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ $$\mathit {other}_{m}$$ $$\mathit {other}_{f}$$ View Large Table 14. Structural estimates. Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Notes: Standard errors in brackets. ***p < 0.01. View Large Table 14. Structural estimates. Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Variable Information −0.4812*** (0.0072) Information × same 0.2409*** (0.0078) Information × different −1.6717*** (0.0078) Female 0.1759*** (0.0046) Soaps 1.4439*** (0.0091) Sports −0.2854*** (0.0194) Female × soaps 0.5584*** (0.0122) Female × sports −0.1839*** (0.0268) Switching cost 3.8089*** (0.0100) Constant −2.5179*** (0.0059) Observations 310,812 Notes: Standard errors in brackets. ***p < 0.01. View Large 5.1. Drop-Off: News to Cadena Our econometric analysis begins with an investigation of how ratings change when a cadena interrupts news programming depending upon the political orientation of the station, under the assumption that viewers prefer to watch like-minded news. Given, as shown previously, that opposition viewers have a higher probability of watching opposition news channels, and, under the assumption that opposition viewers dislike cadenas, we expect viewers of opposition news to be more likely to tune out when cadenas are aired on television, relative to viewers of progovernment news. As argued previously, we hypothesize that viewers watching opposition news will respond more strongly to cadenas when compared to viewers watching news programming on government channels. To test this hypothesis, we estimate the following econometric model of viewer responses to cadenas: \begin{equation} \Delta ^{ic}=\ln \left[\frac{s_{ic}}{s_{in}}\right]=\beta _{i}+\epsilon _{ic}, \end{equation} (1) where s$${ic}$$ represents the measured rating for a cadena aired on channel i and s$${in}$$ is the ratings for the news program that aired just before cadena c on channel i.19 That is, consistent with the theoretical predictions, the drop-off in viewership is measured as the log change in ratings between cadenas and the previous news program for each cadena aired between January 2006 and May 2007.20 On the right-hand side, βi is a channel-specific constant. To test Proposition 1, we use a dummy variable that takes the value of 1 for a private channel and the value of 0 for a public channel. To test the second proposition, we employ a set of dummy variables based on political ideology of the station (i.e., opposition, moderate, and public). Then, we estimate a more flexible specification that uses a separate dummy variable for each channel. Finally, ε$${ic}$$ represents the unobserved determinants of the drop-off in ratings on channel i during cadena c. We begin with a simple comparison of private and public channels, where public channels are the omitted category. Thus, the results are interpreted as reflecting drop-off for the private channel relative to the public channel. As shown in the first column of Table 4, the coefficient on private channels is negative and statistically significant. That is, airing cadenas after news programming on private channels, relative to the public channel, is associated with more viewers tuning out. This provides support for Proposition 1, which predicted that the drop-off in viewership should be more significant for private channels than for public channels. Moreover, the magnitudes of these effects are large, with the drop-off for private channels 45 percentage points larger than the drop-off for public channels. As a robustness check, we next include time fixed effects in order to control for the timing of cadenas. As shown in Table 5, when including fixed effects for the starting hour of cadenas, the results are somewhat stronger, with a 52% reduction in viewership on private channels, relative to public. We next include day of week fixed effects, and, as shown in column (2), the results are again similar to the baseline results. Finally, we include starting hour by day of the week fixed effects. As shown in the final column, there is a 54% reduction in viewership on private, relative to public, in this case. Thus, these baseline results are robust to the inclusion of time fixed effects. Returning to Table 4, we next allow for two separate categories of private channels (opposition and moderate), with public again the omitted category. The coefficients are also large in magnitude and statistically significant for the two categories, opposition and moderate, relative to the public channel. The coefficients in the second column demonstrate that viewers of news on the opposition and moderate channels, relative to viewers of the public channel, are more likely to turn off the television when a cadena is aired. That is, consistent with Proposition 2, which predicted that the drop-off for moderate channels should take an intermediate value, the change in viewership for moderate channels is 19 percentage points higher than the public channel but is 35 percentage points lower than the opposition channels.21 Finally, in the third column of Table 4, we present the results separately for each channel, where the effects should again be considered relative to the public channel VTV. As shown, and for all channels, we find a statistically significant reduction of viewership, relative to the change in viewership of VTV, when a cadena is aired. Consistent with the results in the second column, the effect of switching to an outside option is most significant for Globovision and RCTV, the most extreme channels in terms of opposition to the government. We next explore heterogeneity in these viewer responses along three dimensions. First, we examine whether responses differ during prime time hours, when viewership tends to be higher. As shown in the first column of Table 6, we do find that dropoff is larger during prime time but the interaction between prime time and private is statistically insignificant. Second, we explore whether responses differ during the weekends, when viewership tends to be lower. As shown in the second column, the interaction between weekend and private is again statistically insignificant. Finally, we examine whether responses differ between short and long cadenas, defined as those in excess of 2 h. As shown in the third column, responses are stronger on private channels during long cadenas but these differences are not statistically significant. Overall, these results are consistent with Propositions 1 and 2, which predict that viewers of news on private channels are more likely to turn off the television during cadenas and that the drop-off on the moderate channels during cadenas lies between the opposition channels and the public channel. This behavioral response of shifting to an outside option associated with unanticipated exposure to ideological content that is not like-minded in nature suggests that the impact of political propaganda may be limited. The results are in line with theories of television program choice, which predict that people select television content in order to satisfy their preferences (Youn 1994; Durante and Knight 2012; Yao, Wang, and Chen 2017), whereas, at the same time, suggesting that inertia in television viewership is incomplete (see Moshkin and Shachar 2002; Goettler and Shachar 2001; Perretti and Esteves-Sorenson 2012). 5.2. Other Transitions For comparison purposes, we next extend the analysis to investigate the effect of the reverse experiment: transitioning from cadenas to news programs. Although the formal model did not consider this possibility, it is natural to conjecture that the results should go in the opposite direction, with viewership of news rising on private, relative to public, following a cadena. As shown in Table 7, the coefficient in the first column is positive and statistically significant, documenting that private channels do experience an increase in viewership of 23%, relative to the public channel, when cadenas are followed by a news program. As shown in columns (2) and (3), the effect is driven by opposition channels, especially Globovision, which is the only channel that has a statistically significant coefficient, re-enforcing the idea that viewers of the opposition channel search for ideological content similar to their own ideology. Overall, these results are consistent with notion that viewers have preferences for watching like-minded political content. For comparison purposes, in Table 8, we examine the drop-off in rating when entertainment programs are interrupted by a cadena. We again find similar results to those in the analysis of a change in content from news to cadenas. Nevertheless, as shown in column (2), the results are similar for opposition and moderate channels, and, as shown in column (3), the results are economically significant for all four private channels. Finally, we analyze the change in viewership when moving from news to an entertainment program. Note that, although both involve transitions away from news, transitions from news to entertainment are not directly comparable to transitions from news to cadenas since the former are anticipated and the latter are not. As shown in Table 9, we find that private channels, relative to the public channel, generate a statistically significant 45% increase in ratings when moving from a news program to an entertainment program. This is similar in magnitude to the result for the drop-off when moving from news to cadenas, suggesting that our results may be about viewership of news on different channels per se rather than political ideology. On the other hand, it is not clear that entertainment programming on public channels is comparable to entertainment programming on private channels, which is very popular in Venezuela. Moreover, as shown in columns (2) and (3), the effects are again similar for opposition and moderate channels. The similarity of these results for entertainment across these private channels of differing ideology suggests that our baseline results are driven, at least in part, by channel ideology, rather than other characteristics of news programming on different channels. Taken together, the results for these other transitions suggest that our baseline results relating to channel ideology are not driven by other channel-specific characteristics. 5.3. Cable Television We next consider Propositions 3 and 4 in the context of cable channels, which were not required to broadcast cadenas. Given this, Proposition 3 predicts that the disproportionate drop-off in viewership on the private channel, relative to the public channel, should be more significant for households with cable subscriptions, relative to households without cable subscriptions. Likewise, Proposition 4 predicts that viewership of cable should rise during cadenas, and we test this prediction using data from RCTV International, which began as a cable channel during July 2007. In terms of Proposition 3, we begin by estimating the following regression: \begin{equation} \Delta ^{ic}(\mathit {cable})-\Delta ^{ic}(\mathit {nocable})=\beta _{i}+\epsilon _{ic}, \end{equation} (2) where the drop-off in viewership is now measured separately for cable and noncable households, and, according to Proposition 3, the coefficient for private channels, relative to public channels, should be negative. As shown in Table 10, and consistent with Proposition 3, the drop-off in ratings for those with cable, relative to households without cable, is indeed more significant for private channels, relative to public channels. In columns (2) and (3), we break out this effect by type of channel, finding that the effect is somewhat larger and only statistically significant for moderate channels and is driven in large part by Televen. Taken together, these results demonstrate that the opposition may be exposed to political propaganda to an even lesser degree when a source of opposition programming remains available during cadenas. This implies that viewers who are not able to afford cable, especially those already inclined to support the government, are disproportionately exposed to propaganda. Using ratings data from RCTV International, a cable channel created following the closing of RCTV on broadcast television, we next test Proposition 4, which predicts that RCTV cable ratings should rise during cadenas as viewers use this channel as a source of opposition programming. In particular, we estimate the following regression specification: \begin{equation} \Delta ^{p}(\mathit {RCTV}\,\mathit {cable)}=\beta _{1}\mathit {Change}\,\mathit {in}\,\mathit {cadena}\,\mathit {overlap}^{p}+\epsilon ^{p}, \end{equation} (3) where the left-hand side variable is the percentage change in ratings for program p airing on RCTV International, when compared to the previous program aired on RCTV International. To compute the key right-hand-side variable, we first compute cadena overlap for each RCTV cable show. Cadena overlap is defined as the fraction of minutes for which the RCTV cable show overlapped with a cadena. Thus, cadena overlap varies between zero and one, where the former value is attained if there is no cadena aired at any point of the show, and the latter value is attained if the show overlaps entirely with a cadena. Taking first differences of cadena overlap, we then compute the change in cadena overlap, which ranges in value from negative one to plus one. For this analysis, we use the sample from July 2007 to December 2007, the period in which RCTV is aired on cable. As shown in Table 11, and consistent with Proposition 4, we do find that RCTV cable ratings do rise during cadenas, and the effect is both economically and statistically significant. In particular, considering moving from no overlap to complete overlap (i.e. change in cadena overlap equal to one), we have that ratings on RCTV cable rise by an economically significant 69%. In the second column, we investigate whether these results differ according to the type of programming on RCTV cable. As shown, the results are larger for news programming on RCTV cable, when compared to other types of programming on RCTV cable. More concretely, viewership of RCTV cable news programming increases by 171% when a cadena comes on broadcast television, whereas viewership of non-news programming increases by only 61%. These results provide further support for our hypothesis of viewer choice of like-minded ideological content. 5.4. Content of Cadenas We next use more detailed information about individual cadenas, as provided by Nielsen in the form of short descriptions of the content of each cadena. Using this description and supplementing this with information found online, we create five categories of cadena content, and these are described in what follows. 5.4.1. Foreign Relations Coverage of foreign policy accomplishments, such as visits of presidents, multilateral agreements, and international travel by Chavez. 5.4.2. Delivery Coverage of events involving government promises of the provision of public goods, services, etc. 5.4.3. Elections Vroadcasts focusing on elections, especially coverage of the 2006 Presidential elections and the 2007 constitutional referendum. 5.4.4. Celebrations Coverage of public events, such as the birth of Simon Bolivar, marches, etc. 5.4.5. Information Summary of the progress of the country in several areas, such as economic and political. For cadenas that do not meet one of these definitions, we create a sixth category, other. Table 12 examines the drop-off in rating, separately, for each of these categories on the private channels, compared to the same categories in the public channels, when transitioning from news to cadena. This specification is consistent with the baseline analysis in column (1) of Table 4. The regression also controls for main effects of these categories, not reported in the table. Comparing the magnitude of the coefficients on the interactions, we have that the largest drop-off on private, relative to public, occurs for the categories delivery and elections. The large drop-off for delivery cadenas aired on private television may reflect the fact that many of these broadcasts involve Chavez himself delivering promises of public goods and services to his core voters. Given the targeting of these goods and services, there may be a particular distaste among opposition viewers for these cadenas. Likewise, cadenas about elections are, by their very nature, politically oriented and may have created polarized responses in terms of viewership. Finally, the smaller coefficient on the information category may reflect the fact that both opposition and progovernment viewers find these transmissions to be truly informative about the state of the economy or along other dimensions. 5.5. Summary To summarize, the results of the empirical analysis are consistent with the four key predictions of the model. First, the drop-off in ratings is more substantial for private channels, when compared to the public channel. Second, this effect is concentrated among opposition channels, and results for the moderate channels take an intermediate value. Third, the drop-off in viewership for the private channel is more significant for households with a cable subscription. Fourth, viewership of RCTV International, an opposition cable channel that opened during 2007, rises significantly during cadenas. Finally, we examine heterogeneity according to the content of cadenas, with the largest drop-off of viewership on private channels for cadenas associated with the delivery of public goods and services and for cadenas related to elections. 6. Structural Estimation Building upon this evidence, we next provide estimates of a structural version of the model. We begin by extending the model and the notation to allow for non-news programming and gender-specific preferences over this programming. We then detail several issues in the empirical implementation and describe identification. After presenting the parameter estimates, we use the model to conduct counterfactual experiments. These experiments allow us to quantify the degree to which switching limits exposure to propaganda and ultimately political persuasion among the opposition. 6.1. Approach As in the first extension of the model, we consider three types of stations: government (i = g), moderate (i = m), and opposition (i = o). As before, let $$v$$ ∈ {o, g} index viewer ideology, opposition, and progovernment. Then, viewers receive payoffs equal to θs from same-type programming (cadenas and government news for progovernment viewers and opposition news for opposition viewers) and payoffs equal to θd from different-type programming (cadenas and government news for opposition viewers and opposition news for progovernment viewers). Both opposition and progovernment viewers receive payoffs of θm from moderate news. To estimate switching costs, we also consider the following additional types of non-news programming: soap operas (telenovelas), sports, and other. Following Esteves-Sorenson and Perretti (2012), we measure switching costs via gender-specific preferences over soaps and sports.22 In particular, let k ∈ {m, f} index viewer gender, and let u$${vkip}$$ represent gender-specific systematic payoffs for a viewer with ideology $$v$$ watching programming p on station i. Table 13 summarizes these payoffs. As shown, payoffs from news programming and cadenas are assumed to differ across viewer ideology but not gender, and payoffs from non-news programming, such as sports and soaps, differ across gender but not ideology. Likewise, although preferences for news programming vary across stations, a simplifying assumption is that preferences for sports, soaps, and other programming vary across viewer types but not across stations. In the context of this model, we next derive market shares, separately for viewers of ideology $$v$$ and gender k. We sequence shows within a day according to the time aired (t = 1, 2, 3, …).23 Then, with positive switching costs (η > 0), market shares for a viewer with ideology $$v$$ and gender k watching programming p on station i at time t ($$\sigma _{vkip}^{t}$$), as a function of market shares during the previous time slot ($$\sigma _{vkip}^{t-1}$$), are given by \begin{eqnarray*} \sigma _{vkip}^{t}&=&\sigma _{vkip}^{t-1}\frac{\exp \big(u_{vkip}^{t}\big)}{\exp \big(u_{vkip}^{t}\big)+\sum _{j\ne i}\exp \big(u_{vkjp}^{t}-\eta \big)}\nonumber\\ && +\sum _{l\ne i}\sigma _{vklp}^{t-1}\frac{\exp \big(u_{vkip}^{t}-\eta \big)}{\exp \big(u_{vklp}^{t}\big)+\sum _{j\ne l}\exp \big(u_{vkjp}^{t}-\eta \big)} . \end{eqnarray*} The first term represents the likelihood that a viewer is both watching channel i during the previous time slot (t − 1) and does not switch to another channel at time t. The second term represents the likelihood that a viewer is both watching a different channel (l ≠ i) during the previous time slot and switches from channel l to channel i at time t, incurring switching costs equal to η. This is then summed across all other options. This includes the outside option of not watching television, which, as stated previously, is normalized to provide a systematic payoff of zero. To illustrate the intuition behind these market shares, consider two special cases. First, with high switching costs (η → ∞), market shares do not change between time t − 1 and time t; that is, $$\sigma _{vkip}^{t}=\sigma _{vkip}^{t-1}$$. In this case, inertia is complete, and viewership does not respond to the airing of cadenas. Second, in the absence of switching costs (η = 0), market shares at time t are independent of market shares at time t − 1 and collapse to the standard multinomial logit form: \begin{equation*} \sigma _{vkip}^{t}=\frac{\exp \big(u_{vkip}^{t}\big)}{\exp \big(u_{vkip}^{t}\big)+\sum _{j\ne i}\exp \big(u_{vkjp}^{t}\big)}. \end{equation*} In this case, there is no inertia. Although viewership does respond to the airing of cadenas, the impact lasts for only one period, with viewership during future periods unchanged. In intermediate cases, with moderate switching costs, inertia exists but is incomplete. In particular, a positive shock to viewership of channel i at time t − 1 leads to higher viewership of that channel at time t. For example, if females have a stronger preference for soaps than males, then a soap airing at time t − 1 will, all else equal, tend to increase viewership of that channel for females, relative to males, at time t. This is due to the presence of switching costs, resulting in inertia in viewership. Since our data distinguish between male and female viewers but not between progovernment and opposition viewers, we next aggregate market shares across opposition and progovernment. Recalling that πg represents the fraction of progovernment viewers, we have that market shares among gender k for station i airing programming p equal \begin{equation*} \sigma _{kip}^{t}=\pi _{g}\sigma _{gkip}^{t}+(1-\pi _{g})\sigma _{okip}^{t}. \end{equation*} For the purposes of estimation, these model-based market shares ($$\sigma _{kip}^{t}$$) are then linked to observed market $$(s_{kip}^{t})$$ shares via the following log-odds formulation: \begin{equation*} \ln \left(\frac{s_{kip}^{t}}{1-s_{kip}^{t}}\right)=\ln \left(\frac{\sigma _{kip}^{t}}{1-\sigma _{kip}^{t}}\right)+\epsilon _{kip}^{t}, \end{equation*} where $$\epsilon _{kip}^{t}$$ is assumed to be normally distributed. Then, the parameters of the model (e.g., θd, θm, θs, η) are estimated via maximum likelihood. 6.2. Empirical Implementation and Identification Before presenting estimates of the parameters of this model, we first address three issues regarding empirical implementation. We then provide an intuitive overview of identification. First, although the previous formulation assumed that the sequence of programming (t = 1, 2, 3…) was identical across channels, programming schedules differ across channels within a day. For example, RCTV may air a program from 6 p.m. to 6:30 p.m., whereas Globovision may air a program from 5:30 p.m. to 6:20 p.m. and then another show from 6:20 p.m. to 7 p.m. In this case, for a given program, it is unclear how to define the set of competing shows, those aired on other channels. To do so, we define, for each show, the set of competing shows on other channels as those with the maximal time overlap with the focal program. In the previous example, a show airing from 6 p.m. to 6:30 p.m. on RCTV would compete for viewership with the show airing from 5:30 p.m. to 6:20 p.m. on Globovision, which shares 20 min of programming, rather than the show airing from 6:20 p.m. to 7 p.m., which shares only 10 min of programming. Second, given the recursive formulation stated previously, in which viewership at time t depends upon viewership at time t − 1, one must define initial conditions for market shares. To do so, we assume zero viewership before 6 a.m., when most of the population is sleeping, and ratings are consequently extremely low. That is, we assume that the entire market is consuming the outside option of no television, which provides a systematic payoff of zero, prior to t = 1. This allows us to write viewership during the first time slot (t = 1) as follows: \begin{equation*} \sigma _{vkip}^{1}=\frac{\exp \big(u_{vkip}^{1}-\eta \big)}{1+\sum _{j\ne 0}\exp \big(u_{vkjp}^{1}-\eta \big)}, \end{equation*} where $$u_{vkip}^{1}-\eta$$ is the payoff from switching from the outside option to channel i airing programming p at t = 1 and j = 0 refers to the outside option. In addition to closing the model, this assumption implies no dynamic linkages in viewership between midnight and 6 a.m., allowing us to treat each day as an independent observation. Third, since we do not observe market shares separately for opposition and government viewers, we must aggregate across these groups, as outlined previously. Given this, one must thus measure the fraction of progovernment viewers (πg) in each municipality. To do so, we measure these via municipality-specific vote shares for the opposition party and Chavez, respectively, in the 2006 Presidential election. The intuition for identification is explained in several steps. First, gender-specific preferences over sports and soap operas are identified simply by comparing ratings for these types of programming across male and female viewers. Then, with these estimates of gender-specific programming, switching costs can be identified by examining gender-specific ratings for shows aired on the same channel but after these sports and soaps programs. Finally, with estimates of these switching costs, one can identify ideology-specific preferences over news and cadenas by examining, similarly to the reduced form evidence presented previously, changes in ratings during cadenas that interrupt news programming across different types of stations (opposition, moderate, and progovernment). This identifies preferences over ideological content, as given by θd, θm, and θs.24 6.3. Parameter Estimates Our parameter estimates are provided in Table 14. Note that these coefficients should be considered relative to programming other than news, cadenas, sports, and soaps. This includes categories such as movies and game shows, which receive payoffs equal to the constant term, and the payoff from not watching television is normalized to zero. Following the identification logic given previously, we begin by discussing gender-specific preferences over news programming. As seen, we find overall high viewership for soaps. This is true for men and, consistent with prior evidence (Esteves-Sorenson and Perretti 2012), especially so among female viewers. Likewise, we find slightly lower viewership for sports but especially so among female viewers. These two gender-specific coefficients are both economically and statistically significant, with females, relative to males, having 56% higher viewership for soaps and 18% lower viewership for sports. In addition, females have 18% higher viewership across all categories. As noted previously, by comparing gender-specific ratings on shows immediately following sports and soaps, we can identify switching costs. As shown, these estimated switching costs are also statistically significant, and evidence on their economic significance will be documented in a counterfactual analysis to follow, in which we trace out the dynamic response to cadenas for viewers and channels of differing ideology. Finally, using these estimates of switching costs to identify preferences over ideological content, we have that payoffs from information are associated with lower viewership overall. This is the payoff for both progovernment and opposition viewers from consuming moderate news. As shown, this negative effect is partially offset for same-type information, cadenas, and news on government channels for progovernment viewers and news on opposition channels for opposition viewers. Conversely, payoffs are substantially lower for different-type information, cadenas, and news on government channels for opposition viewers and news on opposition channels for progovernment viewers. Note also the asymmetry between same-type and different-type ideology, with the benefits associated with same-type information (0.2409) smaller than the costs associated with different-type information (1.6717). Taken together, these estimates provide additional support for the hypothesis of preferences for like-minded information. 6.4. Counterfactual Viewership Using these parameter estimates, we then summarize viewership patterns under three scenarios. First, we predict viewership of cadenas as predicted by the structural model separately by channel and separately for progovernment and opposition viewers. Second, we predict viewership patterns under a counterfactual scenario in which cadenas are replaced with news programming. That is, propaganda is replaced by opposition content on opposition channels and moderate content on moderate channels, with no change in content on government channels. Third, we predict viewership patterns under a counterfactual scenario in which cadenas are aired but under which viewers face infinite switching costs and thus cannot switch channels or tune out. This allows us to quantify the degree to which switching limits exposure to government propaganda. Analysis of these three scenarios proceeds in the following four steps. First, we focus on the set of days on which a single cadena was broadcast. This allows us to assume that viewership just before the cadena is identical under the three scenarios. Second, using this sample of days and normalizing the time slot of the cadena to equal zero, we use the estimated model to predict viewership of shows aired throughout the day. For simplicity, we focus on viewership among females and in cities with the full set of available channels (i.e., Caracas and Valencia). Third, we use the model to predict how viewership would have evolved were cadenas to be replaced by news programming, with ideological content depending upon the channel under consideration. Only programming in the focal time slot (t = 0) is altered, and programming during the other time slots is unchanged under the counterfactual. Similarly, and as noted previously, we also consider a counterfactual scenario in which viewers cannot tune out when cadenas come on the air. Table 15 compares viewership patterns under these three scenarios. For progovernment viewers, patterns of viewership are similar under the three scenarios. Viewership combined across the five channels totals 20.4% under our baseline scenario (propaganda), 22.1% under propaganda without switching, and 16.7% under media pluralism. The lower consumption under media pluralism reflects the fact that the media landscape is more opposition-oriented in this case. Table 15. Counterfactual viewership summary. Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 View Large Table 15. Counterfactual viewership summary. Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 Propaganda Media pluralism Propaganda (baseline) (no switching) Progovernment viewers Opposition channels 0.032 0.013 0.031 Moderate channels 0.059 0.058 0.069 Public channel 0.024 0.026 0.022 Combined viewership 0.204 0.167 0.221 Fraction persuaded 0.000 0.011 0.000 Opposition viewers Opposition channels 0.019 0.042 0.045 Moderate channels 0.029 0.056 0.068 Public channel 0.003 0.002 0.003 Combined viewership 0.098 0.197 0.228 Fraction persuaded 0.013 0.007 0.030 Net persuasion 0.006 −0.003 0.013 View Large For opposition viewers, by contrast, consumption is substantially different under the three scenarios. Viewership combined across the five channels totals 9.8% under propaganda, 22.8% under propaganda without switching, and 19.7% under media pluralism. The difference between propaganda with and without switching is driven by significant dropoff on both opposition and moderate channels when a cadena comes on the air, and viewership of the public channel is close to zero under all three scenarios. Taken together, we find that switching significantly limits exposure to propaganda. That is, exposure to propaganda would be more than twice as high for opposition viewers in the absence of switching. 6.5. Implications for Persuasion Given that switching limits exposure to propaganda, it is natural to investigate how this impacts political persuasion. To do so, we next calculate the electoral effects of propaganda under the three scenarios described previously. We use estimates from the literature on persuasion rates, defined as the probability of converting a voter not already persuaded to support the favored candidate based upon media exposure.25 DellaVigna and Gentzkow (2010) summarize persuasion rates in the literature on switching votes as varying between 6% and 20%, and we use the midpoint of 13%. That is, we assume that opposition viewers exposed to propaganda are converted to the government’s side with 13% probability. Likewise, under the media pluralism scenario, we assume that exposure of progovernment viewers to opposition news are converted to the opposition side with 13% probability.26 Finally, we assume that moderate news converts progovernment viewers to the opposition with 6.5% probability (one-half of baseline persuasion) and converts opposition viewers to the progovernment side with 6.5% probability.27 Note that persuasion rates may vary across contexts, and the set of countries considered in DellaVigna and Gentzkow (2010) does not include Venezuela. In addition, persuasion rates should naturally depend upon the types of interventions, with long cadenas having more influence than short cadenas. Given these limitations, we also provide estimates using the lower end of this interval (6%) and the upper end of this interval (20%). As shown in Table 15, we find that having the government control the airways via propaganda leads to 1.3% of opposition viewers persuaded to support the government. Progovernment viewers already support the government and hence there is no persuasion for this group. Given that the opposition represents 45% of the electorate in Caracas, we have support for the government increases by 0.6 percentage points (net persuasion) and thus support for the opposition falls by 0.6 percentage points, a swing of 1.2 percentage points toward the government. Under media pluralism, by contrast, 1.1% of progovernment viewers are persuaded to support the opposition, with the patterns reversed for 0.7% of opposition viewers. These effects largely cancel out, and we have that support for the government falls by 0.3 percentage points. Under propaganda without switching, we have large effects for opposition viewers, with 3.0% switching their support to the government. Comparing these two propaganda scenarios, one with and one without switching, we again demonstrate that switching limits exposure to propaganda and thus also ultimately limits ideological persuasion among opposition viewers. Finally, not reported in this table, we consider the fraction persuaded under both lower and higher persuasion rates. When the persuasion rate is 6% instead of the baseline 13%, we have that 0.6% of opposition viewers are persuaded by cadenas and this increases to 1.4% in the absence of switching. When the persuasion rate is 20%, by contrast, we have that 2.0% of opposition viewers are persuaded by cadenas and this increases to a very large 4.6% in the absence of switching. As noted previously, these estimates are based on persuasion rates from other contexts. Nonetheless, these results demonstrate the potential for behavioral responses in limiting the influence of government propaganda. 6.6. Dynamic Responses Despite substantial behavioral responses, many opposition viewers are exposed to cadenas in our baseline propaganda scenario. Moreover, those who do tune out may consume less opposition news during subsequent time slots if they do not return to opposition channels. We next examine this issue by analyzing the dynamics of viewer responses under two scenarios: propaganda and media pluralism. The results from this exercise are provided in Figure 7, in which we plot viewership during the two shows aired before the cadena, the cadena, and the six shows aired after the cadena. The x-axis is time to cadena and is normalized to equal zero during the cadena time slot. The y-axis is the viewership market share, separately by channel and viewer ideology. The upper panel provides results for ratings on the opposition channel, with progovernment viewers on the left and opposition viewers on the right. The middle and bottom panels provide corresponding results for moderate and government channels. Finally, we consider both market shares under propaganda, as given the solid line, and market shares under the media pluralism counterfactual, as given by the dashed line. Figure 7. View largeDownload slide Predicted viewership under propaganda and counterfactual media pluralism. Figure 7. View largeDownload slide Predicted viewership under propaganda and counterfactual media pluralism. Consistent with the viewership summary described previously, we see a sharp dropoff in viewership among opposition viewers when a cadena comes on the air for both opposition and moderate channels. Moreover, due to the presence of switching costs, differences in viewership, depending upon whether the previous show is a cadena or opposition news, are also apparent during the subsequent time slot (i.e., t = 1), with the counterfactual path of viewership then converging back to the predicted path of viewership several time slots following the cadena. Thus, cadenas have a persistent effect on viewership of opposition channels, with a sustained decrease in viewership by opposition viewers. This can be interpreted as a multiplier effect, under which cadenas reduce viewership of private channels not only during the cadena but also during subsequent time slots. Due to the reduction in exposure to opposition news, opposition viewers may also be less informed in general, making it more difficult for them to hold incumbents accountable. 6.7. Welfare Analysis Finally, we consider consumer welfare under two scenarios: government propaganda and media pluralism. In particular, we measure the overall welfare of opposition and progovernment viewers, respectively, when cadenas come on the air, relative to the counterfactual in which each channel airs news programming. This allows us to measure the welfare gains from media pluralism, defined as moving from an environment with only government programming on all channels to a situation with two channels airing opposition news and two channels airing moderate news. Welfare is measured using the inclusive value, the standard measure in discrete choice models. This is calculated by taking the expected value of the maximal utility over the choice set. Abstracting from gender and taking viewership probabilities at time slot t − 1 as given, the welfare of a viewer with ideology $$v$$ at time t is given by, \begin{equation*} W_{v}^{t}=\sum _{l}\sigma _{vlp}^{t-1}\ln \left[\exp \big(u_{vlp}^{t}\big)+\sum _{j\ne l}\exp \left(u_{vjp}^{t}-\eta \right)\right]. \end{equation*} Within the summation, the term \begin{equation*} \ln \left[\exp \big(u_{vlp}^{t}\big)+\sum _{j\ne l}\exp \big(u_{vjp}^{t}-\eta \big)\right] \end{equation*} represents the value to viewers with ideology $$v$$ watching channel l at time t − 1, where $$u_{vlp}^{t}$$ is the payoff from continuing to watch channel l at time t and $$u_{vjp}^{t}-\eta$$ is the payoff associated with switching at time t from channel l to a different channel j ≠ l. These values associated with watching a given channel at time t − 1 are then aggregated across channels, weighting by viewership at time t − 1. The results from this welfare analysis are presented in Table 16. As shown, welfare for progovernment viewers falls when moving from an environment in which all channels air government programming to media pluralism, an environment with one channel airing government programming, two channels airing moderate programming, and two channels airing opposition programming. This simply reflects the fact that overall ideological content is more opposition-oriented under the counterfactual, relative to the scenario in which only government programming is aired on all five channels. For opposition viewers, by contrast, the pattern is reversed, with an increase in welfare under media pluralism, again reflecting the fact that overall ideological content is more opposition-oriented in this case. Table 16. Welfare analysis. Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 View Large Table 16. Welfare analysis. Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 Propaganda (baseline) Media pluralism Difference Progovernment viewers −0.1841 −0.2218 −0.0377 Opposition viewers −0.2645 −0.1937 0.0717 Aggregate −0.2206 −0.2092 0.0114 View Large Using the 55% share of progovernment viewers, we have that aggregate welfare rises under media pluralism, despite the fact that opposition viewers comprise a minority. This simply reflects the fact that the welfare gains from media pluralism for the opposition exceeds the welfare losses to progovernment viewers.28 In closing, we note three limitations of this welfare analysis. First, we do not have monetary measures of welfare and thus cannot determine whether or not these welfare differences are economically significant.29 Second, this analysis assumes that viewer choices are personal and reflect their underlying preferences. This assumption might be violated if, for example, public employees are expected to watch cadenas and do so in order to enhance their job security. Third, our welfare measures do not account for the consequences of voters using information to hold incumbents accountable. 7. Conclusion In future work, we plan to pursue related topics involving media bias in Venezuela. First, we plan to study the government’s strategic decision regarding the timing of cadenas. For example, does the government air cadenas in order to censor opposition news during key events, such as protests? Second, we plan to examine the bundling of entertainment and news programming. For example, do viewers choose to watch news on channels with popular entertainment programming, and do channels with an ideological objective respond to this type of behavior when developing their programming? Third, although this study has relied on persuasion rates from the existing literature, we hope to measure the persuasive impact of propaganda in the Venezuelan context. Consistent with a preference for like-minded ideological content, we find that viewers respond to high frequency variation in the ideological slant of television programming. These responses are stronger for private channels, when compared to public channels, and for the most ideological channels. The responses are stronger for viewers with larger choice sets, as proxied via cable. Consistent with this result, we also show that viewership of cable increases during cadenas. The results are also stronger for the most polarizing cadenas, those involving the delivery of goods and services and those related to elections. Building upon this evidence, we structurally estimate the model. Based upon counterfactuals, we document that these behavioral responses significantly limit exposure to propaganda among opposition viewers. This may potentially limit the persuasive impact of government propaganda. There have been many changes in the media landscape in Venezuela since our sample period, with the rise of social media a key factor. The intensive use of political propaganda and the homogenization of ideologies in the television channels in Venezuela have made social media particularly popular. The opposition uses social media given the lack of opposition television channels at current. As an example, RCTV began online broadcasting through the Internet, and protesters often find social media an effective means for organizing demonstrations. Likewise, the government now uses social media to deliver propaganda. Nevertheless, television remains a key government vehicle for delivering propaganda since mainstream media continues to reach a large fraction of the population. Notes The editor in charge of this paper was Paola Giuliano. Acknowledgments: For helpful comments, we thank seminar participants at Brown University, Rice University, Carnegie Mellon University, Barcelona GSE Summer Forum, Stony Brook Political Economy Conference, CFPE Conference (Vancouver School of Economics), the New York City Media Seminar, USC Marshall, Workshop in Political Economy (Uppsala), CEMFI, Singapore Management University, University of Calgary, and Washington University. The opinions and statements are the sole responsibility of the authors and do not necessarily represent neither those of the Banco de la República de Colombia nor of its Board of Directors. Footnotes 1 This paper studies three types of behavioral responses by viewers: tuning out, switching, and staying tuned in. Hirschman (1970) identifies Exit, Voice, and Loyalty as three responses to propaganda. In this case, exit is the equivalent of tuning out, since viewers can avoid propaganda completely, but they are also consuming less information overall across all media. Thus, exit may lead viewers to become more disengaged in general with politics, thus becoming less informed and potentially weakening political accountability. Voice is the equivalent of switching since it is a form of protest and people can choose to consume like-minded information. Finally, loyalty is the equivalent of staying tuned in. 2 Recent studies on persuasion include DellaVigna and Kaplan (2007), Enikolopov, Petrova, and Zhuravskaya (2011), George and Waldfogel (2003), Chiang and Knight (2011), Gentzkow, Shapiro, and Sinkinson (2011), Gerber, Karlan, and Bergan (2009), Martin and Yurukoglu (2015), Prat (2014), and Snyder and Stromberg (2010). See also Prat and Stromberg (2013) for a comprehensive overview of this literature. 3 DiTella, Galiani, and Schargrodsky (2012) study the effects of government propaganda against privatization of water services after the 2006 nationalization in Argentina. Qian and Yanagizawa-Drott (2017) document an increase in U.S. news coverage of human rights abuses in countries not aligned with the United States when they rotated onto the United Nations. Security Council during the Cold War. They report similar patterns for reports produced by the U.S. State Department, suggesting an important role for government propaganda. Other literature focuses on the power of propaganda to mobilize the masses. Adena et al. (2015) document the importance of political propaganda to mobilize support for the Nazis. One interesting finding in this study is that a predisposition against propaganda can limit its effectiveness, and our study explores a potential mechanism underlying this finding. Yanagizawa-Drott (2014) provides evidence on the role of propaganda broadcast on radio by the Hutu government during the Rwandan genocide. DellaVigna et al. (2014) document an instance in which propaganda had negative consequences: cross-border exposure to Serbian radio among Croats is associated with anti-Serbian sentiment and anti-Serbian behavior. 4 This section draws upon Wilpert (2007), Corrales and Penfold (2011), Nelson (2009), Republica Bolivariana de Venezuela (2012), and Dinneen (2012). 5 Chang-Tai et al. (2011) document that voters who supported the Presidential recall referendum against Chavez experienced a significant reduction in earnings and employment following the public release of a list of voters who signed the recall petition. 6 For example, private television channels tended to cover only antigovernment protests during the coup and pointed to the government as the cause of violence in the struggle between Pro-Chavez and Anti-Chavez protesters. Once Chavez returned to power, private channels stopped broadcasting news, and a Chavez speech was aired in split-screen to broadcast anti-Chavez protests in parallel with the speech by Chavez. During the strike, the media gave priority to this issue for more than two months, often suspending regular programming for more extensive coverage of the crisis. Even when the protests were significantly weakened, some private media commentators continued to call for Chavez’s resignation in order to end the crisis. 7 Chavez accused the private channels publicly of “inciting rebellion and disrespect 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”. See Reporters Without Borders (2003). 8 New York Times (2007). 9 Besley and Prat (2006) analyze government capture of the media sector. 10 RCTV International was later shut down, closing in 2010. 11 In addition to cadenas, Chavez also hosts a public television program titled “Alo Presidente”, where he promoted the Bolivarian revolution. The show started at 11 a.m. every Sunday and lasted about 5 h (Kitzberger 2010). Frajman (2014) argues that Alo Presidente was a “grand stage for Chavez to promote his position as revolutionary leader and be cheered by crowds of loyal supporters”. 12 The type-1 extreme value distribution assumption is necessary and sufficient for generating logit probabilities. See McFadden (1973) for further details. 13 We do assume, for tractability reasons, that viewers receive a new ε$${vip}$$ draw for the outside option when a cadena comes on the air. This assumption is not crucial to our results given in what follows and can be relaxed. 14 Given our use of average minute rating, the measure of drop off combines viewers who switch off immediately and viewers who switch off part way through. 15 Each member of the household has a separate code, allowing Nielsen to separate viewership within a household according to gender. 16 We group them into three categories: news, entertainment, and cadenas and thus the numbers in the table do not add to 100%. News programs include the categories “Information/Opinion” and “Documentaries”. Entertainment includes “Sports”, “Entertainment”, “Children”, “Games”, “Microseries”, “Miniseries”, “ Movies”, “Series”, and “Soap Operas”. Finally, we leave the category “cadenas” as is. 17 This evidence is consistent with EU-EOM (2006), which shows that VTV and Globovision devoted greater time to political information during 2006 elections and the private channels RCTV, Venevision, and Televen devoted far less time to political information. 18 Likewise, using other measures of political preferences, not reported here, we find that people who watch news on public channels report higher levels of trust in Chavez than people who watch private channels. 19 Our measure of viewership is based upon the percentage of (potential) viewers. But our measures would be identical if we used the absolute number of viewers since drop-off is based upon the percent change when transitioning from news to cadenas. That is, drop-off for channel i is given by Δ$${ic}$$ = ln [s$${ic}$$/s$${in}$$]. Viewer shares for cadenas and news equal s$${ic}$$ = N$${ic}$$/N and s$${in}$$ = N$${in}$$/N, where N$${ic}$$ is the number of viewers of the cadena, N$${in}$$ is the number of viewers of the preceding news program, and N is the population (number of potential viewers). Thus, our dropoff measure would be identical if it were based upon the absolute number of viewers (i.e., s$${ic}$$ = N$${ic}$$ and s$${in}$$ = N$${in}$$) since population appears in both the numerator and the denominator of our measure of dropoff. 20 For this analysis we drop cases where the gap between the end time of the news and the start time of the cadena exceeds 10 min. 21 The difference between these two coefficients is statistically significant at conventional levels. 22 Note that our data also include ratings according to age and socioeconomic status. We choose to focus on gender given the evidence in Esteves-Sorenson and Perretti (2012). 23 As in the baseline theoretical model, viewers receive a new draw from the distribution for unobserved preferences (ε$${vkip}$$) for each new time period/show. 24 Note that this intuition is incomplete, as the model also implies linkages between switching costs and preferences over ideological content via the ideological mechanism. That is, in order to observe a disproportionate drop-off in viewership when a cadena is aired on private channels, it must be the case that switching costs are positive. Otherwise, as noted previously, viewership of cadenas is independent of previous programming, which includes the ideological content of news. Thus, switching costs are identified via both gender-specific preferences over programming and channel-specific changes in viewership when cadenas are aired. 25 In using this definition of persuasion, we abstract from an alternative goal of propaganda: maintaining a base of core supporters. 26 One could also consider asymmetric persuasion rates. If, for example, opposition news is more persuasive than propaganda, then our estimates may overstate the importance of switching among opposition viewers exposed to propaganda. 27 Net persuasion results are very similar when one ignores any persuasive effects of moderate news since the effects for progovernment and opposition viewers tend to cancel out. 28 To better understand why these gains to the opposition under media pluralism exceed the losses to progovernment viewers, consider the following simple example. Suppose there are only two stations and an equal number of opposition and progovernment viewers (i.e., πg = 0.5). With both channels airing government programming, welfare for opposition viewers equals ln [2 exp (θd)], and welfare for progovernment viewers is now ln [2exp (θs)]. Under media pluralism, by contrast, welfare for each group equals ln [exp (θs) + exp (θd)]. One can then show that aggregate welfare rises under media pluralism due to a preference for like-minded information (θd < θs). This simply reflects the fact that the gain to opposition viewers from the introduction of a like-minded station is less than the loss to progovernment viewers from losing one of two like-minded stations. 29 Moreover, the cross-group comparisons rely on the assumption that the variance of the unobservables is the same across groups. References Adena Maja , Enikolopov Ruben , Petrova Maria , Santarosa Veronica , Zhuravskaya Ekaterina ( 2015 ). “ Radio and the Rise of the Nazis in Pre-War Germany .” Quarterly Journal of Economics , 130 , 1885 – 1939 . Google Scholar CrossRef Search ADS Besley Timothy , Prat Andrea ( 2006 ). “ Handcuffs for the Grabbing Hand? Media Capture and Government Accountability .” American Economic Review , 96(3) , 720 – 736 . Google Scholar CrossRef Search ADS Bisbal Marcelino ( 2009 ). “ La Comunicacion Masiva Como Politica de Gobierno de Hugo Chavez Frias .” In Hegemonia y Control Comunicacional , edited by Bisbal Marcelino . Editorial Alfa, Caracas, Venezuela . Chang-Tai Hsieh , Miguel Edward , Ortega Daniel , Rodriguez Francisco ( 2011 ). “ The Price of Political Opposition: Evidence from Venezuela’s Maisanta .” American Economic Journal: Applied Economics , 3 , 196 – 214 . Google Scholar CrossRef Search ADS Chiang Chun-Fang , Knight Brian ( 2011 ). “ Media Bias and Influence: Evidence from Newspaper Endorsements .” Review of Economic Studies , 78 , 795 – 820 . Google Scholar CrossRef Search ADS Corrales Javier , Penfold Michael ( 2011 ). Dragon in the Tropics: Hugo Chavez and the Political Economy of Revolution in Venezuela . Brookings Latin America Initiative Books , Washington DC, USA . DellaVigna Stefano , Enikolopov Ruben , Mironova Vera , Petrova Maria , Zhuravskaya Ekaterina ( 2014 ). “ Cross-Border Media and Nationalism: Evidence from Serbian Radio in Croatia .” American Economic Journals: Applied Economics , 6 , 103 – 132 . Google Scholar CrossRef Search ADS DellaVigna Stefano , Gentzkow Matthew ( 2010 ). “ Persuasion: Empirical Evidence .” Annual Review of Economics , 2 , 643 – 669 . Google Scholar CrossRef Search ADS DellaVigna Stefano , Kaplan Ethan ( 2007 ). “ The Fox News Effect: Media Bias and Voting .” Quarterly Journal of Economics , 122 , 1187 – 1234 . Google Scholar CrossRef Search ADS Dinneen Mark ( 2012 ). “ The Chavez Government and the Battle over the Media in Venezuela .” Asian Journal of Latin American Studies , 25 , 27 – 53 . DiTella Rafael , Galiani Sebastian , Schargrodsky Ernesto ( 2012 ). “ Reality versus Propaganda in the Formation of Beliefs about Privatization .” Journal of Public Economics , 96 , 553 – 567 . Google Scholar CrossRef Search ADS Durante Ruben , Knight Brian ( 2012 ). “ Partisan Control, Media Bias, and Viewer Responses: Evidence from Berlusconi’s Italy .” Journal of the European Economic Association , 10 , 451 – 481 . Google Scholar CrossRef Search ADS Enikolopov Ruben , Petrova Maria , Zhuravskaya Ekaterina ( 2011 ). “ Media and Political Persuasion: Evidence from Russia .” American Economic Review , 101 ( 7 ), 3253 – 3285 . Google Scholar CrossRef Search ADS EU-EOM ( 2006 ). “ Final Report: Presidential Elections Venezuela 2006 .” Tech. rep., European Union Election Observation Mission . Frajman Eduardo ( 2014 ). “ Broadcasting Populist Leadership: Hugo Chavez and Alo Presidente .” Journal of Latin American Studies , 43 , 501 – 526 . Google Scholar CrossRef Search ADS Gehlbach Scott , Sonin Konstantin ( 2014 ). “ Government Control of the Media .” Journal of Public Economics , 118 , 163 – 171 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. ( 2010 ). “ What Drives Media Slant? Evidence from US Daily Newspapers .” Econometrica , 78 , 35 – 71 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. , Sinkinson Michael ( 2011 ). “ The Effect of Newspaper Entry and Exit on Electoral Politics .” American Economic Review , 101 ( 7 ), 2980 – 3018 . Google Scholar CrossRef Search ADS Gentzkow Matthew , Shapiro Jesse M. , Sinkinson Michael ( 2014 ). “ Competition and Ideological Diversity: Historical Evidence from US Newspapers .” American Economic Review , 104 ( 10 ), 3073 – 3114 . Google Scholar CrossRef Search ADS George Lisa , Waldfogel Joel ( 2003 ). “ Who Affects Whom in Daily Newspaper Markets? ” Journal of Political Economy , 111 , 765 – 784 . Google Scholar CrossRef Search ADS Gerber Alan S. , Karlan Dean , Bergan Daniel ( 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 , 35 – 52 . Google Scholar CrossRef Search ADS Goettler Ronald L. , Shachar Roni ( 2001 ). “ Spatial Competition in the Network Television Industry .” RAND Journal of Economics , 32 , 624 – 656 . Google Scholar CrossRef Search ADS Hirschman Albert ( 1970 ). Exit, Voice, and Loyalty: Responses to Decline in Firms, Organizations, and States . Harvard University Press , Cambridge . Kitzberger Philip ( 2010 ). “ The Media Activism of Latin America’s Leftist Governments: Does Ideology Matter? ” GIGA Working Paper 151, German Institute of Global and Area Studies, Hamburg, Germany . Google Scholar CrossRef Search ADS Martin Gregory , Yurukoglu Ali ( 2015 ). “ Bias in Cable News: Real Effects and Polarization .” NBER Working Paper No. 20798, National Bureau of Economic Research, Cambridge, MA, USA. McFadden Daniel ( 1973 ). “ Conditional Logit Analysis of Qualitative Choice Behavior .” In Frontiers in Econometrics , edited by Zarembka P. . Academic Press, New York, USA . Moshkin Nickolay , Shachar Ron ( 2002 ). “ The Asymmetric Information Model of State Dependence .” Marketing Science , 21 , 435 – 454 . Google Scholar CrossRef Search ADS Nelson Brian A. ( 2009 ). The Silence and the Scorpion: The Coup Against Chavez and the Making of Modern Venezuela . Nation Books , New York, USA . Perretti Fabrizio , Esteves-Sorenson Constanca ( 2012 ). “ In Micro-Costs: Inertia in Television Viewing .” The Economic Journal , 122 , 867 – 902 . Google Scholar CrossRef Search ADS Prat Andrea ( 2014 ). “ Media Power .” CEPR Discussion Paper No. 10094, Center for Economic and Policy Research, Washington, DC, USA. Prat Andrea , Stromberg David ( 2013 ). “ The Political Economy of Mass Media .” In Advances in Economics and Econometrics , edited by Daron Acemoglu, Manuel Arellano, and Eddie Dekel . Cambridge University Press, Cambridge , pp. 135 – 187 . Google Scholar CrossRef Search ADS Qian Nancy , Yanagizawa-Drott David ( 2017 ). “ Government Distortion in Independently Owned Media: Evidence from U.S. Cold War News Coverage of Human Rights .” Journal of the European Economic Association , 15 , 463 – 499 . Google Scholar CrossRef Search ADS Reporters Without Borders ( 2003 ). Venezuela: Caught between an authoritarian president and intolerant media . Available at: http://www.refworld.org/docid/47fcdc365.html [accessed 12 March 2018] . Republica Bolivariana de Venezuela ( 2012 ). Abril , Memorias de un golpe de estado . Republica Bolivariana de Venezuela . Snyder James M. , Stromberg David ( 2010 ). “ Press Coverage and Political Accountability .” Journal of Political Economy , 118 , 355 – 408 . Google Scholar CrossRef Search ADS Wilpert Gregory ( 2007 ). Changing Venezuela by Taking Power: The History and Policies of the Chavez Government . Verso Books , London, England . Yanagizawa-Drott David ( 2014 ). “ Propaganda and Conflict: Evidence from the Rwandan Genocide .” Quarterly Journal of Economics , 129 , 1947 – 1994 . Google Scholar CrossRef Search ADS Yao Song , Wang Wenbo , Chen Yuxin ( 2017 ). “ TV Channel Search and Commercial Breaks .” Journal of Marketing Research , 54 , 671 – 686 . Google Scholar CrossRef Search ADS Youn Sug-Min ( 1994 ). “ Program Type Preference and Program Choice in a Multichannel Situation .” Journal of Broadcasting & Electronic Media , 38 , 465 – 475 . Google Scholar CrossRef Search ADS Supplementary Data Supplementary data are available at JEEA online. © The Author(s) 2018. Published by Oxford University Press on behalf of European Economic Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

Journal

Journal of the European Economic AssociationOxford University Press

Published: Apr 7, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off