Government Control of the Media

Transcription

1 Governent Control of the Media Scott Gehlbach and Konstantin Sonin Septeber 2008 Abstract We present a foral odel of governent control of the edia to illuinate variation in edia freedo across countries and over tie, with particular application to less deocratic states. The extent of edia freedo depends critically on two variables: the obilizing character of the governent and the size of the advertising arket. Media bias is greater and state ownership of the edia ore likely when the need for social obilization is large; however, the distinction between state and private edia is saller. Large advertising arkets reduce edia bias in both state and private edia, but increase the incentive for the governent to nationalize private edia. We illustrate these arguents with a case study of edia freedo in postcounist Russia, where edia bias has responded to the obilizing needs of the Krelin and governent control over the edia has grown in tande with the size of the advertising arket. Gehlbach: University of Wisconsin Madison and CEFIR, gehlbach@polisci.wisc.edu. Sonin: New Econoic School, CEFIR, and CEPR, ksonin@gail.co.

2 1 Introduction A substantial literature ties edia freedo to good governance. 1 Less is known about the deterinants of edia freedo itself. Although correlated with the presence of deocratic institutions, political institutions alone do not deterine edia freedo. As Figure 1 illustrates, any nondeocracies have higher levels of edia freedo than any deocracies, and aong the least deocratic countries there is little obvious relationship between political institutions and edia freedo. 2 Moreover, edia freedo often fluctuates within countries even as political institutions reain unchanged. What accounts for variation in edia freedo across space and tie? In this paper, we provide a foral fraework to explore this question. We focus especially on less deocratic states where the governent uses the edia to obilize citizens in support of actions that ay not be in their individual best interest, though soe of our arguents ay also apply to deocracies. We ephasize variation along two diensions of edia freedo: edia ownership, by which we generally ean whether the edia are state-owned or private, and edia bias, which we define as the extent to which the edia isreport the news in favor of governent interests. As we show, edia ownership typically influences edia bias, but edia ownership itself is endogenous to the anticipated bias under state and private ownership. Our theoretical fraework stresses a fundaental constraint facing any governent seeking to influence edia content: bias in reporting reduces the inforational content of the news, thus lowering the likelihood that individuals who need that inforation to ake decisions will read, watch, or listen to it. This constraint operates in two ways. First, excessive edia bias works against the governent s desire for social obilization, as citizens who ignore the news cannot be influenced by it. Second, edia bias reduces advertising revenue, as edia consuption is less when pro-governent bias is large. In general, this reduction in advertising revenue is costly to the governent, regardless of whether the edia are state owned (to the extent that profits and losses of state-owned copanies are internalized) or private (because the governent ust subsidize private owners to copensate for lost revenue). We highlight two variables that influence the operation of this constraint. First, the obilizing character of the governent deterines the value the governent places on the obilization of citizens to take actions that are not necessarily in their individual best interest, such as voting and deonstrating in favor of the incubent regie (Magaloni, 2006; Sipser, 2007) or exposing the corruption of local bureaucrats (Lorentzen, 2006). When the need for social obilization is large, the governent is ore willing to pay the cost (foregone advertising revenue or subsidies) of edia bias. Media bias ay therefore be greater in autocratic states whose leaders ai to transfor society, or under populist governents that retain power through ass public participation, than in kleptocracies or sultanistic regies. 3 Moreover, holding regie type constant, edia bias will be greater in periods 1 Much recent work on the topic is in the political econoy literature. Representative papers include Besley and Burgess (2002); Adserà, Boix, and Payne (2003); Brunetti and Weder (2003); Reinikka and Svensson (2005); Besley and Prat (2006); Prat and Ströberg (2006); and Treisan (2007). 2 Obviously, causality cannot be inferred fro the siple bivariate relationship depicted in Figure 1. 3 This arguent has parallels in Wintrobe s (1990) suggestion that repression will be greater under total- 1

3 Media Freedo AUS BEL DNK NZL NOR TMP SWE CHE FIN FREE JAM AUT CAN DEU NLD USA CRI CYP IRL PRT EST CZE JPN ESP SLB LVA BOL FRA LTU GBR POL GUY MUS CHL BWA BEN MLIKOR ZAF ISRITA TTO GRC HUN PNG SVN URY NAM DOM BRAPAN ARGSVK MNG BGR BFA SEN MDG PHL THA UGA COM ECU MKD SLVNIC IND PARTLY FREE VEN ROM MWI FJI HND HRV GHA MOZMEX TZA KWT PER LSOPRY MAR GAB JOR GNB LKA UKR ARM ALB BIH BGD COL COG GEO GTM PAK MDA GMB DJI IDN NPL NER RUS QAT KGZ KEN KHM ZMB TUR SGP NGA ETH HTI CAF BHR BTN KAZ MYS MRT YEM YUG OMN ZWE CIVGIN LAOAZE EGY CMR TCD AGO SLE SOM TUN LBR SWZ TGO SYR AREVNMERI RWA IRN SAU GNQBLR UZB CHN DZA NOT FREE BDI SDN ZAR TKM AFGTJK LBY IRQ CUB PRK MMR Deocracy Figure 1: Media freedo and deocracy. Media freedo is Freedo House Press Freedo Score, reordered so that higher values correspond to greater edia freedo. Freedo House classifies countries as Free, Partly Free, and Not Free, as indicated. Deocracy is polity2 variable fro Policy IV dataset. Values plotted are averages for when obilization is especially iportant, as during an election capaign. Consistent with cross-country evidence (Djankov, McLeish, Nenova, and Shleifer, 2003), bias is generally greater in state-owned edia, though as obilization increases in iportance, bias in state and private edia converge. Despite this convergence, the governent ay be ore inclined to seize ownership of private edia when obilization is iportant, as it can save the cost of subsidization by controlling the edia directly. Second, the size of the advertising arket, which ay be influenced by such factors as edia technology and econoic regulation, deterines the opportunity cost of lost consuers due to bias in reporting. In the words of a Mexican journalist working in the 1990s at the independent La Jornada, telling the truth is good business (Lawson, 2002, p. 89), though the extent to which that is the case depends on the responsiveness of advertisers to circulation, viewership, or listenership. Consistent with the findings of an eerging epirical and theoretical literature, 4 we show that private edia are generally less biased when the advertising arket is large, as purchasing influence through subsidization or outright bribery itarian than tinpot dictatorships. The distinction between obilizing and non-obilizing dictatorships has its roots in Friedrich and Brzezinski (1965). Przeworski et al. (2000) define obilizing dictatorships as those with at least one political party. In the case study below we discuss the shift toward a ore obilizing governent in Russia as the United Russia party and related organizations developed under Vladiir Putin. 4 Petrova (2008b) provides evidence based on the statistical analysis of data fro the nineteenth-century U.S. Historical accounts include Baldasty (1992) and Starr (2004). For theoretical analysis, see Besley and Prat (2006) and Petrova (2008a). 2

4 is relatively expensive for the governent. We advance on this understanding with two new results. First, we show that growth in the advertising arket can also reduce edia bias under state ownership, though because the governent need not bargain to achieve bias when it directly controls the edia, this effect will be coparatively sall. Second, and ore iportant, we deonstrate that the governent ay seize ownership of the edia for itself when the advertising arket is large. Not only can the governent ipleent a higher level of bias when it owns the edia an advantage that grows with the size of the advertising arket but it can econoize on subsidies and acquire advertising revenue for itself. A surprising iplication of this analysis is that the relationship between edia freedo and the size of the advertising arket ay be nononotonic. Holding ownership constant, growth in the advertising arket reduces edia bias, regardless of whether the edia are private or state-owned. However, the sae growth ay propt the governent to seize direct control of the edia, which leads to a discontinuous jup in edia bias. We illustrate our theoretical odel with a case study of governent control of the edia in postcounist Russia. Largely created fro scratch after the collapse of counis, Russia s advertising arket has grown draatically in recent years, a developent that would see to bode well for edia freedo. Yet by ost accounts, Russia has seen an equally draatic decline in edia independence, as the Krelin has seized direct control of large segents of the Russian edia arket. Our odel suggests two copleentary explanations for this phenoenon. First, Russian president Vladiir Putin seeed ore predisposed to use the edia to obilize society in support of his rule than his predecessor Boris Yeltsin. Second, the Krelin ay have been otivated to prevent private edia, newly flush with advertising revenue, fro asserting their independence. Our odel also provides a theoretical fraework for understanding other developents in the relationship between edia and the state in postcounist Russia, such as the convergence in bias aong private and state-owned edia prior to the 1996 presidential election and the behavior of ore and less coercial stations controlled by the governent in the run-up to the 2007 parliaentary election. Our theory builds on two odeling traditions in the political econoy literature. First, we follow the approach pioneered by Shleifer and Vishny (1994) in odeling a bargaining relationship between a politician and a fir, which in our case corresponds to a edia outlet. As in Shleifer and Vishny, we ephasize the allocation of control rights, which in our setting is the ability to decide what to report. Under state ownership, the governent possesses this right. We refer to this environent as direct governent control of the edia, to ephasize that the governent chooses edia bias directly. In contrast, under private ownership, a private owner possesses the right to choose what to report. We ter this situation indirect governent control, as the governent ay still induce edia bias but ust pay for it by providing subsidies to the private owner. Second, we build on a growing body of work that attepts to explain the origins of edia bias. Existing work stresses the inherent biases of edia owners (Gabszewicz et al., 2001) or journalists (Baron, 2006; Puglisi, 2006); the arket response to consuer deand for bias, which arises either for exogenous reasons (Mullainathan and Shleifer, 2005) or because firs skew their reporting toward consuer priors to build a reputation for quality (Gentzkow 3

5 and Shapiro, 2006); 5 the interaction of arket forces and the internal structure of the edia (Bovitz, Druckan, and Lupia, 2002); and the purchase of journalists by special interests (Petrova, 2008a). A useful distinction soeties ade in this literature is between deandside and supply-side explanations of edia bias. Our paper falls into the latter category: citizens in our odel always prefer less pro-governent bias to ore. Relative to ost of this literature, the key distinction of our approach is that we odel the governent as a strategic actor. Besley and Prat (2006) do consider governent influence over the edia, but their ephasis is the ipact of edia freedo on governent accountability in deocracies. Our approach is copleentary: we analyze the relationship between governent and the edia in an environent that ay best characterize weak deocracies and autocracies. Other odels of edia anipulation in dictatorships include Egorov, Guriev, and Sonin (2006), Debs (2007), Edond (2008), and Lorentzen (2008). Our contribution differs fro these in that we explicitly copare edia bias under private and state ownership and that we endogenize edia ownership, showing in each case how the actions of the governent depend on its obilizing character and the size of the advertising arket. Within the literature on edia bias, our theoretical fraework is ost closely related to Gentzkow and Shapiro (2006). As in their work, we odel Bayesian citizens who ay use inforation reported by the edia when aking a costly decision whose outcoe depends on the state of the world. The questions we explore in this fraework, however, are very different fro those that Gentzkow and Shapiro pose. Our approach is also siilar in soe respects to Petrova (2008a), who like us exaines the tradeoff between advertising revenue and bias. In our case, however, we treat the governent as the relevant special interest, an iportant distinction since the governent ay acquire control of the edia by force. As we show, the governent ay be ore otivated to seize private edia when the advertising arket is large, iplying a relationship between advertising revenue and edia freedo counter to that predicted by Petrova but observed in the Russian case that we discuss. Our adoption of a Bayesian fraework captures in a tractable way the assuption that individuals beliefs can be affected by what they see, read, and hear in the edia. Nuerous studies have exained and generally affired this proposition in the Aerican context; Kinder (1998) and Goldstein and Ridout (2004) provide reviews of the literature. Studies in iature deocracies have found edia effects that, if anything, are stronger than those typically identified in the U.S. These include White, Oates, and McAllister (2005) and Enikolopov, Petrova, and Zhuravskaya (2007) on Russia; Lawson and McCann (2007) on Mexico; and Gentzkow and Shapiro (2004) on the Mideast. The paper proceeds as follows. In Section 2, we exaine edia bias under direct governent control, i.e., when the edia are state-owned. In Section 3, we analyze the case of indirect governent control, i.e., state subsidization of privately owned edia. We endogenize governent control in Section 4, asking when the governent would choose to take over privately owned edia. Section 5 extends the odel to ultiple edia outlets. Although general results are difficult to derive, we show that our key arguents hold for two iportant special cases: when each edia outlet possesses coplete arket power, and when stations 5 The latter paper builds on the insight that rational individuals ay have a preference for biased inforation; see Calvert (1985). 4

6 are perfect substitutes for each other. We illustrate our odel with a case study of edia freedo in postcounist Russia in Section 6. Section 7 concludes. 2 Direct Governent Control of the Media 2.1 Environent Consider a odel with two sets of players: a continuu of citizens of ass one, indexed by i, and a governent that directly controls a news outlet (i.e., the news outlet is state-owned ). As we discuss below, various features of the odel lend theselves best to broadcast edia. For concreteness, we therefore say that citizens watch the news, as when the news outlet is a television station. At a cost noralized to one, each citizen i ay invest in a single project, π i {0, 1}. The private return fro investent depends on the state of the world S {L, H}, where L denotes a low state and H a high state. The governent prefers ore investent to less regardless of the state, and so ay have an incentive to obilize citizens to take actions that are not in their individual best interest. (We use the ter investent in the general sense of a costly action that provides a rando return, and we eploy the ters investent and obilization interchangeably throughout the text.) As we discuss above, this conflict of interest ay best characterize autocracies and weak deocracies, though there are circustances in ature deocracies to which the odel ay also apply. A political exaple of such investent is participation in a state-sponsored rally, where turnout provides legitiacy to the governent but citizens prefer to participate only if the governent is capable of sanctioning nonparticipation. An econoic exaple (one that ay apply to any regie type) is as follows: the governent prefers that econoic actors behave as if the central banker is conservative, whether he is or not, but econoic actors prefer to forego price increases only if he is. In particular, when S = H the project yields a private return r > 1, whereas when S = L the private return is equal to zero. Citizens do not know the state but have a coon prior belief that the state is high with probability θ (0, 1). We assue that θr < 1, which iplies that in the absence of any additional inforation citizens do not invest. Citizens ay update their belief about the state of the } world by watching the news. A news broadcast contains one of two essages, {Ĥ, Ŝ ˆL. Rather than choosing the essage directly, we assue that the governent structures the news operation to attain the desired level of bias, with the essage deterined probabilistically by the structure after the state of the world has been realized. This iplicitly captures the need to delegate responsibility for the news to the reporters, anchors, and editors who ake daily decisions about what to cover and how to cover it. At one Krelin-controlled radio station, for exaple, journalists were told that 50 percent of their stories about Russia should be positive ; the stories were not chosen by the Krelin itself. 6 Iportantly, we assue that this choice is 6 50% Good News Is the Bad News in Russian Radio, New York Ties, April 22, Reington (1989) describes the Soviet edia in siilar ters, noting that the delegation of authority ade it difficult for Gorbachev to quickly liberalize the edia. For a few exceptional stories, however, including the Chernobyl disaster, the Krelin directly chose the essage to be reported. Our odel ay be less applicable to such 5

7 observable by all citizens. This assuption sees especially copelling for broadcast edia, the priary news source for citizens in ost countries. For exaple, CNN progras hosted by Lou Dobbs and Anderson Cooper are easily distinguishable, and the replaceent of conservative coentator Tucker Carlson on MSNBC was viewed by any (including Carlson hiself) as a shift in editorial line. 7 In particular, at the beginning of the gae, prior to realization of the state of the world, the governent publicly chooses an editorial policy β (S) [0, 1] [0, 1], where β (S) is the probability that the edia outlet reports the essage Ĥ when the state is S {L, H}. We abuse notation slightly to denote any citizen s posterior belief ) about the state of the world, conditional on having received the essage (Ŝ Ŝ, by θ. 8 Thus, conditional on having ) watched the news and received the essage (Ŝ Ŝ, any citizen prefers to invest if θ r > 1. Watching the news ay therefore be profitable to citizens if the governent s editorial policy β (S) is such that the news is sufficiently inforative. Potentially offsetting this benefit, each citizen i has an exogenous idiosyncratic opportunity cost of watching the news µ i, where µ i is distributed uniforly on [0, ]. We assue for siplicity that > r 1, which iplies that for all θ the proportion of individuals who watch the news in equilibriu is strictly less than one. We use the indicator variable ω i {0, 1} to denote the decision to watch the news. Our assuption that this opportunity cost is exogenous applies ost clearly to broadcast edia, where there is no purchasing decision to be ade so long as a citizen already possesses a television or radio. In this context, the opportunity cost of watching the news ay reflect a citizen s work schedule, faily obligations, and siilar considerations. For print edia, the opportunity cost would also reflect the purchase price, which would be a choice variable of the edia owner. We choose to leave that extension to future research. Suarizing, the tiing of events is: 1. The governent chooses an editorial policy β (S) [0, 1] [0, 1], which is observed by all citizens. 2. Each citizen decides whether to watch the news, ω i {0, 1}. } 3. The state of the world S {L, H} is realized, with the essage {Ĥ, Ŝ ˆL deterined probabilistically according to β (S). Only citizens who watch the news receive the essage Ŝ. 4. Each citizen decides whether to invest in the project, π i {0, 1}. In general, we assue that the governent ay value viewership as well as the obilization of citizens to invest, to the extent that viewership increases advertising revenue at (and thus ay reduce the cost of running) the state-owned news outlet. We odel this concern in a reduced-for way, assuing that total advertising revenue is proportional to viewership. Let γ denote the size of the arket. Further, assue that the governent s preferences over investent and advertising revenue are separable, such that those preferences cases. 7 See, for exaple, At MSNBC, Tucker Is Out, and David Gregory Is In, New York Ties, March 11, For those who do not watch the news, the posterior belief is equal to the prior belief. 6

8 can be represented by the following utility function: U G = ψ 0 ω i (µ i ) E [π i (Ŝ (S) )] dµ i + γ 0 ω i (µ i ) dµ i. (1) The first ter is proportional to expected investent (recall that, by assuption, citizens choose to invest only if they watch the news, i.e., only if ω i = 1), with expectations over S, and the second ter is advertising revenue. The paraeter ψ denotes the obilizing character of the governent, i.e., the weight that the governent puts on investent relative to advertising revenue. 9 To establish intuition for the behavior of citizens and the governent, we begin by deriving equilibriu edia bias as ψ, i.e., when the governent values γ only obilization. We then exaine the general case where the governent values both obilization and advertising revenue. 2.2 Equilibriu When Governent Values Only Mobilization We solve for a perfect Bayesian equilibriu of this dynaic gae of incoplete inforation, focusing in this section on the special case where the governent values only the obilization of citizens to invest, i.e., to take costly actions that are not necessarily in their individual best interest. To begin, note that the governent wants citizens to believe that S = H, as it is profitable for citizens to invest only when the state is H. Therefore, in equilibriu, it ust be the case that β (H) = However, it cannot be the case in equilibriu that the edia outlet always reports that the state is H, i.e., that it also chooses β (L) = 1, as then the news would be uninforative. Regardless of the essage Ŝ, any citizen who watched the news would choose not to invest; given the opportunity cost of watching the news, all citizens would therefore choose not to watch. Thus, in equilibriu the edia ust truthfully report the state with soe positive probability when the state is L, i.e., the governent ust choose β (L) < 1. To solve for the equilibriu bias β (L), we begin by considering the beliefs and investent behavior of those who watch the news. Using the equilibriu condition that citizens update beliefs on the equilibriu path according to Bayes rule, we can derive the posterior probability that the state is H, conditional on having received the essage Ĥ, as ) θ (Ĥ = θβ (H) θβ (H) + (1 θ) β (L) = θ θ + (1 θ) β (L), (2) where the second equality follows fro β (H) = 1. The higher is edia bias β (L), the less likely citizens are to ) believe that the state is H when they receive the essage Ĥ. Siilarly, we can derive θ (ˆL = 0, which follows trivially fro β (H) = 1. Thus, citizens would never invest after receiving the report ˆL, but ) ight invest after receiving the essage Ĥ if β (L) is sufficiently sall such that θ (Ĥ r > 1. Intuitively, citizens invest after receiving the 9 Although atheatically superfluous in this setting, the use of two paraeters aids interpretation. In Sections 3 5, both paraeters are necessary. 10 This assues that essages have the natural eaning. Otherwise, the equilibria that we describe could be relabeled such that the essage Ĥ is associated with the state L, and vice versa. 7

9 essage Ĥ only if the governent is sufficiently likely to tell the truth when the state is in fact L. In the analysis to follow, we assue preliinarily that this is the case and then show that this condition holds in equilibriu. ) Given the preliinary assuption that θ (Ĥ r > 1, which iplies that citizens invest if (and only if) they receive the essage Ĥ, the expected benefit fro watching the news is [ ) ] [θ + (1 θ) β (L)] θ (Ĥ r 1. The ter on the left is the probability that a citizen receives the essage Ĥ, conditional on watching the news, whereas the ter on the right is the expected payoff fro investing, having received the essage Ĥ. This expression is decreasing in β (L): deand for the news is greatest when bias is least. With the opportunity cost of watching the news µ i distributed uniforly on [0, ], the ass of all individuals who watch the news is given by 0 ω i (µ i ) dµ i = 1 ) ] [θ (Ĥ [θ + (1 θ) β (L)] r 1. (3) Because citizens invest if and only if they receive the essage Ĥ, the probability that any citizen invests, conditional on having watched the news, is equal to the probability that the governent reports Ĥ, θ + (1 θ) β (L), which is increasing in edia bias β (L). Expected investent is then equal to the product of this probability and the ass of citizens who invest: )] ω i (µ i ) E [π i (Ŝ (S) dµ i = 1 [ ) ] [θ + (1 θ) β (L)]2 θ (Ĥ r 1. (4) 0 ) The governent chooses β (L) to axiize Expression 4, given the posterior belief θ (Ĥ (Equation 2): ax β(l) [ 1 [θ + (1 θ) β (L)]2 θ θ + (1 θ) β (L) r 1 This is a concave proble, the solution to which is [ ] β θ (r 2) (L) = ax 0, 2 (1 θ) Equation 5 says that when β (L) > 0, edia bias is greater, i.e., β (L) is larger, the larger is the payoff fro investent r. Intuitively, when the payoff fro aking the right investent decision is relatively large, citizens watch the news even when news content is noisy. In addition, edia bias is (weakly) increasing in θ, which easures the prior belief that the state is high. As Equation 2 shows, edia bias plays a saller role in deterination of the posterior belief that the state is high when θ is large, iplying that bias ust be greater to have the sae arginal ipact on investent. ]. (5) 8

10 ) In deriving Equation 5, we assued preliinarily that θ (Ĥ r > 1. We ay verify this ) ) by substituting β (L) into θ (Ĥ (Equation 2). In particular, θ (Ĥ = 1 when β (L) = 0 ) the essage (Ĥ Ĥ is sent if and only if S = H so that θ r > 1 holds trivially given r > 1, ) which is an assuption of the odel. For β (L) > 0, θ (Ĥ r > 1 is equivalent to θ θ + (1 θ) [ ]r > 1, θ(r 2) 2(1 θ) i.e., θ (r 2) θr > θ + 2θ (r 1) > θ (r 2). 2 Siilarly, we ay show that β (L) is strictly less than 1: θ (r 2) < 2 (1 θ) θr < 2, which always holds given the assuption that θr < Equilibriu When Governent Values Mobilization and Advertising Revenue In Section 2.2, we analyze the special case of the odel where the governent values only investent by citizens. This case illustrates the tradeoff between raising bias to increase investent by those who watch the news and lowering bias to increase the proportion of citizens who watch the news and thus receive the governent s essage. In addition to these concerns, however, governents ay be otivated to a greater or lesser degree to increase advertising revenue for edia under their control. In Russia, for exaple, Channel One and Rossiya, the two ost purely state-owned national television networks, run advertiseents during their prie-tie news broadcasts. Nonetheless, these advertiseents generally appear at the corners of the broadcasts where viewership is less. In contrast, NTV, the soewhat ore coercial network controlled by ajority-state-owned Gazpro, typically runs advertiseents in the iddle of its ain evening news broadcast, suggesting a greater ephasis on advertising revenue. 11 How does a concern for advertising revenue odify the results above? To answer this question, we exaine the general case of the odel, where the governent chooses edia bias β (L) to axiize utility as in Equation 1. Using Equations 3 and 4, we ay write this proble as ax β(l) ψ [θ + (1 θ) β (L)]2 where as before we assue preliinarily that θ (Ĥ [ ) ] θ (Ĥ r 1 + γ [ ) ] [θ + (1 θ) β (L)] θ (Ĥ r 1, ) r > 1. This is a concave proble, the solution to which we provide in the following proposition, where for future reference we use the subscript G to denote direct governent control. 11 The absolute value of advertising revenue is greater on Channel One, but this reflects its broader national distribution. 9

11 Proposition 1. Under direct governent control of the edia (i.e., state ownership), the equilibriu level of bias is [ ] βg ψθ (r 2) γ (L) = ax 0,. (6) 2ψ (1 θ) For βg (L) > 0, edia bias is greater when the governent has a particular interest in obilizing citizens to invest (i.e., when ψ is large), as during an election capaign when state-owned edia are used to increase support for governent-backed parties and candidates. More subtly, Equation 6 suggests that the ipact on βg (L) of a arginal increase in ψ is greater when the governent values advertising revenue ore to begin with (i.e., when ψ is low). 12 As we discuss below, this ay explain the particularly noticeable increase in bias on Gazpro-owned NTV in the run-up to the 2007 Russian parliaentary election. In addition, Equation 6 shows that edia bias is less, the greater is γ, i.e., the larger is the advertising arket. 13 This result supports the arguent that the edia ay be less biased when the advertising arket is large. As we discuss above, however, this arguent is typically ade in the context of private ownership of the edia, a case that we exaine below. Our odel shows that the sae relationship ay hold when the edia are state-owned, so long as the governent values advertising revenue fro edia that it owns. Intuitively, the equilibriu level of bias depends on the degree to which the governent internalizes the negative effect of bias on viewership. That effect is greater when the advertising arket is large, so long as the governent places soe value on advertising revenue. 3 Indirect Governent Control of the Media In the previous section, we assue that the governent has direct control over the news outlet. Even if the owner of the outlet is private, however, the governent ay be able to indirectly control news content, providing various induceents to encourage the private owner to bias coverage away fro the coercial optial editorial policy. Lawson (2002) reports, for exaple, that at the height of its control, Mexico s Institutional Revolutionary Party provided various benefits to private edia in return for favorable coverage, including tax privileges, subsidized newsprint, and cash payents to journalists. As a consequence, a plethora of pro-governent newspapers could operate without serious regard to circulation, coercial advertising, or other noral requisites of financial viability (Lawson, 2002, p. 32). Siilarly, McMillan and Zoido (2004) docuent the regular provision of cash payents and directed governent advertising to the edia by Peruvian secret-police chief Vladiiro Montesinos Torres. Notably, no such payents were necessary to the one state-owned television channel, as Montesinos had control over its content (McMillan and Zoido, 2004, p. 79). 12 To see this, note that for βg (L) > 0, β G (L) γ ψ = 2ψ 2 (1 θ), so that 2 β G (L) ψ = γ 2 ψ 3 (1 θ) < 0. ) 13 Note that because equilibriu bias is less when the governent values advertising revenue, ˆθ (Ĥ is larger ) than in the special case exained in the previous section. Thus, the preliinary assuption ˆθ (Ĥ r > 1 ust hold here, given that it does there. 10

12 Conceptually, the relationship between governent and private owner is analogous to a lobbying proble, though here the governent plays the role of lobby and private owner the role of policy aker. As is standard in the political-econoy literature, we odel the lobbying process process as a enu auction à la Grossan and Helpan (1994, 2001), where the lobby (here, governent) provides a contribution schedule that proises a particular contribution (subsidy) for every policy (bias) that the policy aker (private owner) could choose. Forally, assue that the governent has preferences over citizen investent and contributions represented by the utility function U G = ψ 0 ω i (µ i ) E [π i (Ŝ (S) )] dµ i C, (7) where the first ter is proportional to expected investent and C is a contribution, defined below, fro the governent to the private owner. In this context, the paraeter ψ easures the degree to which the governent values investent (obilization) relative to subsidies. Iplicitly, we assue that subsidies and advertising revenue are denoinated in the sae units, so that ψ has the sae eaning in Equations 1 and 7. Note that the governent does not directly value advertising revenue received by the private owner. The private owner, in contrast, has preferences over advertising revenue and contributions represented by the utility function U P = γ 0 ω i (µ i ) dµ i + C, (8) where the first ter is advertising revenue and C is the subsidy fro the governent to the private owner. Analogous to the governent s preferences, the private owner does not directly value citizen investent. At the beginning of the gae, the governent naes a contribution schedule C (β (S)) that proises a particular subsidy C 0 for all β (S) [0, 1] [0, 1], i.e., for any editorial policy that could be chosen by the private owner. As in Grossan and Helpan (1994, 2001), we assue that this proise is binding. This can easily be otivated either on reputational grounds or because lobbying is a spot-arket transaction with few dynaic considerations, siilar to the exchange of oney for goods in a retail environent. The private owner then chooses β (S) to axiize Expression 8. The reainder of the gae is identical to that in Section 2.1. Thus, the tiing of events is: 1. The governent naes a contribution schedule C (β (S)). 2. The private owner chooses an editorial policy β (S) [0, 1] [0, 1], which is observed by the governent and all citizens, and the governent pays C (β (S)). 3. Each citizen decides whether to watch the news, ω i {0, 1}. } 4. The state of the world S {L, H} is realized, with the essage {Ĥ, Ŝ ˆL deterined probabilistically according to β (S). Only citizens who watch the news receive the essage Ŝ. 5. Each citizen decides whether to invest in the project, π i {0, 1}. 11

13 Given that subsidies enter linearly into both the governent s and private owner s utility functions (i.e., because utility is freely transferable between the two actors), the equilibriu outcoe is jointly efficient between the governent and private owner, i.e., axiizes ψ 0 ω i (µ i ) E [π i (Ŝ (S) )] dµ i + γ 0 ω i (µ i ) dµ i. This is precisely the governent s axiization proble under direct control of the edia, iplying that the equilibriu level of bias is the sae as in Proposition 1. Intuitively, in aking its offer C (β (S)) to the private owner, the governent fully internalizes the ipact of bias on advertising revenue, as it ust copensate the private owner for any advertising revenue lost due to bias in reporting. The sharp prediction that edia bias is the sae under governent and private ownership a consequence of the Coase theore follows fro the assuption that the governent can costlessly transfer utility to the private owner. That assuption ay fail for various reasons. For exaple, because of political considerations the governent ay subsidize the private owner through non-onetary transfers, as when the governent provides transission frequencies to favored enterprises. 14 The opportunity cost to the treasury of providing those transfers ay be greater than the benefit to the private owner. Alternatively, as discussed above, subsidies ay actually be outright bribes to journalists and edia officials. In the latter case, the effort to keep bribes secret ight ipose transaction costs. We follow Besley and Prat (2006) in odeling these considerations in a reduced-for way, assuing that the private owner receives proportion 1 of any subsidy paid by the α governent, where the paraeter α Thus, the private owner s utility (Equation 8 above) now takes the for which is equivalent to U P = γ αγ 0 0 ω i (µ i ) dµ i + C α, ω i (µ i ) dµ i + C. In inducing β (L), the governent therefore puts greater weight on advertising revenue than in the case of direct governent control. The following proposition provides the optiu β (L) under indirect governent control of the edia, where the subscript P denotes private ownership. Proposition 2. Under indirect governent control of the edia (i.e., private ownership and state subsidies), the equilibriu level of bias is [ ] βp ψθ (r 2) αγ (L) = ax 0,. (9) 2ψ (1 θ) 14 In Russia, the allocation of spectru was instruental in the creation of the NTV television network during the 1990s, as well as in the consolidation of Krelin control of the edia during the Putin era. 15 Shleifer and Vishny (1994) also assue that subsidization is inefficient in their canonical odel of bargaining between a politician and a fir. 12

14 The equilibriu level of bias in Proposition 2 differs fro that in Proposition 1 in the ultiplier α on γ. Thus, to the extent that transaction costs prevent efficient bargaining between the governent and the private owner, bias will be less under private than state ownership of the edia. Propositions 1 and 2 show that a arginal change in ψ (the value the governent attaches to obilization) affects edia bias ore for private edia than for state-owned edia, so long as α > 1. Intuitively, the tradeoff between citizen investent and advertising revenue is starker for private edia, such that an increase in the value attached to investent is agnified relative to the case of state ownership. As we discuss below, this ay help explain the convergence in edia bias on private and state-owned edia in the run-up to the 1996 Russian presidential election. More generally, the odel predicts that private and stateowned edia will be siilarly biased in favor of the governent in states that attach great value to citizen obilization. As we show below, however, this does not iply a tolerance of private edia in obilizing states. Rather, the high cost of subsidization encourages governents to seize control of private edia for theselves. With respect to the second paraeter on which we focus, the difference between direct and indirect governent control of the edia is greater, the larger is the advertising arket (easured by γ). To see this clearly, focus on the case where ψθ (r 2) > αγ, so that edia bias is strictly positive even under private ownership. Then the additional bias under state ownership is ψθ (r 2) γ 2ψ (1 θ) ψθ (r 2) αγ 2ψ (1 θ) = γ (α 1) 2ψ (1 θ), which is increasing in γ. This has an iportant consequence for edia freedo: the opportunity cost to the governent of allowing private ownership of the edia, in ters of foregone citizen investent, is greater when the advertising arket is large. As we show in the following section, in such environents, the governent ay therefore be otivated to acquire direct control of the edia. We ay use Equation 9 to derive the subsidy the governent pays the private owner to represent its point of view. To do so, first note that were the private owner to reject the governent s offer, it would choose the level of edia bias ) that axiizes viewership, which is clearly β (L) = 0. Because the posterior belief θ (Ĥ = 1 when β (L) = 0, this iplies advertising revenue of γ [ ) ] [θ + (1 θ) β (L)] θ (Ĥ r 1 = In contrast, advertising revenue in equilibriu is γθ (r 1). (10) γ [ ) ] [θ + (1 θ) γθ (r 1) β P (L)] θ (Ĥ r 1 = γ (1 θ) β P (L), (11) ) where the equality uses the expression for the posterior belief θ (Ĥ given by Equation 2. Thus, the governent ust reiburse the private owner for lost advertising revenue, which is proportional to the equilibriu level of bias βp (L). Using Equation 9 and the assuption that the private owner receives proportion 1 of any subsidy paid by the governent, we can α 13

15 write this as ( ) γ (1 θ) β C (βp (L)) = α P (L) [ = ax 0, αγ ( ψθ (r 2) αγ 2ψ )]. (12) The governent subsidy is strictly increasing in the value the governent places on citizen investent (easured by ψ) for all βp (L) > 0. The greater the obilizing character of the governent, the greater the incentive to subsidize the private owner in return for favorable coverage. In contrast, the relationship between governent subsides and the size of the advertising arket (easured by γ) is nononotonic. To see this, note that an increase in the size of the advertising arket has two effects. First, as the advertising arket increases in size, the governent ust provide a larger subsidy for a given level of bias to copensate the private owner for lost revenue. Second, the governent responds to the higher cost of bias by inducing a saller β (L). The second effect doinates the first when advertising revenue is especially iportant. Indeed, for γ sufficiently large there is no bias, and thus no subsidy, in equilibriu. 4 Endogenous Control of the Media The discussion above treats control of the edia as exogenous. What does the odel say about the deterinants of edia control? To answer this question, first recall that equilibriu edia bias is the sae under state and private ownership when the governent ay costlessly subsidize the private owner. In this case, if we assue that the governent ust purchase a news outlet to acquire control, our odel offers no prediction about who owns the edia. In principle, acquiring direct control provides two benefits to the governent: it saves the cost of subsidization and acquires advertising revenue for itself. However, the private owner would accept nothing less than the value of governent subsidies and advertising revenue in return for relinquishing control, thus eliinating the incentive for the governent to pay for it. As we discuss above, however, various transaction costs ay prevent the governent fro costlessly subsidizing the private owner. In addition, the governent differs fro arket actors in a crucial respect: it can acquire direct control of the edia through force. Although this ay coe at soe cost in political capital and international reputation, that cost is arguably unrelated to the arket value of the news outlet. Forally, assue that the edia outlet is initially privately owned, but that at the beginning of the gae the governent ay transfer the edia outlet to state ownership at soe fixed cost κ > 0. (Alternatively, we ight think of κ as the cost of aintaining governent control, in which case the governent decides whether or not to privatize the edia outlet.) The gae then proceeds as in Section 2 or Section 3, depending on whether the governent has exercised this option. The governent acquires direct control if κ is sall relative to the benefit of taking over the edia outlet. To analyze this tradeoff, consider first the payoff to the governent fro direct control: ψ [θ + (1 θ) β G (L)] 2 [ ) ] [ γθ (r 1) θ G (Ĥ r 1 + γ (1 θ) β G (L) ]. (13) 14

16 The first ter is proportional to total expected investent, given that the governent chooses bias directly, whereas the second ) is equilibriu advertising revenue ) (Equation 11 above). We adopt the notation θ G (Ĥ to denote the posterior belief θ (Ĥ when β (L) = βg (L). In contrast, the payoff to the governent fro indirect control is ψ [ ) ] [θ + (1 θ) β P (L)] 2 θ P (Ĥ r 1 αγ (1 θ) β P (L). (14) The first ter is proportional to total expected investent, given that the governent ust subsidize the private owner to induce bias, whereas the second is the cost of the) subsidy to the governent (Equation 12 above). Analogously, we adopt the notation θ P (Ĥ to denote ) the posterior belief θ (Ĥ when β (L) = βp (L). The additional benefit to the governent fro acquiring direct control is the difference between Equations 13 and 14. To fix ideas, focus on the case where ψθ (r 2) > αγ, which iplies that both βg (L) and β P (L) are strictly greater than zero. Then the additional benefit to the governent of direct control is ( α 2 1 ) [ ] γ 2 γθ (r 1) ψθ (r 2) αγ ψθ (r 2) γ + + αγ γ. 16 (15) 4ψ 2ψ 2ψ The first ter in this expression is the additional investent under state ownership, which results fro the higher level of bias when the governent chooses bias directly. The second ter is total advertising revenue when the news is reported without bias. In equilibriu under private ownership, soe portion of this advertising revenue is replaced with a governent subsidy. Acquiring the news outlet saves the governent that subsidy while providing the share of advertising revenue that reains. The third ter reflects the eliination under state ownership of transaction costs associated with copensating the private owner for lost advertising revenue. The governent chooses to take direct control of the edia outlet when Expression 15 is large relative to κ. We are interested especially in how the incentive to nationalize the edia depends on the obilizing character of the governent (easured by ψ) and the size of the advertising arket (easured by γ). Consider first the ipact of a arginal increase in ψ. As we discuss in the previous section, the additional bias under state ownership diinishes as the governent values investent ore. Thus, the advantage to the governent of direct control for the sake of investent (the first ter in Expression 15) is saller when ψ is large. However, the governent also values direct control for the subsidy it saves in ipleenting its desired level of bias, and this benefit is larger when ψ is large. The following proposition establishes that the second effect outweighs the first, and ore generally establishes that the incentive for the governent to seize direct control of the edia is increasing in ψ, so long as the governent optially chooses bias greater than zero when the edia are state-owned. 16 To see this, note that total investent under state ownership is (ψθr)2 γ 2, whereas total investent under 4ψ private ownership is (ψθr)2 (αγ) 2 4ψ ; that advertising revenue under state ownership is γθ(r 1) γ ψθ(r 2) γ 2ψ ; and that the subsidy under private ownership is αγ ψθ(r 2) αγ 2ψ. 15

17 Proposition 3. If βg (L) > 0, a arginal increase in ψ, which easures the obilizing character of the governent, (generically) increases the incentive for the governent to acquire direct control of the edia. If βg (L) = 0, a arginal increase in ψ (generically) has no ipact on the incentive for the governent to acquire direct control of the edia. Proof. Consider the following three utually exclusive and exhaustive cases: 1. βg (L) > 0 and β P (L) > 0: The derivative of Expression 15 with respect to ψ is ( α 2 1 ) γ 2 which is greater than zero for α > 1. 4ψ 2 + ( α 2 1 ) γ 2 2ψ 2, 2. β G (L) > 0 and β P (L) = 0: Evaluating Equation 13 at β G = ψθ(r 2) γ 2ψ(1 θ) gives (ψθr) 2 γ 2 4ψ + γθ (r 1) γ ψθ (r 2) γ. 2ψ Siilarly, evaluating Expression 14 at βp benefit fro direct control is therefore (ψθr) 2 γ 2 + 4ψ γθ (r 1) (L) = 0 gives ψθ2 ψθ (r 2) γ γ 2ψ The derivative of this expression with respect to ψ is (θr) γ2 4ψ 2 ψθ2 γ2 2ψ 2 θ2 (r 1), (r 1). The additional (r 1). which ( is greater than zero ) if ψθ (r 2) > γ, which is a preise of the proposition βg ψθ(r 2) γ (L) = > 0. 2ψ(1 θ) 3. βg (L) = β P (L) = 0: The additional benefit of direct control is equal to total advertising revenue, γθ(r 1), which is constant in ψ. Now consider the effect of an increase in γ on the governent incentive to acquire control. If α = 1, so that there are no transaction costs associated with subsidization, then the incentive for the governent to acquire control is unabiguously greater when γ is large. Although bias is identical to that under private ownership, the governent can save the cost of subsidization and acquire advertising revenue for itself. More generally, for all α 1, Expression 15 is increasing in γ. Indeed, as the following proposition establishes, the incentive for the governent to acquire direct control of the edia is increasing in the size of the advertising arket, even when there is no edia bias under private and/or state ownership. Proposition 4. The incentive for the governent to acquire direct control of the edia is increasing in the size of the advertising arket (easured by γ). 16

18 Proof. Consider the following three utually exclusive and exhaustive cases: 1. βg (L) > 0 and β P (L) > 0: The derivative of Expression 15 with respect to γ is γ α2 1 2ψ + θ (r 1) + αψθ (r 2) 2α2 γ 2ψ ψθ (r 2) 2γ 2ψ For α = 1, this expression is clearly greater than zero. For α > 1, this expression greater than zero for all γ ( 1 0, [2ψθ (r 1) + (α 1) ψθ (r 2)]) and therefore ( ) α 2 1 for all γ, i.e., for all γ such that βg (L) > 0 and β P (L) > 0. 0, ψθ(r 2) α 2. βg (L) > 0 and β P (L) = 0: The payoff to the governent fro indirect control is constant in γ, so the derivative of the additional benefit of direct control with respect to γ is equal to the derivative of Expression 13 with respect to γ. Evaluating Expression 13 at βg = ψθ(r 2) γ gives 2ψ(1 θ) (ψθr) 2 γ 2 4ψ + γθ (r 1) γ ψθ (r 2) γ. 2ψ The derivative of this expression with respect to γ is which is greater than zero. γ 2ψ + θ (r 1) ψθ (r 2) 2γ, 2ψ 3. βg (L) = β P (L) = 0: The additional benefit of direct control is equal to total advertising revenue, γθ(r 1), which is increasing in γ. The surprising iplication of this analysis is that the relationship between the size of the advertising arket and edia freedo ay be nononotonic. Holding ownership constant, growth in the advertising arket reduces edia bias, regardless of whether the news outlet is owned by the state (so long as the governent places any value on advertising revenue) or a private entity. However, the sae growth ay propt the state to seize direct control of the edia, which leads to a discontinuous jup in edia bias. As we discuss below, such as effect ay have played a role in encouraging Vladiir Putin s Krelin to seize control of Russian edia just as the advertising arket was beginning to provide independent edia with a easure of financial security. 5 Multiple Media Outlets The analysis above considers the case of one edia outlet. To what extent do our results hold when there are ultiple edia outlets? To explore these issues, we ay assue that there are J edia outlets indexed by j, which for concreteness we refer to as stations. The 17

19 J stations siultaneously and publicly choose an editorial policy β j (L) [0, 1]; assue for siplicity that β j (H) = 1 for all j. The process by which these policies are chosen depends on ownership of the station, as in Sections 2 and 3. Following choice of editorial policy, each citizen i decides whether to watch each station j, ω ji {0, 1}. Each citizen i is associated with a vector (µ 1i, µ 2i,..., µ Ji ), where µ ji [0, ] is citizen i s opportunity cost of watching station j. The rest of the gae proceeds as before. Although it is difficult to derive general results for this extended odel, two special (and polar) cases are particularly inforative. First, assue that each station has coplete arket power, with the arket segented aong the J stations. Intuitively, we ay think of broadcast edia whose transission networks do not overlap. Forally, this corresponds to the case where the population of citizens is partitioned into J sets, where citizen i in group j has opportunity cost µ ki = for all k j; we assue that µ ji is distributed uniforly on [0, ] for citizens in group j. Clearly, viewership for any one station is unaffected by that for the others, so this case is analogous to the baseline odel: the equilibriu level of bias is the sae as that in Propositions 1 and 2, and the incentives for the governent to nationalize private edia are precisely those in Section 4. Second, assue that fro any citizen s perspective, stations are perfect substitutes for each other, and that each citizen ay watch no ore than one station. Intuitively, this corresponds to an environent with national broadcast edia that broadcast the news at the sae tie of the day. Forally, we assue that the vector of opportunity costs is distributed uniforly along the ain diagonal of a J-diensional cube, and we ipose the additional constraint that each citizen ay watch only one station. In addition, we assue that citizens choose a station at rando if indifferent aong stations that provide the ost inforation, and that each station is characterized by the sae transaction cost α (if private) and size of advertising arket γ. If there is at least one private station, it ust be the case that there is a coon equilibriu bias β (L) for all stations: given that stations are perfect substitutes for each other, any station that ipleented a β (L) higher than soe other station would have no viewers, iplying that the owner of that station could profitably deviate to the lowest β (L) chosen by the others. 17 Moreover, if β (L) > 0, then the governent ust guarantee any private owner total potential advertising revenue ( γθ(r 1), given by Equation 10), as any private owner could choose soe β (L) < β (L) and capture the entire advertising arket for itself. As Besley and Prat (2006) show, the fact that the governent ust provide each private owner with total potential advertising revenue in return for edia bias iplies that the governent ay instead choose to induce no bias. To see this forally, assue that all stations are private and that there are no transaction 17 In principle, the governent could induce a private station to choose an editorial policy that resulted in no viewers by reibursing it for lost advertising revenue. However, it would have no incentive to do so, as it could save the cost of that subsidy and leave investent unaffected by allowing the station to atch the lowest level of bias chosen by the other stations. Note that if all stations are state-owned, then it need not be the case that stations choose the sae β (L) in equilibriu. Even though stations that choose a higher β (L) have zero viewership, these viewers watch soe other state-owned station, and by assuption the governent is indifferent to the allocation of viewers across state-owned stations. 18

20 costs fro bargaining (i.e., α = 1). Then the governent solves ax β(l) s.t. ψ [ ) ] [θ + (1 θ) β (L)]2 θ (Ĥ r 1 j γ [ J [θ + (1 θ) β (L)] θ (Ĥ ) ] r 1 + C j C j γθ (r 1), j. The governent chooses a coon editorial policy to axiize investent less the cost of subsidies, given the constraint that each private owner be left with at least total potential advertising revenue. (Because each station chooses the sae editorial policy, advertising revenue is divided equally aong the J stations.) But for the fact that there are J constraints, this is precisely the governent s proble in the baseline odel: the ipact of a arginal increase in bias on advertising revenue is the sae as before. Thus, if the governent chooses to induce a positive level of bias, it will be identical to that when there is only one station. However, the governent ay instead choose β (L) = 0, given the cost of inducing stations to aintain a positive level of bias. In particular, if β (L) > 0, then each private owner ust be provided with a subsidy Cj (β γθ (r 1) (L)) = γ [ ) ] J [θ + (1 θ) β (L)] θ (Ĥ r 1, ) where θ (Ĥ is the posterior belief when β (L) = β (L). This iplies total subsidies of ( ) γθ (r 1) Cj (β (L)) = J γ [θ + (1 θ) β (L)] j [ ) ] θ (Ĥ r 1. Clearly, for J arbitrarily large, the cost of subsidizing private stations outweighs any benefit to the governent fro edia bias, iplying that the governent prefers to ipleent β (L) = 0 and pay no subsidies. Iportantly, this effect is greater, the larger is γ. Thus, as in the baseline odel, the cost to the governent of allowing private ownership is greater, the larger is the advertising arket. In suary, the key arguents of this paper hold when there are ultiple edia outlets, at least for the iportant special cases where edia outlets have coplete arket power and (conversely) are perfect substitutes for each other. 6 Illustration: Media Freedo in Postcounist Russia In this section, we briefly discuss the evolution of edia freedo in postcounist Russia. Our ai is not to provide a full history of the role of the edia in Russian politics, a task that others have adirably undertaken. 18 Rather, it is to show how our odel can help to illuinate an iportant case, and so to illustrate the contribution of our theoretical fraework. As we deonstrate, Russia is a particularly useful case to analyze, as edia 18 See especially Mickiewicz (1999) and Oates (2006). 19