Media Mergers and Media Bias. Simon P. Anderson1 and John McLaren2. June 2004 3. Preliminary and Incomplete: Comments highly welcome. ABSTRACT.
Media Mergers and Media Bias Simon P. Anderson1 and John McLaren2 June 2004
Preliminary and Incomplete: Comments highly welcome. ABSTRACT Rational agents will discount information that they know comes from a biased source. This suggests that they cannot be swayed by reported information they know to be biased. However, the capacity for politically biased news organizations to mislead rational agents is restored if consumers are not sure about how much hard information the news organization has, and how much it may be concealing. We use this framework to contribute to the current debate on restrictions of ownership of media. Preliminary results suggest that ownership restrictions may not be needed if the owners of news channels have strong and diverse political motives quite diﬀerent from the general citizenry - for then there would be no mergers contemplated. However, strong political motives that are not too dissimilar from those of the citizenry may lead to mergers that restrict the information divulged, and thus might be proscribed. If political motives are weak, then the merged entity will want to commit to impartial reporting, so that mergers are not harmful to information dissemination. KEY WORDS: Information withholding, market for news, media bias, media mergers. JEL ClassiÞcation: D23, L82 1
114 Rouss Hall, P.O. Box 400182, Department of Economics, University of Virginia, Charlottesville, VA 22904-4182, USA. 2 114 Rouss Hall, P.O. Box 400182, Department of Economics, University of Virginia, Charlottesville, VA 22904-4182, USA. 3 We would like to thank David Ettinger for discussion. The Þrst author gratefully acknowledges funding assistance from the NSF under Grant SES-0137001.
Many stories attest to the power of the Press in shaping political outcomes through the news reported. For example, Allen (1983) shows how a German town was converted to Nazism before the Second World War with the aid of the local newspaper. Today, many media are owned by politically oriented entities. The Italian Premier, Silvio Berlusconi, is a television magnate. The Bouygues group in France, a big telecommunications Þrm with other diverse business interests, is the principal shareholder in the top TV channel, TF1. The French group Dassault (an airplane manufacturer whose CEO is a politician) owns several French newspapers, including “Le Figaro,” “L’express,” and numerous regional titles. Media mergers are an important policy question and the source of heated debate given the FCC’s recent attempt to pass relaxed rules for media Þrms. Some critics argued that relaxed rules would result in reduced diversity of political views available to the public. Unfortunately, analytical economic tools have not been put to much use on this question, which oﬀers substantive theoretical questions and important policy ramiÞcations. This paper studies the question of whether media mergers can result in political distortions in the information that citizens receive and in manipulation of political outcomes. Consistent with standard economic theory, we analyze a model of rational citizens who understand and fully anticipate the bias of the media and Þlter the information received from it accordingly. At the heart of our argument is the idea that consumers of news media do not know how much information is possessed by a given news organization at any time, and so if there is a lack of news coming from the organization that is pejorative to the view of that organization’s ownership, citizens do not know whether that is because of a genuine lack of information or because information is being withheld. This prevents the familiar ‘unraveling’ observed in other models, such as the Milgrom (1981) ‘persuasion game.’ In that game, a sender with private information can send information to or withhold it from a receiver in order to induce the receiver to undertake some action. Because the receiver understands the sender’s preferences, she understands that the sender will send only the information most favorable to his case; in this way, the receiver can deduce all of the sender’s private information in equilibrium 2
(see also the ‘credible claims’ model of Lipman and Seppi (1995)). In the model we present, uncertainty about how much information the sender (the news organization) has will prevent complete deduction of the sender’s information, with the result that who owns the news organization can make a diﬀerence to what the public learns in equilibrium. In order to address these issues, we need a model with several elements. First, there must be some variable, x, whose true value is not known to the public and that is relevant to voting. This could be, for example, the competence or integrity of a particular politician, the true state of the economy, the Þnancial health of social security, the true situation on the ground in Iraq, and so on. Second, it must be possible that veriÞable, documented information uncovered by a news organization can reveal information about the true value of x.4 For simplicity, we will assume that either the news organization uncovers information that can publicly prove the value of x, or it uncovers nothing. Third, there must be a public-sector decision that will be aﬀected by the public’s beliefs about the value of x. Assume that this public sector decision is determined by majority voting (which adds no complication because all voters will be assumed identical). Fourth, in order to explain the existence of private-sector news organizations, there must be a market demand for news. This is tricky, because news naturally has a public-good quality: unless a citizen expects to be a pivotal voter, which is eﬀectively a zero probability event, becoming a more informed voter yields a negligible payoﬀ in the form of improved electoral outcomes. Thus, we need a device to explain why consumers will pay a positive price for a newspaper (or spend valuable time watching the news on television). One route is to assume some entertainment value to news.5 The assumption 4
The most closely related paper, Stromberg (2001), has a slightly diﬀerent informational role for the press. In that model, the press can communicate the policy stands of politicians to the electorate, rather than states of nature. 5 A similar device is used in Stromberg (2001). It presents no diﬃculty to allow papers to have some entertainment value. For example, we could assume an entertainment value of E which is the same for all papers and whose value is independent of the other paper(s) bought. The value E is then simply added to the equilibrium values of prices derived below.
made here is that there are private decisions made by each consumer that can be better informed by use of information on x. Further, we assume that one must purchase a newspaper in order to learn what information its publisher is making public about x. (One might hear informally about a story in the paper from friends, but it is necessary to purchase the paper and read the story carefully in order to understand the information.) For example, perhaps x is the state of the social security program and the private decision is a decision about retirement planning. Alternatively, x could be the state of terrorist threats, and the private decision concerns travel plans, or x could be the health of the public school system and the private decision is the choice of private school or residential location. The point is that a desire to learn about x in order to make a more informed private decision generates a market demand for news, and this then through the voting system aﬀects the direction of the public decision. For concreteness, we assume that all news is propagated by newspapers alone, and that newspapers generate revenue only by the purchase price (of course, neither assumption is realistic). The next section sets out the model. Section 3 determines the information that news organizations will reveal in equilibrium, and what readers infer when nothing is revealed. Section 4 compares equilibrium prices under the diﬀerent market structures (monopoly and duopoly). Section 5 considers when mergers will arise, what the implications are for information dissemination, and the desirable merger policy stance in each case. Section 6 suggests some further directions for research in this vein.
Let x ∈ [0, 1] represent the variable whose true value is not known to the public. Its value is important to individuals in their private choices, and also for their voting choice. Let the exogenous common prior distribution for x be given by a density f and its associated R 1 cumulative distribution function F , both deÞned on [0, 1]. Denote by ρ ≡ 0 xf(x)dx the ex ante mean value R1 of x, and denote by σ 2 ≡ 0 (x − ρ)2 f(x)dx the ex ante variance of x. Let π be the probability that the news organization uncovers proof of the true value of x. We assume that this is a positive constant that is the same for 4
all news organizations, and that information discovery is perfectly correlated for all active news organizations. Let the public sector decision be denoted d pub , and assume that it can take the value 0 or 1. Denote the private decision d priv ∈ [0, 1]. The typical citizen’s preferences are summarized by the following utility function: U
= −α1 (x − dpriv )2 + α2 (x − β)dpub −
pi ni ,
where αi > 0 and β ∈ [0, 1] are constants; pi is the price of newspaper i; and ni is a dummy variable indicating purchase of newspaper i. Clearly, if the citizen knew the value of x, she would want to set dpriv = x. If x > β, the citizen would prefer that the political process set d pub equal to 1 while if x < β, the citizen would prefer that d pub be set equal to zero. More generally, if the posterior Bayesian mean for x is greater (less) than β, the voters will prefer a value of d pub equal to 1 (equal to zero). We normalize the population size to unity. The usual economic objections to monopoly do not apply in this model. This is because all consumers of news are identical, and under a news monopoly each decides simply to buy or not buy the one available newspaper. Without a downward-sloping demand curve, there can be no conventional deadweight loss from monopoly. Thus, the usual economic analysis of antitrust is not relevant. However, we shall see that a new political-economic rationale for antitrust, based on the political manipulation of information, can arise. Suppose that there are two possible publishers, labelled A and B. We write the utility function of publisher i as: U i = αi (x − β i )dpub + pi ni ,
i = A, B,
where αi > 0, β A = 0, and β B = 1. The Þrst term represents the publisher’s interest in the public-policy outcome, and the second represents his or her proÞts. (For simplicity, we ignore both the publisher’s private decision, production costs, and distribution costs, as they have no role in what follows.) Clearly, publisher A would like to see the public decision take the value of 1, regardless of the available information about x, while B would like to see 5
it set equal to 0, regardless of information. The αi parameter measures the strength of this political motive relative to the proÞt motive in determining the behavior of news publishers. All of these parameters are common knowledge. This is important, because it means that consumers of the news can take into account the political motivations of the publishers of the news in deciding which news sources to use. Initially, assume that there is no way a publisher can commit publicly and credibly to non-interference in the operation of his or her news organization. Later, this will be relaxed. For the moment, we will take the structure of the industry as given. This is either a monopoly by publisher A, a monopoly by publisher B, or competition between the two. The choice between these three will later be endogenized. It will be convenient to denote the structure of the industry by a variable s taking the values A, B and C, representing these three structures respectively. The sequence of events is as follows. Each publisher in the market chooses its price pi (simultaneously if they are both functioning), then the state x is either revealed to all publishers in the market (with probability π) or is not revealed to them (with probability (1−π)). In the event that x has been revealed, each publisher then decides whether to print the information or to withhold it. Each consumer then, knowing the biases of the publishers and the prices they are charging but not the content of the newspapers, decides whether or not to purchase a copy of each available newspaper. The Bayesian prior on x is updated with any information revealed in the papers, consumers vote on dpub , and they make their decisions on dpriv . Utilities are then realized. We will Þrst study equilibrium information management for this model, for each exogenous market structure, and then equilibrium pricing. Because of the structure of preferences, for the case of exogenous market structures these can be dealt with separately. The only point to note about the interaction is that given homogeneous consumers and zero production costs for newspapers, prices will be set so that every consumer will purchase a copy of every newspaper available, and so all information printed in any newspaper will go to all consumers. Afterward, we will discuss endogenous market structure.
Equilibrium information management
Each publisher has a very simple decision to make regarding news management. If that publisher receives information on the value of x, the decision is whether to publish that information in the pages of the newspaper controlled by that publisher, or to keep it quiet. It is worth recalling that it is not possible to falsify news, only (sometimes) to hide it; therefore, if a particular value of x is published, readers will accept it as true. It is convenient to deÞne the following notation. For a given market structure s, let g(x; s, β) denote the Bayesian posterior density for x, conditional on no news being published regarding x, where we recall that β denotes the reader’s ideal outcome of x. Let the associated cumulative distribution function be denoted G(x; s, β). In addition, for variables, we will use a tilde to denote the value conditional on no news, and a double tilde to denote the value conditional on the publication of some concrete news (but not conditional on the actual value revealed!) Thus, recalling that ρ denotes the ex ante mean value of x, denote by e ρ(s, β) the mean value of x, conditional on no news being published regarding x, and denote by e e ρ(s, β) the mean value of x, conditional on some concrete news being published regarding x. Similarly, denote by σ e2 (s, β) the variance of x, conditional on no news being published 2 e regarding x, and denote by σ e (s, β) the variance of x, conditional on some 2 e e (s, β) represents the concrete news being published regarding x. The term σ “amount” of information revealed by printing the exact value of x, beyond the mere knowledge that x is in the range that the publisher would print it. This is useful below for evaluating the value of buying a paper.
Competitive news production
Initially, suppose that both A and B operate. In this case, of course, A (who would like the public decision to be 1) will trumpet any information revealing that x > β, while B will bandy any information revealing that x < β. Therefore, since we assume that any news is available to both publishers, all of the information will be revealed, and if there is no hard evidence published 7
either way, the public will know that the reason is that such evidence is not available.6 Thus, in the notation above, ρ(C, β) = e ρ(C, β) = e e ρ(C, β), σ 2 (C, β) = 2 e σ e2 (C, β) = σ e (C, β), and g(x; C, β) = f(x)∀x.
Monopoly news production
Now, suppose that publisher B has been shut down, leaving A as the monopoly news source. Clearly, A would like to convince the electorate that x > β if it is possible to do so, in order to motivate voters to choose dpub = 1. Therefore, if in truth the x variable is greater than β, and if the news organization owned by A Þnds proof of this fact, then it will publish it. This will result in the electorate being certain that x > β, and selection of d pub = 1 by the political process. On the other hand, suppose that x < β, and the news organization owned by A Þnds proof of this fact. In that case, it will withhold the information to leave some doubt in the mind of the electorate. Therefore, news consumers will see no hard information regarding x in the pages of the A newspaper.7 On this basis, they derive their Bayesian ex post distribution for x. There are two possible reasons for the absence of news. Either no news was discovered (an 6
There are also other Nash equilibria to this game. For example, if A is expected to reveal the value of x no matter what it may be, then B will be unable to manipulate public opinion, and will be indiﬀerent between all available strategy choices. Thus, it is a Nash equilibrium for both publishers to reveal all information. However, revelation of information about x that is prejudicial to one’s own preferences regarding dpub is a weakly dominated strategy, and we eliminate such strategies in the equilibrium discussed here. 7
We assume that the news organizations always hide the news unfavorable to their cause. However, there are cases below in which the public decision that is made is unfavorable to them even when they withhold information. In such cases, they are indiﬀerent as to revealing the unfavorable news or not, and, without further restriction, there are multiple equilibria, with the paper revealing any subset of information (consistent with the public decision still being averse). We rule these out by assuming that the news organization holds back unfavorable news. This is the only equilibrium when there is some possibility that voters may waver when not sure of the truth, but are always dogmatic when faced with the truth.
event with a probability of (1 − π)), or else news was discovered but is being withheld. Given the known bias of publisher A to withhold information that x < β the combined probability of these events is ν(A; β) ≡ 1 − π + πF (β). For a value x0 ≤ β, this implies that the probability that x < x0 , conditional on no news, is equal to: G(x0 ; A, β) =
F (x0 ) , ν(A; β)
and for a value x0 > β, the corresponding probability is equal to: G(x0 ; A, β) =
πF (β) + (1 − π)F (x0 ) . ν(A; β)
This implies the Bayesian posterior density is: f (x0 ) ν(A; β) (1 − π)f (x0 ) = ν(A; β)
g(x0 ; A, β) =
if x0 ≤ β; if x0 > β.
It is straightforward to verify that G(x; A, β) > F (x)∀x, so that e ρ(A, β) < ρ. This is the suspicion eﬀect, which works against publisher A’s interests. News consumers always know that A withholds news that cuts against his or her interests. When there is no news reported of a sort that decisively aﬀects public policy debates, people rationally wonder if something might be being hidden from them, and they shade their posterior probabilities accordingly. At the same time, it is easy to see that e ρ(A, β) → ρ as β → 0 and as β → 1. The former case is one in which the public’s preferences are similar to the monopoly publisher’s, so that only in rare events (when x is between zero and β) would the publisher withhold information. Consequently, when β is small, the suspicion eﬀect is weak. The latter case is when the public’s preferences are extremely diﬀerent from those of the monopoly publisher. As a result, it is a rare event when the publisher does not withhold information (that is, when x is between β and 1). Then the public expects the newspaper to be uninformative, so when they read it and see that it is uninformative, not much is deduced from that fact. Thus, in this case as well, paradoxically, the 9
suspicion eﬀect is weak.8 The eﬀect is at its strongest in cases in which the public and the publisher have an intermediate degree of divergence in their preferences. This is illustrated in Figure 1. The publisher has considerable power to mold public opinion due to her ability to withhold information, but because of the rationality of consumers, the monopoly position also comes with the liability that is the suspicion eﬀect. In some instances the latter eﬀect is strong enough that the monopoly power is detrimental to the publisher who holds it. In order to see this, we need to know some properties of the e ρ(A, β) function.
First, it is useful to have the derivative of the function with respect to the parameter β. Recalling that ¶ µZ β Z 1 1 e ρ(A, β) = xf (x)dx + (1 − π) xf(x)dx , ν(A; β) 0 β
with ν(A; β) = 1 − π + πF (β), then
πf(β) ∂ e ρ(A, β) = [β − e ρ(A, β)] . ∂β ν(A, β)
This yields the following result.
Proposition 1 There is a unique value β ∈ (0, ρ) such that β < β implies that e ρ(A, β) > β and β > β implies that e ρ(A, β) < β.
Proof. We know that e ρ(A, 0) = ρ > 0 and e ρ(A, 1) = ρ < 1. Therefore, by continuity of e ρ(A, β), there exists at least one β such that e ρ(A, β) = β. Furthermore, by (2), the function ρ ˜ (.) is decreasing for ρ˜ < β, and increasing for ρ˜ > β, with a zero derivative where ρ ˜ < β. (Think by analogy of the behavior of average costs when marginal cost is rising, with here β playing the role of marginal cost and ρ˜ the role of average cost.) Hence ρ˜ falls initially until it reaches the 45-degree line (see Figure 1), which it crosses with zero 8
The suspicion itself is strong, but its eﬀect is weak because there is little updating of priors.
slope, and then rises without further crossings (since it cannot again satisfy the condition of zero slope at another crossing). This means that the solution, β, is unique and moreover ρ˜ < ρ for all β ∈ (0, 1). In particular, β < ρ. These properties imply that if publisher A’s preferences are not too far from those of the general public (in other words, if β ∈ [0, β)), the political outcome in the event that no news is published is dpub = 1, while if publisher A’s preferences are far from the mainstream (β ∈ (β, 1]), the outcome that ensues following silence is dpub = 0. The former regime is the one in which the public can be successfully manipulated; in the latter regime it cannot. There are two sub-cases to the latter regime, so consider the three cases illustrated by Figure 1. Case I: 0 < β < β. In this case, if voters received no hard news, they would vote for d pub = 1 (since e ρ > β). Thus, we have d pub = 1 with probability 1. In this case, monopoly is of clear political beneÞt to publisher A. Case II: β < β < ρ. In this case, the suspicion eﬀect is weak enough that when voters receive no hard news, they vote for d pub = 0 (since e ρ < β). pub Thus, if x is revealed to be high, we will have d = 1 and if x is revealed to pub be low we will have d = 0 (publisher A will withhold the information but the outcome will still be dpub = 0, since e ρ < β). Thus, in the event that the publisher learns hard information, the outcome is the same as it would have been under competition. On the other hand, in the event that A does not Þnd hard information about x, the suspicion eﬀect leads to dpub = 0, while in the same event under competition it would have led to dpub = 1 (since e ρ < β < ρ). Thus, as regards political outcomes, A is now worse oﬀ under monopoly than under competition. Case III: ρ < β < 1. In this case, the outcome of the political process is exactly the same as it would have been under competition. Voters choose dpub = 0 unless A Þnds hard evidence that x > β.
Clearly, in Case I, A receives a political advantage from possession of a news monopoly, and would be willing to pay something in order to enjoy that situation. This is true despite the full rationality of the public, and its knowledge of the intentions and bias of the publisher. The point is that the power to truncate the information available to the public results in an 11
eﬀect on their decisions in the worst-case situations. On the other hand, in Case II, A would be better oﬀ politically by forfeiting the monopoly. This is the case in which the suspicions of the rational public undo the political intentions of the monopolistic publisher. Note that this is the case in which the public’s tastes are farther from those of A. If the publisher’s tastes are extremely diﬀerent from popular tastes, as in Case III, the monopoly position will make no diﬀerence to the outcome. Thus, the case in which the monopoly position is most useful to the publisher is that in which his or her tastes are most similar to the population as a whole. If they are very dissimilar, no manipulation is possible. If, though, they are moderately diﬀerent, a news monopoly will be politically disadvantageous. The case of a monopoly with publisher B is analogous. In that case, information that x is above β would be withheld. The analogous posterior distribution conditional on no news being published is given by: (1 − π)F (x) ν(B, β) F (x) − πF (β) = ν(B, β)
G(x; B, β) =
(1 − π)f(x) ν(B, β) f(x) = ν(B, β)
g(x; B, β) =
if x < β; if x > β, and
if x < β; if x > β,
where ν(B, β) = 1 − πF (β) is the probability that no news is published. The suspicion eﬀect implies that e ρ(B, β) > ρ, and e ρ(B, β) reaches its maximum at a value β = β > ρ. The picture corresponding to Figure 1 then has e ρ(B, β) rising from e ρ(B, 0) = ρ till it reaches the 45 degree line at β > ρ > β, then falling back down to reach e ρ(B, 1) = ρ.
We will Þrst analyze monopoly pricing, and then move on to duopoly. Throughout, we will denote the equilibrium price of a copy of newspaper i under market structure s and reader preference parameter β by Pi (s, β). Again, assume that publisher A has a monopoly on the news. A news monopolist will charge the highest price consumers are willing to pay, which is of course equal to the expected utility the consumer receives from the information contained in the paper. The only beneÞt for an individual from buying a newspaper is in improving the quality of the consumer’s decisionmaking regarding the private decision, dpriv . From (1), the utility deriving from the private decision is equal to: E[−α1 (x − dpriv )2 |I], where I denotes all the information available to the consumer at the time the decision is made. The Þrst-order condition for this is simply dpriv = E[x|I], so the utility becomes: −α1 σ 2 (I),
where σ 2 (I) denotes the expected variance of x conditional on information I. Thus, the information in the newspaper is useful only to the extent that it reduces the conditional variance of x. In the event that the consumer purchases no newspaper, the decision on dpriv must be made with no information about x, resulting in utility −α1 σ 2 . In the event that the consumer decides to purchase the newspaper, there are two possible outcomes. There may be no relevant news reported in it, in which case the private decision must be made with an ex post variance for x equal to σ e2 (A), yielding utility of −α1 σ e2 (A). This occurs with probability ν(A, β). On the other hand, there may be news about x in the paper, in which case the value of x is known precisely. This results in utility of zero. Consequently, the expected utility from the private decision when the consumer has chosen to purchase a copy of the paper is equal to −ν(A, β)α1 σ e2 (A). Given that the publisher in this situation will set the price so as to extract all of the surplus, we have the following result. 13
Proposition 2 Assume that β ∈ (0, 1). The monopoly equilibrium price of the A newspaper is strictly positive and given by σ 2 (A, β)]. PA (A, β) = α1 [σ 2 − ν(A, β)e
The monopoly equilibrium price of the B newspaper is strictly positive and given by (4) PB (B, β) = α1 [σ 2 − ν(B, β)e σ 2 (B, β)]. Proof. The argument above establishes that these are the appropriate expressions. It remains to show that it must be positive. Consider paper A as a benchmark. The variance of x with no information can be written as follows: σ 2 ≡ E[(x − ρ)2 ] = ν(A, β)E[(x − ρ)2 |NN] + (1 − ν(A, β))E[(x − ρ)2 |N], where N N denotes the event that A does not publish any news about x (either because it has to news to publish or because it is withholding), and N denotes the event that it does publish news about x. We can rewrite this out as: ν(A, β)E[(x2 − 2xρ + ρ2 )|N N] +(1 − ν(A, β))E[(x2 − 2xρ + ρ2 )|N] ¶ µ E[(x2 − 2xe ρ(A, β) + e ρ(A, β)2 ) = ν(A, β) +2xe ρ(A, β) − 2xρ|NN] − e ρ(A, β)2 + ρ2 ) Ã ! 2 2 e e e E[(x − 2xe ρ(A, β) − 2xρ|N] ρ(A, β) + e ρ(A, β) ) + 2xe +(1 − ν(A, β)) 2 2 e −e ρ(A, β) + ρ ) 2 e e (A, β) = ν(A, β)e σ 2 (A, β) + (1 − ν(A, β))σ h i + ν(A, β)(ρ − e ρ(A, β))2 + (1 − ν(A, β))(ρ − e e ρ(A, β))2
Now, rearranging (3), we have:
P 1 (A, β)/α1 = σ 2 − ν(A, β)e σ 2 (A, β) 2 e = (1 − ν(A, β))σ e (A, β) h i + ν(A, β)(ρ − e ρ(A, β))2 + (1 − ν(A, β))(ρ − e e ρ(A, β))2 14
For β > 0, this is strictly positive. Thus, the price of a single monopoly newspaper is always strictly positive as long as the voters are not at an extreme. This is because the newspaper always imparts some (valuable) information. In addition, we can derive some results regarding the limiting behavior of the price as β approaches its limits. First, note that as β → 0, the range of values of x for which A will withhold news (that is, [0, β]) becomes vanishingly small. Therefore, the probability ν that there will be no news in the paper reaches a limit of (1 − π), the probability that there will be no news to report. In addition, the diﬀerence between the densities f and g will become vanishingly small, so σ e2 (A, β) will converge to σ 2 . Therefore, from (3), the price of the newspaper will approach the limit of α1 πσ 2 . It is important to note that this is the value to the consumer of a newspaper with full disclosure, so this is the maximum possible price a newspaper could possibly have. Similarly, as β → 1, ν(A, β) → 1 and σ e(A, β) → σ so, again from (3), the price of the newspaper will converge to zero. The case of the B-monopolist is parallel, with P B (B, β) = α1 [σ 2 − ν(B, β)e σ 2 (B, β)], and P B (B, β) → 0 as β → 0 and P B (B, β) → α1 πσ 2 as β → 1. The point is that the more mainstream are the political views of the monopoly publisher, the less the public will expect that publisher to distort the news, and thus the more informative and valuable the paper will be. As a result, a more politically mainstream publisher will also be more proÞtable. We can now treat duopoly pricing. Prices in this case will be determined by Bertrand competition. Note that this will not in general drive both publishers’ proÞts down to zero, because the two news sources are not perfect substitutes, owing to the diﬀerent political biases of the two publishers. To analyze the prices, Þrst note that in any equilibrium, because production costs for newspapers are assumed away and consumers are homogeneous, each publisher will lower her price enough that all consumers will purchase both papers. Note further that this implies that each paper will have a price no greater than the additional utility derived from reading that paper, given that the consumer is already reading the other paper. Further, the price can be pushed all the way up to this additional utility without losing any customers. This implies that the utility from reading newspaper i is equal to 15
the utility from reading both papers, minus the utility derived from reading only paper j 6= i. Further, by the above discussion, the former utility is equal to −(1 − π)α1 σ 2 , and the latter utility is equal to −α1 ν(j, β)e σ 2 (j, β). Subtracting the latter from the former gives the value below. Proposition 3 The price of a newspaper under duopoly is equal to its incremental information value: σ 2 (j, β) − (1 − π)σ 2 ], Pi (C, β) = α1 [ν(j, β)e
i 6= j, i, j = A, B.
Using the previous analysis of the monopoly prices, we can see that as β → 0, P A (C, β) → πσ 2 and P B (C, β) → 0, while as β → 1, P A (C, β) → 0 and P B (C, β) → πσ 2 . Thus, both under conditions of monopoly and in conditions of competition, a publisher known to be in the political mainstream is proÞtable, while a publisher far out of the mainstream has trouble generating revenues.
Mergers and merger regulation.
This lays the foundation for an analysis of media mergers. The two news organizations will merge whenever the sum of the utilities of A and B under competition is less than the corresponding sum when one of the two acts as a monopoly. Precisely, the structure s that will be chosen will solve: max s
X £ ¤ αi E[x(dpub − β i )|s, β] + Pi (s, β) .
A full characterization is still in progress, but a sample result may be suggestive of the ramiÞcations of the political motives of the publishers. Consider the case in which αA is large relative to αB , and αB is itself suﬃciently larger than 1. In this case, political motives will be paramount, and A is more 16
willing to give up wealth for inßuence than is B. Here, the market structure that emerges will maximize A’s political payoﬀ, αA E[xdpub |s, β]. Using the earlier results on political manipulation of information, it is easy to see what the market structure is for any value of β. (i) First, if β ∈ [0, β), A will prefer to be a news monopoly and will buy B out. (ii) Second, if β ∈ (β, ρ), A will prefer to compete rather than hold monopoly power. This suggests at Þrst glance that no outside interfernce is needed. However, A will also prefer no news to any news, since in that range ρ > β, and so uninformed voters will choose dpub = 1. Thus, A will shut down and pay B to do so as well. (If feasible, A could commit to showing no news at all instead of closing down the whole market). (iii) If β ∈ (ρ, β), A will prefer B to be a monopolist, while B will prefer no news to acting as a monopolist, so A will pay B to act as a monopoly. (iv) Finally, if β ∈ (β, 1], A will operate and will be indiﬀerent to whether or not B operates. The worst outcome here from the point of view of social welfare is (ii), because it leaves voters completely uninformed. The best is (iv), because it provides the competitive level of information. In the other cases, (i) and (iii), we have a news monopoly, which gives positive but distorted news. In all cases except (iv), there is a social welfare motive to regulate mergers (or anti-competitive contracting more generally). Recall that this does not come from deadweight monopoly losses in the conventional sense, because they have been ruled out by construction. Consider now the outcome when revenues are paramount to news groups, and political inßuence is of secondary importance. That is, the news groups do have political preferences, but what they are willing to pay to exercise them is negligible. Notice Þrst that any merger consolidating the two news groups to one entity will be run by the group with higher monopoly revenue. Since all consumers buy one paper in equilibrium, this will be the one with the higher price (see Proposition 2). The price of the newspaper exactly measures the private beneÞts, so the paper with the higher price is also the one that publishes more valuable information. In this sense at least, the consolidation does give the right news group of the two, although consolidation is unlikely to be desirable per se. To see the latter point, consider a symmetric case, with β = 1/2 and f 17
symmetric around 1/2. Denote the common level of a parameter by dropping the arguments (so, for example, υ (A, β) = υ (B, β) = υ). Using the two propositions on pricing, a monopoly earns more than two duopolists as σ 2 − υ˜ σ 2 − 2 (1 − π) σ 2 σ 2 > 2υ˜ or σ 2 (3 − 2π) > 3˜ σ2 .
If π = 0, these two sides are equal (since ρ ˜ = ρ = 1/2, and σ ˜ 2 = σ 2 ), reßecting the fact that newspapers have a zero price regardless of market structure if they convey no information at all. On the other hand, if π = 1, then a monopolist earns more than two duopolists. Under symmetry, the monopolist conveys the exact information half the time, and when it does not, the consumer knows the mean of the true state is the mean of the undisclosed states. Each duopolist just earns the incremental proÞt from the ability to update from the mean of the undisclosed states to the true state. It is therefore not a priori clear whether the monopoly earns more than the two duopolists. When it does, it will want to merge and such a merger will restrict the information imparted. In this case of weak political motivation, the economic incentive to merger dominates. However, if the merged entity can commit to impartial reporting, it will do so. This is because revelation of the full information obtained will lead to higher incremental beneÞts to consumers and so to higher revenues. Then a merger does not have deleterious eﬀects and need not be proscribed. This was not the case under strong political motivation, except in case (iv) when the news organization would again prefer to commit to full disclosure (the extra proÞts breaking its indiﬀerence on pure political grounds).
The analysis above has looked at the incentives to merge, and the case for merger restrictions when news organizations have incentives to withhold information. Another dimension of concern is in terms of “persuasion” of the electorate by the media (see Balan et al., 2004, for an analysis of persuasion 18
competition and merger ramiÞcations). Persuasion, in the Economics of Advertising, is a rather nebulous concept insofar as it is controversial as to how one should evaluate the welfare of those being persuaded (see Dixit and Norman, 1979, for a provocative analysis).9 Our model provides a conduit for analyzing persuasion from the micro-underpinnings of the theoretical model of readership. SpeciÞcally, one could imagine that persuasive powers would impinge on the reader parameter β. The A news organization would like to shift β to the left and so broaden the set of circumstances under which the readers choose dpub = 1. The B organization would like to shift it in the opposite direction. One way to model this conßict is in terms of persuasive eﬀort exerted by the competing news organizations, like a “tug-of-war” between them (see Becker, 1983, for an analysis of lobbying politicians along these lines). The open question remains however in reasonably evaluating the eﬀects on reader beneÞts from persuasive pull.
References  Allen, William Sheridan (1984) “The Nazi seizure of power: The experience of a single German town, 1922-1945.” F. Watts (publisher).  Balan, David J., Patrick DeGraba, and Abraham L. Wickelgren (2004). “Media Mergers and the Ideological Content of Programming.” Working Paper, Bureau of Economics, Federal Trade Commission.  Becker, Gary S. (1983). “A Theory of Competition Among Pressure Groups for Political Inßuence,” Quarterly Journal of Economics, XCVIII (3).  Dixit, A.,and V. Norman (1978). “Advertising and Welfare,” Bell Journal of Economics, 9, 1—17.  Lipman, B. and D. Seppi (1995). “Robust Inference in Communication Games with Partial Provability,” Journal of Economic Theory, 66(2). 9
By contrast, if advertising is purely informative, it is clear how to evaluate the beneÞts from improved information in making better informed decisions. As in this paper, the welfare analysis is uncontroversial.
 Milgrom, Paul R. (1981). “Good News and Bad News: Representation Theorems and Applications.” Bell Journal of Economics 12:2 (Autumn), 380-391.  Mullainathan, Sendhil, and Shleifer, Andrei. (2002). “Media Bias.” NBER Working Paper #9295.  Stromberg, David (2001). “Mass Media and Public Policy.” European Economic Review, 45, 4-6 (May), 652-63.