TY - JOUR
AU - Várdy,, Felix
AB - Abstract Political commentators warn that the fragmentation of the modern media landscape induces voters to withdraw into ‘information cocoons’ and segregate along ideological lines. We show that the option to abstain breaks ideological segregation and generates ‘cross‐over’ in news consumption: voters with considerable leanings towards a candidate demand information that is less biased towards that candidate than voters who are more centrist. This non‐monotonicity in the demand for slant makes voters’ ideologies non‐recoverable from their choice of news media and generates disproportionate demand for media outlets that are centrist or only moderately biased. It also implies that polarisation of the electorate may lead to an ideological moderation in news consumption. Before the advent of cable TV and the internet, the US market for non‐local news was dominated by a few network channels and a handful of newspapers with national reach. The modern media landscape, by contrast, features an overwhelming array of news outlets, catering to virtually every ideological bias and belief. Moreover, the IT revolution has given people the power to customise the news and opinion they consume, tailoring it to fit their preferences. Commentators have argued that this induces individuals to self‐segregate along ideological lines, undisturbed by differing viewpoints or even facts contrary to their world view. It is claimed that, as a result, we live in a society where ‘many people live in information cocoons in which they only talk to members of their own party and read blogs of their own sect’ (Brooks, 2010). Sunstein (2002) expresses a similar sentiment: ‘Our communications? market is rapidly moving [toward a situation where] people restrict themselves to their own points of view – liberals watching and reading mostly or only liberals; moderates, moderates; conservatives, conservatives; Neo‐Nazis, Neo‐Nazis’ (pp. 2 and 3). This limits ‘encounters [that are] central to democracy itself’. Thus, while the proliferation of news outlets and the ability to create ‘Daily Me’ newspapers is, no doubt, a boon to individuals, at a societal level it is potentially baneful. Indeed, by destroying the common experience and understanding that resulted from everybody consulting the same news outlets, the Daily Me phenomenon might fray the social fabric, fan the dreaded ‘culture wars’ and lead to the atomisation of society (Sunstein, 2002, 2009; Jamieson and Cappella, 2010). Relying on binary decision models, formal theory by and large confirms this self‐segregation hypothesis: for purely instrumental reasons, rational voters prefer information sources whose biases conform to their own, both in direction and degree (Calvert, 1985; Suen, 2004). To see why, consider a liberal and a conservative candidate competing in a majoritarian election. The candidates differ sharply in their policy positions. The ‘correct’ policies to address the polity’s problems are determined by a draw from nature. When the state is liberal, the policies espoused by the liberal candidate are more effective, while the reverse is true when the state is conservative. All voters prefer superior outcomes but they differ in their ideologies – i.e. their prior beliefs that the policies proposed by a given candidate are the most effective in addressing the challenges at hand. Prior to voting, voters consult a news outlet, customised to their taste, to determine which candidate offers the better policies. News outlets are biased, characterised by Type I and Type II error, with a centrist outlet having equal errors of each type. What motivates rational voters to choose a biased outlet is the hurdle in overcoming their own ideology. Under compulsory voting, a voter with an ideology favouring the conservative candidate chooses a news outlet more likely to favour the conservative candidate in either state. The reason is that, should this outlet unexpectedly favour the liberal candidate, it offers compelling evidence, sufficient to overcome the voter’s ideology, that the liberal candidate is the correct choice. Thus, in these models, voters exhibit a strong ‘demand for slant’. Indeed, as the population of voters becomes ideologically polarised, relatively centrist media outlets see customers desert in droves. Despite this prediction and the greater polarisation of the US electorate manifesting itself in the Tea Party and Occupy movements, centrist news outlets have not fallen by the wayside. Indeed, programmes such as PBS’s Newshour and ABC’s Good Morning America, among the most centrist of all the national US media outlets according to Groseclose and Milyo (2005), have maintained high ratings.1 In addition, new centrist outlets, such as Yahoo, have successfully entered the scene. Another challenge to the received wisdom that voters only listen to views narrowly aligned to their own is provided by Gentzkow and Shapiro (2011), who find much less ideological segregation in news consumption than suggested by the existing theory. Instead, they observe significant ‘cross‐over’ behaviour, with some relatively conservative voters consulting relatively liberal news media – and vice versa. Similar results were found by Pew (2008, 2009). Thus, there is a disconnect between the dire predictions of the commentariat and the existing theory on the one hand, and the (perhaps more reassuring) reality on the other. In this article, we show that enriching the standard model with the realistic feature of abstention (i.e. the option to stay home from the polls) suffices to reconcile the apparent disconnect. To see how this amendment to the model fundamentally changes the demand for slant, consider the situation of a moderately polarised voter. In the standard model with compulsory voting, such a voter chooses a relatively extreme news outlet and follows its advice at the polls. In the model with abstention, such a voter turns out only when the signal favours his ex ante preferred candidate; if the signal favours the candidate of the opposite ideology, he prefers to stay home. As we show, the crux is that, for this modified plan, a centrist or near‐centrist outlet is in fact better suited than a more extreme outlet. As a result, some polarised voters optimally choose less ideological outlets than their intrinsically more centrist brethren. The intuition for this abstention‐induced non‐monotonicity is as follows. First, centrist and mildly polarised voters continue to behave under voluntary voting as they do under compulsory voting. That is, the former consult and follow a centrist outlet, while the latter follow an outlet with a mild, conforming bias. Next, consider a liberal voter (say) whose prior beliefs are more pronounced. For such a voter to follow his news outlet when it tells him to vote for the conservative candidate, he needs to be consulting an outlet with a strong liberal bias. This comes at the cost of virtually always voting for the liberal candidate, even when the state is conservative, as an outlet with a strong liberal bias rarely comes out in favour of the conservative candidate. Nonetheless, if the voter had to choose between the two candidates, this would be his best option. Suppose, however, that the voter has the option of supporting neither candidate and abstaining. For the usual reasons related to the swing voter’s curse (Feddersen and Pesendorfer, 1996), upon receiving the signal that the state is conservative, abstaining rather than voting for the conservative candidate is indeed an attractive alternative for such a liberal‐leaning voter. What does abstaining do to this voter’s optimal choice of news outlet? That is, how biased does he want his outlet to be if he only follows its recommendation when the advice is to vote for the liberal candidate but abstains when the advice is to vote for the conservative candidate? As he no longer plans on voting for the conservative candidate upon receiving the signal that the state is conservative, his original rationale for consulting an outlet with a strong liberal bias has significantly diminished. In relative terms, he now worries more about voting for the liberal candidate when the state is in fact conservative. Hence, he wants to raise the level of certainty conveyed by the ‘liberal’ signal and, therefore, prefers to consult a strictly more centrist outlet. In other words, the outlet of choice of this liberal‐leaning voter who sometimes abstains is less biased than that of some intrinsically more centrist voter who follows his news outlet’s recommendation in both directions. Notice that, as a result, he also votes for the liberal candidate less often. Our results present a challenge to a researcher seeking to infer voters’ ideologies and level of polarisation from the electorate’s news consumption. With abstention, it is no longer true that the more a voter favours a given side, the more biased the news outlet chosen. Instead, the selected outlet offers a jumble of possibilities. Less ideological voters may select more ideological outlets than their more ideological counterparts and vice versa. Hence, as the population becomes more ideologically polarised, demand for centrist and near‐centrist outlets may in fact increase. Another factor favouring mainstream media is that, unlike their more extreme counterparts, centre‐left and centre‐right news outlets serve multiple constituencies: those who are ideologically aligned with the outlet, and those whose beliefs are more pronounced. The former follow their outlet’s voting recommendation in both directions, while the latter follow it only when it conforms to their prior beliefs and abstain otherwise. As a result, mainstream media benefit from disproportionate demand for their services. Thus, our findings suggest that the oft‐prophesied, imminent demise of mainstream media in a polarising world is perhaps exaggerated (Fineman, 2005; Calderone, 2009). The remainder of the article is organised as follows. In Section 1, we present a simple example illustrating the main point of the article. Section 2 introduces the general model, which we solve in Section 3. In Section 4, we analyse the demand for slant. In Section 5, we discuss some extensions and limitations of the model, while Section 6 reviews the related literature. Finally, Section 7 concludes. Formal proofs are relegated to Appendices. 1. Example The results in this article are driven by the observation that the option to abstain makes the demand for slant non‐monotone in ideology. That is, certain voters with considerable leanings towards a candidate demand information that is less biased towards that candidate than voters who are more centrist. The intuition is illustrated in the following example.2 Consider a voter with the following preferences: he gains u = 1 if he votes for the ‘right’ candidate but loses c = 2 > u if he votes for the ‘wrong’ candidate. In addition, he has the option to abstain, which gives him a utility of 0. From the perspective of the voter, the ‘right’ candidate is determined by the state of nature: in state d, the right candidate is D; in state r, the right candidate is R. The voter’s prior belief that the state is r is equal to θ ∈ [0, 1]. We say that the higher is θ, the more ‘biased’ is the voter towards candidate R. The voter can consult one of two news outlets, labelled r^ and d^ ⁠. Before the election, the outlets publish editorials supporting one of the candidates. Formally, support for candidate d is expressed by sending a signal sd, while support for candidate R is expressed by sending a signal sr. News outlet d^ supports candidate D in state d 75% of the time, while it supports candidate R in state r 65% of the time. Conversely, news outlet r^ supports candidate R in state r 75% of the time, while it supports candidate D in state d 65% of the time. Hence, in both states, outlet d^ supports candidate D more often than does outlet r^, while outlet r^ supports candidate R more often. In an intuitive sense, this makes d^ relatively biased towards D, and r^ relatively biased towards R. Let ρθk^(s) denote the voter’s posterior belief that the state is r upon receiving a signal s from outlet k^ ⁠. By Bayes’ rule, ρθd^(s)>ρθr^(s) for all s ∈ {sd, sr}. That is, a signal s from outlet d^ provides stronger evidence that the state is r than the same signal from outlet r^ ⁠. Which news outlet, if any, does the voter consult, and how does this depend on whether voting is mandatory? When voting is mandatory, the voter must choose between: (i) always voting for D; (ii) always voting for R; and (iii) voting for D after sd and voting for R after sr. We denote these strategies by DD, RR and DR respectively. (Notice that the strategy RD, i.e. voting for R after sd and voting for D after sr, is strictly dominated and, hence, ignored.) When playing DR, the voter also has to decide which news outlet to consult. The voter’s expected utility from voting for D after receiving the signal sd from outlet k^ is Uθk^(D,sd)=1×[1−ρθk^(sd)]−2×ρθk^(sd).(1) Similarly, his utility from voting for R after receiving the signal sr from outlet k^ is Uθk^(R,sr)=1×ρθk^(sr)−2×[1−ρθk^(sr)].(2) The expected utility from playing the voting strategy DR while consulting outlet k^ is then Uθk^(DR)=Uθk^(D,sd)×Pr(sd|k^,θ)+Uθk^(R,sr)×Pr(sr|k^,θ).(3) Here, Pr(s|k^,θ) denotes the probability that the voter assigns to outlet k^ sending the signal s. For instance, Pr(sd|d^,θ)=0.75×θ+0.35×(1−θ) ⁠. The expected utilities from the voting strategies DD and RR, which do not depend on k^ ⁠, are equal to Uθ(DD)=1×(1−θ)−2×θ,Uθ(RR)=1×θ−2×(1−θ).(4) Comparing (3) with (4), we find that under compulsory voting: (i) for θ ≤ 5/18, the voter does not consult a news outlet and always supports D; (ii) for 5/18 < θ ≤ 1/2, he consults outlet d^ and follows its advice at the polls; (iii) for 1/2 < θ ≤ 13/18, he consults and follows outlet r^ ⁠; and (iv) for 13/18 ≤ θ, he once again consults neither outlet and always votes for R. Hence, under compulsory voting, the voter only consults news outlets whose biases conform to his own. This is consistent with Calvert (1985) and Suen (2004). Why does the voter switch news outlet at θ = 1/2? Notice that the two outlets provide different services. While outlet d^ provides a high level of certainty that the state is r when it sends the signal sr, outlet r^ provides a high level of certainty that the state is d when it sends the signal sd. Depending on his prior, a voter attaches different values to these certainties. A voter with θ < 1/2 is predisposed to vote for D. Hence, in relative terms, he cares less about the additional certainty that the state is d provided by signal sd. By contrast, if he plans to vote against his prior and for R after sr, he cares a lot about the additional certainty that the state is r provided by signal sr. He achieves this certainty by consulting outlet d^ ⁠. An analogous argument explains why a voter with θ > 1/2 consults outlet r^ ⁠, if any. When voting is voluntary, the voter’s optimal strategy is more complicated – and interesting. First, abstention expands the set of (non‐dominated) voting strategies with DΦ, ΦR and ΦΦ, which correspond to: (i) voting for D after sd and abstaining after sr; (ii) abstaining after sd and voting for R after sr; and (iii) abstaining after both signals. It is easily verified that, for every θ ∈ [0, 1], there exists at least one news outlet k^∈{d^,r^} such that either Uθk^(D,sd)>0 or Uθk^(R,sr)>0 ⁠. Hence, always abstaining is a dominated strategy for all prior beliefs. In addition, for a liberal‐leaning voter (i.e. θ < 1/2), DR dominates RR and ΦR, while for a conservative‐leaning voter (i.e. θ > 1/2), DR dominates DD and DΦ. We may conclude that, for a liberal voter, only the voting strategies DR, DΦ and DD are relevant, while for a conservative voter the relevant strategies are DR, ΦR and RR. When using a responsive voting strategy – i.e. DΦ, DR or ΦR – a voter also needs to decide which news outlet to consult. Because of symmetry, we may limit attention to, say, the liberal side of the ideological spectrum, i.e. θ ≤ 1/2. Optimal behaviour on the conservative side is analogous. Conditional on consulting outlet k^ ⁠, the expected utility from voting DΦ is Uθk^(D,sd)×Pr(sd|k^,θ) ⁠. Using (1) and the relevant expressions for Pr(sd|k^,θ) ⁠, we find that, conditional on DΦ, a voter with beliefs θ ≤ 1/3 prefers to consult outlet d^ ⁠, while a voter with θ > 1/3 prefers to consult outlet r^ ⁠. Comparing the resulting expected pay‐off with that from the unresponsive voting strategy DD given in (4), we find that a voter with prior beliefs θ > 5/31 prefers DΦ (in combination with the optimal news outlet) to DD, while a voter with θ ≤ 5/31 prefers DD to DΦ. It remains to determine which voters prefer DR over DD and DΦ. Conditional on playing DR, a liberal voter always prefers consulting d^ over consulting r^ ⁠. The intuition for this rational ‘confirmation bias’ is exactly the same as under compulsory voting. Comparing the expected pay‐off from DR with d^ to the pay‐off from DΦ with either d^ or r^ ⁠, we find that a liberal voter prefers DR if and only if θ > 8/17. (A fortiori, he also prefers DR over DD in that region.) Hence, a liberal voter’s optimal strategy as a function of his prior is DDforθ≤531DΦwithd^for531<θ≤13DΦwithr^for13<θ≤817DRwithd^for817<θ≤12. A conservative voter’s optimal strategy is the mirror‐image analogue. Figure 1 provides a graphical representation. Fig. 1. Open in new tabDownload slide Optimal Choice of News Outlet and Voting Strategy as a Function of Prior Belief θ Fig. 1. Open in new tabDownload slide Optimal Choice of News Outlet and Voting Strategy as a Function of Prior Belief θ The voting behaviour is rather intuitive. An extremely liberal voter (θ ≤ 5/31) always votes for D. A less extreme liberal (5/31 < θ ≤ 8/17) votes for D upon receiving sd and abstains upon receiving sr. Finally, a moderate liberal (8/17 < θ ≤ 1/2) votes for D upon receiving sd and votes for R upon receiving sr. The optimal choice of news outlet, by contrast, is rather curious: while voters 5/31 < θ ≤ 1/3 and 8/17 < θ ≤ 1/2 exhibit the familiar rational confirmation bias, voters 1/3 < θ ≤ 8/17 exhibit a rational anti‐confirmation bias. These liberal voters choose to ‘cross‐over’ and consult the conservative outlet r^ ⁠, rather than the liberal outlet d^ ⁠. In other words, in addition to the switch of news outlet at θ = 1/2 which also occurs under compulsory voting, on each side of the political spectrum, we now have two more switches. Let us continue to focus on the liberal side of the spectrum. At the first switch, which occurs at θ = 1/3, only the news outlet changes (from d^ to r^ ⁠) while the voting strategy, DΦ, remains unchanged. By contrast, the second switch, which occurs at θ = 8/17, involves not only a change in news outlet (from r^ back to d^ ⁠) but also a change in voting strategy (from DΦ to DR). The intuition for these two switches is quite different. While the switch at θ = 1/3 is a consequence of the discreteness of the space of available news outlets, the switch at θ = 8/17 is a consequence of the discrete nature of voting itself. As we shall see, this implies that the first switch disappears when we allow for a continuum of news outlets, while the second switch is a robust feature of any voting environment. First, let us focus on the switch at θ = 1/3, where only the news outlet but not the voting strategy changes. The intuition for this switch is quite similar to the one at θ = 1/2. Even though all voters in the interval 5/31 < θ ≤ 8/17 play DΦ, depending on the value of θ, they have different default actions. That is, if they had to decide purely on the basis of their prior, voters to the left of 1/3 would vote for D, while the voters to the right of 1/3 would abstain. This makes these voters concerned about different attributes of the same voting strategy, DΦ. After any signal, a voter whose θ is close to 5/31 more strongly believes that the state is d than a voter whose θ is close to 8/17. Therefore, the former values less than the latter any additional certainty that the state is d after a signal sd. At the same time, in order to abstain, the former needs more reassurance that the state is not d after a signal sr than the latter. Hence, the voter whose θ is close to 5/31 prefers an outlet, d^ ⁠, that provides strong evidence that the state is not d when sending the signal sr, while the voter whose θ is close to 8/17 prefers an outlet, r^ ⁠, that provides strong evidence that the state is d when sending the signal sd. The actual switch in news outlet occurs where the default action changes from D to Φ. The second switch, from r^ back to d^ ⁠, takes place at θ = 8/17 and occurs for a rather different reason. Rather than the change in the default action, it is the change in (responsive) voting strategy – from DΦ to DR – that induces the voter to change his news outlet. To see this, notice that, by its very nature, the strategy space in voting environments is discrete. Hence, any change in voting strategy entails a discontinuous change in the pay‐off function which, in turn, implies a discontinuous change in the value of information. How does the value of information change when the voting strategy changes from DΦ to DR? The jump from abstaining to voting for R after the signal sr makes the voter considerably more ‘concerned’ about the certainty conveyed by the signal sr. The reason is that, instead of taking the relatively safe action of abstaining, he now plans to take the much riskier action of voting against his prior. Suddenly, he attaches a lot more value to having additional certainty that the state really is r after observing sr. In order to achieve this, he is willing to give up some certainty that the state is d after sd. This explains why the switch in voting strategy from DΦ to DR induces the voter to switch from news outlet r^ to news outlet d^ ⁠. Our example is special in three respects. First, there are only two news outlets. Second, u < c; i.e. the benefit of voting for the right candidate is smaller than the cost of voting for the wrong candidate. Third, preferences are ‘pseudo‐expressive’, in the sense that a voter’s pay‐off only depends on whether his own vote matches the state. One may wonder which aspects of the example carry over to more canonical voting environments with multiple news outlets, general cost–benefit ratios and purely instrumental preferences. As we shall see, the non‐monotone switch in news outlet between the DΦ and DR regions is a robust phenomenon, while the monotone switch within the DΦ region is not. (Again, in our discussion, we limit attention to the liberal side of the ideological spectrum. Behaviour on the conservative side is analogous.) The switch within the DΦ region is a consequence of the fact that, while voters’ beliefs are changing smoothly, in the example, voters cannot smoothly adjust their choice of news outlet. Instead, they have to choose between either d^ or r^ ⁠. However, as we have argued in the introduction, the modern media landscape is characterised by a wide array of news outlets, catering to every ideological bias and belief. Hence, in today’s world, people can adjust their choice of news outlet more or less smoothly. When we account for this by allowing for a continuum of outlets spanning the ideological spectrum, the discrete jump within DΦ disappears. Instead, within this interval, there is a smooth and monotone change from liberal to more centrist outlets, as we move from the ideological left towards the centre. As argued above, the discrete and non‐monotone switch at the boundary between DΦ and DR is caused not by the discreteness of the media landscape – indeed, it occurs in spite of it – but by the change in (responsive) voting strategy. Note, however, that when the benefit of voting for the right candidate is greater than the cost of voting for the wrong candidate – i.e. u/c > 1 – no voter ever abstains and the boundary between DΦ and DR does not exist. In that case, because there is no switch from one responsive voting strategy to another, the demand for slant is a monotone function of ideology, and our example has no bite. This raises the question whether u < c or u > c is the ‘right’ assumption. When we study a voting model with fully rational voters and instrumental preferences, (the analogue of) the ratio u/c is endogenous. That is, the probabilities of the pivotal events that give rise to the benefits and costs of casting a vote versus abstaining are determined in equilibrium. Moreover, as explained in detail in Feddersen and Pesendorfer (1996), voters who are (close to) indifferent between D and R suffer from the swing voter’s curse, which implies that they strictly prefer to abstain rather than vote for either candidate. Hence, while in our simple example, we have to assume that u < c in order to generate abstention, in a more general model with instrumental preferences, abstention is endogenously generated. This explains why the non‐monotonicity and discontinuity of the demand for slant are in fact robust phenomena, that extend beyond our simple example to canonical voting environments. 2. Model In Section 1, we studied a simple decision‐theoretic example to illustrate the basic intuition behind cross‐over in news consumption. We now extend the analysis to a fully fledged strategic voting environment. Specifically, we study a two‐candidate election with a Poisson number of voters who have the option to abstain. Preferences are purely instrumental, i.e. voters only care about the outcome of the election and not about their own vote. There are two kinds of voters, partisan and non‐partisan. Partisan voters always prefer one candidate over the other, while non‐partisan voters, who constitute the majority, want the winner of the election to match the state of nature. While all non‐partisans prefer the Democratic candidate in the Democratic state and the Republican candidate in the Republican state, they do, however, disagree as to the prior probability that these states pertain. While some believe that the state is more likely to be Democratic, others believe that the state is more likely to be Republican. We shall say that the former are leaning to, or ‘biased’ towards, the Democratic candidate, while the latter are leaning to or biased towards the Republican candidate. Voters who think that the two states are more or less equally likely are called centrists. As is standard in the literature, before casting a ballot, voters receive a signal as to the state of nature. Our innovation is to allow voters to choose from what kind of news outlet they receive their information. News outlets, of which there are a continuum, send conditionally independent, binary signals that are correlated with the state. These outlets span the ideological spectrum, from strongly biased towards the Democratic candidate at one end, to strongly biased towards the Republican candidate at the other end. An outlet that is biased towards the Democrat is very likely to send the ‘Democratic’ signal when the state is Democratic. In that sense, it has high accuracy in that state. However, high accuracy in the Democratic state comes at the cost of low accuracy in the Republican state. That is, even when the state is Republican, this news outlet is relatively likely to send the Democratic signal. For an outlet that is biased towards the Republican the reverse holds. Finally, a centrist news outlet is one whose accuracy is (close to) the same in the two states. Once voters have consulted their preferred news outlet, if any, they simultaneously cast their ballots or abstain. The candidate who receives the most votes wins the election and pay‐offs are realised. Ties are resolved by a coin flip. Formally, there are two candidates, j ∈ {D, R}, and two states of the world, ω ∈ {d, r}. The number of voters is Poisson distributed with mean υ ( Myerson, 1998,2000). The probability that a voter is partisan is equal to ζ ∈ (0, 1/2) and independent across voters.3 A partisan voter prefers candidate R with probability η and candidate D with the remaining probability, again independently across voters.4 The remaining voters are non‐partisan. Non‐partisans receive a pay‐off of zero when the winner of the election matches the state (i.e. j = D and ω = d, or j = R and ω = r), and incur a loss l = −1 otherwise. We denote a non‐partisan voter’s prior belief that the state is r by θ ∈ [0, 1]. These beliefs are private information and drawn independently from a distribution with cumulative distribution function (CDF) F on [0, 1] that admits a well‐behaved probability density function (PDF) f with no mass points. Heterogeneous priors are sometimes considered controversial (Aumann, 1976,1987; Morris, 1995). Notice, however, that our heterogeneous priors model is isomorphic to a standard common priors model with heterogenous pay‐offs, as in Austen‐Smith and Banks (1996).5 While these formulations are logically equivalent, for ease of exposition and discussion, we stick to the heterogenous beliefs interpretation for the remainder of the article. Voters can cast a ballot for D, a ballot for R, or they can abstain, which we denote by Φ. Before casting a ballot, a voter can collect a signal s ∈ {sd, sr} from one news outlet. News outlets are characterised by accuracies (pd,pr)∈[0,1]2 in state d and state r respectively. Here, pd is the probability that the news outlet sends the (correct) signal sd in state d. Likewise, pr is the probability that the news outlet sends the (correct) signal sr in state r. We identify a news outlet with its accuracy in state r and assume that there is a continuum of outlets, one for each pr∈[0,1] ⁠. The relationship between news outlets’ accuracies in the two states is described by a function G that maps an outlet’s accuracy in state r into its accuracy in state d. That is, pd=G(pr) ⁠. We assume that G is strictly decreasing, twice continuously differentiable and strictly concave. Strict concavity implies that increased accuracy in one state becomes progressively more expensive in terms of reduced accuracy in the other state. It ensures that, for a given voting strategy, a voter’s optimal choice of news outlet is ‘well‐behaved’ (i.e. continuous) in θ and in the ratio of pivotal probabilities. We further assume that G runs from G(0) = 1 to G(1) = 0. This means that a news outlet can achieve perfect accuracy in state d, but only by always sending the signal sd. Similarly, perfect accuracy in state r comes at the cost of perfect inaccuracy in state d. Finally, we let 0 < −G ′(0) < −G ′(1) < ∞. As we show below, this implies that, for non‐degenerate priors, posterior beliefs remain bounded away from 0 and 1. When these assumptions are satisfied, it is easy to show that there always exists a (unique) news outlet, p~>1/2 ⁠, whose accuracy is the same in both states (see Lemma 4 in Appendix B). We call this outlet perfectly centrist, and say that a news outlet is leaning to (or biased towards) R if pr>p~ ⁠. Bias towards D is defined analogously. Summarising, Assumption The accuracy function G : [0, 1]→[0, 1], which maps pr into pd, is a strictly decreasing, strictly concave, twice continuously differentiable bijection with 0 < −G ′(0) < −G ′(1) < ∞.6 For ease of exposition, we impose one additional assumption on G, namely, that it gives rise to ‘increasing elasticities.’ Specifically, denote by δpr,1−pd the elasticity of inaccuracy 1 − pd in state d with respect to accuracy pr in state r. That is, δpr,1−pd measures the percentage rise in inaccuracy in state d associated with a one percent rise in accuracy in state r. Formally, δpr,1−pd=d(1−pd)dprpr1−pd=−G′(pr)pr1−G(pr). Similarly, δpd,1−pr is the elasticity of inaccuracy 1−pr in state r with respect to accuracy pd in state d, i.e., δpd,1−pr=d(1−pr)dpdpd1−pr=G(pr)−G′(pr)(1−pr). When δpr,1−pd and δpd,1−pr are strictly increasing in pr and pd respectively, then each percentage increase in accuracy in one state becomes progressively more expensive in terms of reduced accuracy in the other state, again expressed in percentage terms. It is easy to verify that this property implies strict concavity of G but that the reverse implication does not hold. Strictly increasing elasticities only play a role in the proof of Proposition 1. They ensure that all pay‐off functions satisfy single‐crossing in θ and, thereby, make equilibrium behaviour particularly simple to characterise. Assumption Elasticities δpr,1−pd and δpd,1−pr are strictly increasing in pr and pd respectively.7 In order to better understand the informational implications of choosing a particular news outlet pr ⁠, we now study the posterior beliefs it induces. Denote by ρθpr(s) a voter’s posterior belief that the state is r upon receiving a signal s from news outlet pr ⁠. By Bayes’ rule, ρθpr(sr)=1+1−G(pr)pr1−θθ−1andρθpr(sd)=1+G(pr)1−pr1−θθ−1(5) Differentiating with respect to pr reveals that ρθpr(sr) and ρθpr(sd) are monotonically decreasing in pr (see Lemma 4 in the Appendix B). This is intuitive. Higher pr means that a news outlet is more biased towards R, such that a signal sr conveys less evidence that the state is indeed r. This makes ρθpr(sr) ⁠, the posterior belief that the state is r after sr, decreasing in pr ⁠. At the same time, higher pr also means that the signal sd conveys more evidence that the state is not r. This makes ρθpr(sd) ⁠, the posterior belief that the state is r after sd, also decreasing in pr ⁠. For pr↓0 ⁠, ρθpr(sr) and ρθpr(sd) approach their suprema θ/[θ−G ′(0)(1−θ)] < 1 and θ respectively. For pr↑1 ⁠, ρθpr(sr) and ρθpr(sd) approach their infima θ and θ/[θ − G ′(1)(1 − θ)] > 0 (see Lemma 4 in Appendix B). Because −G ′(0) > 0, upon receiving the signal sr, a voter achieves less than perfect certainty that the state is r even when pr↓0 ⁠. Similarly, because −G ′(1) < ∞, upon receiving the signal sd, a voter achieves less than perfect certainty that the state is not r (i.e. d) even when pr↑1 ⁠.8 Voters receive signals that are conditionally independent. This implies that, in line with the extant literature, voters who consult media outlets with the same accuracy do not necessarily receive the same signal. This assumption is made purely for technical convenience. Assuming conditional independence reduces the level of correlation across voters’ signals. Since our model involves continuous types and continuous media outlets, correcting for this reduction in correlation would require taking care of complex measurability issues. As should be clear from the underlying intuition, this would not affect the main result of our article – i.e. the non‐monotonicity and discontinuity of the demand for slant. However, it would considerably increase the complexity and length of the proofs. The timing of the game is as follows: (i) nature selects the state, the number of voters and the profile of voter types; (ii) each voter observes his own type; (iii) voters privately decide which news outlet to consult (if any) and they draw a private and conditionally independent signal from the selected outlet; and (iv) ballots are cast, the winner is selected according to majority rule and pay‐offs are realised. Ties are resolved by a coin flip. In this game, the optimal behaviour of partisan voters is trivial: they simply vote according to their party affiliation. Indeed, since non‐partisans are the only ones who act strategically, in the remainder, ‘voter’ refers to non‐partisan voter. Non‐partisans must choose which news outlet to consult, if any, and how to vote. For them, a strategy consists of a tuple (P, V). Here, P is a measurable function mapping a voter’s prior θ into a choice of news outlet, pr ⁠. We refer to P as the demand for slant and assume that a voter only consults a news outlet if it leaves him strictly better off. The act of not consulting an outlet is denoted by ∅ ⁠. Hence, P : [0, 1]→[0, 1]∪{∅}. The function V, which we refer to as the voting strategy, maps signals {sd, sr} into a vote for D, a vote for R, or abstention. Hence, V = {sd, sr}→{D, R, Φ}. We denote particular voting strategies as follows: DR refers to voting for D after the signal sd and voting for R after the signal sr. All other combinations of D, R and Φ are similarly defined. For example, DD denotes always voting for D, while ΦR denotes abstaining after sd and voting for R after sr. We study pure‐strategy symmetric Bayesian Nash equilibria of this game. A symmetric Bayesian Nash equilibrium is a profile of identical strategies (P*, V*), one for each voter, such that no voter has a strictly profitable, unilateral deviation. 3. Solving the Model To solve the model, we proceed in two steps. First, we derive a voter’s optimal choice of news outlet conditional on his voting strategy. This gives rise to a set of ‘indirect’ utility functions, one for each voting strategy. Then we compare these indirect utility functions and, for each θ, identify the voting strategy that generates the highest pay‐off. Of course, the complication we have to deal with is that voters’ pay‐offs not only depend on their priors, their choice of news outlet and their voting strategy, but also on the ratios of pivotal probabilities, which are determined in equilibrium. The remainder of this Section is organised as follows. In subsection 3.1, we study pivotal events and probabilities. In subsection 3.2, we derive the optimal choice of news outlet for each voting strategy. Finally, in subsection 3.3, we compare indirect utilities and characterise the equilibrium voting behaviour. 3.1. Pivotal Events and Probabilities A rational voter anticipates that his decision at the ballot box only affects his pay‐off when it changes the outcome of the election. Hence, when deciding how to vote, he conditions on being pivotal. When comparing a vote for D with abstaining, the relevant pivotal events are that R is leading by one vote or that the two candidates are tied. When R is leading by one vote, a vote for D rather than abstaining throws the election into a tie and, hence, raises the probability that D wins by 50%. When the candidates are tied, a vote for D hands the election to D and, therefore, also raises D’s probability of winning by 50%. The pay‐off comparison of voting for R versus abstaining is analogous. In that case, the pivotal events are that D is leading by one vote or that the two candidates are tied. Either way, a vote for R rather than Φ raises R’s chances by 50%. Finally, when comparing a vote for D with a vote for R, the pivotal events consist of all of the above, i.e. R is leading by one vote, D and R are tied, or D is leading by one vote. Notice that when a candidate is leading by one vote, the shift in winning probabilities induced by a vote for R rather than D is once more 50%. By contrast, when the two candidates are tied, a vote for R instead of D shifts the election from a sure win for D to a sure win for R. For a given profile of symmetric strategies (P, V), we now derive expressions for the probabilities of the various pivotal events. Denote by tD(ω) the probability that a randomly drawn voter, who may be partisan or non‐partisan, casts a vote for D in state ω. Denote by nD(ω) the total number of votes for D in state ω. Let tR(ω) and nR(ω) be similarly defined. As proved by Myerson (2000), nD(ω) and nR(ω) are independently distributed Poisson random variables with expectation parameters υtD(ω) and υtR(ω).9 The memorylessness of the Poisson distribution implies that, from the perspective of a given voter i, nD(ω) and nR(ω) also describe the number of votes for D and R cast by all other voters. From i’s perspective, the pivotal events in state ω correspond to |nD(ω)−nR(ω)|≤1 ⁠, i.e. situations where, after all others have cast their votes, one candidate leads by at most one vote. Let Tx(ω) denote the probability that D leads by x∈Z votes. The probabilities of the three pivotal events are then T−1(ω), T0(ω) and T1(ω). Because the number of voters is Poisson distributed, we have T−1(ω)=e−υ[tR(ω)+tD(ω)]∑n=0∞[υtD(ω)]nn![υtR(ω)]n+1(n+1)!. The expressions for T0(ω) and T1(ω) are analogous. Denote the set of all pivotal probabilities by T, i.e. T≡T−1(d),T0(d),T1(d),T−1(r),T0(r),T1(r). Partisan voters guarantee that tD(ω), tR(ω) > 0. This implies that every vote count, (nD,nR), has strictly positive probability of occurring. As pivotal events are nothing but collections of particular vote counts, T is strictly interior, i.e. T ∈ (0, 1)6. Finally, for future reference, let T~−1(ω)≡T0(ω)+T−1(ω) and T~1(ω)≡T0(ω)+T1(ω) ⁠. 3.2. Demand for Slant Conditional on Voting Strategy In this subsection, we derive voters’ optimal choice of news outlet conditional on their voting strategy. First, notice that only DD, DΦ, DR, ΦΦ, ΦR and RR are viable voting strategies. The remaining strategies, RD, ΦD and RΦ, entail the ‘reversing’ of signals. Because signals are informative, such strategies are dominated and, hence, not played in equilibrium. Next, notice that a voter who optimally uses one of the unresponsive voting strategies, DD, ΦΦ or RR, is indifferent between all news outlets and, thus, by assumption, chooses not to consult one. Therefore, it only remains to determine the optimal choice of news outlet for voters using DΦ, DR or ΦR. We denote these optimal outlets by prDΦ(θ) ⁠, prDR(θ) and prΦR(θ) respectively. Conditional on playing DR, a voter’s expected pay‐off is Uθpr(DR)=L(θ)−1−12prT1(r)+(1−pr)T0(r)+12(1−pr)T−1(r)θ−12[1−G(pr)]T1(d)+1−G(pr)T0(d)+1−12G(pr)T−1(d)(1−θ).(6) The first term, L(θ), incorporates the expected pay‐offs (i.e. losses) associated with all non‐pivotal events. By definition, these pay‐offs do not depend on the voter’s choice of news media or voting strategy. The remaining terms incorporate the pay‐offs associated with all pivotal events. Consider, for instance, the last term on the first line of (6), i.e. 12(1−pr)T−1(r)θ ⁠. Its interpretation is as follows: the joint event that the state is r, candidate R leads by one vote, and the voter receives the (wrong) signal sd occurs with probability (1−pr)T−1(r)θ ⁠. As he is playing DR, the signal sd makes the voter cast a ballot for D. This throws the election into a tie and, hence, leads to an expected loss of one half. As a result, under DR, the contribution to the expected pay‐off of this particular event is 12(1−pr)T−1(r)θ ⁠. All other terms have analogous interpretations. Differentiating Uθpr(DR) with respect to pr and rearranging yields the first‐order condition −G′(prDR)=T~−1(r)+T~1(r)T~−1(d)+T~1(d)θ1−θ.(7) It is easily verified that Uθpr(DR) is strictly concave in pr ⁠. Hence, there is at most one prDR that solves (7) and, at that prDR ⁠, the second‐order condition for a maximum is satisfied. The first‐order condition can be easily understood in terms of cost–benefit arguments. To see this, rewrite (7) as 12[T~−1(d)+T~1(d)](1−θ)×−dG(prDR)=12[T~−1(r)+T~1(r)]θ×dprDR. On the right‐hand side, we have the gain in expected pay‐offs induced by a marginal increase in accuracy in state r. This gain is equal to the joint probability that the state is r and the vote is pivotal times the expected gain from voting for R rather than D. On the left‐hand side, we have the loss in expected pay‐offs induced by a marginal decrease in accuracy in state d, which is equal to the joint probability that the state is d and the vote is pivotal times the expected loss from voting for R rather than D. Of course, when the choice of news outlet is optimal, the expected marginal gains are equal to the expected marginal losses. Because pr is a probability, it is bounded between 0 and 1. Define DDθDR to be the type θ such that the solution to the first‐order condition (7) just reaches its lower bound, prDR=0. Similarly, let DRθRR be the type θ such that the solution to (7) just reaches its upper bound, prDR=1 ⁠. In other words, DDθDR and DRθRR are the transition types where DR degenerates into DD and RR respectively. Solving (7) for θ reveals that DDθDR=−G′(0)T~−1(r)+T~1(r)T~−1(d)+T~1(d)+[−G′(0)]andDRθRR=−G′(1)T~−1(r)+T~1(r)T~−1(d)+T~1(d)+[−G′(1)].(8) Hence, the transition types DDθDR and DRθRR exist, are unique, and are strictly interior. Moreover, because −G ′(0) < −G ′(1), we have DDθDR < DRθRR. Recalling that G and Uθpr(DR) are strictly concave in pr ⁠, we may conclude that prDR(θ)=0forθ∈[0,DDθDR](G′)−1−T~−1(r)+T~1(r)T~−1(d)+T~1(d)θ1−θforθ∈(DDθDR,DRθRR)1forθ∈(DRθRR,1)(9) When playing DR, a voter with beliefs θ sufficiently close to 0 maximises his utility by ‘cornering out’ and choosing prDR(θ)=0 ⁠. Hence, such a voter is, in effect, playing DD. Similarly, a voter with θ sufficiently close to 1 maximises his utility by choosing prDR(θ)=1 and, in effect, playing RR. For these voters, there simply do not exist media outlets whose signals are sufficiently convincing to overcome their strong prior beliefs. Along the same lines, we find that prDΦ(θ) and prΦR(θ) are equal to10 prDΦ(θ)=0ifθ∈[0,DDθDΦ](G′)−1−T~−1(r)T~−1(d)θ1−θifθ∈(DDθDΦ,DΦθΦΦ)1ifθ∈[DΦθΦΦ,1].(10) prΦR(θ)=0ifθ∈[0,ΦΦθΦR](G′)−1−T~1(r)T~1(d)θ1−θifθ∈(ΦΦθΦR,ΦRθRR)1ifθ∈[ΦRθRR,1].(11) Here, the transition types {DDθDΦ, DΦθΦΦ} and {ΦΦθΦR, ΦRθRR} are the analogues of the transition types {DDθDR, DRθRR}. They correspond to the points where the responsive voting strategies DΦ and ΦR degenerate into the unresponsive voting strategies DD, ΦΦ and RR. Closed‐form solutions can be found in Lemma 6 in Appendix B. Implicitly differentiating the first‐order conditions with respect to θ reveals that, in the interior, prDR(θ) ⁠, prDΦ(θ) and prΦR(θ) are strictly increasing in θ. Hence, for a given voting strategy, the demand for slant is monotone in ideology. The intuition is essentially the same as in Calvert (1985) and Suen (2004). The more the decision‐maker leans towards one candidate, the stronger evidence he needs in order to vote for the other candidate (or to abstain). Such strong evidence can only be provided by information sources that share his bias. Having derived voters’ preferred demand for slant for a given voting strategy, we now compare the demand for slant across voting strategies. As is evident from (9), (10) and (11), the ordering of prDR(θ) ⁠, prDΦ(θ) and prΦR(θ) crucially hinges on the equilibrium values of the ratios of pivotal probability, [T~−1(r)+T~1(r)]/[T~−1(d)+T~1(d)] ⁠, T~−1(r)/T~−1(d) and T~1(r)/T~1(d) ⁠. In the next Lemma we show that, for large electorates, these ratios can be unambiguously ranked. Lemma 1. For υ sufficiently large, in equilibrium, T~1(r)T~1(d)<T~1(r)+T~−1(r)T~1(d)+T~−1(d)<T~−1(r)T~−1(d). Lemma 1 is intuitive. To see this, notice that [T~1(r)]/[T~1(d)]<[T~−1(r)]/[T~−1(d)] is equivalent to [T~1(r)/T~−1(r)]<[T~1(d)/T~−1(d)] ⁠. In essence, the latter inequality merely states that, comparing across states, the relative likelihood of D leading by one vote versus R leading by one vote is greater in state d than in state r. In turn, this is a consequence of the positive correlation between votes and signals and between signals and states. Finally, the fact that [T~1(r)+T~−1(r)]/[T~1(d)+T~−1(d)] lies in between T~1(r)/T~1(d) and T~−1(r)/T~−1(d) is simply an arithmetic necessity.11 In combination with (9), (10) and (11), Lemma 1 implies: Lemma 2. Suppose prDΦ(θ) ⁠, prDR(θ) and prΦR(θ) are strictly interior. For υ sufficiently large, in equilibrium, prDΦ(θ)>prDR(θ)>prΦR(θ). The result is immediate and a formal proof is omitted. Intuitively, a voter is relatively more concerned about certainty after the signal sd when he is playing DΦ than when he is playing ΦR. Conversely, he is relatively more concerned about certainty after the signal sr when he is playing ΦR than when he is playing DΦ. As a result, the voter consults a more R‐leaning news outlet under DΦ than under ΦR. When playing DR, his concerns about certainty after sd and sr are more balanced, which translates in consulting a paper with intermediate bias. 3.3. Equilibrium Voting Behaviour In the previous Section, we derived voters’ optimal choice of news outlet for a given voting strategy. It remains to compare the induced indirect utilities across voting strategies and, for each θ, determine which voting strategy yields the highest pay‐off. In addition, we must prove that an equilibrium, (P*, V*), indeed exists. As we show below, equilibrium voting strategies can take on one of two forms, which only differ in whether DR is played. When DR is played, voting behaviour is the same as in the example in Section 1. It moves from DD for beliefs θ close to zero, to DΦ, to DR, to ΦR and, finally, to RR for beliefs θ close to 1. With some abuse of notation, we write V* = DD; DΦ; DR; ΦR; RR to denote an equilibrium voting strategy of this form. When DR is not played, V* = DD; DΦ; ΦR; RR. Hence, the voting strategy ΦΦ is never played in equilibrium. More formally, DD; DΦ; DR; ΦR; RR and DD; DΦ; ΦR; RR are defined as follows. Denote by ΘDD* the set of voters θ for whom, in equilibrium, the voting strategy DD is a best response, and let the sets ΘΦΦ* and ΘRR* be analogously defined. Also, denote by ΘDR* the set of voters for whom DR is a best response while consulting a strictly interior news outlet 0<prDR(θ)<1 ⁠, and let ΘDΦ* and ΘΦR* be defined along the same lines.12 We say that a set is strictly smaller than another set (≺) if almost all elements in the former are strictly smaller than almost all elements in the latter. We say that a set is strictly smaller than a scalar (also using ‘≺’) if almost all elements in the set are strictly smaller than the scalar. ‘Strictly greater’ is analogously defined. V* = DD; DΦ; DR; ΦR; RR then means that ΘDD* ⁠, ΘDΦ* ⁠, ΘDR* ⁠, ΘΦR* ⁠, and ΘRR* are ordered intervals with non‐empty interiors that partition the type space [0, 1]. That is, 0≺ΘDD*≺ΘDΦ*≺ΘDR*≺ΘΦR*≺ΘRR*≺1 and ΘDD*∪ΘDΦ*∪ΘDR*∪ΘΦR*∪ΘRR*=[0,1]. The only difference with V* = DD; DΦ; ΦR; RR is that, in this case, ΘDR* has an empty interior. In the next Proposition, we prove existence of equilibrium and formally characterise equilibrium voting strategies. Proposition 1. There exists a pure‐strategy symmetric Bayesian Nash equilibrium, (P*,V*). For υ sufficiently large, the equilibrium voting strategy, V*, takes on one of two forms: 1)DD;DΦ;DR;ΦR;RR2)DD;DΦ;ΦR;RR. 1)DD;DΦ;DR;ΦR;RR2)DD;DΦ;ΦR;RR. To understand why ΦΦ is never played in equilibrium, notice that abstention is a best response if and only if a voter’s posterior belief, ρ, satisfies [T~1(r)/T~1(d)]≤[ρ/(1−ρ)]≤[T~−1(r)/T~−1(d)] ⁠.13 For ΦΦ to be optimal, the voter’s posterior must lie within these bounds after both signals. This is only possible if the prior, θ, also lies within these bounds. In other words, the voter must (i) have a centrist prior; and (ii) read a centrist paper that conveys little information after both signals. Notice, however, that such a voter can profitably deviate by consulting a strongly biased news outlet and play DΦ or ΦR. Most of the time, he will receive a signal that conforms to the outlet’s bias. In that case, he continues to abstain. Sometimes, however, he will receive a signal that goes against the outlet’s bias. Provided the bias is sufficiently strong, a contrarian signal will push the voter’s posterior belief outside the narrow band where abstention is optimal, such that the voter now wants to follow the signal. As this deviation allows the voter to use his information at least some of the time, he is strictly better off than under the strategy of always abstaining.14 The proof of existence follows along the lines of Oliveros (2013). Rather than directly searching for a fixed point in the space of best responses, we search for a fixed point in the space of pivotal probabilities. The idea is that any collection of pivotal probabilities gives rise to a profile of best responses which, in turn, gives rise to a collection of pivotal probabilities. We then show that best responses are continuous functions of pivotal probabilities and vice versa. Hence, the implied mapping of pivotal probabilities into itself is also continuous and we can apply Brouwer’s fixed‐point theorem. This gives us a fixed point in pivotal probabilities, which corresponds to a fixed point in the space of best responses, i.e. an equilibrium. 4. Demand for Slant Having characterised equilibrium voting behaviour, we now study the implications for the demand for slant. Technically, the primary result of this Section is to show that the demand for slant is a non‐monotone and discontinuous function of ideology, θ. The economically relevant implication is that abstention breaks ideological segregation and induces ideological moderation in news consumption. Specifically: (i) voters with considerable leanings towards a candidate consult a news outlet that is less biased towards that candidate than other, intrinsically more centrist voters; (ii) some voters with almost identical prior beliefs optimally consult news outlets with very different ideologies, while voters with very different priors consult outlets with the same ideology. As a result, the electorate does not segregate along ideological lines, and voters’ beliefs and voting behaviour are non‐recoverable from their choice of news outlet; (iii) centrist or moderately biased news outlets serve multiple constituencies who use the same information in different ways. As a result, these outlets benefit from disproportionate demand; and (iv) when the electorate becomes more polarised, its choice of news outlets may become ideologically more moderate. First, we establish the non‐monotonicity and discontinuity of the demand for slant. When V* = DD; DΦ; DR; ΦR; RR, there exists a unique type, DΦθDR, where V* switches from DΦ to DR. From Lemma 2 we know that prDΦ(θ)>prDR(θ) ⁠. Hence, at DΦθDR, the demand for slant is non‐monotone and discontinuous: a voter with beliefs slightly to the right of DΦθDR optimally consults a news outlet that is discontinuously more left‐leaning than a voter slightly to the left of DΦθDR. The intuition is that, upon receiving a signal that contradicts his prior, the former plans on voting against his prior, while the latter merely abstains. This makes the more right‐leaning voter significantly more concerned about certainty after signal sr than the more left‐leaning voter. As a result, the right‐leaning voter consults a more left‐leaning news outlet than the left‐leaning voter. At the other side of the ideological spectrum, the analogous phenomenon occurs: a voter with beliefs slightly to the left of the unique crossing point DRθΦR between DR and ΦR optimally consults a news outlet that is significantly more right‐leaning than a voter slightly to the right of that point. When V* = DD; DΦ; ΦR; RR, there exists a unique type, DΦθΦR, where V* switches from DΦ to ΦR. Lemma 2 implies that prDΦ(DΦθΦR)>prΦR(DΦθΦR) ⁠. Hence, also at DΦθΦR, the demand for slant is non‐monotone and discontinuous. We summarise these observations in the following Proposition. Proposition 2. When voters have the option to abstain, the demand for slant is a non‐monotone and discontinuous function of prior beliefs θ. Non‐monotone discontinuities occur at transition points between responsive voting strategies, i.e. DΦθDR, DRθΦR and DΦθΦR. Schematically, the demand for slant as a function of θ is illustrated in the centre and right panels of Figure 2. In the interior of ΘDΦ* ⁠, ΘDR* and ΘΦR* ⁠, the demand for slant is continuous and monotone in θ. This follows from the continuity and monotonicity of prDR(θ) ⁠, prDΦ(θ) and prΦR(θ) ⁠. To the left of DDθDΦ, voters do not consult a news outlet, as no outlet can dissuade them from voting according to their prior. Just to the right of DDθDΦ, voters do consult a news outlet and, nominally, follow the responsive voting strategy DΦ. However, their preferred outlet has a pr so close to zero that they almost always receive the signal sd. In practice, this makes their voting behaviour almost indistinguishable from that of voters to left of DDθDΦ, who do not consult a news outlet at all. Moving farther to the right, voting becomes more responsive, as voters following the voting strategy DΦ begin to consult more centrist outlets. This continues until we reach either DΦθDR or DΦθΦR. At DΦθDR, the optimal pr jumps down from prDΦ(DΦθDR) to prDR(DΦθDR) (centre panel). At DΦθΦR, the optimal pr jumps down from prDΦ(DΦθΦR) to prΦR(DΦθΦR) (right panel). On the right side of the political spectrum, behaviour is analogous. Fig. 2. Open in new tabDownload slide Demand for Slant as a Function of θ (Schematic), Under Compulsory and Voluntary Voting Fig. 2. Open in new tabDownload slide Demand for Slant as a Function of θ (Schematic), Under Compulsory and Voluntary Voting Notice that discrete and non‐monotone jumps in the demand for slant occur at transition points between one responsive voting strategy and another. Under compulsory voting, such transition points do not exist. The reason is that, without abstention, there is only one (sensible) responsive voting strategy, namely, DR. As a result, the demand for slant under compulsory voting is a continuous and monotone function of beliefs, cornering out towards the extremes (Figure 2, left panel). What are the implications of voters’ demand for slant for ideological segregation in news consumption and for the demand faced by news outlets across the ideological spectrum? As illustrated in Figure 2 (left panel), under compulsory voting, each outlet has a single constituency that, roughly speaking, shares the outlet’s ideology. As a result, voters are perfectly ideologically segregated in their news consumption, and the demand for a particular news outlet is roughly proportional to the number (density) of voters with the same ideology as the outlet. Under voluntary voting, by contrast, some news outlets serve two constituencies. When V* = DD; DΦ; DR; ΦR; RR, moderately biased outlets have readers who follow the outlet’s signal in both directions, as well as readers who follow the signal only when it confirms their prior and abstain otherwise (Figure 2, centre panel). The former more or less share the outlet’s ideology, while the latter are ideologically more extreme. The situation for moderately right‐leaning news outlets is analogous. When V* is of the form DD; DΦ; ΦR; RR, it is centrist outlets that serve multiple constituencies, each of whom follows the outlet’s signal in one direction only (right panel). Finally, notice that polarisation of the electorate may cause ideological moderation in news consumption. Consider an equilibrium voting strategy of the form DD; DΦ; DR; ΦR; RR and, for some δ > 0, transfer probability mass from the interval (DΦθDR, DΦθDR + δ) to (DΦθDR − δ, DΦθDR) and from (DRθΦR − δ, DRθΦR) to (DRθΦR, DRθΦR + δ). Clearly, this makes the electorate more polarised. However, news consumption becomes more moderate, as voters the left of DΦθDR consult more centrist outlets than voters to the right of DΦθDR, and voters to the right of DRθΦR consult more centrist outlets than voters to the left of DRθΦR. When V* is of the form DD; DΦ; ΦR; RR, a similar effect occurs when we transfer probability mass from (DΦθΦR − δ, DΦθΦR) to (DΦθΦR − 2δ, DΦθΦR − δ) and from (DΦθΦR, DΦθΦR + δ) to (DΦθΦR + δ, DΦθΦR + 2δ).15 We may conclude: Proposition 3. When voters have the option to abstain, strict ideological segregation in news consumption breaks down. As a result, centrist or moderately biased news outlets benefit from disproportionate demand. Polarisation of the electorate may cause ideological moderation in news consumption. Political commentators fear that the polarisation of the electorate, in combination with voters’ new‐found ability to customise their news consumption, induces them to live in information cocoons and self‐segregate along ideological lines (Sunstein, 2002, 2009; Brooks, 2010). Proposition 3 suggests that, at least for rational voters, this fear is overblown. 5. Extensions and Limitations In this Section, we discuss some extensions and limitations of our model. 5.1. Non‐monotone Elasticities So far, we have assumed that the elasticities δpr,1−pd and δpd,1−pr are strictly increasing in pr and pd respectively. We now study what happens when we relax this assumption and merely assume that G is strictly concave. In the proof of Proposition 1, increasing elasticities served to establish single‐crossing of the indirect utility functions Uθ(DR) and Uθ(DΦ), as well as of Uθ(DR) and Uθ(ΦR). This guaranteed that ΘDΦ* ⁠, ΘDR* ⁠, and ΘΦR* were connected intervals. Under mere concavity of G, ΘDD* and ΘRR* remain connected. However, in equilibria where DR is played, connectedness of ΘDΦ* ⁠, ΘDR* ⁠, and ΘΦR* may fail. That is, while the ordering of the infima and suprema of ΘDΦ* ⁠, ΘDR* ⁠, and ΘΦR* remains unchanged, the sets themselves can no longer be fully ordered. As a result, as we increase θ, the equilibrium voting strategy may jump back and forth multiple times between DΦ and DR on the left side of the political spectrum and between DR and ΦR on the right side. Obviously, this loss of connectedness makes the characterisation of equilibrium less elegant. For the rest, our results remain essentially unchanged: equilibrium is still guaranteed to exist and ΦΦ is never played. Moreover, the demand for slant continues to be a non‐monotone and discontinuous function of prior beliefs. Indeed, in the absence of connectedness, discontinuities in the demand for slant occur at each and every one of the now potentially many transition points between DΦ and DR, and between DR and ΦR. Of these transitions, the ones from DΦ to DR and from DR to ΦR are non‐monotone. By contrast, ‘reverse’ transitions, i.e. from DR to DΦ and from ΦR to DR, are monotone. Finally, notice that when DR is not played, it does not matter whether δpr,1−pd and δpd,1−pr are increasing. 5.2. Unbounded Likelihood Ratios Another assumption underlying our characterisation of equilibrium was that 0 < −G ′(0) < −G ′(1)< ∞. As we showed in Section 2, this means that there is a limit to how much can be learned from any signal from any news outlet. In other words, even at the extremes, likelihood ratios remain bounded away from zero and infinity. With unbounded likelihood ratios (i.e. −G ′(0) = 0 and −G ′(1) = ∞), no matter how extreme a voter’s (non‐degenerate) prior beliefs, there always exists a sufficiently extreme news outlet whose ‘non‐conforming’ signal contains so much information that it overcomes these priors. Even though an extremist news outlet can be expected to send a non‐conforming signal only extremely rarely, the equally extremist voter is strictly better off consulting this news outlet and voting DR, rather than voting in accordance to his prior. In other words, DD and RR are dominated by DR. As a result, only responsive voting strategies, i.e. DΦ, DR and ΦR, are played in equilibrium. Indeed, with unbounded likelihood ratios, equilibrium voting strategies are of the form DΦ; DR; ΦR or DΦ; ΦR. Hence, the demand for slant remains a non‐monotone and discontinuous function of prior beliefs. 5.3. Multiple News Outlets and Non‐binary Signals In our model, voters receive a signal from, at most, one news outlet. By contrast, Gentzkow and Shapiro (2011) find that many voters choose to consult multiple outlets. This raises the question whether a rational voter receiving signals from multiple outlets would ever cross‐over, or whether cross‐over is an artifact of voters receiving a single, binary signal. While a full analysis is complicated and beyond the scope of this study, the general principle underlying abstention‐induced cross‐over does generalise to voters receiving multiple signals.16 To see this, recall that the discontinuities in the demand for slant are caused by jumps from one responsive voting strategy to another. In the presence of multiple signals, these jumps continue to exist and, in fact, multiply in number. Suppose voters collect two conditionally independent signals. For simplicity, assume that both signals have the same accuracy, pr ⁠. Denote by DΦR the strategy of voting for D upon receiving two sd signals, abstaining upon receiving one sd and one sr signal, and voting for R upon receiving two sr signals. Other letter combinations are defined analogously. Now consider the pair of voting strategies DΦΦ and DΦR and let DΦΦθDΦR denote the voter who, conditional on consulting optimal news outlets, is indifferent between DΦΦ and DΦR. After signals {sr, sr}, a voter just to the right of DΦΦθDΦR votes for R , while a voter just to the left of DΦΦθDΦR merely abstains. In that case, the former is relatively more concerned about certainty that the state is r than the latter. As this is the only respect in which the two strategies differ, a voter to the left of DΦΦθDΦR playing DΦΦ optimally consults a more right‐leaning news outlet than a voter to the right of DΦΦθDΦR playing DΦR . An analogous argument reveals that cross‐over in news consumption does not crucially depend on the binariness of signals either. Suppose that voters are limited to consulting a single news outlet but that signals can take on three values, sd, sϕ or sr. In state r, news outlet pr sends signal sd with probability (1−pr)2 ⁠, sϕ with probability 2pr(1−pr) and sr with probability pr2 ⁠. In state d, the chances of sd, sϕ and sr are G(pr)2 ⁠, 2G(pr)[1−G(pr)] and [1−G(pr)]2 respectively. In this set‐up, DΦR denotes the strategy of voting for D upon receiving the signal sd, abstaining upon receiving sϕ and voting for R upon receiving sr. Other letter combinations are similarly defined. As this model is isomorphic to the two‐signal model discussed above, a voter to the left of DΦΦθDΦR optimally consults a more right‐leaning news outlet than a voter to the right of DΦΦθDΦR. 5.4. Consumption Value We have assumed that voters’ utility from information is purely instrumental. However, for many people, news also provides consumption value. To construct a model where voters enjoy news as a consumption good, we have to take a stand as to how it enters the utility function. We have to determine whether the consumption value derives from the content of the news – i.e. from receiving a particular signal – or whether it derives from the bias of the outlet that provides it. For example, a voter may derive pleasure from hearing that the Republican candidate should be elected, or he may derive pleasure from listening to a news outlet whose reporting tends to coincide with what he believes. In the first case, it does not matter whether a particular signal is reported by CNN or Fox News. In the second case, listening to CNN provides different utility from listening to Fox News, independently of what is reported. In addition, we would have to decide how the consumption value interacts with the informational value of news. Suppose that the consumption value of news is separable from its informational value and derives from receiving a signal that conforms to one’s biases. Then it is easy to show that, in large elections, voters only consult the most extreme media outlets. The reason is simple and similar to Morgan and Várdy (2012). In large elections, the probability of being pivotal is negligible. Hence, the instrumental value of information disappears. Voters now face the simple task of selecting the news outlet that most often provides their preferred signal. Of course, this outlet lies at one of the extremes of the ideological spectrum. Clearly, this is an unsatisfactory prediction. We leave a more thorough analysis for future research. 5.5. Unknown Biases Finally, we have also assumed that voters perfectly know the biases of all news media. While, in practice, voters are roughly aware of the slants of the various media outlets, they may not know how strong these biases are exactly. Hence, one may wonder whether our results carry over to an environment with incomplete information about the precise slants of media outlets. While we have not undertaken a formal analysis, uncertainty about the bias of media outlets should not affect our main result, i.e. the non‐monotonicity and discontinuity of the demand for slant. The reason is that the fundamental cause of the discontinuity – namely, the discrete change in the shape of a voter’s pay‐off function upon a jump from one responsive voting strategy to another – remains present. Also, the direction of the jump should remain unaffected, as it remains true that a voter who switches from DΦ to DR (say) suddenly cares relatively more about certainty after the signal sr. As a result, he now prefers to consult a news outlet that, in expectation, is more biased towards D. 6. Literature Review We now put our findings in the context of the extant literature. In recent years, there has been considerable work analysing the demand for news and its impact on political outcomes (DellaVigna and Gentzkow, 2010; Prat and Strömberg, 2011). In a decision theoretic environment, Calvert (1985) and Suen (2004) show that rational agents optimally demand information that is biased towards their priors. In Oliveros and Várdy (2011), we show that this result extends to the strategic environment of elections with compulsory voting. Gentzkow and Shapiro (2006) provide an additional rationale for voters’ apparent confirmatory bias. In their model, voters are not only uncertain about the state of the world but also about the quality of the various media outlets. As a consequence, outlets that provide information conforming to voters’ prior beliefs are thought to be of higher quality. This gives news media an incentive to pander to the biases and prior beliefs of voters. In our model, we have shown that the media want to pander to a belief distribution that differs from the belief distribution of the population at large. By adopting a centrist or moderately biased position, a news outlet can generate disproportionate demand for its services. Hence, we would expect to see ‘bunching’ in these areas.17 The assumption of confirmation bias has been used extensively in reduced‐form modelling of voters’ demand for information. Examples are Mullainathan and Shleifer (2005) and Baron (2006), who study competition in the market for news via differentiation in price and slant; Chan and Suen (2008), who look at political competition; Krasa et al. (2008), who consider the impact of media bias on election outcomes; and Duggan and Martinelli (2011) and Gul and Pesendorfer (2011), who analyse economic policy selection and optimal slant. Papers such as Strömberg (2004), DellaVigna and Kaplan (2007), Gerber et al. (2007, 2009), Chiang and Knight (2011), Enikolopov et al. (2011) and Durante and Knight (2012) study whether news media affect voting behaviour. They find some evidence to that effect. However, the ideological leanings of media outlets and their customers are not as closely correlated as one might expect. Indeed, Gentzkow and Shapiro (2011) show that news consumption is less ideologically segregated than predicted by existing theories. Similar results were found in surveys by Pew (2008, 2009). The contribution of our article is to uncover a fundamental rationale for this imperfect segregation in news consumption. Our article is closely related to the literatures on abstention and information acquisition in voting. Feddersen and Pesendorfer (1996, 1999) show that voters who are (close to) indifferent between supporting either one of two candidates suffer from a ‘swing voter’s curse’. This gives these voters a strict incentive to abstain. Oliveros (2013) studies the incentives to abstain when information acquisition is endogenous and costly. Martinelli (2006, 2007) also allow for endogenous information acquisition and study the information aggregation properties of elections. In our model, information always aggregates in the limit.18 In that regard, our article is more concerned with political segregation and the challenges of studying the industrial organisation of media markets, than with the effect of media bias on the accuracy of elections. Finally, our article is related to the literature on learning from coarse information. The aforementioned papers by Calvert (1985) and Suen (2004) are important references. Bøg (2008) studies a decision‐maker who can observe the binary choices of other experimenters but not the actual outcomes of their experiments. He discusses how priors, preferences and coarseness of observation interact, such that the decision‐maker sometimes chooses to observe experimenters with different preferences. Gill and Sgroi (2012) study a company about to launch a new product. It can publicly test the product and condition its price on whether the product passes or fails. The ability to adjust the price in response to the outcome of the test reinforces the positive effect of passing, and mitigates the negative effect of failing. This convexifies the firm’s profits and induces the firm to select either the softest or the hardest test. Closer to our paper, Meyer (1991) studies how an employer can optimally generate and use coarse information in promotion decisions that are governed by a sequence of rank‐order contests. If the employer can choose not to promote anyone, Meyer (1991) shows that, for purely informational reasons, it may be optimal to handicap the early leader in the final‐period contest. This handicapping can be interpreted as a kind of ‘cross‐over’ on the part of the employer: despite being positively disposed towards the early leader, the employer chooses to stack the final contest against him. The early leader is promoted only if he wins the final contest. Otherwise, no one is promoted. 7. Conclusions In most elections, voters have the option to stay away from the polls. As we have shown, this seemingly innocuous fact has interesting and perhaps surprising implications for the kind of news media that rational voters choose to consult. In particular, voters with relatively pronounced leanings towards either side of the political spectrum optimally consult more centrist news outlets than other, intrinsically more centrist voters. As a result, relatively centrist outlets benefit from disproportionate demand for their services, serving multiple constituencies who use the same information in different ways. What do our findings imply for the risk that voters self‐segregate along strict ideological lines and end up living in information cocoons? We have shown that rigid ideological segregation is an artifact of compulsory voting, which breaks down when voters have the option to abstain. Because centrist voters naturally demand ‘balanced’ reporting, while moderately biased voters tend to consult news media whose ideological positions are more centrist than their own, rigid ideological segregation is confined to the fringes of the political spectrum. Indeed, our theoretical results are roughly consistent with recent empirical evidence, which suggests that there is far more ‘cross‐over’ in news consumption than commonly believed (Gentzkow and Shapiro, 2011). While voluntary voting reduces ideological segregation and induces ideological moderation in news consumption, it does induce more polarisation at the ballot box. That is, under voluntary voting, people who show up at the polls hold more extreme (posterior) beliefs than the population at large. To see why, recall that voters who receive signals that conform to their bias will go and vote. By contrast, voters who receive signals that go against their bias are more likely to stay home and abstain. As a consequence, abstainers tend to hold more moderate beliefs than active voters and the population at large. In order to study the demand for slant, in this article, we have assumed an exogenous supply of slant. We believe that this is a crucial step in developing a sensible model of the market for political news, which needs to account for the forces presented here. Informed by our finding that centrist or moderately biased media benefit from disproportionate demand, the next step is to endogenise the supply of slant by allowing media outlets to position themselves strategically. Media might be concerned with their circulation, with influencing the outcome of the election or with both. We leave this for future research. Appendix A: Micro‐foundation for G In this Appendix, we provide a micro‐foundation for the accuracy function, G. Specifically, we show that accuracy functions with the assumed properties naturally arise when media outlets observe a continuous signal that they coarsen into a binary voting recommendation using some threshold rule, τ (Suen, 2004). Suppose that the states of nature, ω = d, r, are associated with 0 and ξ respectively, where ξ is some strictly positive constant. Each news outlet observes a conditionally independent, continuous signal σ∈R equal to the true state plus noise, i.e. σ = ω + ɛ. Noise ɛ, which is independent of ω, is described by a CDF H on R that admits a well‐behaved PDF h. Specifically, h is strictly positive, differentiable and single‐peaked around zero. A news outlet with threshold rule τ sends signal sd if and only if σ ≤ τ, and sr otherwise. Each threshold, τ, implies a pair of accuracies [pd(τ),pr(τ)]∈[0,1]2 ⁠, where pd(τ) = H(τ) and pr=1−H(τ−ξ) ⁠. There is a continuum of such media outlets, one for each τ∈R ⁠. The implied accuracy function, G~ ⁠, is then G~(pr)=H[ξ+H−1(1−pr)].(A.1) As an illustration, in the next example we derive the accuracy function implied by Logistic noise. Example Let ξ = 1. If ɛ is Logistically distributed with precision λ, then the implied accuracy function is G~(pr)=11+e−λpr1−pr. In the main text, we have assumed that the accuracy function G : [0, 1] → [0, 1] is a strictly decreasing, strictly concave and twice continuously differentiable bijection with 0 < −G ′(0) < −G ′(1) < ∞ (see Assumption 1). It is easily verified that the implied accuracy function in (A.1) is indeed strictly decreasing and twice continuously differentiable. Moreover, G~ runs from G~(0)=1 to G~(1)=0 ⁠, such that it is a bijection. The following proposition establishes that strict concavity of G~ corresponds to strict log‐concavity of the noise density h. Recall that the densities of many standard distributions, including the Normal, the Logistic and the Extreme Value distribution, satisfy strict log‐concavity (Bergstrom and Bagnoli, 2005). Proposition 4. G~ is strictly concave for all ξ if and only if h is strictly log‐concave. Proof Because G~(pr)=H[ξ+H−1(1−pr)] ⁠, we have G~″(pr)=−−h′(ξ+x)+h(ξ+x)h(x)h′(x)[h(x)]2, where x=H−1(1−pr) ⁠. This expression is strictly negative, such that G~ is strictly concave, iff h(ξ+x)h′(x)>h(x)h′(ξ+x).(A.2) If x = 0 then, by differentiability and single‐peakedness around zero, h′(x) = 0 and h′(ξ + x) < 0. Hence, (A.2) holds. If x > 0, then h′(x) < 0 and h′(ξ + x) < 0. In that case, (A.2) is equivalent to dln[h(x)]dx>dln[h(ξ+x)]dx.(A.3) This inequality holds for all ξ > 0 iff h is log‐concave. Next, suppose that x < 0, such that h′(x) > 0. If h′(ξ + x) < 0, (A.2) holds for sure. If h′(ξ + x) > 0, (A.2) is again equivalent to (A.3), such that strict log‐concavity of h is once more a necessary and sufficient condition. Let γσ denote the likelihood ratio, h(σ−ξ)/h(σ), between state r and state d conditional on signal σ. Because G~′(pr)=−h[ξ+H−1(1−pr)]/h[H−1(1−pr)] ⁠, we have that G~′(0)=−1/γ∞ and G~′(1)=−1/γ−∞ ⁠. Hence, the assumption −∞ < G ′(1) < G ′(0) < 0 corresponds to γσ remaining bounded when σ→∞, and bounded away from zero when σ→−∞. In other words, there is a limit to how much can be learned from any signal from any news outlet. While the Logistic distribution satisfies this property, the Normal distribution does not.19 With the Normal distribution in mind, in Section 5 we discussed what happens when γ∞ = ∞ and γ−∞ = 0, i.e. −G ′(1) = ∞ and −G ′(0) = 0. As we have seen, our results remain essentially unchanged. In the baseline model, we also assumed that elasticities δpr,1−pd and δpd,1−pr are strictly increasing (see Assumption 2). We now show that this corresponds to the PDF h being more log‐concave than its CDF H and DCDF 1 − H. Denote by δ~pr,1−pd and δ~pd,1−pr the elasticities implied by G~ ⁠. Then, Lemma 3. δ~pr,1−pd is strictly increasing in pr∈(0,1) for all ξ if and only if h is strictly more log‐concave than 1 − H, i.e. d2lnh(x)(dx)2<d2ln[1−H(x)](dx)2.(A.4) δ~pd,1−pr is strictly increasing in pd∈(0,1) for all ξ if and only if h is strictly more log‐concave than H, i.e. d2lnh(x)(dx)2<d2lnH(x)(dx)2.(A.5) Proof δ~pr,1−pd is strictly increasing in pr∈(0,1) iff −G~′(pr)1−G~(pr)−G~″(pr)G~′(pr)−1pr<0. In terms of H and h, this inequality can be written as h(ξ+x)1−H(ξ+x)−h′(x)h(x)−h′(ξ+x)h(ξ+x)−h(x)pr<0. where x=H−1(1−pr) ⁠. Using that pr=1−H[H−1(1−pr)] ⁠, the condition for δ~pr,1−pd to be strictly increasing simplifies to h(ξ+x)1−H(ξ+x)−h(x)1−H(x)<h′(x)h(x)−h′(ξ+x)h(ξ+x). This inequality holds for all ξ > 0 iff ddxh(x)1−H(x)<−ddx[h′(x)h(x)] or, equivalently, d2ln[1−H(x)](dx)2>d2lnh(x)(dx)2. The proof for δ~pd,1−pr is analogous. Finally, we show that Normal and Logistic noise distributions satisfy the log‐concavity‐ranking conditions in Lemma 3 – and, thus, generate strictly increasing elasticities – while the Extreme Value distribution does not. Proposition 5. Normal and Logistic noise distributions give rise to strictly increasing elasticities δ~pr,1−pd ⁠, δ~pd,1−pr ⁠. The Extreme Value distribution does not. Proof Notice that, for symmetric densities, the conditions (A.4) and (A.5) coincide. This follows from the fact that, under symmetry, H(x) = 1 − H(−x). Hence, to prove the result for the Normal and the Logistic distribution, it suffices to check only one of these inequalities. Let us focus on the first. For the Logistic distribution with mean μ and scale parameter σ, the result is immediate: d2ln[1−H(x)](dx)2=−ex+μσ(exσ+eμσ)2σ2>−2ex+μσ(exσ+eμσ)2σ2=d2lnh(x)(dx)2. For the Normal distribution with mean μ and standard deviation σ, we have d2lnh(x)(dx)2=−1σ2, while d2ln[1−H(x)](dx)2=−ddxh(x)1−H(x)=−1σddxφ(x−μσ)1−Φ(x−μσ)=−1σ2λ′(x). Here, φ(·) denotes the PDF of the Standard Normal distribution, Φ(·) its CDF, and λ(x) its hazard rate. The result now follows from the well‐known fact that λ′(x) < 1. Finally, for the (non‐symmetric) Extreme Value distribution with location parameter μ and scale parameter σ, d2lnh(x)(dx)2=ex−μσσ2=d2lnH(x)(dx)2. Hence, the inequality (A.5) fails. With the Extreme Value distribution in mind, in Section 5 we discussed what happens when the elasticities δpr,1−pd ⁠, δpd,1−pr fail to be strictly increasing. We saw that, while the equilibrium characterisation becomes less elegant, our results remain essentially unchanged. Appendix B: Proofs This Appendix contains formal proofs of claims and results presented in the main text of the paper. (B.1) Preliminaries For future reference, in this subsection, we prove some useful properties of the G function (Lemma 4); provide expressions for the expected pay‐offs of the various voting strategies (Remark 1); list the first‐order conditions for the optimal choice of news outlet conditional on voting strategy (Lemma 5); and derive closed‐form solutions for the various transition types, VWθXY, where responsive voting strategies degenerate into unresponsive voting strategies (Lemma 6). B.1.1 Properties of G Lemma 4. The accuracy function, G, has the following properties. (i) For all pr∈(0,1) ⁠, pr+G(pr)>1. (ii) −G ′(0) < 1 < −G ′(1). (iii) There exists a unique news outlet, p~>1/2 ⁠, such that G(p~)=p~ ⁠. (iv) The ratios [1−G(pr)]/pr and G(pr)/(1−pr) are strictly increasing in pr∈(0,1) ⁠. For pr↓0 ⁠, they take on their infima −G ′(0) and 1 respectively. For pr↑1 ⁠, they take on their suprema 1 and −G ′(1). Proof We prove the claims one by one. Part 1: Let K(pr)=pr+G(pr) and notice that (i) K(0) = K(1) = 1; and (ii) K is strictly concave. Hence, for all pr∈(0,1) ⁠, it must be that K(pr)>1 ⁠. This proves the result. Part 2: Suppose, by contradiction, that −G ′(0) ≥ 1. In a neighbourhood of 0, concavity of G then implies G(pr)≤G(0)−pr=1−pr ⁠. This contradicts the fact that, by Part 1, pr+G(pr)>1 for all pr∈(0,1) ⁠. The proof that 1 < −G ′(1) is analogous. Part 3: Let M(pr)=G(pr)−pr and notice that M(0) = 1 > 0 > M(1) = −1. By continuity of M, there exists an interior p~ such that M(p~)=0 ⁠, or, equivalently, p~=G(p~) ⁠. Because M′(pr)<0 ⁠, p~ is unique. Finally, by Part 1, p~+G(p~)=2p~>1 ⁠. Hence, p~>1/2 ⁠. Part 4: Differentiating [1−G(pr)]/pr with respect to pr yields ddpr[1−G(pr)pr]=−G′(pr)pr−[1−G(pr)]pr2. For pr>0 ⁠, the sign of d[(1−G)/pr]/dpr turns on the sign of −prG′(pr)−[1−G(pr)] and, for pr>0 ⁠, the latter expression is strictly positive. To see this, notice that by strict concavity of G −G′(pr)pr−[1−G(pr)]>1−G(pr)prpr−[1−G(pr)]=0. Hence, [1−G(pr)]/pr is strictly increasing in pr and takes on its infimum for pr↓0 and its supremum for pr=1 ⁠. Using L’Hôpital’s rule we find that the infimum is equal to −G ′(0), while the supremum is equal to 1. The proof for G(pr)/(1−pr) is analogous. B.1.2 Expected Pay‐offs of Voting Strategies Remark The expected pay‐offs of the various voting strategies are Uθ(DD)=L(θ)−T1(r)+T0(r)+12T−1(r)θ+12T−1(d)(1−θ),Uθ(RR)=L(θ)−12T1(r)θ+12T1(d)+T0(d)+T−1(d)(1−θ),Uθ(ΦΦ)=L(θ)−T1(r)+12T0(r)θ+12T0(d)+T−1(d)(1−θ), UθprDR(DR)=L(θ)−(1−12prDR)T1(r)+(1−prDR)T0(r)+12(1−prDR)T−1(r)θ−12[1−G(prDR)]T1(d)+[1−G(prDR)]T0(d)+1−12G(prDR)T−1(d)(1−θ), UθprDΦ(DΦ)=L(θ)−T1(r)+(1−12prDΦ)T0(r)+12(1−prDΦ)T−1(r)θ−12[1−G(prDΦ)]T0(d)+1−12G(prDΦ)T−1(d)(1−θ),UθprΦR(ΦR)=L(θ)−(1−12prΦR)T1(r)+12(1−prΦR)T0(r)θ−121−G(prΦR)T1(d)+1−12G(prΦR)T0(d)+T−1(d)(1−θ).Here, L(θ) incorporates the pay‐offs associated with all non‐pivotal events. Proof Trivial. B.1.3 First‐order Conditions Lemma 5. The first‐order conditions for prXY(θ) ⁠, XY ∈ {DΦ, DR, ΦR}, to be optimal are −G′(prDR)=T~−1(r)+T~1(r)T~−1(d)+T~1(d)θ1−θ,(B.1) −G′(prDΦ)=T~−1(r)T~−1(d)θ1−θ,(B.2) −G′(prΦR)=T~1(r)T~1(d)θ1−θ.(B.3) Proof The result follows immediately from differentiating with respect to θ the relevant pay‐off expressions given in Remark 1. B.1.4 Transition Types VWθXY Lemma 6. The unique and strictly interior transition types between responsive and unresponsive voting strategies are DDθDΦ=−G′(0)T~−1(r)T~−1(d)−G′(0),DDθDR=−G′(0)T~−1(r)+T~1(r)T~−1(d)+T~1(d)−G′(0)andΦΦθΦR=−G′(0)T~1(r)T~1(d)−G′(0),DΦθΦΦ=−G′(1)T~−1(r)T~−1(d)−G′(1),DRθRR=−G′(1)T~−1(r)+T~1(r)T~−1(d)+T~1(d)−G′(1)andΦRθRR=−G′(1)T~1(r)T~1(d)−G′(1). Proof The responsive strategy (prDR(θ),DR) degenerates into the unresponsive strategy DD at type θ = DDθDR which solves the first‐order condition (7) for prDR=0 ⁠. That is, −G′(0)=T~−1(r)+T~1(r)T~−1(d)+T~1(d)θ1−θ. Solving for θ, we find DDθDR=−G′(0)T~−1(r)+T~1(r)T~−1(d)+T~1(d)−G′(0). The derivations of the other transition types are analogous. (B.2) Proof of Lemma 1 We now turn our attention to Lemma 1, which establishes that [T~1(r)/T~1(d)]<[T~1(r)+T~−1(r)]/[T~1(d)+T~−1(d)]<[T~−1(r)/T~−1(d)] ⁠. First, we determine when abstention is a strict best response. Lemma 7. A voter with posterior belief ρ that the state is r strictly prefers to abstain if and only if T~1(r)T~1(d)<1−ρρ<T~−1(r)T~−1(d). Proof The expected pay‐off of voting for D is L(θ)−T1(r)+T0(r)+12T−1(r)ρ+12T−1(d)(1−ρ), while the expected pay‐off of abstaining is L(θ)−T1(r)+12T0(r)ρ+12T0(d)+T−1(d)(1−ρ). Hence, the expected pay‐off from abstaining is strictly greater than from voting for D if and only if 12[T−1(d)+T0(d)](1−ρ)−12[T−1(r)+T0(r)]ρ<0.(B.4) The first term corresponds to the chance that the state is d and the voter (rightly) casts the decisive vote in favour of D, while the second term corresponds to the chance that the state is r and the voter (wrongly) casts the decisive vote in favour of D. The inequality in (B.4) is equivalent to (1−ρ)/ρ<[T~−1(r)/T~−1(d)] ⁠. Along the same lines, we find that the voter’s pay‐off from abstaining is strictly greater than from voting for R if and only if (1−ρ)/ρ>[T~1(r)/T~1(d)] ⁠. Combining these inequalities yields the result. With the help of Lemma 7, we can now prove Lemma 1. Proof of Lemma We prove the Lemma by showing that, for υ sufficiently large, [T~1(r)/T~1(d)]<[T~−1(r)/T~−1(d)] ⁠. The fact that [T~1(r)/T~1(d)]<[T~1(r)+T~−1(r)]/[T~1(d)+T~−1(d)]<[T~−1(r)/T~−1(d)] then follows as an arithmetic necessity. Recall that T−1(ω)=e−υ[tR(ω)+tD(ω)]∑n=0∞[υtD(ω)]nn![υtR(ω)]n+1(n+1)!. Following Feddersen and Pesendorfer (1999), we use the modified Bessel function Iz(x)=12xz∑n=0∞(14x2)nn!(n+z)!, to get T−1(ω)=e−υ[tR(ω)+tD(ω)]tR(ω)tD(ω)I12υtD(ω)tR(ω). For z ∈ {0, 1}, limx→∞ex/2πx/Iz(x)=1 ⁠. Hence, for large υ, T−1(ω)≈tR(ω)tD(ω)e−v[tD(ω)−tR(ω)]24πvtD(ω)vtR(ω).(B.5) Analogous calculations for T0(ω) and T1(ω) yield T0(ω)≈e−v[tD(ω)−tR(ω)]24πvtD(ω)vtR(ω),(B.6) T1(ω)≈tD(ω)tR(ω)e−v[tD(ω)−tR(ω)]24πvtD(ω)vtR(ω).(B.7) Now suppose that, contrary to the claim in the Lemma, [T~1(r)/T~1(d)]≥[T~−1(r)/T~−1(d)] ⁠. Then, for sufficiently large υ, it must be that 1+tD(r)tR(r)1+tR(r)tD(r)≥1+tD(d)tR(d)1+tR(d)tD(d). Because the function (1+x)/[1+(1/x)] is strictly increasing in x, the last inequality is equivalent to tD(r)tR(r)≥tD(d)tR(d).(B.8) Denote by ΥDR the set of voters who, in equilibrium, strictly prefer [prDR(θ),DR] over DD and RR. Denote by ΥRR the set of voters who weakly prefer RR over [prDR(θ),DR] and DD. Finally, denote by ΥDD the set of voters who weakly prefer DD over [prDR(θ),DR] and RR. Because DDθDR < DRθRR by Lemma 6, we have ΥDD = [0, DDθDR], ΥDR=(DDθDR,DRθRR) and ΥRR = [DRθRR, 1]. Moreover, from Lemma 7 we know that when [T~1(r)/T~1(d)]≥[T~−1(r)/T~−1(d)] ⁠, nobody abstains. Hence, in state r, the probability that a randomly drawn voter votes for R is tR(r)=ζη+(1−ζ)∫ΥDRprDR(θ)f(θ)dθ+∫ΥRRf(θ)dθ. Similarly, in state d, the probability that a randomly drawn voter votes for R is tR(d)=ζη+(1−ζ)∫ΥDR{1−G[prDR(θ)]}f(θ)dθ+∫ΥRRf(θ)dθ. Notice that for all θ∈ΥDR ⁠, 0<prDR(θ)<1 ⁠. The first part of Lemma 4 then implies that prDR(θ)>1−G[prDR(θ)] ⁠. Hence, tR(r) > tR(d) and tD(d) > tD(r), which contradicts the inequality in (B.8). This completes the proof. (B.3) Proof of Proposition 1 This subsection contains the proof of Proposition 1; i.e. the claim that equilibrium exists, and that equilibrium voting strategies are of the form DD; DΦ; DR; ΦR; RR or DD; DΦ; ΦR; RR. First we show that, for any T ∈ (0, 1)6, the pay‐offs of the various unresponsive voting strategies satisfy single‐crossing. We also calculate closed‐form solutions for these crossing points. With slight abuse of notation, we use the same notational convention for crossing points as for transition types between responsive and unresponsive voting strategies, i.e. VWθXY. Lemma 8. Fix some T ∈ (0, 1)6. The pay‐offs of unresponsive voting strategies DD, ΦΦ and RR satisfy single‐crossing in θ. Specifically, (i) Uθ(DD) ≥ Uθ(ΦΦ) iff θ≤DDθΦΦ≡1T~−1(r)T~−1(d)+1, (ii) Uθ(ΦΦ) ≥ Uθ(RR) iff θ≤ΦΦθRR≡1T~1(r)T~1(d)+1, (iii) Uθ(DD) ≥ Uθ(RR) iff θ≤DDθRR≡1T~1(r)+T~−1(r)T~1(d)+T~−1(d)+1. Proof The result trivially follows from the expressions for expected pay‐offs given in Remark 1. Together, Lemmas 6 and 8 allow us to derive the following partial ordering of transition types and crossing points. Lemma 9. Fix some T ∈ (0, 1)6. Transition types and crossing points VWθXY satisfy the following partial order: 0<DΦθΦΦ><DRθRR><ΦRθRR><1,0<DDθΦΦ><DDθRR><ΦΦθRR><1,0<DDθDΦ<DDθDR<ΦΦθΦR<1. Proof From Lemma 1, we know that [T~1(r)/T~1(d)]<[T~1(r)+T~−1(r)]/[T~1(d)+T~−1(d)]<[T~−1(r)/T~−1(d)] ⁠. From the second part of Lemma 4, we know that 0 < −G ′(0) < 1 < −G ′(1) < ∞. The ranking then follows from applying these inequalities to the closed‐form solutions for the various VWθXY derived in Lemmas 6 and 8, and the fact that T ∈ (0, 1)6. We now show that, for any T ∈ (0, 1)6, the pay‐offs of responsive voting strategies also satisfy single‐crossing, provided that δpr,1−pd and δpd,1−pr are strictly increasing. Lemma 10. Fix some T ∈ (0, 1)6. If δpr,1−pd and δpd,1−pr are strictly increasing in pr and pd respectively, then the indirect utilities of the responsive voting strategies DR, DΦ and ΦR satisfy single‐crossing. Specifically, (i) on θ ∈ [DDθDR, DΦθΦΦ], Uθ(DR) − Uθ(DΦ) crosses zero at most once, and from below; (ii) on θ ∈ [ΦΦθΦR, DRθRR], Uθ(DR) − Uθ(ΦR) crosses zero at most once, and from above; (iii) on θ ∈ [ΦΦθΦR, DΦθΦΦ], Uθ(DΦ) − Uθ(ΦR) crosses zero exactly once, and from above. Proof Part 1: From the expressions for UθprDR(DR) and UθprDΦ(DΦ) given in Remark 1, we find UθprDR(DR)−UθprDΦ(DΦ)1−θ=T~−1(r)+T~1(r)2prDR(θ)−T~−1(r)2prDΦ(θ)θ1−θ+T~−1(d)2{1−G[prDΦ(θ)]}−T~−1(d)+T~1(d)2{1−G[prDR(θ)]}. Using the first‐order conditions for prDR(θ) and prDΦ(θ) given in Lemma 5 and simplifying yields Uθ(DR)−Uθ(DΦ)1−θ=T~−1(d)2{prDΦG′(prDΦ)+[1−G(prDΦ)]}−T~−1(d)+T~1(d)2{prDRG′(prDR)+[1−G(prDR)]}.(B.9) Hence, [Uθ(DR)−Uθ(DΦ)]/(1−θ)>0 iff T~−1(d)T~−1(d)+T~1(d)>prDRG′(prDR)+1−G(prDR)prDΦG′(prDΦ)+1−G(prDΦ).(B.10) To prove the claim, we will show that the RHS of (B.10) is strictly increasing in θ. Differentiating with respect to θ and simplifying yields that the RHS of (B.10) is strictly increasing if G″[prDR(θ)]dprDR(θ)dθprDR(θ){1−G[prDΦ(θ)]+G′[prDΦ(θ)]prDΦ(θ)}>G″[prDΦ(θ)]dprDΦ(θ)dθprDΦ(θ){1−G[prDR(θ)]+G′[prDR(θ)]prDR(θ)}.(B.11) From the first‐order conditions for prDR(θ) and prDΦ(θ) it follows that G″(prDR)dprDR(θ)dθ=G′(prDR)1θ(1−θ),G″(prDΦ)dprDΦ(θ)dθ=G′(prDΦ)1θ(1−θ). Substituting these expressions for G ′′ back into (B.11) and rearranging, we find that the RHS of (B.10) is strictly increasing in θ if −G′(prDR)prDR1−G(prDR)<−G′(prDΦ)prDΦ1−G(prDΦ).(B.12) Notice that [−G′(pr)pr]/[1−G(pr)]=δpr,1−pd ⁠, which is strictly increasing by assumption. Moreover, from Lemma 2, we know that prDR(θ)<prDΦ(θ) ⁠. Hence, the inequality in (B.12) indeed holds. We may conclude that [Uθ(DR)−Uθ(DΦ)]/(1−θ) – and therefore, Uθ(DR)−Uθ(DΦ) – crosses zero at most once, and from below. Part 2: The proof is analogous to Part 1. Part 3: The expression for [Uθ(DR)−Uθ(DΦ)]/(1−θ) in (B.9) and its analogue for [Uθ(DR)−Uθ(ΦR)]/(1−θ) yield Uθ(DΦ)−Uθ(ΦR)1−θ=Uθ(DR)−Uθ(ΦR)1−θ−Uθ(DR)−Uθ(DΦ)1−θ=T~1(d)+T~−1(d)2[G′(prDR)+1]−T~−1(d)2{prDΦG′(prDΦ)+[1−G(prDΦ)]}−T~1(d)2[G(prΦR)+G′(prΦR)(1−prΦR)]. The first‐order conditions (B.1), (B.2) and (B.3) imply that −G′(prDR)=T~−1(d)T~−1(d)+T~1(d)[−G′(prDΦ)]+T~1(d)T~−1(d)+T~1(d)[−G′(prΦR)]. Hence, Uθ(DΦ)−Uθ(ΦR)1−θ=T~−1(d)2[G(prDΦ)+G′(prDΦ)(1−prDΦ)]+T~1(d)2[1−G(prΦR)+G′(prΦR)prΦR].(B.13) To prove the claim, we will show that the RHS of (B.13) is strictly increasing in θ. Differentiating the RHS with respect to θ gives T~−1(d)2(1−prDΦ)G″(prDΦ)dprDΦdθ+T~1(d)2prΦRG″(prΦR)dprΦRdθ.(B.14) From the first‐order conditions (B.2) and (B.3), it follows that G″(prDΦ)dprDΦdθ=−T~−1(r)T~−1(d)1(1−θ)2, G″(prDΦ)dprDΦdθ=−T~−1(r)T~−1(d)1(1−θ)2, Substituting these expressions back into (B.14), we find that the derivative of the RHS of (B.13) is −12(1−prDΦ)T~−1(r)+prΦRT~1(r)1(1−θ)2<0. We may conclude that [Uθ(DΦ)−Uθ(ΦR)]/(1−θ) is strictly decreasing in θ on [ΦΦθΦR, DΦθΦΦ]. Hence, Uθ(DΦ) − Uθ(ΦR) crosses zero at most once, and from above. To see that a crossing indeed takes place, notice that at θ = ΦΦθΦR, prΦR=0<prDΦ ⁠, where the inequality follows from Lemma 2. Hence, at ΦΦθΦR, Uθ(DΦ) > Uθ(ΦR) = Uθ(ΦΦ). Similarly, at θ = DΦθΦΦ, prDΦ=1>prΦR ⁠, which implies that Uθ(DΦ) = Uθ(ΦΦ) < Uθ(ΦR). The intermediate value theorem then proves the result. Next, we prove that, for any T ∈ (0, 1)6, DR and ΦΦ never coexist. Lemma 11. Fix some T ∈ (0, 1)6. If, for this T, ΦΦ is played by a positive measure of voters θ ∈ [0, 1], then the measure of voters playing DR is zero. Proof We prove the Lemma by showing that when DR is played for a particular T ∈ (0, 1)6, then ΦΦ is not played. A necessary condition for DR to be played is that, conditional on sd, D is weakly better than Φ. Similarly, conditional on sr, R must be weakly better than Φ. The expected pay‐off of voting for D – respectively, R – conditional on s is Uθpr(D|s)=L(θ)−T1(r)+T0(r)+12T−1(r)ρθpr(s)+12T−1(d)[1−ρθpr(s)],Uθpr(R|s)=L(θ)−12T1(r)ρθpr(s)+12T1(d)+T0(d)+T−1(d)[1−ρθpr(s)], while the expected pay‐off of abstaining is Uθpr(Φ|sd)=L(θ)−T1(r)+12T0(r)ρθpr(s)+12T0(d)+T−1(d)[1−ρθpr(s)]. Hence, Uθpr(D|sd)≥Uθpr(Φ|sd) if and only if ρθpr(sd)1−ρθpr(sd)≤T~−1(d)T~−1(r). Using (5), we can rewrite this as θ1−θ≤G(pr)1−prT~−1(d)T~−1(r). By Lemma 4, G(pr)/(1−pr) is strictly increasing in pr ⁠. Hence, θ1−θ≤limpr↑1G(pr)1−prT~−1(d)T~−1(r)=−G′(1)T~−1(d)T~−1(r).(B.15) Analogous calculations show that if Uθpr(R|sr)≥Uθpr(Φ∣sr) ⁠, then θ1−θ≥limpr↓01−G(pr)prT~1(d)T~1(r)=−G′(0)T~1(d)T~1(r).(B.16) Together, (B.15) and (B.16) imply that −G′(0)=T~1(d)T~1(r)≤−G′(1)=T~−1(d)T~−1(r).(B.17) Next, notice that a necessary condition for ΦΦ to be played is that DΦθΦΦ<ΦΦθΦR, or, equivalently, DΦθΦΦ1−DΦθΦΦ=−G′(1)T~−1(d)T~−1(r)<−G′(0)T~1(d)T~1(r)=ΦΦθΦR1−ΦΦθΦR.(B.18) Finally, observe that (B.18) contradicts (B.17). This proves the Lemma. The following Lemma establishes that the voting strategy ΦΦ is never played in equilibrium. (Notice that, in contrast to the previous lemmas, Lemma 12 below is true only in equilibrium, and not for arbitrary T ∈ (0, 1)6). Lemma 12. ΦΦ is not played in equilibrium. Proof We prove the Lemma by showing that, in equilibrium, ΦΦθΦR < DΦθΦΦ. This is equivalent to showing that T~1(d)T~1(r)T~−1(r)T~−1(d)<−G′(1)−G′(0).(B.19) The expressions for Tx(ω), x ∈ {−1, 0, 1}, in (B.5), (B.6) and (B.7) imply T~1(d)T~1(r)T~−1(r)T~−1(d)=tR(r)tR(d)tD(d)tD(r).(B.20) Substituting (B.20) into (B.19) gives that ΦΦθΦR < DΦθΦΦ if and only if tR(r)tR(d)tD(d)tD(r)<−G′(1)−G′(0).(B.21) Next, notice that tR(r)tR(d)=ζη+(1−ζ)[∫ΘDR*pDR(θ)dF(θ)+∫ΘΦR*pΦR(θ)dF(θ)+∫ΘRR*dF(θ)]ζη+(1−ζ)∫ΘDR*{1−G[pDR(θ)]}dF(θ)+∫ΘΦR*{1−G[pΦR(θ)]}dF(θ)+∫ΘRR*dF(θ). By strict concavity of G, we have G(p)<G(0)+G′(0)p, which yields 1−G(p)>−G′(0)p. Therefore, tR(r)tR(d)<ζη+(1−ζ)∫ΘDR*pDR(θ)dF(θ)+∫ΘΦR*pΦR(θ)dF(θ)+∫ΘRR*dF(θ)ζη+(1−ζ)−G′(0)[∫ΘDR*pDR(θ)dF(θ)+∫ΘΦR*pΦR(θ)dF(θ)]+∫ΘRR*dF(θ)<1−G′(0).(B.22) where the last inequality follows from the fact that −G ′(0) < 1. An analogous argument reveals that tD(d)tD(r)<−G′(1).(B.23) Combining (B.22) and (B.23) implies (B.21). This completes the proof. Before proceeding with the actual proof of Proposition 1, we introduce some more notation. Fix some T ∈(0, 1)6 and let ΘDD be the set of voters θ ∈ [0, 1] for whom the strategy DD is a best response conditional on T. That is, for given pivotal probabilities, the indirect utility from DD is (weakly) greater than the indirect utility from any other strategy. Let ΘRR and ΘΦΦ be analogously defined. Define ΘDR, to be the set of voters θ ∈ [0, 1] for whom the strategy (prDR(θ),DR) is a best response and 0<prDR(θ)<1 ⁠. Let ΘDΦ and ΘΦR be defined analogous to ΘDR. Proof of Proposition 1. We prove the proposition in three steps. In Step 1 we show that, for fixed T ∈ (0, 1)6, voting strategies take on one of three forms, namely, (1) DD; DΦ; DR; ΦR; RR, (2) DD; DΦ; ΦR; RR, or (3) DD; DΦ; ΦΦ; ΦR; RR. Using the structure this imposes, in Step 2, we prove that an equilibrium always exists. Finally, in Step 3, we invoke Lemma 12 to conclude that, in equilibrium, only voting strategies of forms (1) and (2) can occur. Step 1: Fix some T ∈ (0, 1)6. Lemma 11 establishes that when ΦΦ is played, then DR is not played. This implies that we may restrict attention to three possible cases, namely: (1) ΦΦ is played and DR is not played; (2) DR is played and ΦΦ is not played; (3) Neither ΦΦ nor DR are played. Case 1: Suppose that ΦΦ is played and DR is not played. In that case, we have to consider the voting strategies DD, DΦ, ΦΦ, ΦR and RR. From Lemma 8, we know that the pay‐offs of the unresponsive voting strategies DD, ΦΦ and RR satisfy single‐crossing in θ. Specifically, voter θ prefers DD over ΦΦ iff: θ≤DDθΦΦ≡1T~−1(r)T~−1(d)+1; he prefers RR over ΦΦ iff θ≥ΦΦθRR≡1T~−1(r)T~−1(d)+1; and he prefers DD over RR iff θ≤DDθRR≡1T~1(r)+T~−1(r)T~1(d)+T~−1(d)+1. As to the crossing point between the responsive strategies DΦ and ΦR, notice the following. Because ΦΦ is played, it must be that DΦθΦΦ < ΦΦθΦR. Hence, at the point where DΦ and ΦR generate equal pay‐offs, both voting strategies have already degenerated into ΦΦ. Therefore, DΦ and ΦR do not cross. From Lemma 9 and the fact that DΦθΦΦ < ΦΦθΦR, we know that transition types and crossing points can be ranked as follows: 0<DDθDΦ<DDθΦΦ<DΦθΦΦ<ΦΦθΦR<ΦΦθRR<ΦRθRR and DDθΦΦ<DDθRR<ΦΦθRR. This implies that ΘDD(T) = [0, DDθDΦ], ΘDΦ(T) = (DDθDΦ ,DΦθΦΦ), ΘΦΦ(T) = [DΦθΦΦ ,ΦΦθΦR], ΘΦR(T) = (ΦΦθΦR ,ΦRθRR) and ΘRR(T)=[ΦRθRR,1] ⁠. Hence, if ΦΦ is played and DR is not played, V(T) takes on the form DD; DΦ; ΦΦ; ΦR; RR. Case 2: Suppose that DR is played and ΦΦ is not played. In that case, we have to consider the voting strategies DD, DΦ, DR, ΦR and RR. From Lemma 9, we know that 0<DDθDΦ<DDθDR<DDθRR<DRθRR<ΦRθRR<1. This implies that inf ΘDR > DDθDΦ and sup ΘDR < ΦRθRR. In other words: ΘDR begins strictly to the right of where ΘDΦ begins and ends strictly to the left of where ΘΦR ends. By Lemma 10, Uθ(DR) − Uθ(DΦ) crosses zero at most once and from above, while Uθ(DR) − Uθ(ΦR) crosses zero at most once and from below. Because, by assumption, the interior of ΘDR is not empty, the crossing points DΦθDR and DRθΦR must indeed exist and DΦθDR < DRθΦR. This implies that ΘDD(T) = [0,DDθDΦ], ΘDΦ(T) = (DDθDΦ, DΦθDR], ΘDR(T) = [DΦθDR, DRθΦR], ΘΦR(T) = [DRθΦR, ΦRθRR) and ΘRR(T)=[ΦRθRR,1] ⁠. Hence, if DR is played and ΦΦ is not played, then V(T) takes on the form DD; DΦ; DR; ΦR; RR. Case 3: Suppose that neither DR nor ΦΦ are played. In that case, we have to consider the voting strategies DD, DΦ, ΦR and RR. Because neither ΦΦ nor DR are played, it must be that ΦΦθΦR ≤ DΦθΦΦ. (Else, ΦΦ would be played by all θ ∈ [DΦθΦΦ, ΦΦθΦR].) The third part of Lemma 10 then implies that the crossing point, DΦθΦR, between DΦ and ΦR exists, is unique, and lies between ΦΦθΦR and DΦθΦΦ. Combining this with the ranking in Lemma 9, we may conclude that 0<DDθDΦ<DΦθΦR<ΦRθRR<1. This implies that ΘDD(T) = [0, DDθDΦ], ΘDΦ(T) = (DDθDΦ, DΦθΦR), ΘΦR(T) = (DΦθΦR, ΦRθRR) and ΘRR(T)=[ΦRθRR,1] ⁠. Hence, if neither DR nor ΦΦ are played, then V(T) takes on the form DD; DΦ; ΦR; RR. Step 2: We prove the existence of equilibrium by showing that there exists a fixed point in pivotal probabilities, T* ∈ (0, 1)6, and that it gives rise to a fixed point in best‐responses, i.e. an equilibrium. From (9), (10) and (11) we know that prDR(θ;T) ⁠, prDΦ(θ;T) and prΦR(θ;T) are continuous functions of T. As a result, the indirect utility functions, Uθ(DR; T), Uθ(DΦ; T), Uθ(ΦR; T), Uθ(DD; T), Uθ(RR; T) and Uθ(ΦΦ; T) are also continuous in T. The probability that a random voter casts a vote for R in state r is tR(r;T)=ζ(1−η)+ζ∫ΘDR(T)prDR(θ)f(θ)dθ+∫ΘΦR(T)prΦR(θ)f(θ)dθ+∫ΘRR(T)f(θ)dθ. The expressions for tD(d; T), tD(r; T) and tR(d; T) are analogous. Recall from Step 1 that, for all T ∈ (0, 1)6, the sets ΘXY(T), {X,Y} ∈ {D, Φ, R}2, partition the interval [0, 1]. Moreover, the boundaries between them, which consist of transition types and crossing points XYθVW, are continuous in T. This implies that all tX(ω; T), where X ∈ {D, R} and ω ∈ {d, r}, are also continuous in T. Finally, notice that the probability of each vote count is a continuous function of tX(ω; T). This makes T a continuous function of itself. Denote the implied mapping by Γ. Since {ζ, η} ∈ (0, 1)2, tX(ω; T) > 0. Hence, there exists a 0 < α < 1/2 such that, for all T ∈ (0, 1)6, Γ(T) ∈ [α, 1 − α]6. Applying Brouwer’s fixed‐point theorem to the mapping Γ: [α, 1−α]6→[α, 1−α]6 implies that Γ admits a fixed point T* = Γ(T*) ∈ [α, 1−α]6. Since voters’ best responses are pinned down by prDR(θ;T) ⁠, prDΦ(θ;T) ⁠, prΦR(θ;T) and ΘXY(T), {X, Y} ∈ {D, Φ, R}2, a fixed point T* gives rise to a fixed point in best responses and, hence, an equilibrium of the voting game. Step 3: Finally, from Lemma 12 we know that, in equilibrium, ΦΦ is not played. Hence, at T*, V* can only be of the form DD; DΦ; DR; ΦR; RR or DD; DΦ; ΦR; RR. This completes the proof of Proposition 1. Footnotes 1 " For morning show ratings, see http://www.mediabistro.com. For a discussion of the success of PBS’s Newshour, see Hagey (2011). 2 " The example is special in several respects. First, preferences are ‘pseudo‐expressive’, in the sense that a voter’s pay‐off only depends on whether his own vote matches the state. Second, the cost of voting for the ‘wrong’ candidate is greater than the benefit of voting for the ‘right’ candidate. Third, there are only two media outlets and their biases are symmetric. Towards the end of this Section, we discuss in detail which aspects of the example carry over to more canonical voting environments. 3 " All results can be extended to the case where ζ = 0 but some extra care is needed to prove existence. Details can be provided upon request. 4 " Because the number of voters is Poisson distributed, we could accommodate ‘partisans’ who always prefer to abstain. This would not change the results. 5 " To transform our model into a model with homogeneous 50–50 priors and heterogeneous pay‐offs, simply redefine voters’ losses associated with the winner not matching the state to be li,d = −2(1 − θi) in state d and lr,i = −2θi in state r. Conversely, to move from a homogenous prior ϑ and heterogeneous losses (li,d,li,r) to heterogeneous priors and homogenous, unit losses, redefine voters’ priors to be θi = ϑlr,i/[(1 − τ)ld,i+τlr,i]. More generally, our model can accommodate heterogeneity in both pay‐offs and beliefs, as long as preferences are ‘non‐adversarial’, i.e. co‐monotone: that is, non‐partisans agree that candidate D is preferred in state d, while candidate R is preferred in state r. To see this, suppose non‐partisan voter i earns αi when D is elected in state d, βi when D is elected in state r, γi when R is elected in state d and δi when R is elected in state r. When comparing a vote for D, a vote for R and abstaining, only the pay‐off differences ld,i ≡ γi − αi and lr,i ≡ βi − δi matter. Provided that preferences are co‐monotone such that ld,i,lr,i < 0 for all i, this implies that we can redefine voters’ priors to be θ′ = θlr,i/[(1−θ)ld,i+θlr,i] and proceed as before. 6 " For maximum generality, we have simply postulated an accuracy function G and endowed it with certain plausible properties. This naturally raises the question what kind of information technology and behaviour on the part of media outlets generates accuracy functions with the assumed properties. In Appendix A, we provide a micro‐foundation for G and show that such accuracy functions naturally arise when media outlets observe a noisy continuous signal that they coarsen into a binary voting recommendation, using some threshold rule (Suen, 2004). 7 " A micro‐foundation for Assumption 2 can be found in Appendix A. In Section 5 we relax this assumption and show that our results remain essentially unchanged. 8 " In Section 5, we relax the assumption of bounded derivatives and show that our results remain essentially unchanged. 9 " Recall that υ is the parameter of the Poisson distribution describing the total number of voters. Hence, υ is equal to the expected number of voters. 10 " Expressions for Uθpr(DΦ) and Uθpr(ΦR) can be found in Remark 1 in Appendix B. The corresponding first‐order conditions for prDΦ(θ) and prΦR(θ) can be found in Lemma 5. 11 " As shown by Feddersen and Pesendorfer (1996, 1999), voters with posterior beliefs ρ such that [T~1(r)]/[T~1(d)]<ρ/(1−ρ)<[T~−1(r)]/[T~−1(d)] suffer from the swing voters’ curse. That is, whichever candidate they vote for, these voters are more likely to push the election in the wrong direction than in the right direction. Hence, they are strictly better off abstaining. 12 " We insist on strict interiority of prDR(θ) in the definition ΘDR* to avoid overlap of ΘDR* with ΘDD* and ΘRR* ⁠. Without this additional requirement, significant but spurious overlap would occur because DR reduces to DD whenever prDR(θ)=0 ⁠, while it reduces to RR when prDR(θ)=1 ⁠. Similar issues would arise with ΘDΦ* and ΘΦR* ⁠. 13 " See Lemma 7 in Appendix B for a formal proof of this claim. 14 " Because ΦΦ is not played in equilibrium, voters only abstain as part of a responsive strategy. Hence, abstainers are not ‘uninformed’, in the sense that they have, in fact, consulted a news outlet. Cf. Feddersen (2004): ‘Because uninformed independents abstain and informed independents vote, Feddersen (2004) provides an informational explanation for why better educated individuals are more likely to vote’. The lesson is that, if the media landscape is sufficiently broad, the association between abstaining and being uninformed breaks down. 15 " Somewhat paradoxically, the electorate is better informed under voluntary than under compulsory voting, in the sense that more voters consult a news outlet. To see this, notice that types θ ∈ (DDθDΦ, DDθDR) only consult an outlet when they have the option to abstain but remain uninformed when they must cast a ballot. Intuitively, the option to abstain allows for more ‘nuanced’ voting behaviour, which increases the value of information. On a more technical level, recall that the swing voter’s curse implies that a voter who is indifferent between D and R strictly prefers to abstain. This implies that, at DDθDR, the voting strategy DΦ does strictly better than both DD and DR. As the pay‐off difference between DD and DΦ is decreasing in θ, the indifference point between DΦ and DD must lie strictly to the left of DDθDR. For the same reason, ΦRθRR > DRθRR. 16 " Fuchs and Oliveros (2011) study the implications of sequential information on the demand for slant in a model without abstention. They show that sequentiality also creates incentives for cross‐over, albeit through a different mechanism. 17 " Besley and Prat (2006) study an altogether different source of media bias. They argue that the government’s desire to control the flow of information can induce media bias by capture. 18 " A proof is available from the authors upon request. 19 " The Extreme Value distribution constitutes a hybrid: γ−∞ = 0, while γ∞ < ∞. References Aumann , R.J. ( 1976 ). ‘ Agreeing to disagree ’, Annals of Statistics , vol. 4 ( 6 ), pp. 1236 – 9 . Google Scholar Crossref Search ADS WorldCat Aumann , R.J. ( 1987 ). ‘ Correlated equilibrium as an expression of bayesian rationality ’, Econometrica , vol. 55 ( 1 ), pp. 1 – 18 . Google Scholar Crossref Search ADS WorldCat Austen‐Smith , D. and Banks , J.S. ( 1996 ). ‘ Information aggregation, rationality, and the condorcet jury theorem ’, American Political Science Review , vol. 90 ( 1 ), pp. 34 – 45 . Google Scholar Crossref Search ADS WorldCat Baron , D. ( 2006 ). ‘ Persistent media bias ’, Journal of Public Economics , vol. 90 ( 2 ), pp. 1 – 36 . Google Scholar Crossref Search ADS WorldCat Bergstrom , T. and Bagnoli , M. ( 2005 ). ‘ Log‐concave probability and its applications ’, Economic Theory , vol. 26 ( 2 ), pp. 445 – 69 . Google Scholar Crossref Search ADS WorldCat Besley , T. and Prat , A. ( 2006 ). ‘ Handcuffs for the grabbing hand? Media capture and government accountability ’, American Economic Review , vol. 96 ( 3 ), pp. 720 – 36 . Google Scholar Crossref Search ADS WorldCat Bøg , M. ( 2008 ). ‘ Whom to observe? ’, Working Paper, Erasmus University, Rotterdam. Brooks , D. ( 2010 ). Getting Obama right , New York Times, 12 March 2010. Calderone , M. ( 2009 ). ‘ News gets worse for the mainstream media ’, Politico.com. 28 October 2009. Calvert , R.L. ( 1985 ). ‘ The value of biased information: a rational choice model of political advice ’, Journal of Politics , vol. 47 ( 2 ), pp. 530 – 55 . Google Scholar Crossref Search ADS WorldCat Chan , J. and Suen , W. ( 2008 ). ‘ A spatial theory of news consumption and electoral competition ’, Review of Economic Studies , vol. 75 ( 3 ), pp. 699 – 728 . Google Scholar Crossref Search ADS WorldCat Chiang , C.F. and Knight , B. ( 2011 ). ‘ Media bias and influence: evidence from newspaper endorsements ’, Review of Economic Studies , vol, 78 ( 3 ), pp. 795 – 820 . Google Scholar Crossref Search ADS WorldCat DellaVigna , S. and Gentzkow , M. ( 2010 ). ‘ Persuasion: empirical evidence ’, Annual Review of Economics , vol. 2 ( 2 ), pp. 643 – 69 . Google Scholar Crossref Search ADS WorldCat DellaVigna , S. and Kaplan , E. ( 2007 ). ‘ The Fox News effect: media bias and voting ’, Quarterly Journal of Economics , vol. 122 ( 3 ), pp. 1187 – 234 . Google Scholar Crossref Search ADS WorldCat Duggan , J. and Martinelli , C. ( 2011 ). ‘ A spatial theory of media slant and voter choice ’, Review of Economic Studies , vol. 78 ( 2 ), pp. 640 – 66 . Google Scholar Crossref Search ADS WorldCat Durante , R. and Knight , B. ( 2012 ). ‘ Partisan control, media bias and viewers’s responses: evidence from Berlusconi’s Italy ’, Journal of The European Economic Association , vol. 10 ( 3 ), pp. 451 – 81 . Google Scholar Crossref Search ADS WorldCat Enikolopov , R. , Petrova , M. and Zhuravskaya , E. ( 2011 ). ‘ Media and political persuasion: evidence from Russia ’, American Economic Review , vol. 101 ( 7 ), pp. 3253 – 85 . Google Scholar Crossref Search ADS WorldCat Feddersen , T. ( 2004 ). ‘ Rational choice theory and the paradox of not voting ’, Journal of Economic Perspectives , vol. 18 ( 1 ), pp. 99 – 112 . Google Scholar Crossref Search ADS WorldCat Feddersen , T. and Pesendorfer W. ( 1996 ). ‘ The swing voter’s curse ’, American Economic Review , vol. 86 ( 3 ), pp. 408 – 24 . OpenURL Placeholder Text WorldCat Feddersen , T. and Pesendorfer , W. ( 1999 ). ‘ Abstention in elections with asymmetric information and diverse preferences ’, American Political Science Review , vol. 93 ( 2 ), pp. 381 – 98 . Google Scholar Crossref Search ADS WorldCat Fineman , H. ( 2005 ). The ‘media party’ is over . NBC News. 13 January 2005. Fuchs , W. and Oliveros , S. ( 2011 ). ‘ Optimal test choice anticipating future information ’, Working Paper, UC Berkeley. Gentzkow , M. and Shapiro , J.M. ( 2006 ). ‘ Media bias and reputation ’, Journal of Political Economy , vol. 114 ( 2 ), pp. 280 – 316 . Google Scholar Crossref Search ADS WorldCat Gentzkow , M. and Shapiro , J.M. ( 2011 ). ‘ Ideological segregation online and offline ’, Quarterly Journal of Economics , vol. 126 ( 4 ), pp. 1799 – 839 . Google Scholar Crossref Search ADS WorldCat Gerber , A. , Gimpel , J.G., Green , D.P. and Shaw , D.R. ( 2007 ). ‘ The influence of television and radio advertising on candidate evaluations: results from a large scale randomised experiment ’, Working Paper, Yale University. Gerber , A.S. , Karlan , D. and Bergan , D. ( 2009 ). ‘ Does the media matter? A field experiment measuring the effect of newspapers on voting behavior and political opinions ’, American Economic Journal: Applied Economics , vol. 1 ( 2 ), pp. 35 – 52 . Google Scholar Crossref Search ADS WorldCat Gill , D. and Sgroi , D. ( 2012 ). ‘ The optimal choice of pre‐launch reviewer ’, Journal of Economic Theory , vol. 147 ( 3 ), pp. 1247 – 60 . Google Scholar Crossref Search ADS WorldCat Groseclose , T. and Milyo , J. ( 2005 ). ‘ A measure of media bias ’, Quarterly Journal of Economics , vol. 120 ( 4 ), pp. 1191 – 237 . Google Scholar Crossref Search ADS WorldCat Gul , F. and Pesendorfer , W. ( 2011 ). ‘ Media and policy ’, Working Paper, Princeton University. Hagey , K. ( 2011 ). ‘ Jim Lehrer leaving daily anchor role at news hour ’, Technical report, Politico.com. 11 May 2011. Jamieson , K.H. and Cappella , J.N. ( 2010 ). Echo Chamber: Rush Limbaugh and the Conservative Media Establishment , New York : Oxford University Press . Krasa , S. , Bernhardt , D. and Polborn , M. ( 2008 ). ‘ Political polarization and the electoral effects of media bias ’, Journal of Public Economics , vol. 92 ( 5–6 ), pp. 1092 – 104 . OpenURL Placeholder Text WorldCat Martinelli , C. ( 2006 ). ‘ Would rational voters acquire costly information? ’, Journal of Economic Theory , vol. 129 ( 1 ), pp. 225 – 51 . Google Scholar Crossref Search ADS WorldCat Martinelli , C. ( 2007 ). ‘ Rational ignorance and voting behavior ’, International Journal of Game Theory , vol. 35 ( 3 ), pp. 315 – 94 . Google Scholar Crossref Search ADS WorldCat Meyer , M.A. ( 1991 ). ‘ Learning from coarse information: biased contests and career concerns ’, Review of Economic Studies , vol. 58 ( 1 ), pp. 15 – 41 . Google Scholar Crossref Search ADS WorldCat Morgan , J. and Várdy , F. ( 2012 ). ‘ Mixed motives and the optimal size of voting bodies ’, Journal of Political Economy , vol. 120 ( 5 ), pp. 986 – 1026 . Google Scholar Crossref Search ADS WorldCat Morris , S. ( 1995 ). ‘ The common prior assumption in economic theory ’, Economics and Philosophy , vol. 11 ( 2 ), pp. 227 – 53 . Google Scholar Crossref Search ADS WorldCat Mullainathan , S. and Shleifer , A. ( 2005 ). ‘ The market for news ’, American Economic Review , vol. 95 ( 4 ), pp. 1031 – 53 . Google Scholar Crossref Search ADS WorldCat Myerson , R.B. ( 1998 ). ‘ Population uncertainty and Poisson games ’, International Journal of Game Theory , vol. 27 ( 3 ), pp. 375 – 92 . Google Scholar Crossref Search ADS WorldCat Myerson , R.B. ( 2000 ). ‘ Large Poisson games ’, Journal of Economic Theory , vol. 94 ( 1 ), pp. 7 – 45 . Google Scholar Crossref Search ADS WorldCat Oliveros , S. ( 2013 ). ‘ Abstention, ideology and information acquisition ’, Journal of Economic Theory , vol. 148 ( 3 ), pp. 871 – 902 . Google Scholar Crossref Search ADS WorldCat Oliveros , S. and Várdy , F. ( 2011 ). ‘ Unfair and unbalanced: in praise of media bias ’, mimeo, UC Berkeley. Pew ( 2008 ). ‘ The state of the news media 2008: an annual report on American journalism ’, Technical report, Pew Research Center for the People and the Press. Pew ( 2009 ). ‘ Partisanship and cable news audiences ’, Technical report, Pew Research Center for the People and the Press. Prat , A. and Strömberg , D. ( 2011 ). ‘ The political economy of mass media ’, manuscript, CEPR Discussion Paper No. 8246. Strömberg , D. ( 2004 ). ‘ Radio’s impact on public spending ’, Quarterly Journal of Economics , vol. 119 ( 1 ), pp. 189 – 221 . Google Scholar Crossref Search ADS WorldCat Suen , W. ( 2004 ). ‘ The self‐perpetuation of biased beliefs ’, Economic Journal , vol. 114 ( 495 ), pp. 377 – 96 . Google Scholar Crossref Search ADS WorldCat Sunstein , C.R. ( 2002 ). Republic.Com. Princeton, NJ : Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Sunstein , C.R. ( 2009 ). Republic.Com 2.0. Princeton, NJ : Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Author notes " For very helpful comments and suggestions we thank the editor Martin Cripps, an anonymous referee, David Ahn, Bob Anderson, Ned Augenblick, Ernesto Dal Bo, Rui de Figueiredo, William Fuchs, Faruk Gul, Ben Hermalin, Matias Iaryczower, Adam Meirowitz, Bob Powell, Keith Takeda, Justin Tumlinson and, especially, Mike Golosov, John Morgan and Alessandro Pavan. © 2014 Royal Economic Society
TI - Demand for Slant: How Abstention Shapes Voters’ Choice of News Media
JF - The Economic Journal
DO - 10.1111/ecoj.12169
DA - 2015-09-01
UR - https://www.deepdyve.com/lp/oxford-university-press/demand-for-slant-how-abstention-shapes-voters-choice-of-news-media-ArBQ1atQPN
SP - 1327
VL - 125
IS - 587
DP - DeepDyve
ER -