Counterterrorism Policy-Making, Partisanship, and the Electoral Consequences of Terrorism

Transcription

1 Counterterrorism Policy-Making, Partisanship, and the Electoral Consequences of Terrorism Livio Di Lonardo New York University March 14, 2016 Abstract The prevention of terrorist attacks is an important policy concern for many governments. In democracies, officials also fear the electoral consequences of successful attacks. As a result, counterterrorism policy-making and electoral concerns are tightly intertwined. To understand the implications of this link, I develop a game-theoretic model and show that left-wing incumbents respond to terror threats more aggressively than their right-wing counterparts in order to convince voters that they can be trusted in the fight against terrorism. Terrorist attacks improve right-wing incumbents reputation, while they worsen the reputation of left-wing incumbents. When the future terror threat is likely to be high, voters ignore right-wing incumbents reputation, reelecting them independently of their performance. Finally, I extend the model to consider the strategic implications of having counterterrorism policies unobservable to the electorate and of terrorist threats that are persistent over time. Keywords: Counterterrorism, Elections, Terrorist Threats, Security, Civil Liberties. For helpful comments and discussions I would like to thank Brett Casper, Jon Eguia, Tim Feddersen, Justin Fox, Sanford Gordon, Indridi Indridason, Dimitri Landa, Adam Przeworski, Kris Ramsey, Peter Rosendorff, Ken Shotts, Alastair Smith, Scott Tyson, and seminar audiences at New York University, the 2014 Annual Meeting of the Midwest Political Science Association, the 2014 Annual Meeting of the European Political Science Association, the 2016 Annual Meeting of the Southern Political Science Association, and the 2016 Annual Meeting of the International Studies Association. I am particularly indebted to Tiberiu Dragu for his support with this project. Wilf Family Department of Politics, New York University, 10012, New York, NY. ldl297@nyu.edu 1

2 In the wake of the 9/11 terrorist attacks, security from terrorism has become a central issue of the political debate in many democratic countries. By brutally exposing the vulnerability of Western democracies to terrorist strikes on their domestic soil, these terrorist acts have generated a deep sense of fear within the public, who has demanded additional security from terrorism and has since repeatedly indicated terrorism prevention as one of the top policy priorities. The later events of 2004 in Madrid, of 2007 in London, and of 2015 in Paris, have represented periodic and tragic reminders of how the fight against terrorism still remains an open issue and have reinforced its relevance in the domestic political arena. In light of the emergence of terrorism prevention as a major political concern in many democratic systems, it is important to understand the impact that terrorism violence has on democratic institutions and processes, beyond the tragic loss of human lives. In this paper I tackle this fundamental question focusing on the effect that the occurrence or the threat of terror acts have on some of the defining aspects of a liberal democracy: the electoral process, the policy-making process, and the protection of citizens individual rights and liberties. These questions are not new to the large literature on terrorism, and many scholars have investigated separately various aspects of the important implication that terrorist violence has on the life of democratic regimes (e.g. Bueno de Mesquita 2007, Dragu and Polborn 2014). 1 However, trying to uncover and understand in isolation the different ways in which terrorism affects the functioning of democratic institutions is an endeavor that, although useful, is most likely bound to generate seemingly puzzling and possibly conflicting results. The literature on the link between terrorist attacks and electoral results provides a fitting example of these challenges. Some scholars have shown that terrorism hampers the electoral 1 More generally, the extant literature on terrorism has tackled several questions regarding both the causes and the consequences of terrorism, including the effect of repressive policies on security and terrorism support (Bueno de Mesquita 2005b; Bueno de Mesquita and Dickson 2007), the provision of counterterrorism (Dragu 2011; Hodler and Rohner 2012; Powell 2007; Rosendorff and Sandler 2004), the role of the media in promoting terrorism (Rohner and Frey 2007; Wilkinson 1997), and the negotiations between governments and terrorist groups or factions (Bueno de Mesquita 2005a; Kydd and Walter 2002; Lapan and Sandler 1988). For a broad review of this literature see Bueno De Mesquita (2008). 2

3 success of incumbents (e.g. Bali 2007; Montalvo 2011; Gassebner, Jong-A-Pin and Mierau 2008), while other researchers have shown how terrorist attacks might lead to an increase in support for incumbents, with partisan affiliations mediating the relationship between terrorist attacks and incumbents electoral performance (e.g. Berrebi and Klor 2008; Getmansky and Zeitzoff 2014; Koch and Tkach 2012). As such, the following questions still remain largely unanswered: who benefits and who suffers electorally from having a terrorist attack while in office? Under what conditions? And why? Similarly, the literature on counterterrorism policy-making has uncovered that the incumbents partisanship affect the anti-terror measures that are adopted as a response to the terror menace. Even if left-wing parties are usually perceived to have a softer approach to counterterrorism, while right-wing parties tend to be thought as tougher on terrorism (e.g Berrebi and Klor 2006; Koch and Cranmer 2007), several studies show that left-wing incumbents enact more draconian anti-terror measures than right-wing incumbents (Epifanio 2011; Plümper and Epifanio 2014; Pokalova 2014). But why do left-wing incumbents tend to curtail civil liberties to a larger extent than their right-wing counterparts? What mechanisms are behind these surprising findings (Berrebi and Klor 2006, 904)? I argue that, in order to provide a satisfactory answer to these questions and to develop an in-depth understanding of the far-reaching consequences of terrorism violence in democracies, we need a comprehensive analysis of the strategic incentives underlying the behavior of the key actors at play. In fact, the emerging electoral relevance of terrorism prevention creates an incentive for accountable elected officials to adopt new counterterrorism measures that not only reflect the need to balance security and liberties, but that also account for the voters reaction to such measures, which in turn is shaped by the different reputations that political parties tend to have regarding their approach to the fight against terrorism. 2 2 The only paper to take into account the interaction between counterterrorism policy-making and the electorate reaction to terrorist attacks are the theoretical contributions of Berrebi and Klor (2006), Bueno de Mesquita (2007), Hodler and Rohner (2012), and Plümper and Epifanio (2014), which differ from my model 3

4 To account for the strategic feedback between voters reaction to terrorism violence, counterterrorism policy-making, and the electoral ambitions of elected representatives, I develop a game-theoretic model where a reelection-seeking incumbent is in charge of counterterrorism provision, and terrorist attacks can occur before the election. Voters want politicians to prioritize security in times of high terrorist threat and to prioritize liberties in times of low terrorist threat. Politicians differ with respect to their partisan affiliation, and, in turn, about their views on the security-liberty trade-off: right-wing politicians tend to be biased towards security, while left-wing politicians tend to have a bias towards defending liberties. Voters are uncertain about the actual preferences of the candidates for office, (i.e. whether or not each politician is biased) but using the candidates partisan affiliation they are able to discern the direction of each candidate s potential bias. 3 The model yields three main results. First, regarding the policy responses to terror threats, I show how incumbents use counterterrorism policy-making to further their electoral ambitions. This strategic behavior comes either at the expense of citizens security or their civil liberties, depending on the incumbent s partisan affiliation. Specifically, for right-wing incumbents, who have a reputation for being potentially biased towards excessive security, the task is to convince the electorate that they are not willing to curtail civil liberties unless necessary for security reasons. For left-wing incumbents, who are known to be potentially biased towards an excessive protection of civil liberties, the objective is to convince voters that they are not going to blindly defend such liberties if a severe terror threat manifests. To achieve their objective, incumbents will act against type and adopt policies that are in a variety of ways. 3 My model builds on a large literature on electoral accountability. Similar to several other papers (e.g. Ashworth and Bueno de Mesquita 2014, Canes-Wrone and Shotts 2007, Fox 2007, Fox and Stephenson 2014, and Maskin and Tirole 2004), I build a model in which an incumbent politician has private information about her policy preferences. Unlike most of these models, I posit that politicians differ with respect to their partisan affiliation, which the voter observes. As I will describe bellow, partisanship, along with the voter s perception of the state of the world, greatly affects the voter s reelection decision, and, under certain conditions, it renders the incumbent s congruence in terms of policy preference irrelevant. 4

5 supposed to defy their reputation, thus leading left-wing incumbents to enact more aggressive anti-terrorism measures than right-wing counterparts. Second, regarding the effect of terror acts on electoral outcomes, I show how the electoral consequences of successful terrorist attacks crucially depend on the incumbent s partisanship and the perceived terrorist threat level. In particular, I show how right-wing incumbents either benefit or remain unaffected by the occurrence of terrorist attacks, while left-wing incumbents see their electoral prospects either hurt or unchanged by the occurrence of terror acts. The driving force is the direction in which each candidate is potentially biased and their desire to appear unbiased to the electorate. For a right-wing incumbent, the occurrence of an attack leads voters to believe that her approach was not excessively aggressive, otherwise, all else equal, an attack would have been less likely to succeed. Conversely, for a left-wing incumbent, the occurrence of an attack leads voters to believe that her approach to the fight against terrorism was too lenient and her reputation as an unbiased incumbent is damaged. This logic is able to account for the empirical variation in the impact that a terrorist attack has on an incumbent s reelection prospects. Third, I show how even an incumbent believed to be biased is not doomed to fail in her reelection bid. Depending on voters outlook on the future threat level, voters have different standards for reelecting incumbents with different partisan affiliation. When the future terrorist threat is likely to be high, electing a right-wing politician is de facto costless, since this politician is biased towards policies that in such a high threat environment are actually optimal from the voters standpoint. Conversely, left-wing incumbents suffer in electoral terms when voters are particularly pessimistic about the future threat: it is too risky to elect somebody who might turn out to be unwilling to restrict liberties even in times of concrete danger for national security. The opposite is true when the future threat from terrorism is low. Thus, voters might well elect a candidate who is ex ante more likely to be biased, if the direction of her bias matches the future optimal policy, given the nature of 5

6 the threat coming from terrorists. Even if terrorist attacks affect an incumbent s reputation for being biased, this might ultimately not have an effect on the electoral result when the perceived future threat is either very high or very low. Finally, I extend the model to consider the possibility that the counterterrorism policies adopted by the incumbents remain unobservable to the electorate. An increase in the level of transparency of counterterrorism policies increases the likelihood of suboptimal counterterrorism provision, regardless of the incumbent s partisan affiliation. The strategic suboptimal design of a counterterrorism strategy relies on the ability of incumbents to use policy-making to signal congruence between the voters preferences and theirs. This ability to signal congruence depends on the transparency of the antiterrorism measures, and thus higher levels of transparency provide incumbent politicians with an incentive to try and convince voters of their lack of bias, while low transparency levels push incumbents to choose counterterrorism policies that are optimal from the voters standpoint. The results I present have three merits. First, the results reconcile and explain several empirical regularities: a) Leftist incumbents tend to enact more aggressive counterterrorism policies than rightist incumbents; b) there is mixed evidence on the effect of terrorist attacks on incumbents electoral prospects; c) the terrorist threat level affects incumbents differently depending on their partisan affiliation. Second, the results have implications for optimal institutional design with respect to counterterrorism policy-making, by providing a new rationale for keeping the policy making process and its outcomes outside of public sight. Third, the results have implications on how to design empirical investigations that aim to assess the effect of terrorist threats and terrorist attacks on incumbents electoral fortunes. Specifically, terrorist attacks (or their absence) provide useful information to the electorate about the underlying threat level, and, in turn, about the incumbent s suitability to fight terrorism without ideological biases. When the electorate already possesses precise information about threat levels and enacted antiterrorism measures, the informational value 6

7 of terrorist attacks decreases and we should not expect electoral results to be affected by terror acts. Additionally, if the perceived threat level is very high, a good or poor performance in terms of terrorism prevention can be offset by the reputational effects that partisan affiliations carry with them. It is then crucial for empirical work to take into account the informational environment in which voters make their electoral decisions and the key role of partisanship in shaping the relationship between terrorist attacks and electoral fortunes. The Model I consider a two-period model where the actors are an incumbent (I), a challenger (C) and a representative voter (V). 4 In each period, Nature determines the level of the terrorist threat θ, which can be either high (i.e. θ = h) or low (i.e. θ = l). The incumbent observes the threat level θ, while the voter does not observe θ and his prior belief is Pr(θ = l) = π [0, 1]. By assuming that the incumbent observes the threat level, I am able to isolate the distortionary effect of electoral incentives on the incumbent s counterterrorism strategy, thus excluding the confounding effect that uncertainty about the terrorist threat might have on national security decisions. The incumbent politician chooses a counterterrorism policy x, to combat the terrorist threat faced by the society. She can either choose a moderate policy, x = m, or an aggressive one, x = a. 5 Before the election, a terrorist attack can take place. The voter, after observing if an attack happened or not, chooses whether to reelect the incumbent, r = 1, or opt for the challenger, r = 0. After the election, the new incumbent chooses a new counterterrorism strategy, a new terrorist attack can happen, and the game ends. 4 I refer to the voter as he and to a politician as she. 5 Examples of aggressive policies are interception of phone calls and s of US citizens by the NSA outside the limits set by the FISA; data-mining projects; monitoring of constitutionally protected speech in mosques, suspension of the right of habeas corpus for terrorist suspects; and granting the FBI access to records of library borrowings (Posner 2006). 7

8 All players face a trade-off between security from terrorism and civil liberties. The voter wants the incumbent politician to choose, in each period, a counterterrorism policy that best suits the pressure of contemporary events. In times of danger, the concern for public safety outweighs the concern for civil liberties and the voter wants the incumbent to take resolute actions in order to curb the terrorist threat, even if this means sacrificing some individual rights. In safer times, the balance shifts the other way and the voter wants the incumbent to engage in the fight against terrorism without curtailing his civil liberties. 6 Different types of politicians have differing views about the trade-off between security and liberty, with some politicians biased toward security, some toward liberties, and some sharing the unbiased view of the voter. Politicians are also interested in holding office, and they view the choice of a counterterrorism strategy not only as a tool to fight terrorism, but also as an instrument to secure reelection. Terrorist Threat and Successful Attacks The occurrence of a terrorist attack depends on the counterterrorism policy chosen by the incumbent, the terrorist threat level, and other factors outside the incumbent s control. Given this probabilistic link between threat, policy, and occurrence of an act of terror, the voter is able to use a terrorist attack or the lack thereof as a valuable source of information on the incumbent s counterterrorism strategy. Denote by t(θ, x) = Pr(T = 1 θ, x) the probability that a terrorist attack happens (i.e. T = 1) given that the threat level is θ and the counterterrorism policy chosen by the incumbent is x. For any threat level, the aggressive policy is always more effective than the moderate policy at reducing the probability of a terrorist strike. Additionally, the aggressive policy has a larger impact in curbing the chances of a successful terrorist strike when the 6 Davis and Silver (2004) show that the greater the citizens sense of threat, the lower their support for civil liberties, regardless of their ideological positions. 8

9 threat is high than when it is low. Let θ = t(θ, m) t(θ, a) be the effectiveness of the aggressive counterterrorism policy at reducing the probability of a successful terrorist strike when the level of the threat is equal to θ. To capture these features, I assume h > l > 0. 7 If an attack occurs (T = 1), all players pay a cost α R +. The expected cost from terrorism when the counterterrorism measure adopted is x and the threat level is θ is given by αt(θ, x). Therefore, the benefit that an aggressive policy produces in terms of terrorism prevention with respect to a moderate policy is equal to α θ. Politicians A politician in office observes the threat level θ and then selects a counterterrorism policy x {m, a}. Along with the level of the terrorism threat, Nature selects the incumbent s and the challenger s type, ω I and ω C. A politician s type is determined by the cost she incurs to enact an aggressive counterterrorism policy, which implies a curtailment of civil liberties. Each politician can be either unbiased (ω = un), hawkish (ω = ba) or lenient (ω = bm). 8 An unbiased politician, who shares the voter s policy preferences, pays a cost c un = c > 0 from curtailing civil liberties, a hawkish type pays a lower cost c ba < c, while a lenient type pays a higher cost c bm > c. 9 From a pure policy perspective, an unbiased type prefers matching the policy to the threat level, a hawkish type always prefers aggressive counterterrorism measures, and a lenient type always prefers moderate counterterrorism measures, regardless of the terrorist threat level. Even though politicians types are private information, their partisan affiliation is known. Right-wing politicians and left-wing counterparts tend to have different approaches toward 7 This implies that the function t : {l, h} {m, a} [0, 1] has increasing differences in θ and x, considering the ordering h > l and a > m. 8 Throughout the paper, I will use interchangeably hawkish and biased toward aggressive policies. Similarly, I will use interchangeably lenient and biased toward moderate policies. 9 Even if a politician s type corresponds to her cost for curtailing civil liberties, for clarity I denote explicitly types by ω {un, ba, bm}. Additionally, while formally a type in this model is a pair (θ, ω), for simplicity I refer only to ω as a politician s type. 9

10 the security-liberty trade-off. While right-wing officials are more willing to err on the side of security, left-wing ones tend to have a more uncompromising stance on defending civil liberties. To model this, I consider two possible pairs of types, Ω R = {un, ba} and Ω L = {un, bm}. I refer to these type pairs as the partisanship of the candidate. Specifically, I call right-wing a politician whose type belongs to Ω R, while I call left-wing a politician whose type belongs to Ω L. Hence, a right wing politician is either unbiased or hawkish, while a left-wing politician is either unbiased or lenient. I assume that the incumbent and the challenger have different partisanship (one is right-wing, while the other is left-wing) so as to capture the basic competitive dynamic between two candidates who are perceived to have different attitudes toward the balance between security and liberties. 10 Finally, when holding office, politicians receive a positive payoff B > 0, capturing ego rents and other material gains. The Voter The voter decides whether to retain the incumbent or to elect the challenger. In terms of policy preferences, the voter is identical to an unbiased politician. Specifically, he incurs a cost c > 0 if the incumbent enacts an aggressive counterterrorism policy. Given θ, if the incumbent chooses the aggressive policy, the voter s per period expected utility is E[u V (a, θ)] = c αt(θ, a). If the incumbent chooses the moderate policy, in expectation the voter receives utility E[u V (m, θ)] = αt(θ, m). The voter prefers the moderate counterterrorism policy in a low threat environment and the aggressive counterterrorism policy in a high 10 I abstract away from the possibility that politicians might differ not only with respect to their views about the security-liberty trade-off, but also with respect tot their ability to observe the true state of the world. That is, I assume politicians are homogeneous with respect to competence. This allows me to isolate clearly how partisan reputations affect the impact of terrorist attacks and threats on electoral outcomes. For a formal analysis of how terrorist attacks affect electoral outcomes when the incumbent s ability to prevent terrorism is the voters primary concern, see Di Lonardo and Dragu (2016). For a general electoral accountability model where incumbents vary with respect to both policy preferences and ability, see Fox and Shotts (2009). 10

11 threat environment and thus prefers unbiased incumbents to biased ones. However, he faces challenges in selecting unbiased politicians as the politicians types are private information. What helps the voter in this task is his ability to observe the partisanship of each candidate. Knowing whether each candidate is right-wing or left-wing allows the voter to restrict his uncertainty from a triple of types Ω = {un, ba, bm} to pairs of types, Ω R = {un, ba} or Ω L = {un, bm}. Still, the voter remains uncertain about each politician s actual type within each pair. 11 Formally, within each pair, the voter has a prior on the unbiasedness of the politician, Pr(ω i = un ω i Ω R ) = Pr(ω i = un ω i Ω L ) = p. To further clarify the information available to the voter at each stage of the game, he does not know the actual level of the terrorist threat and he does not know the incumbent s and challenger s types, while he knows whether each politician is right-wing or left-wing. Additionally, the voter observes whether or not a terrorist attack has occurred and the counterterrorism policy chosen by the incumbent. Timing and Equilibrium Concept To summarize, the timing of the game is as follows: Period One 1. Nature determines: the terrorist threat level θ {h, l}; the incumbent s type, ω I Ω I, and the challenger s, ω C Ω C. 2. The incumbent observes θ and chooses a policy x {m, a}. 3. A terrorist attack happens with probability t(θ, x). 4. The voter observes the enacted policy and the occurrence or lack of an attack (x, T ) and chooses whether or not to reelect the incumbent, r {0, 1}. 11 Parties usually develop particular images and reputations which are based on the history of the policy stances they have taken and the performances they have achieved while in office (Fiorina 1991). Given these party images, voters use candidates partisan affiliation to infer some information about the candidates ideological stances (Aldrich 1995; Cox and McCubbins 2007; Franklin 1991; Snyder Jr. and Ting 2002). 11

12 Period Two 6. Nature determines the terrorist threat level θ {h, l} The incumbent observes θ and chooses a policy x {m, a}. 8. A terrorist attack happens with probability t( θ, x) and the game ends. I solve for the pure strategy perfect Bayesian equilibria of the game that survive the Equilibrium Domination criterion (Cho and Kreps 1987), which I refer to as equilibria. Informally, an equilibrium consists of a strategy for the incumbent and the challenger and a strategy and system of beliefs for the representative voter. The voter s beliefs about the incumbent s type must be derived, whenever possible, using Bayes rule, on the basis of the occurrence of a terrorist attack or lack thereof, his prior beliefs, and, possibly, the nature of the counterterrorism policy enacted by the incumbent. 13 For simplicity of exposition, I posit that biased types always choose their preferred policy and I focus on the behavior of the unbiased types, who share the voter s policy preferences and yet, under some conditions, will fail to be a faithful agent of the voter. 14 I say there is an optimal provision of counterterrorism when an unbiased type conditions her policy choice on the actual terrorist threat level in the same way that a perfectly informed voter would. Conversely, I will speak of underprovision (overprovision) of counterterrorism when an unbiased incumbent chooses a moderate (aggressive) policy even in times of high (low) threat. Finally, I assume, with no effect on the main results, that whenever indifferent between the incumbent and the challenger, the voter reelects the incumbent. 12 In the baseline model I assume the threat level to be independently drawn each period. In the Extension and Robustness section I consider the case where the threat level remains constant across periods. 13 A formal definition of the equilibrium is given in the Appendix. 14 This simplifying assumption is present in several other accountability models, see Ashworth and Bueno de Mesquita (2014), Fox and Stephenson (2011), and Svolik (2013). The results I describe in the following section are consistent with equilibrium behavior in a more complete game where biased types are fully strategic. For a formal statement and proof see the last section of Appendix B. 12

13 Analysis In the second period, without any accountability mechanism in place before the end of the game, the politician in office will enact her preferred policy. If an unbiased politician is in office, she will choose the counterterrorism measures that best adapt to the level of the terrorist threat; if a biased politician is in office, she will either always choose aggressive counterterrorism measures or always choose moderate counterterrorism policies, depending on the direction of her bias. Recall that π is the probability the threat is low, α is the cost associated with a successful terror attack, c the cost of civil liberties curtailment, and t(θ, x) is the probability that a terrorist attack occurs when the politician in office faces a threat level equal to θ and has enacted a policy x. Let γ w be the expected utility that the voter receives in the second period from having in office a politician of type ω. If the politician is unbiased we have γ un = E[u V (ω = un)] = π( αt(l, m)) + (1 π)( c αt(h, a)); if the politician is hawkish we have, γ ba = E[u V (ω = ba)] = π( c αt(l, a)) + (1 π)( c αt(h, a)); and finally if the politician is lenient we have, γ bm = E[u V (ω = bm)] = π( αt(l, m)) + (1 π)( αt(h, m)). Let us now explore the voter s electoral choice. The voter reelects the incumbent whenever the incumbent provides him with a higher expected second-period utility than the challenger. When the incumbent is right-wing and the challenger is left-wing, this logic implies that the 13

14 voter reelects the incumbent when E[u V (I)] = (1 µ x,t )γ un + µ x,t γ ba pγ un + (1 p)γ bm = E[u V (C)], (1) where µ x,t = P r(ω = ba x, T ) is the posterior belief about a right-wing incumbent being hawkish given that the policy observed by the voter is x and the terrorism occurrence variable takes value T. When instead the incumbent is left-wing and the challenger is right-wing, this logic implies that the voter reelects the incumbent when E[u V (I)] = (1 ˆµ x,t )γ un + ˆµ x,t γ bm pγ un + (1 p)γ ba = E[u V (C)], (2) where ˆµ x,t = P r(ω = bm x, T ) is the posterior belief about a left-wing incumbent being lenient. Let us now determine the optimal behavior for the incumbent politician. First, notice that γ un is also the policy payoff that an unbiased incumbent would receive in the second period if she was reelected. As such, in case of reelection the cumulative benefit of reelection is given by the sum of the policy payoff and the ego rents. For convenience, denote such cumulative benefit of holding office by B γ un + B. Denote by E[u ω (x θ)] the expected utility for an incumbent of type ω to choose policy x when the state of the world is θ. We have that an unbiased incumbent chooses counterterrorism policy a over policy m if E[u un (a θ)] E[u un (m θ)], that is if, c αt(θ, a) + B v(a) αt(θ, m) + B v(m), (3) where v(x) is the probability with which the voter reelects the incumbent when he observes counterterrorism policy x. As such, we have that v(x) = t(θ, x)r(x, 1) + (1 t(θ, x))r(x, 0). 14

15 Rearranging equation (3) we obtain, α }{{ θ } security benefit B [v(m) v(a)] }{{} potential electoral costs + }{{} c, (4) reduced liberties which is easily interpretable. Namely, the incumbent chooses a over m if the benefits an aggressive policy produces in terms of enhanced security outweigh the costs of curtailing civil liberties and the potential costs that such a choice might generate in terms of electoral prospects. This points toward a clear trade-off that incumbents face between ensuring security from terrorism, protecting civil liberties, and winning office, a trade-off that I examine more in detail in the following section. The Security-Liberty-Reelection Trade-off Accountable politicians craft policies taking into account the effect that these policies could have on their careers. Therefore, trying to understand why incumbents adopt the measures they do without placing this policy-making process within the electoral cycle is an exercise likely to produce unsatisfactory results. Given the optimal behavior of incumbents and of the voter described above, I study how the equilibrium provision of counterterrorism is affected by the incumbent s partisanship and her reelection ambitions. Proposition 1 (Equilibrium Characterization) Consider the case of a right-wing (leftwing) incumbent. For any prior on the threat level, (a) there exists an optimal provision equilibrium iff the effectiveness of aggressive policies is high (low) enough or the probability of a low threat is low (high) enough. (b) there exists an underprovision ( overprovision) equilibrium iff the effectiveness of aggressive policies is low (high) enough and the probability of a low threat is high (low) enough. 15

16 Moreover, these are the only equilibria of the game. The incumbent politician wants to convince the voter she is unbiased, in order to secure reelection. She can influence the voter s beliefs through the adoption of counterterrorism measures. Therefore, the provision of counterterrorism will change depending on the effectiveness of antiterrorism policies at preventing terrorism (i.e. θ ). An unbiased right-wing incumbent is aware that to increase her electoral prospects she has to convince the voter she is not hawkish. To show her lack of bias, she would have to chose moderate measures. However, when the country faces a serious threat to national security, adopting such a strategy would translate into a relatively high probability of a successful terrorist attack. In other words, the incumbent faces a dilemma: better reelection prospects entail a higher chance of experiencing a terror attack. So she will compare the benefits produced by a moderate policy in terms of a higher chance of reelection and the costs in terms of an increased probability of a successful attack along with the ones associated with a curtailment of liberties. When the effectiveness of aggressive policies at preventing terrorism is low, choosing moderate polices instead of aggressive ones does not cause a big enough increase in the probability of a terrorist attack to outweigh the electoral benefits generated by such a choice. This induces an unbiased incumbent to always choose a moderate counterterrorism policy to signal to the voter her lack of bias, independently from the actual level of the terrorist threat. Instead, when aggressive policies are particularly effective at curbing the probability of a successful attack, the incumbent is unwilling to sacrifice security in order to ensure for herself a higher probability of reelection. Adopting more aggressive antiterrorism measures improves security from terrorism significantly, thus providing the right incentives for the unbiased incumbent to chose optimal counterterrorism policies even if this entails costs in terms of future electoral prospects. For an unbiased left-wing incumbent, the concern is not to appear too lenient in the 16

17 fight against terrorism. The best way to convince the voter she is not lenient is to adopt an aggressive policy. However, when the threat is low, there is a trade-off between electoral concerns and the desire to protect civil liberties. When the effectiveness of aggressive policies at preventing terrorism is high enough, the benefits of choosing aggressive policies in terms of increased electoral prospects and increased security from terrorism outweigh the costs in terms of the unnecessary curtailment of civil liberties. Policies that are too effective at preventing terror attacks lead to overprovision of counterterrorism, in that they tend to be overused. For an unbiased incumbent to provide counterterrorism optimally, it has to be the case that the security-enhancing effect of aggressive policies, coupled with the electoral incentive to signal her type to the voter, is not enough to outweigh the concern for protecting the individual liberties and rights, given the absence of a serious and imminent threat. Hence, while very effective antiterrorism policies leads left-wing incumbents to overprovide counterterrorism, antiterrorism policies that are not very effective lead right-wing incumbents to underprovide couterterrorism. As such, Proposition 1 highlights how even a desirable feature of counterterrorism policies, namely their effectiveness at preventing terrorist attacks, could lead to less than optimal outcomes for the electorate. 15 The Effect of Terrorist Attacks on Incumbents Reputation We have discussed the signaling mechanism behind the link between counterterrorism policymaking and electoral concerns. Let us now explore the inferences that the voter makes 15 In the suboptimal provision equilibrium (both with underprovision and overprovision), in the first period, incumbents will choose a counterterrorism policy that goes against their policy preferences so as to increase their reelection prospects, but they will not do that in the second period because of the absence of a second election. One might be concerned that, if continued reelection was possible, it would be hard for a right-wing party to maintain its reputation for being tougher on terrorism, since a right-wing politicians might actually underprovide counterterrorism in order to show his unbiasedness. However, after first-period underprovision, a right-wing incumbent would have no incentive to keep underproviding counterterrorism in the following periods because she has already convinced the voter, through the first-period underprovision, that she is unbiased. That is, unbiased right-wing politicians will not underprovide for more than one period on the equilibrium path and this, coupled with the biased approach of hawkish right-wing politicians, will preserve the right-wing party s reputation for being tougher on terrorism than the left-wing opposition. 17

18 after the occurrence or absence of a terrorist attack. Analyzing how the voter uses the informational content of terrorist attacks and policy choices is crucial to better understand and explain the variation in the electoral effects of terror acts uncovered in the empirical literature. More precisely, we are interested in the conditions under which a terror act improves or hinders the incumbents electoral prospects. I provide such conditions in the following result. For convenience, I refer to the voter s posterior belief that the incumbent is unbiased as the incumbent s reputation. Proposition 2 (Effect of terrorist attacks on incumbents reputation) Consider the case of a right-wing (left-wing) incumbent. In any equilibrium: (a) if the voter observes a moderate policy, the incumbent s reputation is not affected (weakly worsened) by the occurrence or the absence of a terrorist attack; (b) if the voter observes an aggressive policy, the incumbent s reputation weakly improves (is not affected) after a terrorist attack. This result shows how a right-wing incumbent tends to benefit while a left wing tends to suffer, in terms of reputation, from the occurrence of a terrorist attack. The result offers a possible explanation for the heterogeneous empirical findings regarding the link between terrorist attacks and the performance of incumbents. Specifically, it highlights the important role played by the incumbent s partisanship and the enacted antiterrorism policy. Let us discuss separately the case of a right-wing incumbent and the case of a left-wing incumbent. Right-wing Incumbents. Recall that a hawkish incumbent always chooses aggressive counterterrorism measures, thus curtailing civil liberties regardless of the threat level. When the voter observes a moderate counterterrorism policy (i.e. part (a)), he immediately knows that the incumbent is unbiased, since a hawkish incumbent would have never chosen the 18

19 moderate policy. Along with observing the incumbent s policy choice, the other source of information for the voter is whether or not an attack occurred. When the policy choice tells the voter all he needs to know in order to identify the incumbent s type, the occurrence or absence of a terrorist attack has no bearing on his evaluation of the incumbent. Therefore the incumbent s reputation is not affected by the occurrence or absence of an attack. When the voter observes an aggressive counterterrorism policy (i.e. part (b)), he is led to believe the incumbent is more likely to be hawkish than he initially expected. This is the case both when a terrorist attack has happened and when it has not happened. In the underprovision equilibrium the unbiased incumbent always chooses the moderate policy. So, when the voter observes aggressive antiterror measures, he immediately knows that the incumbent is hawkish. Since the policy choice provides all the information the voter needs in order to identify the incumbent s type, the occurrence or lack of an attack will not affect the incumbent s reputation. In the optimal provision equilibrium, observing an aggressive policy leads the voter to believe the incumbent is more likely to be biased than he initially expected. In fact, a hawkish incumbent always curtails civil liberties, while an unbiased incumbent only does so when the threat is high. Still observing an aggressive policy is not enough evidence to determine with certainty that the incumbent is hawkish: the voter does not know the underlining level of the terrorist threat, and in the event it was high, he knows that both an unbiased and a hawkish incumbent would have chosen the aggressive policy. Because of this uncertainty, the occurrence or absence of an attack will provide useful information to help the voter assess the incumbent s type. In particular, and perhaps surprisingly, the occurrence of an attack will partially improve the incumbent s reputation vis a vis the voter. The reason for this is that a terrorist attack leads the voter to be more confident that the threat was high, and as such, he becomes more inclined to think that an aggressive policy was the result of a unbiased type attempting to protect the country during perilous times. 19

20 Combining this case with the underprovision equilibrium case (where the incumbent reputation is unaffected), we establish that observing an attack when the aggressive policy is known to be in place weakly improves the incumbent s reputation. Left-wing Incumbents. Let us now consider the case of a left-wing incumbent, starting from part (a). Observing a moderate policy leads the voter to negatively revise his belief about the incumbent s leniency. This is the case regardless of the outcome of the fight against terrorism. In the overprovision equilibrium, when the voter observes a moderate policy, he is certain that the incumbent is biased, because in such an equilibrium only a biased type would ever chose moderate antiterror measures. As such, the incumbent s reputation is not affected in any way by the occurrence or absence of an attack. In the optimal provision equilibrium, observing moderate counterterrorism policies leads the voter to believe the incumbent is more likely to be lenient than expected. This is due to the fact that a lenient incumbent always chooses moderate policies, while an unbiased incumbent only does so when the threat is low. However, since the voter does not know the threat level, he cannot determine with certainty that the incumbent is lenient. Because of this uncertainty, the occurrence or absence of an attack will provide useful information to help the voter assess the incumbent s type. Specifically, the occurrence of an attack will partially worsen the incumbent s reputation in the eyes of the electorate. This is because a terrorist attack prompts the voter to think that the threat must have been high. With this revised assessment, he becomes more inclined to believe that a moderate policy must have been enacted by a lenient type failing to protect the country during perilous times. Combining this case with the overprovision equilibrium case (where the incumbent reputation is unaffected), we establish that observing an attack when the moderate policy is known to be in place weakly worsens the incumbent s reputation. 20

21 If the voter observes an aggressive policy (i.e. part (b)), he concludes immediately that the incumbent is unbiased. In fact, both in the overprovision and the optimal provision equilibrium, only an unbiased left-wing politician ever chooses aggressive policies. This inference does not depend on the occurrence or absence of a terrorist attack, and thus the incumbent s reputation is unaffected by acts of terror. The Effect of Threats and Partisanship on Electoral Outcomes We have assessed how an incumbent s performance on terrorism prevention affects her reputation. However, even incumbents with a reputation for being biased in their counterterrorism approach can still be reelected in office. In fact while the incumbent s reputation is an important electoral factor, it does not determine automatically and single-handedly the voter s electoral behavior. As mentioned at the beginning of the analysis, the voter decides between reelecting the incumbent and electing the challenger on the basis of the expected utility that each of these candidates can provide him in the future. Incumbents with different potential policy biases produce, through their future policy choices, different expected costs. As such, along with the likelihood of a politician being biased, the direction of a politician s potential bias is a crucial factor in the voter s electoral decision. Importantly, depending on the future threat, being biased toward a particular set of policies might represent an electoral advantage for an incumbent. When his perception of the threat is skewed in one particular direction (i.e. he is relatively confident the threat is going to be high or low), the voter is willing to trade-off absolute lack of bias with bias going in the right direction. Recall that a strategy for the voter specifies a voting decision for every possible scenario that the voter has observed up until the electoral stage. Each scenario can be summarized by a pair (x, T ) composed of what the voter has observed about the enacted counterterrorism policy and the outcome of the fight against terrorism. Specifically, there are for possible 21

22 pairs: (m, 1), (m, 0), (a, 1), (a, 0). With this in mind, I define the restrictiveness of the voter retention rule to be the number of policy-outcome pairs (x, T ) after which the incumbent is not granted reelection. Additionally, I call the voter s retention rule maximally restrictive if the voter reelects the incumbent if and only if he believes she is unbiased, while I call the voter s retention rule maximally permissive if the voter always reelects the incumbent independently from the policy chosen and the outcome of the fight against terrorism. The restrictiveness of the voter s retention rule depends on the direction of the incumbent s potential bias and on the future threat level, which I show in the following result. Proposition 3 (Effect of terrorist threat on electoral prospects) Consider the game with a right-wing (left-wing) incumbent. In any equilibrium, the voter retention rule is: (a) weakly more (less) restrictive as the probability of facing a low threat increases; (b) maximally restrictive (permissive) if the probability of facing a low threat is high enough; (c) maximally permissive (restrictive) if the probability of facing a low threat is low enough. When the future terrorist threat is very likely to be high, the voter is concerned about the possibility of electing a politician biased toward moderate policies. Such a politician would jeopardize the security of the society by adopting counterterrorism measures that are not aggressive enough. This logic clearly helps a right-wing incumbent, who is always reelected when the threat is very likely to be high, independently from the policy enacted and from the results in terms of terrorism prevention. Even if the voter is aware that the incumbent is very likely to be biased toward aggressive policies, he is better off keeping her in office than electing a potentially lenient challenger. While it is true that in the (unlikely) case the threat turns out to be low, a hawkish type will overprovide counterterror. In the most probable case that the threat is extremely serious, she will enact the aggressive policies needed to protect citizens from the occurrence of a successful attack. 22

23 On the contrary, when the incumbent is left-wing and the future threat is very likely to be high, the voter is willing to reelect her only if he is sure about her unbiasedness. If there is still uncertainty about the incumbent s type, her reelection might produce large costs for the society. Therefore, when the prior on the threat level is particularly pessimistic, the voter will opt for the right-wing challenger, unless he knows the incumbent is unbiased. As the probability of facing a high terrorist threat decreases, having a right-wing politician in office becomes increasingly less desirable than having a left-wing one. A low threat does not need to be addressed with policies that require a curtailment of civil liberties, and a hawkish politician would not refrain from enacting these intrusive measures. This means that as the probability of facing a high threat decreases, the electoral prospects of a rightwing incumbent worsen while those of a left-wing incumbent improve. Finally, when the level of the future terrorist threat is very likely to be low, the voter needs to guard against the possibility of electing a politician biased toward aggressive policies. As such, the voter is willing to reelect a right-wing incumbent only if he is sure that this politician is actually unbiased. Conversely, the voter finds it optimal to always reelect a left-wing incumbent, regardless of her performance in office. In fact, a politician biased toward moderate policies will fail to optimally choose counterterrorism measures only when the threat posed by terrorists is high, which is a low probability event. Substantively, the result in Proposition 3 shows how partisanship fundamentally mediates the relationship between the terrorist threat level and the incumbents electoral fortunes. On the one hand, candidates who have a reputation for being potentially inclined toward an aggressive approach to fighting terrorism largely benefit from the prospect of a high future threat. On the other hand, candidates perceived to be potentially biased toward an inflexible protection of citizens civil liberties stand to suffer, in electoral terms, from the expectation of a serious future threat. 23