Economic recessions are one of the most robust

Learnng to Love Democracy: Electoral Accountablty and the Success of Democracy Mlan W. Svolk Unversty of Illnos at Urbana-Champagn Ths artcle explans why dssatsfacton wth the performance of ndvdual poltcans n new democraces often turns nto dsllusonment wth democracy as a poltcal system. The demands on electons as an nstrument of poltcal accountablty are much greater n new than establshed democraces: poltcans have yet to form reputatons, a condton that facltates the entry nto poltcs of undesrable canddates who vew ths perod as ther one-tme opportunty to get rch. After a repeatedly dsappontng government performance, voters may ratonally conclude that all poltcans are crooks and stop dscrmnatng among them, to whch all poltcans ratonally respond by actng lke crooks, even f most may be wllng to perform well n offce f gven approprate ncentves. Such an expectaton-drven falure of accountablty, whch I call the trap of pessmstc expectatons, may precptate the breakdown of democracy. Once poltcans establsh reputatons for good performance, however, these act as barrers to the entry nto poltcs of low-qualty poltcans. The resultng mprovement n government performance renforces voters belef that democracy can delver accountablty, a process that I assocate wth democratc consoldaton. These arguments provde theoretcal mcrofoundatons for several promnent emprcal assocatons between the economc performance of new democraces, publc atttudes toward democracy, and democratc stablty. Economc recessons are one of the most robust predctors of the breakdown of democracy. Between 1848 and 2008, a democracy was more than twce as lkely to revert to dctatorshp durng an economc declne than durng a perod of economc growth. Ths assocaton s especally relevant for young and poor democraces: when democraces break down, nne n ten do so before they are 20 years old or when ther annual GDP per capta s less than $4,900. 1 Fgure 1 llustrates ths assocaton by separately plottng the hazards of a democratc breakdown durng economc recessons and expansons. The rse of the Naz Party n the Wemar Republc amdst the hardshp of the Great Depresson and Russa s return to authortaransm under Vladmr Putn followng the chaos of ts transton to democracy under Bors Yeltsn are merely two promnent examples of the exstental danger that economc downturns present for young democraces. Yet why democratc breakdowns frequently follow economc downturns s far from obvous. After all, democratc theorsts from Madson (2010b) to Dahl (1971) emphasze that unlke n a dctatorshp voters n a democracy have the unque opportunty to hold poltcans accountable. Hence f Russan voters are dssatsfed wth ther democratc experence n the 1990s, they should blame poltcans, not the poltcal system they should blame Yeltsn, not democracy. Democratc theory thus fals to account for a major consequence of a dsappontng government performance n new democraces: wdespread dsllusonment wth democracy followed by a return to authortaransm. In ths artcle, I explan ths puzzle. My arguments clarfy why the demands on electons as an nstrument of poltcal accountablty are much greater n new than establshed democraces. In any democracy, the success of electoral accountablty crtcally depends on two Mlan W. Svolk s Assocate Professor of Poltcal Scence, Unversty of Illnos at Urbana-Champagn, 420 Davd Knley Hall, 1407 W. Gregory Dr., Urbana, IL 61801 (msvolk@llnos.edu). I would lke to thank Larry Bartels, Lsa Blaydes, Carles Box, Damarys Canache, José Chebub, Jeff Cohen, Alex Debs, Larry Damond, Tberu Dragu, Georgy Egorov, Jon Egua, Scott Gehlbach, Navd Hassanpour, Jude Hays, Phl Keefer, Carre Konold, Beth Leech, Beatrz Magalon, Nkolay Marnov, Mathew McCubbns, James Melton, Monka Nalepa, Mattas Polborn, Robert Powell, Raner Schwabe, Francesco Sobbro, Bonne Wer, Matt Wnters, and partcpants at semnars at the Unversty of Illnos, IMT Lucca, Northwestern, Prnceton, Stanford, Unversty of São Paulo, Yale, and the APSA and MPSA conferences for helpful comments and Dan Koev and Mchael Martn for research assstance. The data and smulaton code from ths artcle can be found at the author s webste, http://publsh.llnos.edu/msvolk/. 1 GDP per capta s measured n 1990 nternatonal Geary-Khams dollars and comes from Maddson (2008). Haggard and Kaufman (1995) provde an early analyss of the relatonshp between economc recessons and the breakdown of democracy. Ths assocaton s robust to controllng for varous covarates and estmaton technques; see, e.g., Bernhard, Nordstrom, and Reenock (2001), Box (2003), Chebub (2007), Kapsten and Converse (2008), Przeworsk et al. (2000), and Svolk (2008). Amercan Journal of Poltcal Scence, Vol. 57, No. 3, July 2013, Pp. 685 702 C 2013, Mdwest Poltcal Scence Assocaton DOI: 10.1111/ajps.12005 685

686 MILAN W. SVOLIK FIGURE 1 NonparametrcEstmatesofthe Hazard of Democratc Breakdowns durng Economc Recessons (Annual Declne n GDP per Capta) and Expansons (Annual Increase n GDP per Capta), 1948 2007 Hazard rate of a democratc breakdown (Nelson Allen estmator, smoothed) 0.01.02.03.04 Recesson Expanson 0 10 20 30 40 Age of democracy (years) nterdependent expectatons: voters belef that poltcans wll respond to electoral ncentves and poltcans expectaton that voters wll punsh them for a poor performance n offce. In new democraces, however, poltcans have yet to form reputatons, a condton that facltates the entry nto poltcs of undesrable canddates who vew ths perod as ther one-tme opportunty to get rch. As a result, a dfferent, partcularly vcous set of expectatons may emerge after a repeatedly dsappontng government performance. Voters may conclude that all poltcans are crooks and that any attempt to dscrmnate among them based on ther performance s therefore a waste of tme. In turn, poltcans wll ratonally respond by actng lke crooks, even f most of them may be wllng to performwellnoffcefgventheappropratencentves. When such pessmstc expectatons fuel a mutually renforcng cycle of voter apathy and poor government performance, electons fal as an nstrument of poltcal accountablty. I call ths expectaton-drven falure of accountablty the trap of pessmstc expectatons. When t occurs, t undermnes the publc s wllngness to defend democracy aganst attempts to subvert t, thus elmnatng a key check on poltcans or groups wth authortaran ambtons. Ths artcle thus proposes a mechansm by whch a falure of electoral accountablty rather than a redstrbutve conflct precptates the breakdown of democracy. I develop these arguments wth the help of a theoretcal model that offers several new nsghts nto the relatonshp between government performance n new democraces, publc atttudes toward democracy, and democratc stablty. Frst, t provdes mcrofoundatons for the process by whch repeated popular dssatsfacton wth the performance of ndvdual poltcans turns nto doubts about the value of democracy as a poltcal system, thereby precptatng ts breakdown. Second, the present model also clarfes the process by whch electons succeed as an nstrument of accountablty n new democraces. We shall see that gradually mprovng poltcal reputatons play a key selecton role n ths process: they dscourage undesrable canddate types from enterng nto poltcs, thus strengthenng publc support for democracy and facltatng a democracy s consoldaton. And thrd, the present model accounts for a number of emprcal assocatons promnently establshed n the lterature on democratc transtons, especally those concernng the effect of government performance and economc development on publc atttudes towards democracy and democratc stablty. In developng these arguments, I emphasze two features of posttranston poltcs that present a dstnct challenge to electoral accountablty. As I antcpated, new democraces are exceptonally vulnerable to the entry nto poltcs of undesrable canddate types who vew ths perod as ther one-tme opportunty to get rch, and therefore the threat of removal from offce cannot deter them from explotng t for personal gan. A classc formulaton of ths concern comes from Madson (2010a, 52), who warns aganst men of factous tempers, of local prejudces, or of snster desgns, [who] may, by ntrgue, by corrupton, or by other means, frst obtan the suffrages, and then betray the nterests, of the people. I dstngush between such bad canddates and normal canddates who may be tempted to explot offce for personal gan but are wllng to behave f faced wth approprate ncentves. Bad canddates vew the frst few electons n a new democracy as an opportune moment for ther entry nto poltcs because poltcans have yet to form reputatons for performng well, and nsttutons that may otherwse lmt ther entry poltcal partes, the meda, or cvl socety are only developng. 2 A second feature that brngs the present model closer to the poltcal realtes n new democraces s costly montorng: voters have the choce to ether gnore nformaton about a poltcan s performance or seek and evaluate t 2 Indeed, research on new democraces ndcates that poltcal partes, whch may otherwse lmt the entry of bad canddates nto poltcs, are frequently the vehcles for ther ambtons (Grzymała- Busse 2007). On the large number of mostly unknown canddates and partes that typcally compete n posttranston electons, see Brch (2003), Manwarng and Scully (1995), and Tavts (2008).

LEARNING TO LOVE DEMOCRACY 687 at a cost. 3 Whle the cost of acqurng unbased nformaton about poltcans performance may be nonneglgble n any democracy, t s partcularly pronounced n new democraces. Transtons to democracy typcally occur n countres where voters lack prevous democratc experence, large-scale poltcal and economc reforms occur smultaneously, and a hstory of government-controlled meda facltates the ncumbent s nfluence over news coverage. 4 The assumpton of costly montorng thus reflects the heghtened dffcultes that voters confront when evaluatng poltcans performance n new democraces. 5 Jontly, these new features generate a dynamc that hghlghts the nterdependence between a new democracy s performance, mass support for democracy, and ts survval. Voters must use electons n order to both motvate normal canddates to perform well and to weed out bad canddates. Yet because canddate reputatons take tme to develop, voters n a new democracy fnd t partcularly challengng to dstngush between bad and normal canddates. I show that dependng on ts early performance, a new democracy may follow two sharply dvergent poltcal trajectores. On the frst of these trajectores, popular dssatsfacton wth the performance of ndvdual poltcans turns nto doubts about the value of democracy as a poltcal system. After a seres of polcy falures, voters ratonally and potentally erroneously conclude that poltcs has been captured by bad canddates, and therefore dscrmnatng among poltcans based on ther performance s no longer worth the effort. In turn, a self-fulfllng cycle of voter skeptcsm and poor government performance 3 Ths opton to reman ratonally gnorant (Downs 1957) s consstent wth a large lterature on the role of nformaton n the democratc process accordng to whch voters habtually, and often ntentonally, do not seek out relevant poltcal facts (see, e.g., Lupa and McCubbns 1998). 4 On meda control n new democraces, see Lawson (2002), Enkolopov, Petrova, and Zhuravskaya (forthcomng), and Gehlbach (2010); on economc reforms, see Shlefer and Tresman (2000) and Frye (2010). Besley and Prat (2006) and Gehlbach and Sonn (2011) explore the consequences of government control of the meda for accountablty theoretcally; Besley and Burgess (2002), Adserà, Box, and Payne (2003), and Snyder and Strömberg (2010) study the relatonshp between voters access to nformaton about government performance and accountablty emprcally. On meda control n dctatorshps, see Egorov, Gurev, and Sonn (2009) and Lorentzen (2008). 5 The magntude of the montorng cost may n turn be medated by poltcal nsttutons, see, e.g., Powell and Whtten (1993), Duch and Stevenson (2008), Hellwg and Samuels (2008), and Cleary (2010). ensues: even normal poltcans act lke crooks because voters do not expect any poltcan to act any better. 6 Ths trajectory hghlghts the vulnerablty of new democraces to poor government performance. Even f they are entrely due to bad luck, polcy falures can have a perncous effect on mass support for democracy. In fact, followng Easton (1965), a large lterature has examned the proposton that wdespread dssatsfacton wth ndvdual governments may translate nto declnng support for the poltcal system as a whole. 7 Yet the mechansm underlyng ths correlaton has often been elusve n exstng accounts. After all, the alternatve to democracy s dctatorshp, and t s unclear why even n a democracy afflcted by dsappontng government performance voters would gve up the voce that democratc electons afford them. The arguments here suggest an explanaton: once a democracy descends nto the trap of pessmstc expectatons, even voters who are ntally optmstc about the ablty of electons to motvate accountable behavor ratonally conclude that ther partcular democracy rather than democracy as an abstract deal cannot delver governance that s more responsve than that under a dctatorshp. The above logc also clarfes the connecton between a new democracy s performance and the vablty of publc opposton to elte transgressons aganst democracy (see, e.g., Fearon 2011; Wengast 1997). In a semnal artcle, Wengast examned the problem of publc coordnaton on lmts aganst such transgressons, emphaszng that these lmts become self-enforcng when ctzens hold [them] n hgh enough esteem that they are wllng to defend them by wthdrawng support from the soveregn when he attempts to volate these lmts. (1997, 251) 6 There s ample evdence for such apathetc poltcal atttudes. Carothers (2002, 10), for nstance, laments that n many new democraces poltcal eltes from all the major partes or groupngs are wdely perceved as corrupt, self-nterested, and neffectve. The alternaton of power seems only to trade the country s problems back and forth from one hapless sde to the other. Among the publc, these atttudes are exemplfed by comments lke I don t thnk about poltcs at all....nobody beleves n poltcans any more or beleves that any government wll do a better job. (See James P. Gallagher, Russan Apathy Runs Deep as Electons Near, Voters Lose Fath, Chcago Trbune, November 13, 1995.) Statements such as all crows under heaven are equally black (O Bren and L 2006, 125 26) or t makes no sense to replace a full tger wth a hungry wolf (L and O Bren 1996, 34) llustrate popular dsenchantment wth vllage electons n Chna. 7 See, e.g., Booth and Selgson (2009), Canache (2002), Norrs (1999), and Rose, Mshler, and Munro (2006).

688 MILAN W. SVOLIK When a new democracy falls nto the trap of pessmstc expectatons, a crtcal mass of the publc loses such hgh enough esteem for democracy and n turn fals to restran a leader or group wth authortaran aspratons. The publc s wllngness to check transgressons aganst democracy erodes as mass dsllusonment wth government performance under democracy turns nto ndfference to alternatve forms of government. Durng the Cold War, such a clmate of publc poltcal ndfference was frequently credted wth nvtng mltary coups. 8 But more recently, the modal path to democratc breakdown s one on whch an elected ncumbent gradually encroaches on cvl lbertes, emasculates the opposton, but stops short of abolshng electons. The present framework thus suggests an explanaton for the emergence of regmes that have been alternatvely referred to as compettve or electoral authortaran (Bunce and Wolchk 2010; Levtsky and Way 2010; Schedler 2006). Yet the present model also explans why some new democraces follow another poltcal trajectory, one along whch gradually mprovng reputatons ncreasngly serve as barrers to the entry of bad canddates nto poltcs and ultmately brng about the consoldaton of democracy. As normal canddates develop reputatons for performng well, bad canddates who explot offce for personal gan and n turn more often presde over polcy falures see ther chances of returnng to offce declne due to ncreasng compettveness and eventually leave poltcs. The resultng mprovement n government performance renforces voters confdence n the ablty of electons to delver accountablty to the extent that even after a successon of polcy falures they wll no longer conclude that all poltcans are crooks and democracy does not work. I call a democracy that s no longer at rsk of descendng nto the trap of pessmstc expectatons a consoldated democracy. To put t metaphorcally, n a consoldated democracy, voters expectatons and poltcans performance postvely renforce each other to the extent that a few bad apples can no longer spol the bunch. Ths dynamc s consstent wth research that vews democratc consoldaton as a process of atttudnal change that results n wdespread popular acceptance of democracy (Almond and Verba 1963; Damond 1999; 8 Nordlnger, for nstance, wrtes that performance falures lead to the deflaton of governmental legtmacy wthn the poltczed stratum of the cvlan populaton. It s ths factor that encourages and allows the offcers to act upon ther nterventonst motves... (1977, 64). On the other hand, the overthrow of legtmate governments wll spark mass protests, general strkes, rots, sporadc volence, and possbly armed resstance....solders thus rarely overthrow legtmate governments because they do not want to brng on, or deal wth, the dsorderly and volent behavoral expressons of popular censure (94 95). Lnz and Stepan 1996). Lnz and Stepan, for nstance, promnently characterzed consoldated democraces as those where a strong majorty of publc opnon, even n the mdst of major economc problems and deep dssatsfacton wth ncumbents, holds the belef that democratc procedures and nsttutons are the most approprate way to govern collectve lfe....(1996, 16) Here, once a democracy consoldates, voters correctly nterpret polcy falures as the result of the uncertanty nherent n polcymakng rather than the capture of electoral poltcs by canddates bent on explotng offce for personal gan. I show that consoldaton more lkely occurs n democraces where suffcently attractve, nonpoltcal careers are avalable, a result that may account for the strong emprcal assocaton between wealth and democratc stablty. 9 By contrast, n poor democraces, wherepoltcsmaybe themostproftablegamentown, even ncreasngly compettve electons wll not dscourage bad canddates from runnng for offce. The present model thus outlnes a mechansm by whch wdespread popular acceptance of democracy emerges endogenously. As ths dscusson suggests, I conceve of electons as frst of all an nstrument of accountablty, thus departng from Box (2003) and Acemoglu and Robnson (2005), who vew electons as a mechansm for resolvng conflcts over the redstrbuton of wealth. Accordng to ther arguments, democracy fals n economcally unequal countres because the preferred way of resolvng redstrbutve conflcts n these socetes s volence rather than electons. 10 By contrast, I emphasze that the central promse of democracy s to delver good governance by holdng poltcans accountable for ther performance and argue that democraces are partcularly vulnerable to a breakdown when electons fal to delver on ths promse. 11 The present approach explans the falure of democracy n countres wthout acute redstrbutve conflcts such 9 See Lpset (1959), Przeworsk et al. (2000), and Box and Stokes (2003); cf. Acemoglu et al. (2009). 10 Ahlqust and Wbbels (2012), Houle (2009), and Freeman and Qunn (2012) assess these arguments emprcally. 11 Certanly, the benefts of democracy go beyond the promse of electoral accountablty and also nclude cvl lbertes, the rule of law, and checks and balances, to name just a few. Proposton 3 and nequalty (6) n the next secton clarfy why the wllngness of a sgnfcant fracton of the publc to defend these addtonal benefts may be crtcally mpared once voters conclude that ther democracy cannot delver accountablty. See also Lzzer and Persco (2004), who examne an accountablty ratonale for transtons to democracy.

LEARNING TO LOVE DEMOCRACY 689 as Russa where poor governance after the transton to democracy led to wdespread dsenchantment wth democratc poltcs and, n turn, ambvalence about alternatve, authortaran forms of government (Colton and McFaul 2003; Rose, Mshler, and Munro 2006). The model of democratc breakdown and consoldaton that I present n the next secton also contrbutes to recent research on the challenges to electoral accountablty n new democraces. 12 In a related set of models, Casell and Morell (2004) and Messner and Polborn (2004) hghlght that poor democratc performance may be the consequence of the comparatve advantage that low-qualty canddates have n competng for poltcal offce; Keefer and Vlacu (2007) explan why poltcans n new democraces turn to clentelstc polces when they lack the credblty to make electoral promses to delver publc goods; Brender and Drazen (2009) study the role of fscal manpulaton n shorng up publc support n a new democracy; and Merowtz and Tucker (forthcomng) show that the publc may lose an ncentve to protest aganst a poorly performng government after learnng about the (dsmal) qualty of poltcans n a new democracy. The analyss here bulds on Myerson (2006), who ponts out that electoral accountablty may fal f forward-lookng voters hold pessmstc expectatons about any poltcan s performance. I explan why followng repeatedly dsappontng government performance, voters and canddates ratonally abandon a focal equlbrum n whch electons have the potental to delver accountablty for one n whch electons fal to do so. By relatng ths equlbrum shft to government performance, the model n the next secton dentfes emprcally verfable crcumstances under whch voters skeptcsm about whether electons can delver poltcal accountablty becomes self-fulfllng. In the next secton, I present the formal model. I then outlne ts mplcatons for the emprcal study of democratc breakdown and consoldaton and show that t accounts for several key fndngs n the lterature. I conclude by dscussng the mplcatons of my arguments for the contrbuton of consttutonal desgn to the consoldaton of new democraces. Detaled proofs of all techncal results as well as a self-standng statstcal analyss of some predctons from the present model can be found n a supplementary appendx at the author s webste. An Accountablty Model of Democratc Consoldaton and Breakdown Consder the followng electoral accountablty game between a voter and two canddates ( = 1, 2). At the begnnng of any perod t = 1, 2,...,, oneofthetwo canddates s the ncumbent and ether behaves n offce or explots offce. The latter acton mples polces that beneft the ncumbent at the expense of the voter, and we assume that any ncumbent prefers explotng to behavng n offce. More precsely, explotng offce s costless for an ncumbent, whereas behavng s costly, c > 0. Addtonally, an ncumbent prefers to reman n offce n the next perod rather than be replaced by the challenger. I normalze the canddates payoffs from beng n offce and out of offce to w>0 and 0, respectvely. The canddates and the voter dscount future payoffs by a dscount factor, (0, 1), respectvely. I wll notatonally dstngush between and only when dscussng player-specfc dscount factors. As I ndcated earler, two assumptons n ths model reflect the dstnct challenges to electoral accountablty n new democraces. Frst, each canddate can be one of two types, {B, N}. Canddate s bad, = B, f he would prefer explotng offce to behavng even f a permanent removal from offce (reelecton) were a sure consequence of the former (latter). Thus, for a bad canddate, w>(w c)/(1 ), or equvalently f c > w. On the other hand, canddate s normal, = N, fthe threat of a certan removal from offce deters hm from explotng t, w c. Thus, we can alternatvely thnk of c as the beneft from explotng offce and of the bad canddate as one for whom the beneft from explotng offce n the present term s more attractve than a reelecton for another term. 13 But note that the normal type of a canddate s no angel: f not threatened wth a removal from offce,hepreferstoexplottratherthanbehave.importantly, a canddate s type s hs prvate nformaton: the voter does not know the canddates types, and each canddate knows hs own type but not the type of the other canddate. The second key assumpton n ths model s costly montorng. The voter does not observe the ncumbent s acton drectly; nstead, she ether gnoresor montors the ncumbent s performance, and the latter entals a cost 12 See Besley (2006) and Persson and Tabelln (2000) for a revew of electoral accountablty models. For emprcal research on accountablty n new democraces, see Tucker (2006), Keefer (2007), Roberts (2008), Pacek, Pop-Eleches, and Tucker (2009), and Gehlbach, Sonn, and Zhuravskaya (2010). 13 The dfference between the two canddate types can therefore be thought of as a summary of the poltcally relevant dfference among a contnuum of canddate types that dffer n the ntensty of ther concern for reelecton as opposed to the beneft from explotng offce.

690 MILAN W. SVOLIK m > 0. 14 Only f the voter montors the ncumbent s performance, she observes a polcy outcome O, whch s ether a polcy success or falure, O {S, F }. The polcy outcome s an nformatve but mperfect sgnal of the ncumbent s acton: f the ncumbent behaves, the voter observes a polcy success wth probablty b andapolcy falure wth probablty 1 b ; f the ncumbent explots offce, the correspondng probablty of a success s e.i assume that 0 < e < 1/2 < b < 1, so that polces may fal even f the ncumbent behaves but are less lkely to do so than when the ncumbent explots offce and vce versa. In order to keep the formal analyss as smple as possble, I let e = 1 b and assume that the voter does not dscrmnate between the two canddates n her decson whether to montor the ncumbent s performance. At the end of any perod, the voter ether reelects or replaces the ncumbent wth the challenger. The voter prefers a polcy success to a falure, wth the correspondng payoffs s > 0and0. 15 Snce b > e, the voter would lke to motvate the normal ncumbent to behave n offce by condtonng reelecton on a polcy success. However, f the voter gnored the ncumbent s performance, she must decde whether to reelect the ncumbent wthout any nformaton about the polcy outcome. Electoral Accountablty wthout Bad Canddates Before examnng equlbrum outcomes n ths accountablty game, consder a smpler settng wth only the normal canddate type. Equlbrum outcomes n ths smpler settng wll serve as a benchmark aganst whch outcomes n the more complex settng wth both normal and bad canddates can be compared. When only normal canddates are present, the threat of a removal from offce followng a polcy falure wll motvate them to behave as long as both the voter and the canddates are suffcently patent. More specfcally, the followng strateges consttute an equlbrum: n any perod, ncumbents behave n offce whle the voter montors ncumbent performance and condtons reelecton on a polcy success, as long as the voter montored and 14 On the other hand, the canddates observe the voter s actons as well as polcy outcomes, but each canddate only observes hs own acton. 15 I only consder a sngle, representatve voter n order to focus on the problem of motvatng desrable canddate performance. In a settng wth many voters, the preferences of all voters about the ncumbent s acton n offce would be dentcal. In effect, I am assumng that any sngle voter sees electons as an opportunty to act strategcally vs-à-vs the canddates rather than other voters. condtoned reelecton on a polcy success n any prevous perod. Otherwse, ncumbents explot offce whle the voter gnores polcy outcomes and uses a reelecton rule that s unrelated to actual polcy outcomes. 16 To see that these strateges consttute an equlbrum, suppose that the voter montored ncumbent performance and condtoned reelecton on a polcy success n any prevous perod and consder ncumbent s expected dscounted payoff u I (b)frombehavng, u I (b) = w c + [ b u I (b) + (1 b)u C ], (1) where u C s s expected dscounted payoff when he s the challenger whle ncumbent j behavesnoffce, u C [ = b u C + (1 b )u I (b)]. (2) Solvng (1) and (2) for u I (b), we obtan u I (b) = (1 b )(w c) (1 )[1 ( b e )]. Incumbent prefers behavng to explotng as long as behavng yelds a greater expected dscounted payoff than a one-shot devaton to explotng, u I (b) w + [ e u I (b) + (1 e)u C ], or equvalently, as long as c ( b e )w =. Observe that the threshold dscount factor s postve, ncreasng n c and e,decreasngnw and b, and less than 1 as long as w>c/( b e ) = w. Thus, ncumbents have a stronger ncentve to behave when they value offce hghly, when the cost of behavng s low, and when the polcy outcome s a more nformatve sgnal of the ncumbent s acton. Now consder the voter s ncentve to montor ncumbent performance and condton reelecton on a polcy success. When she does so and ncumbents behave n offce, her expected dscounted payoff s v = b s m + v = bs m 1. By contrast, ncumbents wll explot offce followng any perod n whch the voter gnored ncumbent performance, yeldng to the voter the expected dscounted payoff v = es 1. (3) 16 The restrcton to pure strateges and grm trgger punshments smplfes the analyss by avodng reelecton rules that condton on the hstory of play n possbly complex ways; see, e.g., Schwabe (2011). Banks and Sundaram (1993) and Myerson (2006) employ grm trgger strateges n repeated models of electoral accountablty.

LEARNING TO LOVE DEMOCRACY 691 In turn, the voter prefers to montor ncumbent performance and condton reelecton on a polcy success as long as m v b s + v, or equvalently ( b e )s =. Note that the voter s threshold dscount factor s postve, ncreasng n m and e,decreasngns and b, and smaller than 1 as long as s > m/( b e ) = s. Thus, the voter s wllng to montor ncumbent performance and condton reelecton on a polcy success when she values polcy successes hghly, when the cost of montorng s low, and when the polcy outcome s a more nformatve sgnal of the ncumbent s acton. 17 To summarze, we see that n ths benchmark settng, the threat of a removal from offce after a polcy falure motvates canddates to behave n offce as long as the canddates and the voter are suffcently patent, and, and value the offce and polcy successes hghly enough, w w and s s. In order to restrct attenton n the remander of the artcle to nterestng scenaros, I assume that these condtons on patence as well as the assocated lower bounds on w and s hold. Accordngly, we may say that wth only the normal canddate types, electoral accountablty succeeds when,, w w,ands s. 18 Proposton 1. In a subgame perfect equlbrum wth only normal types, ncumbents behave n offce whle the voter montors and condtons reelecton on a polcy success, as long as,, w w, s s, and the voter montored and condtoned reelecton on a polcy success n any prevous perod. Proof. Follows drectly from the text. 17 When the voter gnores ncumbent performance, her postelecton payoff wll be nformatve about the polcy outcome, but she wll not be able to use that nformaton to dscplne the ncumbent because of the ensung grm trgger regme. For nstance, we can vew the eventual realzaton of the payoff as the outcome of a reform that occurs too late for the voter to use t at the tme of votng. An alternatve formulaton that avods such post-electon nformatveness would assume that unless the voter montors ncumbent performance, she does not know whether the ncumbent s acton could have affected the polcy outcome n the frst place (and the parameter that determnes ths vares across perods). 18 As s typcal for repeated games, ths good equlbrum s not unque; there s also a bad equlbrum n whch ncumbents explot offce whle the voter gnores polcy outcomes. Throughout ths artcle, however, the former can be plausbly consdered focal when t exsts, snce the advent of democratc nsttutons s typcally heralded wth enthusasm. By contrast, the trap of pessmstc expectatons equlbrum ntroduced n the next secton s unque. The arguments n ths artcle help us understand the reversal of such ntal optmsm. The Trap of Pessmstc Expectatons Consder now whether electoral accountablty succeeds once we let each canddate be ether a normal or bad type, whch s prvately observed by each canddate. The tmng of moves s as follows. At tme t = 0, nature determnes the type of each canddate and determnes (wth equal probablty) whch canddate wll be the ncumbent n perod one. Independently, each canddate wll be normal wth probablty 0 (0, 1) and bad wth probablty 1 0. The followng stage game ensues n each subsequent perod t = 1, 2,...,. Frst, the ncumbent ether behaves n offce or explots offce. Second, nature determnes whether the polcy outcome s a success or falure. Thrd, the voter ether montors or gnores the ncumbent s performance and observes the polcy outcome f and only f she montored the ncumbent s performance. Fnally, the voter ether reelects or replaces the ncumbent. Denote by t( t 1, O t 1 ) the voter s belef that canddate s normal at the begnnng of perod t 1 gven her belef about canddate s type and the polcy outcome n the prevous perod, t( t 1, O t 1 ) =, O t 1 ). The voter updates her belef t( t 1, O t 1 ) about ncumbent s type usng Bayes rule; 1 0 and 0 2 are the voter s ntal, possbly dstnct belefs about each canddate s type. 19 If the voter montors the polcy outcome, then after observng a polcy success n perod t 1, the voter s belef at the begnnng of perod t that the ncumbent s normal wll ncrease to Pr( = N t 1 t ( t 1, S ) = b t 1 b t 1 ( + e 1 t 1 On the other hand, f the voter observes a polcy falure n perod t 1, her belef that the ncumbent s normal wll declne to t ( t 1, F ) = (1 b ) t 1 (1 b ) t 1 ). + (1 e ) ( 1 t 1 If the voter gnores the polcy outcome n any perod t 1, then a normal ncumbent explots offce n any followng perod and the voter s belef about both canddates remans constant across perods, t( t 1 Fnally, canddates update ther belefs about each other s type accordng to Bayes rule as well, but they observe the polcy outcome n any perod. We may call the voter s belef about canddate s type t canddate s reputaton. )., O t 1 ) = t 1. 19 Such dfferences n ntal belefs may exst n countres where canddates have had an opportunty to acqure a poltcal reputaton pror to the transton to democracy, ether because the dctatorshp that preceded the transton allowed for some electoral competton (as n Mexco) or because t was preceded by a perod of democracy (as n Argentna).

692 MILAN W. SVOLIK When convenent, I wll denote the two canddates reputatons by (k) 1 and (l) 2.Thendexesk and l refer to the ordered set of the voter s possble belefs about canddates 1 and 2 nstead of tme perods. Thus, for canddate 1, (k) 1 = 1 t( (k 1) 1, S) = 1 t( (k+1) 1, F ), and the transton from (k) 1 to (k+1) 1 corresponds to a onestep ncrease n canddate 1 s reputaton. 20 Can the voter motvate normal ncumbents to behave n offce by condtonng reelecton on a polcy success? Consder a scenaro n whch after a successon of polcy falures the voter concludes that both canddates are most lkely of the bad type. That s, t s close to zero for = 1, 2. Because montorng polcy outcomes s costly, at whch the voter concludes that montorng ncumbent performance s not worth ts cost, snce both canddates are most lkely bad and wll explot offce regardless of the voter s actons. Antcpatng ths, even normal ncumbents wll explot offce. Therefore, after a successon of polcy falures, electoral accountablty mayfal even f both canddates are normal and wllng to behave n offce f suffcently motvated. I now examne when the above dynamc depcts a perfect Bayesan equlbrum of ths accountablty game. Snce a bad ncumbent always explots offce, we only need to consder strateges for the normal type of the ncumbent and the voter. When canddate 1 s the ncumbent n perod t and the voter s belefs about the canddates there wll be a level of t are (k) 1 and (l) 2, her payoff from montorng ncumbent 1 s performance and condtonng reelecton on a polcy success s v 1( (k) 1, (l) ) 2 = (k) 1 v 1N + ( 1 (k) ) 1 v 1B, (4) where v 1N and v 1B are the voter s expected dscounted payoffs from the normal and the bad type of ncumbent 1, respectvely. When the normal ncumbent behaves n offce, the voter s expected dscounted payoff s v 1N = m + b (s + v 1N ) + (1 b ) v 2( (k 1) 1, (l) ) 2, where v 2 ( (k 1) 1, (l) 2 ) s the voter s payoff when canddate 2 replaces 1 n offce n perod t + 1 after a polcy falure. Because a bad ncumbent always explots offce, the voter s expected dscounted payoff from a bad ncumbent 1 s v 1B = m + e (s + v 1B ) + (1 e ) v 2( (k 1) 1, (l) ) 2. Intutvely, the voter s payoff from montorng ncumbent performance and condtonng reelecton on a polcy success n (4) s ncreasng n each canddate s reputaton. As the canddates reputatons declne, the 20 For nstance, for b = 0.7, e = 0.3, and the ntal belef 0 1 = 0.5 (the parameters n the llustraton n Sectons 2.5 and 3), the (ncomplete) ordered set of the voter s possble belefs about canddate 1s{..., 0.07, 0.16, 0.3, 0.5, 0.7, 0.85, 0.93,...}. ) enters offce, electoral accountablty fals: the voter no longer montors ncumbent performance and therefore cannot condton reelecton on actual polcy outcomes. To keep the analyss smple, I assume that the ncumbent s reelected wth the probablty 1/2 f ths voter s expected per-perod payoff approaches e s m, whch s less than e s, the expected per-perod payoff that the voter would obtan f she gnored ncumbent performance. Suppose canddate 1 s the ncumbent n the current perod. 21 Then for some (l) 2,therewllbea threshold belef (k) 1 = 1 ( (l) 2 ) such that f the ncumbent s reputaton drops to (k 1) 1,thevoterprefersto gnore ncumbent performance, 1 ( (l) 2 ) = mn (k) 1 such that v 1( (k) 1, (l) ) 2 >v. (5) Recall from (3) that v = e s /(1 ) s the payoff that the voter obtans when she gnores the ncumbent s performance and both canddates explot offce. In the supplementary appendx, I show that the ncumbent s threshold reputaton 1 ( (l) 2 ) s weakly decreasng n the challenger s current reputaton (l) 2.That s, the stronger canddate 2 s reputaton, the worse canddate 1 must perform before the voter gves up on montorng ncumbent performance. In new democraces, one canddate s success n offce s a postve externalty from whch all canddates beneft. Furthermore, the threshold reputaton 1 ( (l) 2 )may not exst when (l) 2 s above some threshold 2 > 0. In other words, the voter wll not gve up on montorng ncumbent performance when canddate 1 s n offce, as long as her belef that canddate 2 s normal s hgh enough. The voter must beleve suffcently strongly that both canddates are crooks before she gves up on montorng ncumbent performance. Once a canddate wth a reputaton below the threshold 1 ( (l) 2 occurs. 22 Importantly, once the threshold 1 ( (l) 2 )has been crossed, the voter s skeptcsm about canddate behavor s self-fulfllng: even a normal ncumbent who would be wllng to behave n offce f the voter condtoned reelecton on a polcy success wll choose to explot offce. Accordngly, we may call ths equlbrum the trap of pessmstc expectatons. 21 The arguments when canddate 2 s the ncumbent are analogous. 22 In a rcher model, the voter may have the opton to abstan and votng may ental a small cost, gvng the voter an ncentve to abstan should the ncumbent s reputaton drop below the threshold 1 ( (l) 2 ). Alternatvely, canddates may start competng on ssues that are less demandng on the voter s attentveness, such as charsma, clentelstc practces, or populst polcy proposals; see, e.g., Ktschelt (2000) and Stokes (2005) for a dscusson of the negatve consequences of these practces for electoral accountablty.

LEARNING TO LOVE DEMOCRACY 693 Proposton 2. In a perfect Bayesan equlbrum, () If 1 t 1 ( (l) 2 ), t 2 2 ( (k) 1 ), and as long as the voter montored and condtoned reelecton on a polcy success n any prevous perod, normal ncumbents behave n offce, and the voter montors ncumbent performance and condtons reelecton on a polcy success. Otherwse, normal ncumbents explot offce and the voter gnores ncumbent performance and uses a reelecton rule that s unrelated to polcy outcomes. () The belef thresholds 1 ( (l) 2 ) and 2 ( (k) 1 ) are weakly decreasng n (l) 2 and (k) 1, respectvely. m () If s > e ( b e ), the belef thresholds 1 ( (l) 2 ) and 2 ( (k) 1 ) do not exst for suffcently hgh values of (l) 2 and (k) 1, respectvely. Proof. See the supplementary appendx. The Breakdown of Democracy The above analyss mples that a sequence of polcy falures may result n a complete falure of electoral accountablty. Once n the trap of pessmstc expectatons, voters no longer beleve that ther democracy can delver better governance than a dctatorshp. In turn, they no longer have a stake n defendng democracy aganst attempts to subvert t, thus gvng canddates an ncentve to acqure or mantan power by nondemocratc means. 23 Afalure of electoral accountablty may therefore precptate the breakdown of democracy. I now examne an extenson of the above accountablty game that models ths ntuton. Suppose that at the begnnng of any perod t = 1, 2,..., and before any of the exstng moves, canddates smultaneously decde whether to subvert or comply wth democracy. If both canddates comply, the stage game contnues as prevously. But f an attempt to subvert democracy occurs, the voter ether acquesces or defends democracy, the latter beng assocated wth a cost d > 0. If the voter acquesces, the game ends n a breakdown of democracy and the voter receves the dscounted payoff v. If only one canddate subverts democracy, the canddate who subverts becomes a dctator and obtans the per-perod payoff w n any subsequent perod, whle the other canddate obtans the payoff 0. If both canddates subvert, then each becomes a dctator wth the probablty 23 Ths lne of reasonng thus parallels Wengast s argument that n order to survve, a consttuton must have more than phlosophcal or logcal appeal; ctzens must be wllng to defend t (1997, 251). Here, such defense most plausbly takes the form of a postelectoral protest, as n Fearon (2011), Merowtz and Tucker (forthcomng), and Tucker (2007). See also Casper and Taylor (1996), who emphasze the role of the mass publc durng transtons to democracy. 1/2. On the other hand, f the voter defends democracy, any attempt to subvert democracy fals, any canddate who attempts to subvert democracy leaves the game and receves a negatve dscounted payoff u < 0, a new canddate enters the game n hs place, and the stage game contnues as prevously. Ths extenson mples that once a democracy enters the trap of pessmstc expectatons, both the ncumbent and challenger subvert democracy, the voter acquesces, and the game ends n a breakdown of democracy. After accountng for the cost of defendng democracy, the threshold 1 ( (l) 2 )n(5)becomes 1 B ( (l) 2 ) = mn t 1 such that v1( (k) 1, (l) ) 2 d >v. (6) That s, as long as ncumbent 1 s reputaton s above 1 B( (l) 2 ), the voter has an ncentve to defend democracy for a suffcently small cost d, because her expected payoff under democracy v 1 ( (k) 1, (l) 2 ) s greater than her payoff v under dctatorshp. Accordngly, we may call 1 B( (l) 2 ) n (6) the breakdown threshold. Proposton 3. In a perfect Bayesan equlbrum, f 1 t B 1 ( (l) 2 ), t 2 B 2 ( (k) 1 ), d < ( b e )s m, and as 1 long as the voter montored and condtoned reelecton on a polcy success n any prevous perod, both canddates comply wth democracy, and the voter defends democracy. Otherwse, both canddates subvert democracy and the voter acquesces. Proof. See the supplementary appendx. Reputaton Buldng and Democratc Consoldaton Propostons 2 and 3 mply that as long as the ncumbent s reputaton s above the breakdown thresholds 1 B( (l) 2 ) and 2 B( (k) 1 ), the voter montors ncumbent performance and condtons reelecton on a polcy success, thus motvatng normal ncumbents to behave n offce. In the long run, however, any democracy wll fall nto the trap of pessmstc expectatons: wth a postve probablty, a sequence of polcy falures occurs that s suffcently long to brng even a normal ncumbent s reputaton under the breakdown threshold. Ths happens because we fxed the dentty of the two canddates who compete for offce at the begnnng of the game. I now relax that assumpton by lettng canddates enter and ext poltcs. Suppose that at the begnnng of any perod t = 1, 2,..., and before any other moves n the game, each of two canddates decdes whether to run for offce or ext poltcs. Any canddate who exts obtans a per-perod

694 MILAN W. SVOLIK ), a bad challenger 2 wll ext, and no bad challenger wll run for offce agan. Ths occurs because a bad ncumbent, who gets reelected less often than a normal one, sees hs chances of returnng to offce declne as hs compettor develops a reputaton for beng the normal type. Once the ncumbent s reputaton crosses the consoldaton threshold, equlbrum strateges n ths game wll be dentcal to those n the benchmark game wthout bad types, and, crucally, electoral accountablty wll succeed. Consstent wth our earler analyss, however, before the consoldaton threshold s crossed, the voter may conclude after repeated polcy falures that both canddates are bad, and ths democracy may fall nto the trap of pessmstc expectatons and break down. To examne these ntutons, consder how the ncumbent s reputaton affects the challenger s ncentves to stay n poltcs. The expected dscounted payoff of a challenger of ether type s decreasng n the ncumbent s reputaton. For nstance, the payoff of a bad challenger 1s payoff x > 0 from an alternatve career. New canddates are drawn at random from a canddate pool wth a postve share of normal types > 0. 24 The tmng of the moves n the rest of any perod remans unchanged. Ths natural extenson has some surprsng but ntutve mplcatons. When suffcently attractve nonpoltcal careers are avalable, good reputatons serve as barrers to the entry of bad canddates. More specfcally, once ncumbent 1 s reputaton grows past the consoldaton threshold 1 ( (l) 2 u C 1B ( ) ( (k) 1, (l) 2 = [( ) (l) 2 b u C 1B ( (k) 1, (l+1) 2 ( )] +(1 b )u1b I (k) 1, (l 1) 2 ( )[ ( ) + 1 (l) 2 e u C 1B (k) 1, (l+1) 2 ( )]) +(1 e )u1b I (k) 1, (l 1) 2, and the payoff of a normal challenger 1 s analogous. Intutvely, both types of challenger expect to return to offce sooner f the ncumbent s a bad type because the latter explots offce and s therefore less lkely to get 24 In a rcher model, the voter may have the opton to replace the challenger wth a canddate from the pool should the challenger s current reputaton t j drop below the voter s belef about the qualty of the pool and update her belefs about the qualty of the pool based on her belefs about the canddates. The key force that accounts for the ext of bad canddates n ths secton a bad canddate antcpates to stay out of offce longer when competng aganst a normal canddate would also operate n such a rcher model. However, the need to condton canddates contnuaton payoffs on the voter s evolvng belef about the qualty of the pool sgnfcantly reduces the tractablty of the model wthout yeldng addtonal nsghts. reelected. Meanwhle, the expected dscounted payoff of a bad ncumbent 1 s ( ) ( ( ) u1b I (k) 1, (l) 2 = w + e u1b I (k+1) 1, (l) 2 ( )) +(1 e )u C 1B (k 1) 1, (l) 2 ; the payoff of a normal challenger 1 s analogous. Because a bad ncumbent s less lkely to get reelected, he expects to spend less tme n offce than a normal ncumbent. Jontly, the above expressons for the ncumbent s and challenger s payoffs mply that a normal challenger expects a greater dscounted payoff from runnng than a bad challenger as the ncumbent s reputaton approaches 1. When canddate 2 s the challenger, we have lm (k) 1 1 u C 2N ( (k) 1, (l) 2 ) > lm (k) 1 1 u C 2B ( (k) 1, (l) 2 ). (7) Inequalty (7) ndcates that once the ncumbent s reputaton (k) 1 s suffcently strong, there wll be a range of dscounted ext payoffs u(ext) = x/(1 ), for whch a bad challenger prefers to ext but a normal challenger prefers to run. A bad challenger 2 wll ext f u(ext) > lm (k) 1 1 u C 2B ( (k) 1, (l) 2 ). On the other hand, a normal challenger 2 wll run even as the ncumbent s reputaton (k) 1 approaches 1 as long as u(ext) < lm (k) 1 1 u C 2N ( (k) 1, (l) 2 ). Suppose therefore that lm (k) 1 1 u C 2B ( (k) 1, (l) 2 ) < u(ext) < lm (k) 1 1 u C 2N ( (k) 1, (l) 2 ) and denote the hghest value of ncumbent 1 s reputaton at whch bad challenger 2 prefers runnng to extng by 1 ( (l) 2 ), ( ) ( ) 1 (l) 2 = max (k) 1 such that u C 2B (k) 1, (l) 2 u(ext). Then as soon as ncumbent 1 s reputaton grows past 1 ( (l) 2 ), a bad challenger 2 exts, and any bad canddate whom nature may draw to run wll ext as well. 25 Denote the perod n whch ths occurs by T.Thenfromperod T onward, only normal canddates run for offce, and the voter s consstent belef s t = 1for = 1, 2. Ths equlbrum dynamc mples that once ncumbent 1 s reputaton grows past 1 ( (l) 2 ), electoral accountablty succeeds: the voter correctly nterprets any polcy falure as the result of bad luck rather than a bad ncumbent s polcy choce. Consequently, ths democracy can no longer enter the trap of pessmstc expectatons and break down. Accordngly, we may call the sets of belefs 1 ( (l) 2 ) and 2 ( (k) 1 )theconsoldaton thresholds,thetmet the 25 In the unlkely case that the ncumbent wth t 1 > 1( (l) 2 )sa bad type, he wll ext as soon as he s replaced n offce snce the new ncumbent wll be normal for sure.

LEARNING TO LOVE DEMOCRACY 695 consoldaton tme, and any state past perod T a consoldated democracy. By contrast, the set of states n whch canddate reputatons are at or between the breakdown and consoldaton thresholds may be called a transtonal democracy. FIGURE 2 Breakdown and Consoldaton Thresholds (Crcles and Squares, Respectvely) When Canddate 1 or 2 Is the Incumbent (Sold and Dashed Lnes, Respectvely) Proposton 4. In a perfect Bayesan equlbrum, ( n any perod t T, f ) u(ext) lm t 1 uc jb ( t 1, t 2 ), lm t 1 uc jn ( t 1, t 2 ) for j, and as long as the voter montored and condtoned reelecton on a polcy success n any prevous perod, () the voter s consstent belef s t = 1 for = 1, 2, () both normal and bad ncumbents run, and () normal challengers run but bad challengers ext. π 2 l 0.97 0.93 0.84 0.7 0.5 0.07 0.16 0.3 0.5 0.7 0.84 0.93 0.97 0.93 0.84 0.7 0.5 Proof. See the supplementary appendx. 0.3 0.3 An Illustraton Accordng to Propostons 2 4, the voter s belefs about the type of canddates that she s facng and ultmately about whether democracy can delver accountablty evolve endogenously. In any perod, the ncumbent s type and equlbrum strategy determne the most lkely polcy outcome, whch n turn shapes the voter s belef about the ncumbent s type. Eventually, however, ths process results n one of two outcomes: the breakdown or the consoldaton of democracy. I now llustrate ths dynamc wth smulatons of the equlbrum behavor predcted by ths model. Suppose that s = 1, m = 0.1, w = 1, c = 0.1, d = 0.5, x = 0.35, b = 0.7, e = 0.3, and = = 0.95. In Fgure 2, I plot the breakdown thresholds 1 ( (l) 2 )and 2 ( (k) 1 ) and the consoldaton thresholds 1 ( (l) 2 )and 2 ( k) 1 ). The thn dotted lnes plot the (ncomplete) set of canddate reputatons, whle crcles and squares connected by thck sold and dashed lnes plot the breakdown and consoldaton thresholds when canddate 1 or 2 s the ncumbent, respectvely. As stated by Proposton 2, the breakdown threshold when canddate 1 s the ncumbent 1 ( (l) 2 )sweaklydecreasngnthechallenger sreputaton (l) 2 and may not bnd f the challenger s reputaton s suffcently strong. We see that democraces n whch ntal belefs about canddates are so pessmstc that they le below both breakdown thresholds wll start n the trap of pessmstc expectatons and are doomed to break down. Meanwhle, those democraces n whch ntal belefs are so optmstc that they le above one of the consoldaton thresholds wll consoldate mmedately. 0.16 0.07 0.03 0.07 0.16 0.3 0.5 0.7 0.84 0.93 π 1 k 0.16 0.07 0.03 Note: Thn dotted lnes plot the (ncomplete) set of canddate reputatons that may occur n equlbrum. For most real-world cases, however, an emprcally plausble set of ntal belefs s nether so pessmstc as to result n an mmedate breakdown of democracy nor so optmstc as to lead to an mmedate consoldaton of democracy these are transtonal democraces. For nstance, f the voter s ntal belefs about the two canddates are 1 0 = 0 2 = 0.5, then t takes at least three consecutve polcy successes for a democracy to consoldate. Meanwhle, t takes at least fve consecutve polcy falures for a democracy to fall nto the trap of pessmstc expectatons and then break down. Note that any canddate s three consecutve polcy successes wll be suffcent for consoldaton, but t takes more than two falures by each canddate before a democracy breaks down. That s, a sngle successful canddate can brng about the consoldaton of democracy, but wth suffcently optmstc ntal belefs, t wll take several falures by each canddate before a democracy breaks down. The equlbrum behavor n ths accountablty game thus mples a partcular dstrbuton of tme to democratc consoldaton and breakdowns. I examne ths and related emprcal predctons n the next secton.