The Democratic Peace and the Wisdom of Crowdsfaculty.ucmerced.edu/...democratic_peace_and_the_wisdom_of_crowds.pdfThe Democratic Peace and the Wisdom of Crowds BRAD L. LEVECK University

The Democratic Peace and the Wisdom of Crowds

BRAD L LEVEC K

University of California

NEI L NARANG

University of California

This article proposes a new theory for the democratic peace that highlights a previously unexplored advantage enjoyed by de-mocracies in crises We argue that because democracies typically include a larger number of decision-makers in the foreignpolicy process they will produce fewer decision-making errors in situations of crisis bargaining Thus bargaining amonglarger groups of diverse decision-makers will fail less often In order to test our hypothesis we use data from experiments inwhich subjects engage in ultimatum bargaining games We compare the performance of individuals small groups and for-eign policy experts against the performance of larger groups of decision-makers We find strong support for the idea that col-lective decision-making among larger groups of decision-makers decreases the likelihood of bargaining failure

Introduction

Few phenomena in the field of international relations receivethe same level of academic attention as the finding that de-mocracies tend to resolve their conflicts with one anotherthrough means short of war This well-established patternmdashthe democratic peacemdashhas two parts first and most fa-mously the existence of few if any clear cases of war betweenestablished democracies (Chan 1984 Kant [1795] 1969Maoz and Abdolali 1989 Weede 1984 1992) second andsomewhat more controversially evidence that democraciesare no less war-prone overall than other kinds of states(Bremer1992 1993 Dixon 1993 1994 Lake 1992 Small andSinger 1976) In other words democracies rarelymdashif evermdashfight each other but because they fight as many warmdashon av-eragemdashas other states it follows that they frequently findthemselves in wars against nondemocratic states

These findings are of such potential importance to poli-cymakers that scholars have over the last several decadessubjected them to numerous empirical checks Overallthese tests support the existence of a democratic peace(Gartzke 1998 2000 Kacowicz 1995 Lemke and Reed1996 Maoz and Abdolali 1989 Maoz and Russett 1993

Oneal and Russett 1999 2001 Rousseau et al 1996Russett 1993 Russett Oneal and Davis 1998 Small andSinger 1976 Thompson and Tucker 1997 Dafoe 2011)1

As Levy notes ldquothe absence of war between democraticstates comes as close as anything we have to an empiricallaw in international relationsrdquo (1989 270)

Perhaps not surprisingly theories of the democraticpeace continue to proliferate alongside empirical tests inpart because of the difficulty in accounting for the appar-ent dyadic nature of the observation What is it about theirinstitutions that facilitate peaceful relations among demo-cratic states Drawing on a now well-established literatureon the advantages of group decision-making we propose anew theory for the democratic peace We highlight a previ-ously underexplored advantage that democracies may havein crisis bargaining Specifically we argue that democraticstates have diverse collections of independently decidingindividuals This will likely lead democracies to producefewer decision-making errors than states that place moreforeign policy decision processes in the hands of smallerand more homogenous groups of individualsmdashwhether in-dividual leaders or even foreign policy experts

We test these expectations via a simple experimentaldesign that isolates one key difference between demo-cratic and autocratic decision-making democracies typi-cally have a larger group of decision-makers involved inthe foreign policy process Closely matching our experi-mental conditions with both the assumptions of the bar-gaining model of war and the ldquowisdom of the crowdsrdquoliterature we find strong support for the idea that collec-tive decision-making decreases the likelihood of bargain-ing failure Across experimental conditions larger groupsof decision-makers consistently outperform individuals insituations of ultimatum bargaining whether they are

Brad L LeVeck is an assistant professor of political science at the Universityof California Merced

Neil Narang is an assistant professor of political science at the University ofCalifornia Santa Barbara Formerly he was a Council on Foreign RelationsInternational Affairs Fellow serving as a senior adviser in the Office of theSecretary of Defense for Policy

Authorsrsquo note For their helpful comments and feedback on this project wewould like to thank Scott Wolforth David Lake Dan Neilson Rachel SteinJessica Stanton James Fowler Robert Trager Iyad Rahwan Michal TomzJessica Weeks Rebecca Morton the Human Nature Group at the University ofCalifornia San Diego the Scalable Cooperation Group at the MassachusettsInstitute of Technology Media Lab the University of Virginia InternationalRelations Workshop the University of California Los Angeles InternationalRelations Workshop and the Security Hub at the Orfalea Center at Universityof California Santa Barbara We also thank Daniel Nexon and the editorialteam at International Studies Quarterly along with three peer reviewers for theircomments edits and helpful suggestions

1There may be thousands of books and articles on the democratic peacemdashtoo many to review here See Rosato (2003) and Dafoe (2011) for more thor-ough reviews of the theoretical and empirical challenges to the democraticpeace finding

Brad L LeVeck and Narang Neil (2017) The Democratic Peace and the Wisdom of Crowds International Studies Quarterly doi 101093isqsqx040VC The Author (2017) Published by Oxford University Press on behalf of the International Studies AssociationAll rights reserved For permissions please e-mail journalspermissionsoupcom

International Studies Quarterly (2017) 0 1ndash14

Downloaded from httpsacademicoupcomisqadvance-article-abstractdoi101093isqsqx0404757452by University of California Merced useron 19 December 2017

matched against a smaller group of individuals (ie in amixed dyad) or other similarly large groups The findingsimply that existing theories of the democratic peace that ap-peal to shared normative values accountability or transpar-ency may be correct but also incomplete as simply aggre-gating decision-makersrsquo bargaining choices through avoting institution replicates two key features of the demo-cratic peace finding in a controlled experimental setupdemocratic dyads avoid costly bargaining failure more thanautocratic or mixed dyads and democracies do no worsethan other regime types in terms of bargaining outcomes

Theory

The Wisdom of Crowds

In the opening anecdote of his popular book The Wisdom ofthe Crowds Surowiecki (2005) illustrates a classic example ofhow crowds may be wise At a 1906 county fair in PlymouthEngland British scientist Francis Galton came across aweight-judging competition in which members of a gatheringcrowd lined up to place wagers on the weight of a fat ox Thebest guess won the prize Seven hundred and eighty-seven di-verse individuals (including expert butchers and farmers andnonexpert clerks) tried their luck at guessing the oxrsquos weightin an attempt to win prizes When the contest was overGalton borrowed the tickets from the organization and ana-lyzed the guesses hoping to show that the average voter wascapable of very little Adding the contestantsrsquo estimates to-gether and calculating the mean Galton used this number torepresent the collective wisdom of the Plymouth crowd actingas if the crowd voted as a single person Given the mixture ofthe crowd which included relatively ldquosmartrdquo guesses fromexperts with relatively ldquodumbrdquo guesses from nonexpertsGalton undoubtedly expected the guesses would be way offThe crowd guessed the ox would weigh 1197 pounds The ac-tual weight of the ox was 1198 pounds In Surowieckirsquos wordsldquothe crowdrsquos judgment was essentially perfectrdquo (2005 xiii)

What Galton discovered in averaging the guesses of thePlymouth crowd was a phenomenon now reproduced inmultiple real-world and experimental settingsmdashthat undercertain conditions groups of independent decision-makerscan be remarkably smart even smarter than the smartestmembers within that group While it was certainly true thatthe ldquodumbestrdquo members of the Plymouth crowd performedconsiderably worse than the so-called ldquoexpertsrdquo as Galtonpredicted (each individual in the group was off by an aver-age of nearly fifty-five pounds with a standard deviation ofroughly sixty-two pounds) their guesses appeared wrong invery different ways Some individuals dramatically overesti-mated the weight of the ox and others dramatically underes-timated its weight In averaging a diverse set of individualguesses the errors canceled out and thus produced a collec-tively wise decision In other words even if most peoplewithin a group are not particularly well informed or rational(lacking the ability and desire to make sophisticated cost-benefit calculations) when those imperfect judgments areaggregated together our collective intelligence is oftentimessuperior to the smartest of decision-makers (Tetlock 2005)

The importance of this finding for studying the behav-ior of political and social groups was not lost on GaltonIn particular the analogy to a democracy where peopleof radically different abilities and interests each get onevote suggested itself immediately In Galtonrsquos wordsldquo[t]he average competitor was probably as well fitted formaking a just estimate of the dressed weight of the oxas an average voter is of judging the merits of most polit-ical issues on which he votesrdquo (Surowiecki 2005 xii)

Despite his own belief that power in society should be-long to a select few with the best qualities for breedingGalton later conceded that ldquothe result seems more cred-itable to the trustworthiness of a democratic judgmentthan might be expectedrdquo (Surowiecki 2005 xiii)

Not all crowds are wise however And over timemdashasresearchers examined the implications of Galtonrsquos findingsacross various social contextsmdashthey gradually refined a the-ory of collective intelligence to include certain key criteriaContemporary theorists emphasize that collective accuracydepends on a combination of both individual accuracy anddiversity Specifically collective accuracy can be character-ized by the simple mathematical identity below (Page 2008Hong and Page 2004 2009 2012)

Collective accuracy frac14 average accuracy thorn diversity

Average accuracy in this equation refers to the averagemagnitude of each individualrsquos error Diversity refers tohow different individual guesses are on average What thefirst term in this simple equation makes clear is thatcrowds must know something about the issue at hand Ifindividuals know nothing about an issue and are wildlywrong then the crowd will still tend toward incorrectdecisions as well After all rockets are designed bygroups of engineers not laypeople On the other handif a number of individuals do know something about theproblem at hand but are prone to making differenttypes of errors then aggregating their views can helpmake an accurate decision because different errors willcancel one another out As we discuss below it is plausi-ble that democratic decision-makers are both accurateand diverse enough to give democracies an advantage inforeign policy decision-making

In addition to these general rules scholars in the psy-chology literature have also identified a number of specificconditions under which groups are unlikely to performbetter (Cason and Mui 1997 Bone Hey and Suckling 1999Rockenbach Sadrieh and Barabara 2001 Cox and Hayne2006 Puncochar and Fox 2004 Kerr MacCoun andKramer 1996)2 For example worse decision-making mayemerge when designated leaders promote conformity andself-censorship which can lead to group-think (Sniezek1992 Kleindorfer Kunreuther and Schoemaker 1993Mullen et al 1994) Similarly problems can arise whengroups polarize the attitudinal judgments of their mem-bers (Davis 1992 Kerr MacCoun and Kramer 1996 Casonand Mui 1997) Importantly however many of these con-ditions do not apply in our experimental setup and thereare also good reasons to believe that democratic decision-making is less vulnerable to many of these harmful condi-tions We describe these reasons in detail below

The Wisdom of Crowds in Democracies VersusAutocracies

If a diverse group of independently deciding individualscan be collectively wisemdashand this may be behind some ofdemocraciesrsquo ability to formulate superior policy deci-sionsmdashit is surprising that more attention has not beenpaid to this particular democratic advantage in foreignpolicy decision-making3 Perhaps democracies by

2In the interest of space we review the results of these papers in the sup-plementary appendix

3One exception is an important study by Reiter and Stam (2002) who ap-ply a similar logic to a different empirical puzzle why democracies win the

2 The Democratic Peace and the Wisdom of Crowds

aggregating predictions from a diverse population of in-telligent agents may outperform a team comprised ofeven the best-performing agents That is it might be thecase that democracies have an advantage in foreign policydecision-making when compared against alternative insti-tutional forms like autocracies that aggregate informationfrom a smaller less diverse set of ldquoexpertrdquo individuals

Even though foreign policy decision-making in democ-racies is oftentimes dominated by a relatively small groupof educated elites (Saunders 2011 Hafner-BurtonHughes and Victor 2013 Hafner-Burton LeVeck andVictor 2017) there are still compelling reasons to believethat democracies draw on a larger more diverse set ofviews on average when making decisions about warbargaining First by holding periodic elections citizenscan express their views on which leader or mix of repre-sentatives is best suited to conduct international affairsIndeed existing evidence suggests that citizens whilehardly experts in foreign policy do hold broadly in-formed opinions on such matters see clear differences be-tween the candidates on issues of foreign policy and votepartially on the basis of these factors (Aldrich 1999)Citizens may therefore elect representatives who take aparticular approach to foreign policy such as whether astate should take a more hawkish or dovish approach tomatters of interstate conflict (DeNardo 1995) At thesame time they may leave the details of how to best imple-ment a given approach to elected representatives and thebureaucrats they oversee (Lupia and McCubbins 19942003) The diverse approaches of different elected offi-cials (many of whom have some input into the foreignpolicy decision-making process) may act like the diverseheuristics and interpretations found in recent models ofcollective wisdom (Hong and Page 2004 2009) Secondcitizens in democracies can more efficiently express ap-proval or disapproval for their leaderrsquos policies throughpublic polls Again these polls may aggregate citizensrsquo di-verse views on the wisdom of a particular approach to for-eign policy Third democracies tend to have freer mar-kets with exchanges that can react almost instantly toinform leaders about the expected outcome of a particu-lar policy choice (Gartzke 2007 Wolfers and Zitzewitz2009) These market signals can act like weighted votesfrom market investors Finally democracies tend to estab-lish different domestic institutions with diverseapproaches or perspectives on foreign policy For in-stance in the United States the Departments of State andDefense have different intelligence sources decision-making structures and personnel4 Yet both institutionsmay have input on how to deal with a particular adversary

Together these information aggregation mechanismsallow for more diverse groups of independently deciding

individuals to process information separately and expresstheir own independent assessment on foreign policy mat-ters Thus existing studies support the comparative-staticclaim that democratic decision-making ismdashon averagemdashrelatively more pluralistic than autocratic decision-makingdue to these mechanisms of accountability This is trueeven though the decision to go to war in a democracy likethe United States may ultimately rest with only a smallgroup of leaders gathered in a ldquosituation roomrdquoFurthermoremdasheven when aggregating similar beliefsacross similar numbers of individualsmdashparticipants in au-tocracies often lack the incentive to tell leaders the truth(Reiter and Stam 2002) And although elites may often in-fluence or manipulate the preferences of citizens in de-mocracies (challenging the assumption of independence)(Zaller 1992 Lenz 2012) existing studies suggest thatdemocratic decision-making is influenced by a more di-verse set of opinions on average relative to autocraticstates5

Even at the level of elite decision-makingmdashoutside thedirect influence of everyday citizensmdashthere is little contro-versy in the academic literature that democracies tend tohave a larger group of decision-makers involved in theforeign policy process At the broadest level the Polity IVindex measuremdashon which the democratic peace phe-nomenon is basedmdashis primarily driven by the variableXCONST (Gleditsch and Ward 1997) which in a largepart codes the number of actors across institutions thatconstrain policy-making by the executive The variabletherefore reflects the fact that democratic policy-makingis typically influenced by a larger number of indepen-dent actors Similarly the The Political Constraint Index(POLCONIII) (Henisz 2000) used in some robustnesschecks of the democratic peace (Tsebelis and Choi2009) measures the raw number of institutional vetoplayers and their relative independence in terms of pref-erences and ideological viewpoints6 As we review furtherin the supplementary appendix there is also evidencethat these veto players have some influence over foreignpolicy not just domestic policy

There is also plenty of qualitative evidence to supportthe assumption that democracies contain a larger morediverse group of individual decision-makers on averageFor example in categorizing foreign policy decision-making across states over time Hermann and Hermann(1989) show that autocratic regimes are almost perfectlycorrelated with ldquoPredominant Leaderrdquo or ldquoSingle Grouprdquodecision units that ldquowill be relatively insensitive to discrep-ant advice and datardquo (365) while foreign policy-making indemocratic regimes is correlated with ldquoMultipleAutonomous Actorsrdquo7

Even in the United States where the executive branchis thought to enjoy a great deal of autonomymdashparticularlyover decisions to go to warmdashthere nevertheless exists a ro-bust and well-documented interagency process as a

wars they initiate Reiter and Stam argue that democracies ldquoare better at fore-casting war outcomes and associated costsrdquo because they ldquobenefit from moreand higher quality informationrdquo (2002 23) and thus only initiate winnablewars They argue ldquothe unitary nature of dictatorships forgoes democraticadvantages from the market-place of ideas that provide broad checks on a sin-gle leaderrdquo (2002 25) Reiter and Stam build from Schultz (1999) who alsoraises the prospect that democracies are more strategic about what conflictsthey enter Here we explore whether this advantage helps democracies fore-cast the reservation price of opponents in crisis bargaining and whether itoffers a partial explanation for the democratic peace (ie bargaining successrather than war outcomes)

4In other words even though cabinet membersrsquo views may be correlatedby a shared ideology or by a desire to gain favor with an ideological leader(Saunders 2011) in many contexts ideology will not induce perfectcorrelation

5For example consider that even when partisan media like Fox News orMSNBC heavily influences citizensrsquo views (1) even these opposing views arelikely to create diversity in opinion with errors that cancel out and (2) somecomponent of citizensrsquo opinions still remains statistically independent (ieunexplained) by these ldquoeliterdquo opinions (Levendusky 2009) The experimentbelow can be understood to capture this independent component

6In Supplementary Appendix Table A6 we compare democracies and au-tocracies along both variables quantitatively and show that democracies are sys-tematically characterized by a larger more diverse group of independentlydeciding individuals on average

7Geddes (1999 2003) and Weeks (2012 2014) have also detailed intricatedecision-making processes across different types of autocratic regimes

BRAD L LEVECK AND NEIL NARANG 3

mechanism for collective decision-making At multiple lev-els the US interagency process draws together a diversecollection of independently deciding actors from acrossmultiple agencies with distinctmdashsometimes parochial of-ten times conflictingmdashinterests and beliefs based on inde-pendent characterizations of the international system(Raach and Kass 1995 Marcella 2004 Gorman andKrongard 2005)8

Detailed historical accounts illustrate how this inter-agency process can aggregate a large and diverse numberof views In his seminal article ldquoConceptual Models andthe Cuban Missile Crisisrdquo Allison (1969 63) provides whatis perhaps the most well-known example of how US for-eign policy outputs are ldquothe consequences of innumerableand oftentimes conflicting smaller actions by individualsat various levels of bureaucratic organizations in service ofa variety of only partially compatible conceptions of na-tional goals organizational goals and political objectivesrdquoSpecifically Allison shows that Kennedy struggled toweigh different and sometimes conflicting recommenda-tions from his closest advisors drawn from different agen-cies with different perspectives The moves appearedldquoresultant of collegial bargainingrdquo (Allison 1969 691)from a ldquoconglomerate of semifeudal loosely allied organi-zations each with a substantial life of its ownrdquo (Allison1969 698) As Allison notes ldquothe nature of problems offoreign policy permits fundamental disagreement amongreasonable men concerning what ought to be doneAnalyses yield conflicting recommendations Separate re-sponsibilities laid on the shoulder of individual personali-ties encourage differences in perceptions and priorities More often however different groups pulling in differ-ent directions yield a resultant distinct from what anyoneintendedrdquo (Allison 1969 707) In the US governmentthese actors include ldquochiefsrdquo the president secretaries ofstate defense and treasury director of the CIA jointchiefs of staff and since 1991 the special assistant fornational security affairsrdquo (709)

Allisonrsquos account of the decision to implement a block-ade of Cuba during the crisis provides an excellent illus-tration of how inputs from numerous diverse view-pointsmdasheven from within the executive branch wheremembers often have a shared ideology (Saunders 2011)mdashcan have a significant impact on crisis bargaining As de-scribed by Allison Senators Keating Goldwater CapehartThurmon and others initially attacked Kennedy for hisldquodo nothing approachrdquo while McGeorge Bundy thepresidentrsquos assistant for National Security Affairs assertedthat there was no present evidence that the Cuban andSoviet Government would attempt to install a major offen-sive capability (Allison 1969 712) Meanwhile ColonelWright and others at DIA believed that the Soviet Unionwas placing missiles in Cuba This information fell on thediverse crowd of advisers differently (Allison 1969 713)Kennedyrsquos principal advisors including Secretary ofDefense McNamara McGeroge Bundy TheodoreSorenson and the presidentrsquos brother Robert Kennedyconsidered two tracks do nothing and taking diplomaticaction (Allison 1969 714) However the joint chiefs of

staff advocated for a military invasion of Cuba (Allison1969 714) According to Allison ldquothe process by whichthe blockade emerged is a story of the most subtle and in-tricate probing pulling and hauling [and] leading guid-ing and spurringrdquo Initially Allison notes ldquothe Presidentand most of his advisers wanted the clean surgical airstrikerdquo (Allison 1969 714) Remarkably however despitethe presence of a sizeable minority preferring an airstrike the president ultimately opted for a blockade afterconsidering the advice of McNamara and Robert Kennedy(Allison 1969 714) Reflecting on the influence of the di-verse opinions of his advisors the presidentrsquos brotherclaimed that ldquothe fourteen people involved were very sig-nificantrdquo (Allison 1969 714)

In stark contrast to the Kennedy administrationrsquos han-dling of the Cuban Missile Crisis the overwhelming con-sensus among diplomatic historians on the Cuban MissileCrisis is that Kennedyrsquos counterpart in the Cuban MissileCrisis the Soviet leader Nikita Khrushchev drew from amuch smaller group of advisors than KennedyFurthermore Khrushchev systematically ignored theadvisers that he did consult with during the crisis if theyeven felt safe to express their true beliefs at all (Fursenkoand Naftali 1998 2007 Taubman 2003 Dobbs 2008)Beyond the case of the Cuban Missile Crisis Hermannand Hermann (1989) use four case studies to demonstratehow autocratic regimes made the decision to initiate or es-calate war after periods of failed negotiations due to theirrelative insensitivity to discrepant advice and data In amore recent example Saddam Hussein repeatedly ig-nored the advice of his military advisers and scientists(many of whom appeared afraid to express dissent in thefirst place) many of whom correctly estimated that therate of Iraqrsquos nuclear program ran a high risk of trigger-ing war (Horowitz and Narang 2014 Braut-Hegghammer2016) This further illustrates how autocracies may beworse at incorporating knowledge dispersed among multi-ple actors even when those actors hold key advisory rolesin government

The Wisdom of Crowds and the Democratic Peace

The possibility that a more diverse collection of indepen-dently deciding individuals characteristic of democraticstates might be superior to nondemocracies in predictivetasks has important implications for the democratic peacefinding Existing theories of the democratic peace tend toargue that democratic institutions facilitate peaceful rela-tions among states in two ways first democratic institu-tions can help align the interests of leaders with their citi-zens and second democratic institutions may improvethe quality of information conveyed by states during crisisbargaining9

The first of these explanations begins with the idea thatdemocratic institutions tend to hold leaders accountablefor the costs of war10 War can be an extremely costly andrisky process for citizens They pay the psychological andmaterial costs of fighting in the form of lives lost andhigher taxes However political leadersmdashwho ultimatelymake the decision to wage warmdashrarely suffer these coststhemselves If leaders expect to enjoy the benefits of8Indeed despite the presence of a dedicated intelligence community

organizations in the US federal government maintain their own intelligenceagencies They do this precisely to arrive at independent assessments andavoid group-think for example the Department of Defense operates theDefense Intelligence Agency (DIA) the State Department operates theBureau of Intelligence Research and the Treasury Department operates theOffice of Intelligence Analysis etc

9For a survey of behavioral and normative theories of the democraticpeace dating to Kantrsquos Liberal Peace see Rosato (2003) and Dafoe (2011)See Stevenson (2016) for a review of normative theories

10See Rosato 2003 for a general review of the literature in support of thismechanism

victory with little to no exposure to the costs of wagingwar they will prove more inclined to fight a risky warrather than negotiate a compromise

According to this view representative forms of govern-ments better align the interests of the ruler with the ruledby periodically holding leaders accountable to their citi-zenry (Doyle 1997 24ndash25 Russett 1993 38ndash39) Becausedemocratic institutions make leaders more sensitive to thecosts of war they thereby decrease the probability thatleaders will fight for personal gain (Maoz and Russett1993 Russett 1996) If war is costlier for democratic lead-ers they should be less willing to risk war on average com-pared to leaders of nondemocratic statesmdashwho can affordto gamble with othersrsquo lives and resources This height-ened sensitivity to the costs of war may also explain whydemocracies fight with nondemocracies more often Ifdemocratic leaders are less willing to pay the cost of warautocratic states should challenge democracies more fre-quently and demand greater concessions during diplo-matic negotiations thereby increasing the risk of war

A second popular explanation focuses on how demo-cratic institutions may influence crisis bargaining betweenstates Building off the bargaining model of war (Fearon1995) this argument rests on the idea that war resultsfrom bargaining failure due to credible commitmentproblems or the effects of private information on negotia-tions It wagers that something about democratic institu-tions must solve these problems Thus democracies aremore likely to find mutually beneficial bargains that avoidthe costs of war In particular proponents of this argu-ment suggest that democracies may be better able to re-solve the informational problem that arises when sideshave private information about their costs of war relativeto the issues at stake For example democratic decision-making processes are often more open and transparentespecially in cases where different representatives argue ornegotiate over foreign policy in public forums (Schultz1998 2001) This greater transparency of democraticdecision-making allows opposing states to better assess thetrue capabilities and resolve of democratic states (Schultz1998)11

While both of these arguments suggest plausible mech-anisms that might account for the democratic peaceneither one addresses the possibility that democracy mayproduce superior foreign policy decision-making pro-cesses The first argument simply suggests that leaders rep-resenting democracies are pacific because democraticinstitutions more directly expose them to the costs of warThis should bias democracies toward peace in generalbut does little to explain whymdashif democratic institutionsheighten leadersrsquo sensitivity to the costs of war which inturn causes nondemocracies to exploit their pacific ten-dency to make greater demandsmdashdemocracies do notperform worse on average than other kinds of states incrisis bargaining situations (Bueno de Mesquita et al1999) That is no evidence implies that nondemocraticstates generally extract greater concessions from demo-cratic states over time because the latter are more inclinedto back down

The second argument incorporates our understandingof crisis bargaining It acknowledges that all partiesmdashre-gardless of regime typemdashhave an incentive to avoid warBut it also wagers that democracies are better able to

convey their own capabilities and resolve to opponents Ittherefore implies that democracies are less likely to bechallenged in the first place when possible adversariesperceive them to have high levels of resolve But thisargument may be incomplete It treats the role of thedemocratic decision-making process as strictly passivemdashasallowing an opponent to better assess a democratic statersquosreservation price But it ascribes no distinct advantages todemocratic foreign policy decision-making itself

Our argument is substantially different In contrast toprevious theories of the democratic peace we propose analternative mechanism through which democracies maybe able to resolve the informational problems that lead tobargaining failure For the reasons outlined above weposit that democracies are better able to aggregate and in-terpret noisy signals gathered during a crisis in a way thatcancels out decision-making errors

Consider the simplest model of crisis bargaining as out-lined by Fearon (1995) In this setup two states (S1 andS2) have divergent preferences over the division of someissue space represented by the interval Xfrac14 [01] whereeach statersquos utility is normalized to a zero to one utilityspace S1 prefers issue resolutions closer to one while S2

prefers resolutions closer to zero Supposing states fight awar S1 prevails with probability p 2 [01] and gets tochoose its favorite outcome closer to 1 S1rsquos expected util-ity is pu1(1)thorn (1 p)u1(0) c1 or p c1 S2rsquos expectedutility for war is 1thorn p c2 The parameters c1 and c2 repre-sent the costs for fighting a war to each side along withthe value of winning and losing on the issues at stakeImportantly the costs of fighting open up a range of bar-gained solutions between each statersquos reservation pricep c1 and pthorn c2 that both sides should strictly prefer topaying the costs of war (Narang 2017 Narang and Mehta2017 Mehta and Narang 2017) Structured this way thepuzzle becomes about why sides ever fail to identify a ne-gotiated settlement within this range ex ante knowingthat war is always inefficient ex post

Fearon suggests that coherent rationalist explanationsfor war will fall into one of two categories sides can fail toreach a bargain because (1) they have private informationwith incentives to misrepresent or (2) because sides areunable to credibly commit themselves to follow throughon the terms of the agreement According to the first ex-planation sides have asymmetric information about theirown capabilities p and resolve c and they have an incen-tive to overrepresent (or underrepresent) their ability onthese dimensions to their opponent in order to secure abetter settlement As a result while the costs of fightingopen up a range of negotiated settlements both sides pre-fer to war the incentive to bluff may lead sides to delaysettlement in favor of fighting in order to accrue enoughinformation to formulate reliable beliefs about theiropponentrsquos strength (Slantchev 2003 Narang 20142015)

In situations of incomplete information war (bargain-ing failure) can occur in Fearonrsquos model if State 1 overes-timates State 2rsquos cost of going to war and therefore makesan offer that is too small for State 2 to accept On theother side of the decision war can also occur if State 2underestimates its own costs of war and chooses to onlyaccept offers that State 1 would not reasonably proposeIn each of these cases decision errors can happen be-cause decision-makers have uncertainty about key parame-ters and they can only estimate these parameters withsome error However it is possible that the error made byone decision-maker within a state may be different from

11A related informational mechanism domestic audience costs has alsoreceived significant attention in the crisis bargaining literature See Fearon(1994) Tomz (2007) Weeks (2008)

that of another For example while one decision-makermight overestimate the other statersquos cost of going to waranother decision-maker could err in the opposite direc-tion If such views are aggregated the errors could cancelout

In the next section we describe a version of the classicultimatum game and we use this model as the basis for anexperimental research design in which we test the propo-sition that regimes with more decision-makers experiencefewer instances of costly bargaining failure (analogous towar) and achieve outcomes that are at least as good as theoutcomes achieved by regimes with fewer decision-makers

Methodology and Results

Using observational data to identify the effect of informa-tion aggregation mechanisms on war bargaining outcomesis difficult for a number of reasons First asymmetric in-formation presents the same problem for the analyst thatit does for states in the international system a statersquos reser-vation price for war is private information that is rarelyrevealed This makes it difficult to know how close onestatersquos offers are to another statersquos reservation price forcostly conflict This is especially true for the majority ofcrisis bargaining scenarios because offers rarely triggerwar Even in the rare cases where crisis bargainingdevolves into war it is impossible to know with any cer-tainty just how much one statersquos offer fell short of anotherstatersquos threshold for avoiding conflict

Second in an uncontrolled environment it is difficultto ascertain what information individual decision-makershad access to and exactly how that information was fil-tered through executive decision-making processesFuture work needs to trace the precise process by whichsignals about opponents are aggregated and how these ag-gregated signals influence state decision-makers But thisapproach is not ideal for clearly answering the more pri-mary question of whether aggregation can influence bar-gaining in the manner predicted by existing theoriesSuch questions are better answered in an environmentwhere the researcher can carefully control what informa-tion actors have access to and how that information isaggregated

An Experiment

To examine the question of whether information aggrega-tion can improve bargaining outcomes we look at datafrom laboratory bargaining games Specifically we look ata variant of the ultimatum game (Guth et al 1982) which(as we further explain below) mimics key features of warbargaining12 The game is played between two players aproposer and a responder who bargain over a fixed pie ofone hundred monetary units (mu) The proposer makesan integer offer Sp 2 [0100] which is the portion of thepie she proposes keeping for herself The responder si-multaneously makes a demand Sr 2 [0100] which is theminimum portion of the pie they will accept without

rejecting the proposerrsquos offer The monetary payoffs forthe proposer and responder are the following

ethSp 100 SpTHORN if 100 Sp Sr

eth0 0THORN if 100 Sp lt Sr

In other words if the proposerrsquos offer exceeds or equalsthe responderrsquos demand then the pie is split according tothe proposerrsquos offer If the offer falls short of the demandthen the offer is rejected and both parties receive zeromu

If proposersrsquo and respondersrsquo utility is strictly increasingin the amount of money they personally receivemdashandthey both have mutual knowledge of this factmdashthen theunique subgame perfect Nash equilibrium for theultimatum game is for proposers to offer zero and forresponders to accept zero because they are indifferent be-tween accepting and rejecting If this theoretical expecta-tion holds this might make the ultimatum game a pooranalogy to the bargaining model of war because only theproposer is strictly worse off when an offer of zero is madeand rejected However the existence of this strategy pro-file does not present a major problem for testing our the-ory This is because as a practical matter individuals inthe ultimatum game almost never propose zero or set zeroas their minimum acceptable offer across real world set-tings (Camerer 2003) Thus empirically these potentialoffersmdashwhile theoretically possiblemdashhave no practical ef-fect on our results below13

The infrequency of proposals that offer zero in theultimatum game is likely due to the fact that respondersexhibit aspects of real world bargaining that are crucialfor our particular question they have positive but variableminimum acceptable offers (Camerer 2003 Henrich et al2001) This is because subjects derive utility from otherthings besides monetary payoffsmdashlike satisfying norms offairness or feelings of spite So while the responder cannotpossibly gain a higher payoff by demanding more this isonly true in terms of monetary payoffs In terms of playersrsquoutility for monetary splits things are often different Thismeans that responders can rationally demand more thanzero and proposers can anticipate this by offering somepositive amount to avoid bargaining failure Numerousexperiments have shown that respondersrsquo varied thresh-olds for rejecting an offer do not purely reflect a mistakebut rather some actual differences in playersrsquo utility fordifferent monetary splits (Camerer 2003 Andreoni andBlanchard 2006)

Crucially heterogeneity in demands creates uncertaintyfor proposers regarding what offers will and will not trig-ger costly bargaining failure In this regard the experi-ment is analogous to many models of war bargaining un-der asymmetric information such as Fearon (1995) orPowell (1999) where the proposer makes a single take-it-or-leave-it offer under uncertainty about an opponentrsquoscosts of war (ie opponent type) Such decision-makingerrors are analogous to a leader underestimating its oppo-nentrsquos willingness to fight Rejection in our game is analo-gous to a costly outside option such as war which both

12We use the ultimatum game instead of the games used by Tingley andWang (2010) and Tingley and Walter (2011) which allow the experimenter tomanipulate respondersrsquo cost of bargaining failure We did this for two practi-cal reasons First compared to the laboratory it is more difficult to ensurethat subjects in online experiments fully understand complex instructions(Rand 2012 176) We therefore chose the ultimatum game in part because itwas the simplest game that met our requirements Second there now existhundreds of experiments conducted using the ultimatum game including in-ternational policy elites We could therefore examine how well crowds per-formed relative to individual experts

13Indeed individuals in our experiment vote to propose zero just morethan 4 percent of the time but in most cases these votes do not manifest inobserving a proposal of zero because the votes occurred as part of a group inwhich votes for larger proposals bring the actual observed frequency of pro-posals that offer zero to substantially less than 1 percent

players wish to avoid in favor of some mutually acceptablebargain

While the ultimatum game is a workhorse of laboratorystudies on bargaining our innovation is to systematicallymanipulate the number of decision-makers on each sideand see how this affects the rate of costly bargaining fail-ure Other articles have looked at what happens when sub-jectsrsquo views on how to play the ultimatum game are aggre-gated by deliberation (Bornstein and Yaniv 1998) andvoting (Elbittar Gomberg and Sour 2011) However nostudy to date has examined what happens to the rate ofbargaining success when the number of decision-makerson each side is systematically varied Our experiment doesthis with respect to voting which is a common way for ag-gregating decisions

Even though previous studies of individual bargainingin the ultimatum game suggest that decision-makers avoidbargaining failure a large fraction of the time (Camerer2003) it is far from guaranteed that aggregating subjectsrsquoviews will further increase the proportion of successfulbargains in a population For one subjects may have in-formed views about how to bargain with other individualsbut may be relatively uninformed when it comes to bar-gaining with groups of different sizes Second the size ofa group itself may diminish individual decision-makersrsquoincentives to make wise decisions (Downs 1957) Making awise vote takes mental effort but that effort can be poten-tially rendered moot by other votersrsquo decisions (Downs1957 Popkin 1991) Furthermore simply knowing thatyou are part of a group may make one more aggressivetoward other out groups such as the group you are bar-gaining with (Tajfel and Turner 1979) this aggressionmight plausibly lead to increased bargaining failureWhether these potential pitfalls of collective decision-making can be overcome by its advantages is an empiricalquestion which we test

H1 Our hypothesis is that decisions aggregated fromlarger groups of proposers and responders will lead tofewer instances of bargaining failure and higher earn-ings compared to smaller groups and individuals

To test this we modified an experiment by Rand et al(2013) where we asked proposers and responders to playa single round of the ultimatum game described above14

In the original experiment each proposer submitted asingle offer while each responder submitted a single de-mand simultaneously Experimenters then paireddemands and offers at random and paid subjects accord-ingly Thus each proposer had an incentive to make aproposal that would yield the highest expected earningswhen played against a random (anonymous) responderThe expected success of each proposerrsquos offer in the ex-periment can be calculated based on how often the popu-lation of responders would reject it and how many mone-tary units each proposal would have earned on average

In our modification to this experiment we comparethe success of offers and demands made by small groupsof three individuals to the success of offers and demandsmade by much larger groups of nine individuals Thesesmaller groups of size three in the experiment are analo-gous to autocracies which tend to have a smaller number

of decision-makers included in the policy-making processLarger groups of size nine are taken as analogous to moredemocratic polities where more individuals are typicallyinvolved in the policy-making process We use a group sizeof three for autocracies because it is the smallest size thathas a well-defined majority Henceforth we refer to smallgroups as autocracy and large groups as democracy Ofcourse all the caveats with this stylized operationalizationstill apply (see External Validity section below) We use agroup size of nine because it represents one of the largesttreatment ldquodosagesrdquo we could implement while still havingenough observations to test our directional hypothesis(that larger groups of decision-makers decrease the rateof bargaining failure) However in SupplementaryAppendix Figure 1 we test whether our results are partic-ularly sensitive to using nine players (as opposed tosmaller groups of five or seven) We find evidence thatour results are robust to these differences

We determined a grouprsquos proposal to the other side inthe following manner each individual in a group simulta-neously and anonymously submitted a vote for what theirgroup should offer to the other side We then took themedian offer submitted in the group to represent thegrouprsquos actual proposal For example say that in a groupof three individuals voted to offer seventeen eighteenand twenty-four The grouprsquos actual offer would be eight-een While this procedure certainly does not capture theintricacies of foreign policy decision-making in a democ-racy or any other state it is akin to a decision rule wherethe median voterrsquos preference is decisive and thus itapproximates a number of real-world collective decision-making bodies such as voting in elections (Downs 1957)or Congress (Krehbiel 1998) Specifically aggregationprocesses like this one can be understood as similar to citi-zens voting for politicians with a particular level of hawk-ishness or dovishness representation across bureaucraciesin interagency meetings (Allison 1969 Janis 1972) or con-gressional votes over war authorizationwar funding dur-ing crisis bargaining While there are many significant dif-ferences across each of these aggregation mechanismsthey all collect a large number of diverse viewpoints andaggregate them into a single number or outcome that caninfluence or determine foreign policy

Of course the downside of our stylized procedure isthat it abstracts away from the intricacies of any one ofthese mechanisms However the upside is that it capturesour key independent variable in a way that is tractable andrelatively easy to interpret We further discuss concernsover the external validity of this mechanism in a subse-quent section below

It is also worth noting that in the absence of delibera-tion groupness in our experiment emerges from informingindividuals about whether or not they played in a groupbefore making their votes Thus individuals cast theirvote in expectation of it becoming aggregated Thereforeour treatment induced any behavioral changes that wouldarise from subjects knowingly voting as part of a group toinfluence the final proposal And despite the presence ofdeliberation in the real world (and the attendant risk ofattenuating the wisdom of the crowds) our discussionabove illustrates that the risk of group-think from deliber-ation is much more severe in autocracies whereldquopredominant leaderrdquo or ldquosingle grouprdquo decision units areldquorelatively insensitive to discrepant advice and datardquo(Hermann and Hermann 1989 366) Therefore whileour voting mechanism does not fully capture some of thedynamics that might emerge from deliberation it does

14It is possible that crowds might have additional advantages that wouldemerge in a more dynamic setting Future experiments might explore groupadvantages in learning

preserve the fact that democratic deliberation typicallyinvolves a larger number of more independent inputs

We posted this experiment online and recruited 1409subjects through the internet labor market AmazonMechanical Turk15 We paid subjects $050 as a show-upfee simply for participating in the experiment We ran-domly assigned subjects as players on Side A or Side B Wetold players that Side Arsquos task was to propose to Side Bhow much of $040 should go to each member of Side Band how much should go to each member of Side A Forexample each member of Side B might get $010 imply-ing that each member of Side A would get $03016 Side Bwould decide what minimum amount satisfied an accept-able offer If Side Arsquos offer to Side B met or exceededSide Brsquos minimum acceptable offer then we paid bothplayers the bonuses according to the proposed divisionOtherwise no member of either side earned a bonus

We defined the total size of the pie in terms of whateach member received so that the individual stakes of thedecision remained constant across conditions In otherwords changing the group size across conditions did notchange the absolute amount of a fixed prize that each in-dividual in a group could receive While we made this de-cision primarily to improve the experimentrsquos internal va-lidity (by isolating the effect of aggregation rather than anindividualrsquos stake in the decision) it does have a realworld analogue Whereas the benefits of any bargain aretypically more diffuse in large populations when the stakesare strictly material there are many conflicts where onepolity might impose a different way of life on citizens inanother country (Lake 1992) In these situations citizensand other decision-makers might place the same value ontheir own way of life regardless of how many other citizensexist in the country

To ensure comparability of our study to existing studieswe began by first randomly assigning 232 of the subjects(out of 1409) to a baseline condition of a single proposermaking a take-it-or-leave-it offer to a single responder (thecanonical ultimatum game) We then randomly assignedeach of the remaining 1177 subjects to one of our fourexperimental conditions

1 A small group of three proposers making a take-it-or-leave-it offer to a small group of three responders(autocracyautocracy)

2 A small group of three proposers making a take-it-or-leave-it offer to a large group of nine responders(autocracydemocracy)

3 A large group of nine proposers making a take-it-or-leave-it offer to a small group of three responders(democracyautocracy)

4 A large group of nine proposers making a take-it-or-leave-it offer to another large group of nine respond-ers (democracydemocracy)

We informed subjects that the voting mechanism forgroup decision-making would simply be the highest offerthat gained a majority support as described above A sum-mary of the conditions is shown below in Table 1

For each of our experimental conditions we estimatedhow well each side would do on average both in terms ofavoiding bargaining failure and in terms of how muchindividuals earned by randomly drawing 1000 samples(with replacement) of k group members from the N sub-jects who participated in that experimental condition Forinstance in the democracydemocracy condition we ran-domly drew a set of nine proposers out of all the subjectsin the pool assigned to this condition and another set ofnine responders assigned to this condition We wouldthen measure whether bargaining succeeded or failed bywhether proposers collectively made an offer greater thanor equal to what the responders collectively demandedTo obtain standard errors for this estimator we used thenonparametric bootstrap running our procedure over3000 samples of the data

Results

We began by confirming that we could replicate past stud-ies of one-on-one bargaining between individuals in theultimatum game using the 232 subjects in our baselinecondition Similar to past studies our results show thatindividuals avoid bargaining failure approximately 75percent of the time (Camerer 2003) Specifically individu-als in this baseline condition of our experiment avoidedbargaining failure 765 percent of the time (95 percentconfidence interval [CI] [070 to 083])

Next we examined each of our main experimental con-ditions Figure 1 shows the estimated mean outcome ineach condition with bootstrapped standard errors from3000 subsamples of the data Moving from left to rightalong the X-axis are the four experimental conditionsCondition 1 is labeled autocracyautocracy condition 2 islabeled autocracydemocracy condition 3 is labeleddemocracyautocracy and condition 4 is labeleddemocracydemocracy

In Panel A of Figure 1 the Y-axis represents the percent-age of times bargaining succeeded ormdashin our analogymdashthe percentage of time subjects avoided the costly rever-sion outcome of war In Panel B the Y-axis represents theaverage earnings of proposers in each condition We in-vestigated playersrsquo earnings to distinguish our hypothesisthat groups in situations of ultimatum bargaining are col-lectively wise (by making more efficient proposals thatmore closely predict the reservation price of their oppo-nent) from the alternative possibility that groups exhibit alower rejection rate simply because they bargain in a morerisk-averse and inefficient way (with groups consistently of-fering more generous proposals in order to secure apeaceful settlement at any cost)

Beginning with the autocracyautocracy condition atthe far left of Panel A our results show that small groupsof three do no better with respect to the percentage oftimes bargaining succeeds compared to the baselinecondition described above in which individuals faced

Table 1 Four ultimatum bargaining experimental conditions

Side BAutocracy(3 Responders)

Democracy(9 Responders)

Side A Autocracy(3 Proposers)

Condition 1(Nfrac14 124 110)

Democracy(9 Proposers)

15See the supplementary appendix for further details on our recruitmentprocedure

16The size of the pie is always shown as $040 We used numerical exam-ples in the instructions to illustrate how the $040 would be divided as a resultof the proposal but the hypothetical payoffs used were drawn randomly so asnot to systematically bias playersrsquo strategies

individuals and bargaining succeeded roughly 75 percentof the time (761 95 percent CI [070 to 083])Consistent with the wisdom-of-the-crowds hypothesis how-ever we find that mixed dyads in which even one side rep-resents a large group of nine perform significantly betterin situations of ultimatum bargaining compared to dyadswith two small groups Autocracydemocracy dyads avoidconflict 873 percent of the time (95 percent CI [079 to096]) and democracyautocracy dyads avoid conflict 904percent of the time (95 percent CI [085 to 096]) Alsoconsistent with our theory democratic dyads perform thebest avoiding bargaining failure 967 percent of the time(95 percent CI [093 to 100]) In other words ultimatumbargaining between democracies rarely if ever fails

In Panel B we investigate earnings across the four con-ditions for the reasons outlined above These findingsmirror the result in Panel A with mixed dyads earning sig-nificantly more than autocratic dyads and democraticdyads earning more than even mixed dyads on averageDemocratic dyads earned on average 194 cents comparedto autocratic dyads in which individuals earn 159 centson average This suggests that proposals of large groupsare better calibrated to the demands of responders whichappears consistent with the hypothesis that democraciesare ldquowiserrdquo and also appears consistent with the finding inobservational studies that democracies do not performworse on average in crisis bargaining situations (Bueno deMesquita et al 1999) These higher earnings do notemerge because larger groups on average make substan-tially more generous offers Instead higher earningsemerge because aggregation averages out overly aggressiveoffers from individuals that would normally trigger bar-gaining failure and also offers that would be far toogenerous17

Why Is the Result Not Strictly Dyadic

The results above clearly replicate the important dyadicaspect of the democratic peace finding democracies

almost never fight each other However it is not obviousfrom Figure 1 whether our results replicate the more con-troversial finding that democracies are no less war proneoverall which implies that mixed dyads should be morewar prone than even autocratic dyads (Gleditsch andHegre 1997)18 In the supplementary appendix we dis-cuss two potential reasons why decision aggregation mayappear to have a monotonic effect in our experiment buta dyadic effect in the real world First mixed dyads mayhave an overall higher rate of dispute initiation that fullyoffsets the benefits of aggregation within a crisis Secondfactors not present in our experiment could lead the dif-ferent types in mixed dyads to have systematically biasedviews about how to bargain with another type and thiscould cause aggregation to actually produce worse bar-gaining outcomes in mixed dyads

Additional Tests

A second aspect of the wisdom-of-the-crowds hypothesisposits that crowds of individuals can even outperform ex-pert individuals in predictive tasks (Tetlock 2005) Abovewe discussed the possibility that democracies by aggregat-ing predictions from a larger number of decision-makersmay outperform even relatively skilled experts in bargain-ing scenarios that mimic key aspects of war bargaining Toinvestigate this we compared the performance of demo-cratic dyads in our experiment to three types of individu-als The first type is inexperienced individuals These areindividuals from our baseline condition who in a post-experiment survey reported that they had never played agame similar to our ultimatum game scenario19 The sec-ond type of individuals that we compared to democraticdyads represented experienced individuals who reported thatthey had played a similar game in the past (50 percent ofthe subjects in our baseline condition) The third type ofindividuals represented international policy elites This sampleincluded 102 international foreign policy elites recruited toplay an ultimatum game in a previous study by LeVeck et al

Figure 1 Bargaining failure and earnings across treatments

17The median offer from autocracies and democracies was both twentyand the mean was both seventeen If we condition on bargaining success de-mocracies and autocracies earn roughly the same amount in our experimentThis replicates other findings in the literature which suggest that democraciesdo not do appreciably worse in the bargains they successfully conclude shortof war (Bueno de Mesquita et al 1999)

18See Gleditsch and Hegre (1997) for a summary of the controversy overand mixed results for a monadic democratic peace

19Specifically inexperienced individuals did not answer ldquoyesrdquo to the follow-ing post-experiment question have you ever played a similar game where oneplayer proposes how to split a monetary prize and another player decideswhether to accept or reject the offer

(2014) These elites had significant real-world experience inactual international bargaining

Figure 2 compares the results of each type of individualagainst the performance of democratic dyads along thesame two dimensions Panel A shows the percentage oftime bargaining succeeded and Panel B shows the aver-age earnings of proposers in each condition20 Beginningin Panel A the results show that experienced individualsand international policy elites avoid bargaining failuremore than inexperienced individuals However this differ-ence failed to reach statistical significance at conventionallevels At the same time the results in Panel B show thatboth groups of expert individuals earn significantly morethan inexperienced individuals Meanwhile the results inboth panels strongly confirm the wisdom-of-the-crowds hy-pothesis Democratic dyads comprised of both experi-enced and inexperienced individuals dramatically out-perform even experts on both measures These results areconsistent with the findings of Tetlock (2005)

Finally we investigated a third aspect to determinewhich factors are actually driving the observed behavior inour experiment We do this because our aggregationmechanism may actually aggregate two distinct factors be-havioral norms and knowledge about what the other sidersquosminimum acceptable offer will be (Camerer 2003)Because our theory focuses on the second elementbeliefs we isolated that component to see if our main hy-pothesis holds Supplementary Appendix Figures A2 andA3 shows an even stronger dyadic effect when we isolatethe influence of beliefsmdashmeaning larger groups performparticularly well at guessing the threshold when bargain-ing with larger groups

External Validity

A common concern with the use of laboratory experi-ments in political science has to do with the use of under-graduates as a convenience sample The concern is thatundergraduates are neither representative of elitedecision-makers nor the general population from whichthey are drawn As Renshon (2015) notes such concernsare neither new nor unique to political science as

psychologists have long worried about the fieldrsquos relianceon college students in drawing conclusions that may notbe externally valid Renshon reviews a series of productiveresponses to these concerns including attempts to repli-cate findings across different populations with mixedresults In some studies professionalsexperts behavedsimilarly to nonprofessionalsnonexperts (Glaser Langerand Weber 2005) while in other cases the results substan-tially differed (Tyszka and Zielonka 2002 Mintz Reddand Vedlitz 2006) For example Hafner-Burton et al(2014) and LeVeck et al (2014) found interesting differ-ences between elites and student subjects across a varietyof strategic games including the ultimatum game

In many ways we address this potential threat betterthan even the nascent experimental literature on crisisbargaining In our experiment we compare the behaviorof individuals and groups in situations of ultimatum bar-gaining drawn from two different samples subjects drawnfrom a more general population on Amazon Turk and asample of political elites We find important differencesand surprising similarities across the different samples dis-cussed above

A second and related concern is that subjectsmdashboth stu-dents and elites alikemdashwould behave differently in real-life situations when compared to the lab This could bebecause subjects are not fully motivated to engage in theexperiment or because the experiment omitted factors inthe real world that may cause them to behave differently(similar to omitted variable bias when making inferencesin observational studies) The latter is a constant risk withthe use of experiments across all fields For example inthe biological sciences scientists debate whether effectsfrom ldquotest tuberdquo experiments conducted in vitro are likelyto generalize to highly complex living organisms in vivoWhen studying decision-making processes it is possiblethat important factors like experience high stakes andemotions are relevant in the real world even if not cap-tured in the setup of the experiment

In the case of our experiment there are at least twosimplifications that may induce different results in the lab-oratory when compared to the real world First a reason-able case can be made that the voting mechanism in theexperiment does not capture the intricacies of foreignpolicy decision-making in a democracy or any other stateThis is true Our voting rulemdashwhich calculates a grouprsquos

Figure 2 Group vs expert performance

20The elite sample from LeVeck et al (2014) played for a larger monetaryprize We have therefore rescaled earnings to match the prize used in ourstudy

proposal as the median proposal submitted in thegroupmdashpurposely abstracts from factors like coalitionalbargaining within states democratic deliberation and theinfluence of elite opinion leaders We do not assert thatall decision-makers are completely independent in anyreal-world decision but rather that to the extent individ-ual inputs are at least somewhat independent our treat-ment manipulation captures this independent compo-nent A second simplification is that the bargainingscenario that our subjects in the experiment face is muchsimpler and lower stakes than the real-world bargainingscenarios faced by leaders21 This is also true In closelymatching our experiment to the assumptions of the bar-gaining model of war we abstract from the multidimen-sional nature and high stakes of international crisisbargaining

However compelling evidence exists to suggest thatlarger groups may still outperform individuals even if thesituation becomes more complex22 or if the stakes areraised in the domain of foreign policy On the one handaverage accuracy in the model of collective accuracy out-lined is likely low among the general population withrespect to designing a rocket On the other hand in thedomain of foreign policy Tetlock (2005) has shownthatmdashassuming nonspecialists have some baselineknowledge of foreign affairsmdashldquowe reach the point ofdiminishing marginal predictive returns for knowledgedisconcertingly quicklyrdquo (Tetlock 2005 59) in predictingwhat will happen in a particular region That is averageaccuracy of foreign affairs is typically at a sufficient levelamong ldquoattentive readers of the New York Times in lsquoread-ingrsquo emerging situationsrdquo (Tetlock 2005 233) to expectthat even individual specialists are not significantly morereliable than groups of nonspecialists We expect thatwhen the wisdom of the crowds is harnessed in the realworld that the larger more diverse group of indepen-dently deciding individuals generally has some baselineaccuracy Moreover laboratory evidence suggests thathigher stakes have a fairly minimal effect on behavior inthe ultimatum game (Camerer and Hogarth 1999)

A third concern may be that a ldquofairrdquo or ldquoacceptablerdquo of-fer is much clearer in the ultimatum gamemdashnamely afifty-fifty splitmdashthan in the real world where a fair oracceptable division can be much more ambiguous andcontingent on factors that nonexperts know little about(history power regime type etc) If true the structure ofthe ultimatum game may bias against the importance ofexpertise by providing a clearer focal point around whichthe offers of nonexpert proposers and responders canmore easily converge when compared to the real world

This concern is certainly possible and it is an interestingarea for future research However we note that even inour relatively simple and controlled experiment experi-enced individuals actually do perform better than inexperi-enced individuals suggesting that the ultimatum game isnot so simple that expertise is rendered meaninglessInstead our results confirm that individual expertise helpsbut they show that aggregation helps even more This find-ing mimics related research showing that larger and more

diverse groups of nonexperts can outperform expertseven on complex issues related to foreign policy (Tetlock2005) Furthermore while the norm of fifty-fifty divisionsis well-known there is good reason to suspect that it is notthe only widely known norm relevant for crisis bargainingFor example work by Tomz and Weeks (2013) shows thatcitizens in different democratic statesmdashthe United Statesand the United Kingdommdashshare many norms that are rel-evant to reducing the risk of conflict between democraciesIt is possible that processes of aggregation could help dis-till which of these norms are most relevant to a particularcrisis and further reduce the chance of bargaining failureand war between democratic states

Therefore despite the fact that each of these three con-cerns is reasonable we believe the level of realism in ourexperiment is appropriate for the specific hypotheses weseek to test In general we agree with McDermott thatmdashrather than emerging a property of any individual experi-mentmdashldquoexternal validity follows as replications acrosstime and populations seek to delineate the extent towhich conclusions can generalizerdquo (2011 28) Futurestudies can and should identify theoretically relevantconditions along which our experiment differs from thereal world and testmdashas part of a broader research pro-grammdashwhether the inclusion of these factors moderatesthe effects identified here

Conclusion

The evidence gathered from our experiments is ofcourse preliminary There remains much more work thatcan be done to develop and evaluate our core argumentSuch work might include further studies that systemati-cally manipulate how information is distributed acrossindividuals the identity of bargainers as well as the pre-cise mechanism by which information is aggregatedOther studies may look at observational data to see howaggregated signals (LeVeck and Narang 2016 Narang andLeVeck 2011) such as market movements or polls actu-ally influence democratic decision-making Finally de-mocracies and autocracies vary systematically in the cali-ber of various aggregation mechanismsmdashsuch as thedepth of markets or how informed their publics areMeasures of this variation might be linked to measures ofwar-bargaining outcomes

However we believe the findings presented here aresignificant In bargaining scenarios that mimic key aspectsof war bargaining aggregated offers from larger groupssystematically outperform the offers made by smallergroups and individuals Furthermore part of the informa-tion aggregated appears to involve individualsrsquo knowledgeof what they themselves would do if placed in theiropponentsrsquo shoes This may help them actually predictthe responses of their opponents Thus the democraticpeace may partially arise because democracies aggregatesignals from diverse individuals which increases the chan-ces of some of those individuals matching the characteris-tics of decision-makers in the other statemdashand thereforeanticipating the strategies and responses of those deci-sion-makers

These results notwithstanding we think it important toemphasize an important limitation to our inferences Tobe clear we do not claim that democracies always makebetter decisions in every situation Indeed we see numer-ous cases in which democratic decision-makers committedgrave errors in crisis bargaining For instance it is welldocumented that the United States made several errors in

21For example in a multidimensional policy space aggregating diversepreferences across multiple actors may result in a single foreign policy pro-posal that is ideologically incoherentmdashwith more hawkish measures on somedimensions and more dovish measures on others

22In fact results from Hong and Page (2004) suggest the oppositeGroups of diverse individuals have a particular advantage in more complexdecisions

estimating the capabilities and resolve of Saddam Husseinin the run-up to the Iraq War in 2003 (Gordon andTrainor 2006 Lake 2010) Such examples suggest thatmdasheven if democracies can make collectively wiser decisionscompared to nondemocracies on averagemdashthey are cer-tainly not immune from making decision errors in partic-ular cases However we note that in many famous casesof miscalculation by democracies the actual reason forthe miscalculation appears to stem from restricteddecision-making where a narrow group of similar-mindedleaders engaged in an echo chamber (Janis 1972) and ef-fectively excluded the diverse views of numerous individu-als (Packer 2005 50ndash60 Daalder and Lindsay 2003 46ndash47Mann 2004 351ndash53)

More broadly our results may suggest policy implica-tions beyond the domain of crisis bargaining includingsituations of international cooperation on issues likehealth development or the global environmentAlthough there is generally broad support for greater eco-nomic development global health global peace and acleaner environment uncertainty over the costs and bene-fits of cooperation can often lead citizens to hold diverseviews on whether and how to cooperate For exampleRomano (2011) shows that most Americans assume thatdevelopmental aid accounts for 27 percent of the nationalbudget when it is actually less than 1 percent (Narang2013 Narang 2016 Narang and Stanton 2017) Similarlyindividuals appear to hold diverse opinions about the riskof global health epidemics (Leach et al 2010) the processof collective security (Dellmuth 2016) and global climatechange (Keohane and Ostrom 1995 Ostrom 2009Stevenson 2013 Stevenson and Dryzek 2014) However aswe show with bargaining it may be possible that in situa-tions of international cooperation the errors made byone decision-maker may cancel out the error made by an-other and produce a collectively wise policy decisionacross domains (Landemore 2012a 2012b 2013)

Supplementary Information

Supplementary Information is available at httpsdataverseharvardedu and the International Studies Quarterlydata archive

References

ALLISON GRAHAM 1969 ldquoConceptual Models and the Cuban Missile

Crisisrdquo American Political Science Review 63 (3) 689ndash718ALDRICH JOHN H 1999 ldquoPolitical Parties in a Critical Erardquo American

Politics Quarterly 27 (1) 9ndash32ANDREONI JAMES AND EMILY BLANCHARD 2006 ldquoTesting Subgame

Perfection Apart from Fairness in Ultimatum Gamesrdquo Experimental

Economics 9 (4) 307ndash21BONE JOHN JOHN HEY AND JOHN SUCKLING 1999 ldquoAre Groups More (or

Less) Consistent Than Individualsrdquo Journal of Risk and Uncertainty

18 (1) 63ndash81BORNSTEIN GARY AND ILAN YANIV 1998 ldquoIndividual and Group Behavior in

the Ultimatum Game Are Groups More lsquoRationalrsquo Playersrdquo

Experimental Economics 1 (1) 101ndash8BRAUT-HEGGHAMMER MALFRID 2016 Unclear Physics Why Iraq and Libya

Failed to Build Nuclear Weapons Ithaca NY Cornell University PressBREMER STUART A 1992 ldquoDangerous Dyads Conditions Affecting the

Likelihood of Interstate War 1816ndash1965rdquo Journal of Conflict

Resolution 36 (2) 309ndash41mdashmdashmdash 1993 ldquoDemocracy and Militarized Interstate Conflict 1816ndash1965rdquo

International Interactions 18 (3) 231ndash49

BUENO DE MESQUITA BRUCE JAMES D MORROW RANDOLPH M SIVERSON AND

ALASTAIR SMITH 1999 ldquoAn Institutional Explanation of theDemocratic Peacerdquo American Political Science Review 93 (4) 791ndash807

CAMERER COLIN F 2003 Behavioral Game Theory Experiments in Strategic

Interaction Princeton NJ Princeton University PressCAMERER COLIN F AND ROBIN M HOGARTH 1999 ldquoThe Effects of

Financial Incentives in Experiments A Review and Capital-Labor-Production Frameworkrdquo Journal of Risk and Uncertainty 19 (1ndash3)7ndash42

CASON TIMOTHY N AND VAI-LAM MUI 1997 ldquoA Laboratory Study of GroupPolarisation in the Team Dictator Gamerdquo Economic Journal 107(444) 1465ndash83

CHAN STEVE 1984 ldquoMirror Mirror on the Wall Are the FreerCountries More Pacificrdquo Journal of Conflict Resolution 28 (4)617ndash48

COX JAMES C AND STEPHEN C HAYNE 2006 ldquoBarking up the Right TreeAre Small Groups Rational Agentsrdquo Experimental Economics 9 (3)209ndash222

DAALDER IVO H AND JAMES M LINDSAY 2003 America Unbound The Bush

Revolution in Foreign Policy Washington Brookings Institution PressDAFOE ALLAN 2011 ldquoStatistical Critiques of the Democratic Peace

Caveat Emptorrdquo American Journal of Political Science 55 (2) 247ndash62DAVIS JAMES H 1992 ldquoSome Compelling Intuitions about Group

Consensus Decisions Theoretical and Empirical Research andInterpersonal Aggregation Phenomena Selected Examples 1950ndash1990rdquo Organizational Behavior and Human Decision Processes 52 (1)3ndash38

DELLMUTH LISA M 2016 ldquoThe Knowledge Gap in World PoliticsAssessing the Sources of Citizen Awareness of the United NationsSecurity Councilrdquo Review of International Studies 42 (4) 673ndash700

DENARDO JAMES 1995 The Amateur Strategist Intuitive Deterrence Theories

and the Politics of the Nuclear Arms Race Cambridge CambridgeUniversity Press

DIXON WILLIAM J 1993 ldquoDemocracy and the Management ofInternational Conflictrdquo Journal of Conflict Resolution 37 (1) 42ndash68

mdashmdashmdash 1994 ldquoDemocracy and the Peaceful Settlement of InternationalConflictrdquo American Political Science Review 88 (1) 1ndash17

DOBBS MICHAEL 2008 One Minute to Midnight Kennedy Khrushchev and

Castro on the Brink of Nuclear War New York VintageDOWNS ANTHONY 1957 An Economic Theory of Democracy New York

Columbia University PressDOYLE MICHAEL W 1997 Ways of War and Peace Realism Liberalism and

Socialism 276 24ndash25 New York NortonELBITTAR ALEXANDER ANDREI GOMBERG AND LAURA SOUR 2011 ldquoGroup

Decision-Making and Voting in Ultimatum Bargaining AnExperimental Studyrdquo BE Journal of Economic Analysis amp Policy

11 (1) 1ndash31FEARON JAMES D 1994 ldquoDomestic Political Audiences and the Escalation

of International Disputesrdquo American Political Science Review 88 (3)577ndash92

mdashmdashmdash 1995 ldquoRationalist Explanations for Warrdquo International Organization

49 (3) 379ndash414FURSENKO ALEKSANDR AND TIMOTHY NAFTALI 1998 ldquoOne Hell of a Gamble

Khrushchev Castro and Kennedy 1958ndash1964 New York WW Nortonamp Company

mdashmdashmdash 2007 Khrushchevrsquos Cold War The Inside Story of an American

Adversary New York WW Norton amp CompanyGARTZKE ERIK 1998 ldquoKant We All Just Get Along Opportunity

Willingness and the Origins of the Democratic Peacerdquo American

Journal of Political Science 42 (1) 1ndash27mdashmdashmdash 2000 ldquoPreferences and the Democratic Peacerdquo International Studies

Quarterly 44 (2) 191ndash212mdashmdashmdash 2007 ldquoThe Capitalist Peacerdquo American Journal of Political Science 51

(1) 166ndash91GEDDES BARBARA ldquoAuthoritarian Breakdown Empirical Test of a Game-

Theoretic Argumentrdquo Paper presented at the 95th Annual Meetingof the American Political Science Association Atlanta GASeptember 1999

mdashmdashmdash 2003 Paradigms and Sandcastles Theory Building and Research Design

in Comparative Politics Ann Arbor University of Michigan PressGLASER MARKUS LANGER THOMAS AND WEBER MARTIN 2005

ldquoOverconfidence of Professionals and Lay Men IndividualDifferences Within and Between Tasksrdquo Working Paper Accessed

January 20 2015 Available at httpsideasrepecorgpxrssfbmaa05-25html

GLEDITSCH NILS PETTER AND HAVARD HEGRE 1997 ldquoPeace and DemocracyThree Levels of Analysisrdquo Journal of Conflict Resolution 41 (2)283ndash310

GLEDITSCH KRISTIAN S AND MICHAEL D WARD 1997 ldquoDouble Take AReexamination of Democracy and Autocracy in Modern PolitiesrdquoJournal of Conflict Resolution 41 (3) 361ndash83

GORDON MICHAEL R AND BERNARD E TRAINOR 2006 Cobra II The InsideStory of the Invasion and Occupation of Iraq New York Pantheon

GORMAN MARTIN J AND ALEXANDER KRONGARD 2005 A Goldwater-Nichols Actfor the US Government Institutionalizing the Interagency ProcessWashington Defense Intelligence Agency

GeuroUTH WERNER ROLF SCHMITTBERGER AND BERND SCHWARZE 1982 ldquoAnExperimental Analysis of Ultimatum Bargainingrdquo Journal ofEconomic Behavior amp Organization 3 (4) 367ndash88

HAFNER-BURTON EMILIE M BRAD L LEVECK DAVID G VICTOR AND JAMES HFOWLER 2014 ldquoDecision Maker Preferences for International LegalCooperationrdquo International Organization 68 (4) 845ndash76

HAFNER-BURTON EMILIE M BRAD L LEVECK AND DAVID G VICTOR 2017ldquoNo False Promises How the Prospect of Non-Compliance AffectsElite Preferences for International Cooperationrdquo InternationalStudies Quarterly 61 (1) 136ndash149

HAFNER-BURTON EMILIE M D ALEX HUGHES AND DAVID G VICTOR 2013ldquoThe Cognitive Revolution and the Political Psychology of EliteDecision Makingrdquo Perspectives on Politics 11 (2) 368ndash86

HENISZ WITOLD J 2000 ldquoThe Institutional Environment for EconomicGrowthrdquo Economics amp Politics 12 (1) 1ndash31

HENRICH JOSEPH ROBERT BOYD SAMUEL BOWLES COLIN CAMERER ERNST

FEHR HERBERT GINTIS AND RICHARD MCELREATH 2001 ldquoIn Search ofHomo Economicus Behavioral Experiments in 15 Small-ScaleSocietiesrdquo American Economic Review 91 (2) 73ndash78

HERMANN MARGARET G AND CHARLES F HERMANN 1989 ldquoWho MakesForeign Policy Decisions and How An Empirical InquiryrdquoInternational Studies Quarterly 33 (4) 361ndash87

HONG LU AND SCOTT PAGE 2004 ldquoGroups of Diverse Problem Solverscan Outperform Groups of High-Ability Problem SolversrdquoProceedings of the National Academy of Sciences of the United States ofAmerica 101 (46) 16385ndash89

mdashmdashmdash 2009 ldquoInterpreted and Generated Signalsrdquo Journal of EconomicTheory 144 (5) 2174ndash96

mdashmdashmdash 2012 ldquoThe Micro-Foundations of Collective Wisdomrdquo In CollectiveWisdom Principles and Mechanisms edited by Helene Landemoreand Jon Elster 56ndash71

HOROWITZ MICHAEL C AND NEIL NARANG 2014 ldquoPoor Manrsquos atomicbomb exploring the relationship between ldquoweapons of massdestructionrdquo Journal of Conflict Resolution 58(3) 509-535

JANIS IRVING L 1972 Victims of Groupthink a Psychological Study of Foreign-Policy Decisions and Fiascoes Oxford Houghton Mifflin

KACOWICZ ARIE M 1995 ldquoExplaining Zones of Peace Democracies asSatisfied Powersrdquo Journal of Peace Research 32 (3) 265ndash76

KANT IMMANUEL (1795) 1969 Perpetual Peace Reprint New YorkColumbia University Press

KERR NORBERT L ROBERT J MACCOUN AND GEOFFREY P KRAMER 1996ldquoBias in judgment Comparing Individuals and GroupsrdquoPsychological Review 103 (4) 687

KEOHANE ROBERT O AND ELINOR OSTROM 1995 ldquoIntroductionrdquo In LocalCommons and Global Interdependence edited by Robert O Keohaneand Elinor Ostrom 1ndash26 London Sage

KLEINDORFER PAUL R HOWARD C KUNREUTHER AND PAUL H SCHOEMAKER1993 Decision Science An Integrative Perspective New YorkCambridge University Press

KREHBIEL KEITH 1998 Pivotal Politics A Theory of US Lawmaking ChicagoUniversity of Chicago Press

LAKE DAVID A 1992 ldquoPowerful Pacifists Democratic States and WarrdquoAmerican Political Science Review 86 (1) 24ndash37

mdashmdashmdash 2010 ldquoTwo Cheers for Bargaining Theory Assessing RationalistExplanations of the Iraq Warrdquo International Security 35 (3) 7ndash52

LANDEMORE HELENE 2012a ldquoCollective Wisdom Old and Newrdquo InCollective Wisdom edited by Helene Landemore and Jon Elster1ndash20 Cambridge Cambridge University Press

mdashmdashmdash 2012b ldquoWhy the Many Are Smarter Than the Few and Why ItMattersrdquo Journal of Public Deliberation 8 (1) Article 7 httpwwwpublicdeliberationnetjpdvol8iss1art7 hc_locationfrac14ufi

mdashmdashmdash 2013 Democratic Reason Politics Collective Intelligence and the Rule ofthe Many Princeton NJ Princeton University Press

LEACH MELISSA IAN SCOONE AND ANDREW STIRLING 2010 ldquoGoverningEpidemics in an Age of Complexity Narratives Politics andPathways to Sustainabilityrdquo Global Environmental Change 20 (3)369ndash77

LEMKE DOUGLAS AND WILLIAM REED 1996 ldquoRegime Types and Status QuoEvaluations Power Transition Theory and the Democratic PeacerdquoInternational Interactions 22 (2) 143ndash64

LENZ GABRIEL S 2012 Follow the Leader Chicago University of ChicagoPress

LEVECK BRAD L D ALEX HUGHES JAMES H FOWLER EMILIE HAFNER-BURTON AND DAVID G VICTOR 2014 ldquoThe Role of Self-Interest inElite Bargainingrdquo Proceedings of the National Academy of Sciences 111(52) 18536ndash41

LEVECK BRAD L AND NEIL NARANG 2016 ldquoHow International ReputationMatters Revisiting Alliance Violations in Contextrdquo InternationalInteractions 43 (5) 1ndash25

LEVENDUSKY MATTHEW 2009 The Partisan Sort How Liberals becameDemocrats and Conservatives became Republicans Chicago Universityof Chicago Press

LUPIA ARTHUR AND MATTHEW D MCCUBBINS 1994 ldquoLearning FromOversight Fire Alarms and Police Patrols Reconstructedrdquo Journal ofLaw Economics and Organization 10 (1) 96

mdashmdashmdash 2003 The Democratic Dilemma Can Citizens Learn What They Need toKnow Cambridge Cambridge University Press

MCDERMOTT ROSE 2011 Internal and External Validity In Handbook ofExperimental Political Science edited by James N Druckman DonaldP Green James H Kuklinski and Arthur Lupia 27ndash41 New YorkCambridge University Press

MAOZ ZEEV AND BRUCE RUSSETT 1993 ldquoNormative and Structural Causesof Democratic Peace 1946ndash1986rdquo American Political Science Review87 (3) 624ndash38

MAOZ ZEEV AND NASRIN ABDOLALI 1989 ldquoRegime Types and InternationalConflict 1816ndash1976rdquo Journal of Conflict Resolution 33 (1) 3ndash36

MANN JAMES 2004 Rise of the Vulcans The History of Bushrsquos War CabinetNew York Penguin

MARCELLA GABRIEL 2004 ldquoNational Security and the Interagency ProcessrdquoUS Army War College Guide to National Security Policy and Strategy 239260 httpswwwresearchgatenetprofileChas_Freemanpublication265101496_CHAPTER_3_NATIONAL_SECURITY_AND_THE_INTERAGENCY_PROCESSlinks56cc4ada08ae5488f0dcf2a9pdf

MEHTA RUPAL AND NEIL NARANG 2017 ldquoDo Nuclear Umbrellas Increasethe Diplomatic Influence of Client States A Theory and EmpiricalEvidencerdquo Presented at the Department of Defense WashingtonDC

MINTZ ALEX STEVEN B REDD AND ARNOLD VEDLITZ 2006 Can WeGeneralize from Student Experiments to the Real World inPolitical Science Military Affairs and IR Journal of ConflictResolution 50 (5)757ndash76

MULLEN BRIAN TARA ANTHONY EDUARDO SALAS AND JAMES E DRISKELL 1994ldquoGroup Cohesiveness and Quality of Decision Making AnIntegration of Tests of the Groupthink Hypothesisrdquo Small GroupResearch 25 (2) 189ndash204

mdashmdashmdash 2016 Forgotten Conflicts Need versus Political Priority in theAllocation of Humanitarian Aid across Conflict Areas InternationalInteractions 42(2)189ndash216

NARANG NEIL 2013 Biting the Hand that Feeds An OrganizationalTheory Explaining Attacks Against Aid Workers in Civil ConflictPaper presented at Princeton University

mdashmdashmdash 2015 ldquoAssisting Uncertainty How Humanitarian Aid CanInadvertently Prolong Civil Warrdquo International Studies Quarterly 59(1) 184ndash95

mdashmdashmdash 2014 ldquoHumanitarian Assistance and the Duration of Peace afterCivil Warrdquo Journal of Politics 76 (2) 446ndash60

mdashmdashmdash 2017 ldquoWhen Nuclear Umbrellas Work Signaling Credibility inSecurity Commitments through Alliance Designrdquo Presented at theUniversity of California Conference on International CooperationSanta Barbara

NARANG NEIL AND BRAD L LEVECK ldquoSecuritizing International SecurityHow Unreliable Allies affect Alliance Portfoliosrdquo Presented at theAmerican Political Science Association Conference Seattle WA2011

NARANG NEIL AND JESSICA A STANTON 2017 ldquoA Strategic Logic ofAttacking Aid Workers Evidence from Violence in AfghanistanrdquoInternational Studies Quarterly 61(1) 38ndash51

NARANG NEIL AND RUPAL N MEHTA 2017 ldquoThe UnforeseenConsequences of Extended Deterrence Moral Hazard in a NuclearClient Staterdquo Journal of Conflict Resolution

ONEAL JOHN R AND BRUCE RUSSETT 1999 ldquoThe Kantian Peace ThePacific Benefits of Democracy Interdependence and InternationalOrganizations 1885ndash1992rdquo World Politics 52 (1) 1ndash37

mdashmdashmdash 2001 ldquoClear and Clean The Fixed Effects of the Liberal PeacerdquoInternational Organization 55 (2) 469ndash85

OSTROM ELINOR 2009 ldquoA general framework for analyzing sustainabilityof social-ecological systemsrdquo Science 325 (5939) 419ndash422

PACKER GEORGE 2005 The Assassinsrsquo Gate America in Iraq New YorkFarrar Straus and Giroux

PAGE SCOTT E 2008 The Difference How the Power of Diversity Creates BetterGroups Firms Schools and Societies (New Edition) Princeton NJPrinceton University Press

POPKIN SAMUEL L 1991 The Reasoning Voter Chicago University ofChicago Press

POWELL ROBERT 1999 In the shadow of power States and strategies in interna-tional politics Princeton University Press

PUNCOCHAR JUDITH M AND PAUL W FOX 2004 ldquoConfidence in Individualand Group Decision Making Whenrdquo Two Headsrdquo Are Worse ThanOnerdquo Journal of Educational Psychology 96 (3) 582

RAACH GEORGE T AND ILANA KASS 1995 National Power and the InteragencyProcess Washington National Defense University

RAND DAVID G 2012 ldquoThe Promise of Mechanical Turk How OnlineLabor Markets Can Help Theorists Run Behavioral ExperimentsrdquoJournal of Theoretical Biology 299 (21) 172ndash79

RAND DAVID G CORINA E TARNITA HISASHI OHTSUKI AND MARTIN NOWAK2013 ldquoEvolution of Fairness in the One-Shot AnonymousUltimatum Gamerdquo Proceedings of the National Academy of Sciences 110(7) 2581ndash6

REITER DAN AND ALLAN C STAM 2002 Democracies at War Princeton NJPrinceton University Press

RENSHON JONATHAN 2015 ldquoLosing Face and Sinking Costs ExperimentalEvidence on the Judgment of Political and Military LeadersrdquoInternational Organization 69 (3) 659ndash95

ROCKENBACH BETTINA ABDOLKARIM SADRIEH AND BARABARA MATHAUSCHEKldquoTeams Take the Better Risksrdquo Working paper University ofErfurt 2001

ROMANO ANDREW 2011 ldquoHow Ignorant Are Americansrdquo Newsweek March20 Accessed March 24 2016 httpwwwnewsweekcomhow-ignorant-are-americans-66053

ROSATO SEBASTIAN 2003 ldquoThe Flawed Logic of Democratic PeaceTheoryrdquo American Political Science Review 97 (4) 585ndash602

ROUSSEAU DAVID L CHRISTOPHER GELPI DAN REITER AND PAUL K HUTH1996 ldquoAssessing the Dyadic Nature of the Democratic Peace 1918ndash88rdquo American Political Science Review 90 (3) 512ndash33

RUSSETT BRUCE 1993 Grasping the Democratic Peace Principles for a Post-Cold War World Princeton NJ Princeton University Press

RUSSETT BRUCE JOHN R ONEAL AND DAVID R DAVIS 1998 ldquoThe Third Legof the Kantian Tripod for Peace International Organizations andMilitarized Disputes 1950ndash85rdquo International Organization 52 (3)441ndash67

SAUNDERS ELIZABETH NATHAN 2011 Leaders at War How Presidents ShapeMilitary Interventions Ithaca NY Cornell University Press

SCHULTZ KENNETH A 1998 ldquoDomestic Opposition and Signaling inInternational Crisesrdquo American Political Science Review 92 (4)829ndash44

mdashmdashmdash 1999 ldquoDo Democratic Institutions Constrain or InformContrasting Two Institutional Perspectives on Democracy andWarrdquo International Organization 53 (2) 233ndash266

mdashmdashmdash 2001 Democracy and Coercive Diplomacy Cambridge CambridgeUniversity Press

SLANTCHEV BRANISLAV L 2003 ldquoThe Principle of Convergence in WartimeNegotiationsrdquo American Political Science Review 97 (4) 621ndash632

SMALL MEL AND J DAVID SINGER 1976 ldquoThe War-Proneness ofDemocratic Regimes 1816ndash1965rdquo The Jerusalem Journal ofInternational Relations 1 (4) 50ndash69

SNIEZEK JANET A 1992 Groups Under Uncertainty An Examination ofConfidence in Group Decision Making Organizational Behavior andHuman Decision Processes 52(1) 124ndash155

STEVENSON HAYLEY 2013 Institutionalizing Unsustainability The Paradox ofGlobal Climate Governance Berkeley University of California Press

mdashmdashmdash 2016 ldquoThe Wisdom of the Many in Global Governance AnEpistemic-Democratic Defense of Diversity and InclusionrdquoInternational Studies Quarterly 60 (3) 400ndash12

STEVENSON HAYLEY AND JOHN S DRYZEK 2014 Democratizing Global ClimateGovernance Cambridge Cambridge University Press

SUROWIECKI JAMES 2005 The Wisdom of Crowds New York Random HouseLLC

TAJFEL HENRI AND JOHN C TURNER 1979 ldquoAn Integrative Theory ofIntergroup Conflictrdquo The Social Psychology of Intergroup Relations33(47) 74

TAUBMAN WILLIAM 2003 Khrushchev The Man and His Era New YorkWW Norton amp Company

TETLOCK PHILIP 2005 Expert Political Judgment How Good Is It How CanWe Know Princeton NJ Princeton University Press

THOMPSON WILLIAM R AND RICHARD TUCKER 1997 ldquoA Tale of TwoDemocratic Peace Critiquesrdquo Journal of Conflict Resolution 41 (3)428ndash54

TINGLEY DUSTIN H AND BARBARA F WALTER 2011 ldquoCan Cheap TalkDeter An Experimental Analysisrdquo Journal of Conflict Resolution 55(6) 996ndash1020

TINGLEY DUSTIN H AND STEPHANIE W WANG 2010 ldquoBelief Updating inSequential Games of Two-Sided Incomplete Information AnExperimental Study of a Crisis Bargaining Modelrdquo Quarterly Journalof Political Science 5 (3) 243ndash55

TOMZ MICHAEL 2007 ldquoDomestic Audience Costs in InternationalRelations An Experimental Approachrdquo International Organization61 (4) 821ndash40

TOMZ MICHAEL R AND JESSICA LP WEEKS 2013 ldquoPublic Opinion and theDemocratic Peacerdquo American Political Science Review 107 (4) 849ndash65

TSEBELIS GEORGE AND SEUNG-WHAN CHOI ldquoDemocratic Peace Revisited ItIs Veto Playersrdquo Presented at the American Political ScienceAssociation Conference in Toronto 2009

TYSZKA TADEUSZ AND PIOTR ZIELONKA 2002 Expert Judgments FinancialAnalysts Versus Weather Forecasters Journal of Psychology andFinancial Markets 3 (3)152ndash60 httpdxdoiorg101207S15327760JPFM0303_3

WEEDE ERICH 1984 ldquoDemocracy and War Involvementrdquo Journal ofConflict Resolution 28 (4) 649ndash64

mdashmdashmdash 1992 ldquoSome Simple Calculations on Democracy and WarInvolvementrdquo Journal of Peace Research 29 (4) 377ndash83

WEEKS JESSICA L 2008 ldquoAutocratic Audience Costs Regime Type andSignaling Resolverdquo International Organization 62 (1) 35ndash64

mdashmdashmdash 2012 ldquoStrongmen and Straw Men Authoritarian Regimes and theInitiation of International Conflictrdquo American Political Science Review106 (2) 326ndash47

mdashmdashmdash 2014 Dictators at War and Peace Ithaca NY Cornell UniversityPress

WOLFERS JUSTIN AND ERIC ZITZEWITZ 2009 ldquoUsing Markets to InformPolicy The Case of the Iraq Warrdquo Economica 76 (302) 225ndash50

ZALLER JOHN R 1992 The Nature and Origins of Mass Opinion (CambridgeStudies in Public Opinion and Political Psychology) CambridgeCambridge University Press

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

matched against a smaller group of individuals (ie in amixed dyad) or other similarly large groups The findingsimply that existing theories of the democratic peace that ap-peal to shared normative values accountability or transpar-ency may be correct but also incomplete as simply aggre-gating decision-makersrsquo bargaining choices through avoting institution replicates two key features of the demo-cratic peace finding in a controlled experimental setupdemocratic dyads avoid costly bargaining failure more thanautocratic or mixed dyads and democracies do no worsethan other regime types in terms of bargaining outcomes

Theory

The Wisdom of Crowds

In the opening anecdote of his popular book The Wisdom ofthe Crowds Surowiecki (2005) illustrates a classic example ofhow crowds may be wise At a 1906 county fair in PlymouthEngland British scientist Francis Galton came across aweight-judging competition in which members of a gatheringcrowd lined up to place wagers on the weight of a fat ox Thebest guess won the prize Seven hundred and eighty-seven di-verse individuals (including expert butchers and farmers andnonexpert clerks) tried their luck at guessing the oxrsquos weightin an attempt to win prizes When the contest was overGalton borrowed the tickets from the organization and ana-lyzed the guesses hoping to show that the average voter wascapable of very little Adding the contestantsrsquo estimates to-gether and calculating the mean Galton used this number torepresent the collective wisdom of the Plymouth crowd actingas if the crowd voted as a single person Given the mixture ofthe crowd which included relatively ldquosmartrdquo guesses fromexperts with relatively ldquodumbrdquo guesses from nonexpertsGalton undoubtedly expected the guesses would be way offThe crowd guessed the ox would weigh 1197 pounds The ac-tual weight of the ox was 1198 pounds In Surowieckirsquos wordsldquothe crowdrsquos judgment was essentially perfectrdquo (2005 xiii)

What Galton discovered in averaging the guesses of thePlymouth crowd was a phenomenon now reproduced inmultiple real-world and experimental settingsmdashthat undercertain conditions groups of independent decision-makerscan be remarkably smart even smarter than the smartestmembers within that group While it was certainly true thatthe ldquodumbestrdquo members of the Plymouth crowd performedconsiderably worse than the so-called ldquoexpertsrdquo as Galtonpredicted (each individual in the group was off by an aver-age of nearly fifty-five pounds with a standard deviation ofroughly sixty-two pounds) their guesses appeared wrong invery different ways Some individuals dramatically overesti-mated the weight of the ox and others dramatically underes-timated its weight In averaging a diverse set of individualguesses the errors canceled out and thus produced a collec-tively wise decision In other words even if most peoplewithin a group are not particularly well informed or rational(lacking the ability and desire to make sophisticated cost-benefit calculations) when those imperfect judgments areaggregated together our collective intelligence is oftentimessuperior to the smartest of decision-makers (Tetlock 2005)

The importance of this finding for studying the behav-ior of political and social groups was not lost on GaltonIn particular the analogy to a democracy where peopleof radically different abilities and interests each get onevote suggested itself immediately In Galtonrsquos wordsldquo[t]he average competitor was probably as well fitted formaking a just estimate of the dressed weight of the oxas an average voter is of judging the merits of most polit-ical issues on which he votesrdquo (Surowiecki 2005 xii)

Despite his own belief that power in society should be-long to a select few with the best qualities for breedingGalton later conceded that ldquothe result seems more cred-itable to the trustworthiness of a democratic judgmentthan might be expectedrdquo (Surowiecki 2005 xiii)

Not all crowds are wise however And over timemdashasresearchers examined the implications of Galtonrsquos findingsacross various social contextsmdashthey gradually refined a the-ory of collective intelligence to include certain key criteriaContemporary theorists emphasize that collective accuracydepends on a combination of both individual accuracy anddiversity Specifically collective accuracy can be character-ized by the simple mathematical identity below (Page 2008Hong and Page 2004 2009 2012)

Collective accuracy frac14 average accuracy thorn diversity

Average accuracy in this equation refers to the averagemagnitude of each individualrsquos error Diversity refers tohow different individual guesses are on average What thefirst term in this simple equation makes clear is thatcrowds must know something about the issue at hand Ifindividuals know nothing about an issue and are wildlywrong then the crowd will still tend toward incorrectdecisions as well After all rockets are designed bygroups of engineers not laypeople On the other handif a number of individuals do know something about theproblem at hand but are prone to making differenttypes of errors then aggregating their views can helpmake an accurate decision because different errors willcancel one another out As we discuss below it is plausi-ble that democratic decision-makers are both accurateand diverse enough to give democracies an advantage inforeign policy decision-making

In addition to these general rules scholars in the psy-chology literature have also identified a number of specificconditions under which groups are unlikely to performbetter (Cason and Mui 1997 Bone Hey and Suckling 1999Rockenbach Sadrieh and Barabara 2001 Cox and Hayne2006 Puncochar and Fox 2004 Kerr MacCoun andKramer 1996)2 For example worse decision-making mayemerge when designated leaders promote conformity andself-censorship which can lead to group-think (Sniezek1992 Kleindorfer Kunreuther and Schoemaker 1993Mullen et al 1994) Similarly problems can arise whengroups polarize the attitudinal judgments of their mem-bers (Davis 1992 Kerr MacCoun and Kramer 1996 Casonand Mui 1997) Importantly however many of these con-ditions do not apply in our experimental setup and thereare also good reasons to believe that democratic decision-making is less vulnerable to many of these harmful condi-tions We describe these reasons in detail below

The Wisdom of Crowds in Democracies VersusAutocracies

If a diverse group of independently deciding individualscan be collectively wisemdashand this may be behind some ofdemocraciesrsquo ability to formulate superior policy deci-sionsmdashit is surprising that more attention has not beenpaid to this particular democratic advantage in foreignpolicy decision-making3 Perhaps democracies by

2In the interest of space we review the results of these papers in the sup-plementary appendix

3One exception is an important study by Reiter and Stam (2002) who ap-ply a similar logic to a different empirical puzzle why democracies win the

An Experiment

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

An Experiment

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

An Experiment

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

An Experiment

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

An Experiment

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

Results

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

Additional Tests

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

External Validity

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

Conclusion

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

References

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

sqx040-FM1

sqx040-FN1

sqx040-FN2

sqx040-FN3

sqx040-FN4

sqx040-FN5

sqx040-FN6

sqx040-FN7

sqx040-FN8

sqx040-FN9

sqx040-FN10

sqx040-FN11

sqx040-FN12

sqx040-FN13

sqx040-FN14

sqx040-FN15

sqx040-FN16

sqx040-FN17

sqx040-FN18

sqx040-FN19

sqx040-FN20

sqx040-FN21

sqx040-FN22

Documents

The Democratic Peace and the Wisdom of Crowdsfaculty.ucmerced.edu/...democratic_peace_and_the_wisdom_of_crowds.pdfThe Democratic Peace and the Wisdom of Crowds BRAD L. LEVECK University