Elster Analytic Narratives

8/12/2019 Elster Analytic Narratives

http://slidepdf.com/reader/full/elster-analytic-narratives 1/12

Review: Rational Choice History: A Case of Excessive AmbitionAuthor(s): Jon ElsterSource: The American Political Science Review, Vol. 94, No. 3 (Sep., 2000), pp. 685-695Published by: American Political Science AssociationStable URL: http://www.jstor.org/stable/2585842 .Accessed: 25/04/2011 14:49

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=apsa . .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access toThe American Political Science Review.

http://www.jstor.org

http://www.jstor.org/action/showPublisher?publisherCode=apsa

http://www.jstor.org/stable/2585842?origin=JSTOR-pdf

http://www.jstor.org/page/info/about/policies/terms.jsp



http://www.jstor.org/page/info/about/policies/terms.jsp

http://www.jstor.org/stable/2585842?origin=JSTOR-pdf




American Political Science Review Vol. 94, No. 3 September 2000

AnalyticNarratives yBates,Greif,Levi,Rosenthal, ndWeingast:A Review nd ResponseAnalytic Narratives. By Robert H. Bates, Avner Greif,

Margaret Levi, Jean-Laurent Rosenthal, and BarryWeingast. Princeton, NJ: Princeton UniversityPress, 1998. 296p. $65.00 cloth, $22.95 paper.

^ he rational choice revolution in political sciencebegan in American politics in the 1970s, firstinfluenced international relations in the 1980s,

and made its way to comparative politics during the1990s. As it moved beyond its base in Americanpolitics, rational choice theory confronted comparativeand historical questions of regime transition, socialconflict, democratic stability, economic development,and international governance. The application of ra-tional choice theory to these macro concerns createsanalytic problems that were less relevant when the

approach was applied primarily to microquestions.Robert Bates, Avner Greif, Margaret Levi, Jean-Lau-rent Rosenthal, and Barry Weingast have written amuch discussed book, Analytic Narratives, hat developsthese novel applications. Jon Elster, a seminal rationalchoice theorist who also has written extensively oncomparative and historical problems, believes thatAn-alytical Narratives expresses hopes for rational choicetheories of comparative and historical processes thatexceed what the approach is capable of producing.Bates and his colleagues reject Elster's critique anddefend their goal of a genuine social science capable ofaddressing arge-scale phenomena. Is rational choice his-tory a case of excessive ambition, or is it the logical nextstep in the rational choice revolution in political science?

Rational hoice History: Caseof Excessive mbitionJON ELSTER Columbia University

In Pathologies of Rational Choice, Donald Greenand Ian Shapiro (1994) claim that rational choicetheory has poor empirical support. Analytic Narra-

tives (AN) is among other things an attempt to rebut

this claim (p. 231).1 In my opinion, Bates et al. do notsucceed in their attempts to explain complex historicalphenomena in terms of rational choice. Although mycomments may seem harsh, certain statements in theIntroduction and the Conclusion invite the applicationof strict criteria. The authors claim that "the chaptersin this volume ... engage the concerns of many disci-plines and should therefore command a broad audi-ence" (p. 3). Similarly, "because the chapters engagesuch fundamental issues, they are of general signifi-cance" (p. 230); moreover, "they furnish deep insightsin particular cases" (p. 232).

I proceed in three steps. First, I discuss each chapterto argue that the authors make serious errors ofcommission that undermine the credibility of the anal-yses. Second, I raise some general questions concern-ing the application of rational choice models to large-scale historical phenomena. I argue that the

Jon Elster is Robert K. Merton Professor of Social Science, Colum-bia University, New York, NY 10027.

The author is grateful to Brian Barry, John Ferejohn, DiegoGambetta, Stephen Holmes, David Laitin, Bernard Manin, PasqualePasquino, John Roemer, and four anonymous referees for commentson an earlier draft.1 Unless otherwise stated, all page references are to this volume. Inassessing the chapters by Bates and Levi, I also draw on some of theirother writings. Although Greif refers to a forthcoming book-lengthversion of his chapter in AN and Weingast to an unpublishedbook-length manuscript, I did not have access to these. Some of theissues I raise in the text may be addressed by these more extensivediscussions.

contributors to AN make important errors of omission,such as failure to provide microfoundations or to offerevidence about the beliefs and intentions of the actors.Third, I try to draw some conclusions about the

feasibility of the AN project.

INDIVIDUAL CHAPTERS

Greif's ChapterThe study by Avner Greif of the political economy oflate-medieval Genoa offers a periodization of Genoesepolitics between 1099 and (roughly) 1350. From 1099to 1154, an economically suboptimal but peacefulequilibrium obtained. Each clan invested in militarystrength to deter the other from becoming a controllingclan, with the result that fewer resources were available

for economic cooperation abroad. (Following Greif, Iassume throughout that there were two clans only.)They might still cooperate in economic ventures, but ata less than optimal level. Between 1154 and 1164, anexternal threat did away with the need for investing inmilitary strength, which made more resources availablefor economic cooperation. From 1164 to 1194, therewas semipermanent interclan warfare. After 1194, thepresence of an external administrator of the city, thepodesta', brought about peace and "spectacular eco-nomic performance" (p. 58). I focus on the secondperiod (1154-64) and the final period (after 1194). Asthe chapter is opaquely written and marred by severe

misprints (e.g., in the statement of Condition III on p.51), I had to work hard to figure out what is being said.If I got it wrong, Greif must share the blame.

Greif's argument about the optimal nature of the

685



Analytic Narratives September 2000

second period is based on the premise that the externalpower will attack if and only if there is interclanwarfare. Consider first the "if" part of the premise.Because the clans know that if they fight each other thefuture will look bleak, they need not invest in militarystrength to deter each other. Without an external

threat, they may have to invest to raise the cost to theother of attacking. With an external threat, the benefitsof attack are so small that no deterrence is needed.Consider next the "only if" part of the premise. If thereis no need to fear an external attack as long as theyabstain from fighting each other, the clans can devoteall their resources to economic cooperation. Thisseems implausible. A strong external power mightattack, with some positive probability, even if the clansmanage to keep peace. Greif acknowledges as much onpage 32. To deter attack and/or to increase theirchances of winning a war against the external enemy,the clans would presumably invest some resources in

military strength. Greif acknowledges as much: "Rivalclans within various Italian city-states often cooperatedin confronting external threats" (p. 42). This militarycooperation must have come at the expense of eco-nomic cooperation. Greif seems to ignore, as far as Ican see, that both investment to deter or fight anexternal enemy and investment to deter a rival clanmust come at the expenses of resources available foreconomic cooperation. Although the presence of theexternal enemy does away with the need to invest forthe latter purpose, it creates a need to invest for theformer. In the absence of more information, the neteffect of an external enemy on military investment is

indeterminate.I also was not persuaded by Greif's claim that thepodesta' restored peace and efficiency in the finalperiod. Consider first the argument that the presenceof a third party-the external podesta'-prevented theendemic interclan warfare that characterized the thirdperiod. Greif captures the strategic interaction be-tween the two clans and the podesta by means of twogames. In the analysis of the "collusion" game betweenone clan and the podesta' against the other clan, heasserts that "after collusion has occurred, the amountby which the clan would reward a podesta woulddepend on the podestd's ability to confront that clan

militarily" (p. 49), because if he could not gain byconfronting it, then the clan could not credibly promiseto reward him.

The collusion game is embedded in a larger "podes-teria game," in which the first clan has the choicebetween challenging and not challenging the second;the second, if challenged, has a choice between fightingand not fighting; and the podesta', if the second clan ischallenged, has three choices-collude with the firstclan, prevent the first clan from taking control (thismeans the podesta' tries to take control for himself),and do nothing. For specific parameter values thisgame has a subgame-perfect equilibrium in which clan

1 does not challenge; clan 2 fights if challenged; and thepodesta prevents if clan 1 challenges and clan 2 fights,and, if clan 2 does not fight, either colludes or doesnothing. The podestai has to be strong enough to be

686

pivotal with a sufficiently high probability but not sostrong that the first clan could credibly promise toreward him in a collusion against the second. (I sim-plify.)

One empirical question is whether the strength ofthe podesta' was within the relevant interval. Greif tells

us he had twenty soldiers (p. 53). This is a smallnumber, but if the clans were of nearly equal strength,he might still be pivotal. Greif notes this fact, but failsto provide evidence about the relative strength of theclans. A more serious problem is that Greif does notexplain how the podesta' generated not only peace butalso prosperity. In the second period, prosperity sup-posedly occurred because the clans could afford todisarm-they had nothing to gain by fighting eachother. No mechanism is provided, however, for disar-mament in the fourth period. The military strength ofthe clans is taken as given throughout the analysis, andGreif offers no argument that would exclude higher

levels of military investment-and fewer resourcesavailable for cooperation-than in the first period. (Wemight also ask whether the high wage paid to thepodesta' cut heavily into these resources, but this isprobably a minor factor.)

Rosenthal's ChapterThe second chapter is a study by Jean-LaurentRosenthal of the political economy of European abso-lutism. His explanandum is nothing less than thedivergent courses of French and English absolutism inthe seventeenth and eighteenth centuries. To focus my

comments, I limit myself to one part of that explanan-dum: "why per capita taxation was significantly lower inFrance or Spain than in England or the Low Coun-tries" (p. 73). (Actually, I follow Rosenthal in limitingmyself to the comparison between France and Eng-land.) The explanation relies on a model that linksrates of taxation with gains or losses from wars fi-nanced by taxation., In brief, Rosenthal argues that taxrates were higher in countries with a unified taxsystem-in which either "the elite" or the king had fullcontrol over taxation-than in countries where fiscalauthority was divided between them. The reason is thattaxes are raised mainly for the purpose of conducting

wars, the spoils from which are shared between theelite and the crown. In countries where fiscal authorityis divided, neither actor can fully internalize the bene-fits from the taxes it raises; hence, we would expect taxrates to be lower than in a unified country.

Rosenthal assumes that "fiscal resources were usedprimarily to prepare for and prosecute war" (p. 69).Although he does not spell out what "primarily"means, an estimate is that "war expenditures oftenrepresented 80% of public expenditures" in early mod-ern Europe (Berenger 1996, 623). Whether one isjustified in using 100% as a proxy for 80%, asRosenthal does (p. 102), is not obvious to me.

Rosenthal then proceeds to specify the objectives andconstraints of the two actors. Both want to maximizethe return from war. Both are constrained by the costof raising resources for warfare. In a numerical exam-



American Political Science Review Vol. 94, No. 3

pie, the cost of raising taxes is equivalent to 5% of thenational product and equals the amount of taxes raised.The outcome of war affects the two actors differently,for two reasons. First, they "share the returns fromwinning and losing according to the extent of theirfiscal control of the domestic economy" (p. 103).Second, if the war ends in a draw, which in onespecification of the model happens in almost half thecases (p. 72), the elite bears all the costs of conductingwar. The first mechanism is responsible for the free-rider problem mentioned above, which induces bothparties to suboptimal funding of wars. The secondimplies that the elite might have reason to be evenmore reluctant than the king.

The idea of spoils divided in proportion to fiscalcontrol is unsubstantiated, empirically and theoreti-cally. Rosenthal does not cite any evidence that spoilswere in fact divided in this way. He also does notconsider alternative hypotheses, for example, that thespoils were divided in proportion to taxes raised forwarfare. In one specification of the model (p. 72) theelite raises no taxes although it controls 10% of taxa-tion and hence receives 10% of the spoils. Why wouldthe king give away 10% of the fruits of victory to anelite that in no way contributed to it? In fact, whywould the king ever give away anything? As the exec-utive in charge of conducting the war, he was presum-ably in a position to retain all gains for himself. To saymerely that, "after all, the king cannot hog the profitsfrom war if he expects support from the elite" (p. 71) isto confuse ex ante and ex post decisions and to ignorethe question of credible commitments. In any case, theargument fails if the elite does not raise any taxes forwar.

The idea that the elite bears all the costs from a drawis partly defended as a substantive proposition: "Theelite is ... assumed to bear a cost from draws becauseit is responsible for staffing the army and the adminis-tration, both of which are made to work harder duringhostilities" (p. 70). But that invites the questionwhether these costs would not also be incurred in caseof victory or defeat. Partly (and inconsistently), how-ever, the assumption is defended as a "convenient" (p.70, n. 7) device to create "the key policy tension in themodel: there are some wars that the king wishes to fight

but that the elite does not." Rosenthal notes that thereare "other ways of creating this difference," which donot, however "alter the substantive results of theanalysis" (p. 70). But would it not be more appropriateto identify the reasons the king was more belligerentthan the elite and model them directly?

The mathematical models in which these ideas arespelled out are described at three levels of abstraction.First, Rosenthal asserts that if one tries to work out theimplications of the causal links described above, "it ispossible to solve for equilibria in a general setting, butobtaining comparative static results requires additionalassumptions" (p. 71, n. 9). Second, therefore, the

mathematical appendix uses specific functional formsto model the relations between taxes and the cost ofraising them, and between taxes raised and the proba-bility of the war ending in victory, defeat, or draw.

Third, in the text Rosenthal draws on numerical illus-trations of this model with specific parameter values,asserting that the "results stand up to specificationchanges" (p. 71). He makes no similar assertion, how-ever, with respect to the choice of functional forms.Why should the reader have any confidence in theresults?

Let us look more closely at one central result,concerning the partial explanandum described above."In the model when either party controls the fiscalsystem, the tax rate is about 6.3 percent, whereas in thecase in which the regime is evenly divided the tax ratefalls below 5 percent, a drop in revenue of nearlyone-fifth. This finding goes a long way toward explain-ing why per capita taxation was significantly lower inFrance ... than in England" (p. 73). In a footnoteRosenthal adds: "The observed differences in ratesbetween Great Britain and France are much largerthan those found here, a finding that may well bedriven by the fact that the elite was not a unified taxauthority in France." I find this statement both uncom-fortably ad hoc and hard to square with a later state-ment: "Given that the current model predicts thevariation of tax rates properly, it may well be unneces-sary to be concerned with elite politics" (p. 89).

In my opinion, the model does not go any way at alltoward explaining the observed differences. The fit ispoor (as Rosenthal admits), and the model is veryartificial (as I have argued). In addition to the objec-tions mentioned above, I cannot see any reason thesame parameter values should obtain for all countries.One might expect, for instance, that the resourcesneeded to raise a given amount of revenue (and for agiven BNP) would be smaller in England than in thesomewhat more sparsely populated France. Also, canone assume that all armies are equally good at convert-ing money into probability of winning? This objectionobtains both for purposes of comparative statics andfor the purpose of determining the level of mobiliza-tion in the enemy.

Levi's ChapterThe third chapter, by Margaret Levi, is a study ofnineteenth-century conscription in France, the United

States, and, more briefly discussed, Prussia. The ex-planandum is "the disappearance of various ways ofbuying one's way out of military service if conscripted:commutation (a fee paid to government) and substitu-tion and replacement (payment to someone else totake one's place)" (p. 109). The explanation relies on"the micromotives of four sets of key actors: govern-ment policy-makers, the army, legislators and constit-uents" (p. 111). As Levi notes later, a fifth actor playsa considerable role in explaining the disappearance ofbuying-out and its replacement by universal conscrip-tion, namely, the largely disenfranchised poor, who"rioted against commutation, making the practice suf-

ficiently costly that government acceded to their de-mands" (p. 135). Although the explanation is presentedas "derived from rational choice and relying on thelogic of game theory" (p. 144), the chapter does not

687




draw on game theory at all and deviates from standardrational choice theory in one important respect.

Let me first address some empirical shortcomings ofthe section on France. I was surprised to read that afterThiers gave a speech in the 1848 constituent assemblyadvocating the maintenance of replacement, "the voteof the Assembhee Nationale was 663 against and 140 infavor of his position" (p. 129). In fact, the vote was 663against and 140 in favor of a proposal by Deville to banreplacement. The book-length account by Levi (1997,95) has the correct version, so the version in AN mustbe due to an accident of compression. Other mistakesor ambiguities, however, are common to the book andto the chapter in AN. They concern the timing of theintroduction of universal (male) suffrage, the timing ofthe abolition of buying-out, and the causal link betweenthe two.

Levi inconsistently dates the introduction of univer-sal suffrage in France to 1875 (p. 111) and to 1884 (p.130), but the dates of 1848 or 1871 would be equally ormore plausible. In the wake of the February revolution,the constituent assembly of 1848 was elected by uni-versal suffrage. The constitution that it adopted alsoincluded a provision of universal suffrage. It was rein-troduced in 1871, after defeat in the war against Prussiaand the fall of Louis Napoleon. As for the changes inmilitary service, Levi writes that "the military defeats of1870 and the Commune of Paris were to sound thedeath of replacement" (p. 129). Hence, on her chro-nology, there could be no causal link between universalsuffrage and universal conscription: "Since full maledemocracy was not achieved until 1884, direct politicalpressure by those who had come to believe themselvesmost harmed by the inequities of the system cannotexplain the change in governmental norms about mil-itary service" (p. 130). Note, however, that she assertsthe opposite in an earlier and more theoretical section:"The preferences of the government policymakers andthe low-wealth population are the same, thus helpingto account for the policies that emerge when thefranchise is significantly extended" (p. 118). A similarpassage (p. 115) is quoted below. As these statementsare not explicitly made about France, I postponediscussing them until later.

Chronology and the facts suggest a close connection

between the two reforms. With regard to 1871, thechronology is consistent with the idea that abolition ofreplacement and introduction of universal suffragewere part of the same ground swell-not that the latterwas the cause of the former, but that the two sprangfrom the same popular sentiment. I know too littleabout this episode, though, to affirm that there was alink of this kind. I feel more confident in asserting astrong link during the first part of the 1848 revolution.Levi's treatment of this event is confusing. Althoughshe correctly asserts (in her 1997 book) that theconstituent assembly voted against a ban on replace-ment, she also says in AN that "the members of the

Assemnbke Constituante did overwhelmingly declaresupport of the abolition of replacement" (p. 128). Thetwo statements can be reconciled by noting the drasticchange of mind of the assembly after the June insur-

688

rection of the Paris workers. In the first draft by theconstitutional commission, presented to the assemblyon June 19, 1848 (four days before the insurrection),Art. 107 declared starkly: "Le remplacement est inter-dit." After the insurrection, the assembly canceled thismeasure and a number of other radical proposals,notably the right to work and progressive taxation. Inthis case, it seems clear that universal suffrage anduniversal subscription were part of the same groundswell, broken only when the bourgeoisie came to seethe workers as a threat rather than an ally.

My main complaints with the chapter, however, arelogical rather than empirical. I do not agree with Levi'sconstruction of the preferences of the various socialgroups concerning military service. First, "the tradi-tional elites and wealthy ... would prefer a profes-sional army and, second, a system in which they couldbuy their way out through commutation since, inprinciple, price is no object. Their third choice wouldbe a system of substitutes or replacements, whichwould require them to incur some search costs as wellas pay a fee" (p. 114). This seems illogical: Search costsand fee add up to a price, which may or may not behigher than in the second case. In the second case thefee is set by the government, in the third by the market,and there is no a priori reason market fee plus searchcosts should exceed the government-set fee. Also, if"price is no object," why mention search costs at all?Levi also claims that "the preference ordering ofunemployed and low-skilled workers and landless peas-ants should be, first, a volunteer army, then substitu-tion, then universal conscription, and, last commuta-tion" (p. 114). I do not see why this group would preferuniversal conscription over commutation. If commuta-tion coexists with lottery, as it did throughout thisperiod, then they would prefer this system (which gavethem a chance of not being selected) to universalconscription.

Levi writes in summing up:If [substitution nd replacement end to dominate univer-sal conscription or everyone], why does so much of thepopulation begin to object to directly purchased exemp-tions? It may be that, in fact, commutation, ubstitution,and replacement are not Pareto optimal. As governmentrequires more men to serve, substitution nd commutation

may increase the probability f having o serve for thosewho can afford neither of those options. The more peoplewho buy themselves ut of the pool of eligibles, he deeperinto the pool the army reaches. Thus it may not, after all,be an efficient ystem (pp. 114-5).

But this argument applies only to commutation. Also,even commutation is Pareto optimal, although, unlikereplacement, it is not a Pareto improvement comparedto lottery-based conscription.

These preferences are constructed on the assump-tion that people assessed the alternatives "on purelyeconomic grounds" (p. 114). Levi argues, however, thatfairness considerations also entered into the formation

of preferences. The existing arrangementmay also begin to be perceived as an inequitable ystem.Those who cannot afford o purchase ubstitutes ace thepossibility of having o become soldiers when they would






pass. This feature of the situation is more consistentwith one-sided blackmail than with a quid pro quo.

Because I am not a specialist on the period, I cannotoffer a persuasive alternative to Weingast's account.Given the facts that he adduces, however, I cannot seethat the convention-based explanation is superior toone based on a succession of bargains. The allegedconvention is left dangling. What is the mechanism byvirtue of which it enabled credible commitments? In avolume that generally tries to spell out the game-theoretic foundations of credibility, it is surprising tosee that Weingast does not specify the sanctions thatsupported the convention. The obvious answer is that ifthe North had tried to impose its will on the South inmatters of slavery, the South would have broken out ofthe Union. But that standing fact could equally havesustained a succession of deals.

Bates's ChapterThe final chapter, by Robert Bates, is an account of theoperations of the International Coffee Organization(ICO) from 1962 to 1989. The main explanandum isthe stability of the producer-consumer agreementagainst free riders on the consumer side. Why did notlarge American coffee roasters, such as Folgers, breakthe agreement by purchasing coffee from illicit sourcesat lower prices? An additional question raised in thechapter concerns the internal organization of the ICO.Here, the explanandum is the allocation of quotas toproducer countries.

Bates begins by asking why there was any need for acartel at all. Why did not the dominant producer,Brazil, try simply to deter other producers from enter-ing the market? Although Selten's chain-store paradoxshows that entry deterrence fails when there is fullinformation and a finite number of successive entrants,deterrence can work when there is private informationor an infinite number of periods. Bates does notaddress the strategic use of private information todevelop a reputation for toughness to deter entrants(Fudenberg and Tirole 1991, 370). He asserts, wrongly,that reputation effects arise in infinite-horizon games(p. 204). Instead, he argues that the entry of Africanproducers into the coffee market during World War II,whom Brazil could not exclude after the war, turnedthe market into a one-shot game. Once the Africanproducers had incurred the high fixed costs of plantingcoffee trees, Brazil could not credibly threaten to drivethem out of business by underselling them: "Theywould be reluctant to uproot their plantations, andinstead would simply reduce their inputs of labor" (p.205). The argument is obscurely presented and relieson an economic model that is unstated rather thanexplicit.

Next, Bates tells the story of U.S. involvement inICO. The apparent paradox-why a consumer nationwould enter into an agreement to raise prices-isexplained by security reasons. After the rise of Castro,"the governments of Brazil and Colombia were able tolink the defense of coffee prices to the defense ofhemispheric security" (p. 205). Consumer concerns and

690

electoral preoccupations in the House of Representa-tives delayed U.S. adherence to the agreement butcould not prevent it. Although Bates takes the delay asproviding an argument against the realist school ofinternational relations, one might equally well take theultimate acceptance of the agreement as support forthat school.

To explain the stability of the cartel, Bates focuseson the consumer side. In his book-length study of thesubject, Bates (1997, 145-50) argues against the prop-ositions that producers were kept in line by Brazilianhegemony, by U.S. enforcement (but see below ), or byredesign of the pricing system to deter free riding.Instead, he argues that stability among the consumersinduced stability among the producers. The key ele-ment in the explanation is the discount on bulk pur-chases that Brazil and Colombia offered to large Amer-ican roasters. The passage linking the discount to thestability of the agreement goes as follows:

[The] reasoning focuses attention on the relationshipbetween those who demanded he agreement (the largecoffee producers) and those who organized ts supply thecoffee roasting firms). Examining their behavior in acollusive equilibrium, we find that when a large roaster,like General Foods, helped the U.S. government to policeillicit shipments of coffee from the competitive fringe-say,Central America-it drove up the price of those coffees.The differential etween he price of Colombian nd othercoffees then declined. Given the differentials n qualitybetween Colombian and other mild coffees, buyersswitched o Colombian offee. As more buyers urned toColombian coffee, Colombia's dependence on any onegiven buyer of its coffee declined. The strategies pursuedby the parties n the collusive equilibrium hereby oweredthe costs to producers of implementing hreats to theroasting irms.The reduced costs to Colombia f cancelingits contract with any given firm rendered Colombia'sthreats .. more credible. The collusive agreement couldtherefore become self-enforcing p. 215, emphasis added).

Once again, the economic model is verbal and implicitrather than explicit, which makes it hard to judge thevalidity of the explanation. It may be true, or it maynot.

The phrase that I emphasized differs sharply from apassage in which Bates (1997, 147) denies the efficacyof U.S. enforcement:

From time to time, the United States followed a secondtactic. It signaled o the "competitive ringe" hat shouldthey not adhere o the coffee agreement, hey would not beeligible for benefits rom the Alliance for Progress. Whilethe United States used its aid program o offer induce-ments and sanctions, ittle evidence suggests that thesemeasures affected the behavior of nations in the coffeemarket. Rather, he evidence suggests hat such blandish-ments lacked credibility.

Like the chapter in AN, the book nevertheless assertsthe role of the United States in policing illicit ship-ments (Bates 1997, 152).

One might try to eliminate the contradiction by asuitable interpretation of the claim that the roasters"helped" the United States police illicit shipments.Perhaps the roasters interfered directly in, say, theCentral American countries where such shipments




might originate? In that case, the roasters wouldachieve what the United States could not do on its ownby credible threats. This suggestion must be rejected,however, in light of the statement that, "from Colom-bia's point of view, the discount in the price of coffee toGeneral Foods bought compensating advantages: thehelp of a large roaster in securing the United States'enforcement of the International Coffee agreement,thereby checking opportunistic behavior by the com-petitive fringe-and raising the average price of coffee"(p. 214). On a normal reading of this phrase, theroasters ensured U.S. enforcement rather than substi-tuted for it. I have read and reread Bates's chapter inAN as well as his book, but I may have missedsomething. Even if I have, the need to go throughtortuous exegesis to detect exactly what is being as-serted does not support the claim that this is an"analytic" narrative. For me, to be analytic is above allto be obsessed with clarity and explicitness, to putoneself in the place of the reader and avoid ambiguity,vagueness, and hidden assumptions. As I have indi-cated, not many, if any, of the chapters in AN live up tothis demand.

In a final section, Bates discusses the internal oper-ations of the ICO. Even when supplemented by otherwritings (Bates 1997; Bates and Lien 1985; Lien andBates 1987), the argument is opaque. Bates asserts thatone can explain the 1982 adoption of quotas by theICO in terms of the voting power, as measured by theShapley value, of individual members. The mechanismfor proposing quotas and the constraints on suchproposals remain, however, mysterious. In one article(Lien and Bates 1987), there is a reference to aproposal that failed and to a proposal that succeeded,but there is no mention of who made them. Bycomparing equation 4 in AN (p. 225) and equation 10in Lien and Bates (1987), we can infer that thesuccessful proposal must have been made by Colombia.(I cannot understand, however, how equation 4 canrelate "Colombia's Shapley value to its quota" [p. 225].Presumably, it regresses the quotas of all countries onthe Colombian proposal against their Shapley values.)It might seem (p. 222) that the failed proposal wasmade by Brazil, but this is inconsistent with the figuresoffered. On the one hand, under Brazil's proposal, it"would secure over 33 percent of the market" (p. 222);on the other hand, under the failed proposal, Brazilwould obtain 30% of the market (Lien and Bates 1987,631). Or perhaps these percentages are not compara-ble; I cannot tell.

Bates also asserts (p. 224) that producer countriescould make a claim for a quota according to theiraverage exports in one of two three-year periods in therecent past. Which of these two periods they chose wasup to them; presumably all countries chose the periodin which they had the larger share. If average coffeeexports are measured in absolute numbers of tons, thenthe sum of the individual claims might in theory exceed

the total to be allocated. If measured in percentageshare of exports, then they must exceed the total wheneach country can choose the most favorable period.Bates never tells us, however, how much the claims

constrained the proposals. Presumably, they must haveleft some scope for bargaining-but how much?

Although the Shapley value is uncorrelated with thevotes cast on the two successive proposals, and uncor-related with the allocative quotas in the proposal thatfailed, it is correlated with the allocative quotas in thesuccessful proposal.

The resolution of the paradox may lie in the following:quotas are established n stages. In the ICO, a proposal sfirst negotiated, hen proposed, and then voted on. It isduring he period of negotiations hat each nation exer-cises its bargaining power to shape the proposed alloca-tion. When the proposal s subsequently oted on, how-ever, the nations hen appear imply o determine whetherthey do better or worse under the proposed quota thanunder the status quo (Lien and Bates 1987, 635).

The claim is not that the Shapley value allows us toexplain the outcome of prevote bargaining (and of the

subsequent vote). It is the weaker claim that if theoutcome of prevote bargaining is correlated with theShapley value, then the proposal will gain the requisitemajority. Is this analysis an application of "rationalchoice theory" (Bates and Lien 1985, 555, 559)? Apartfrom the indeterminacy just noted, the use of theShapley value to explain the allocation of quotas is adeviation from the use of noncooperative game theoryin the rest of the book and in the earlier parts of Bates'schapter. The microfoundations of cooperative gametheory are by definition shaky.

I am also puzzled by the following statement:

Under the rules of the agreement, bigger producers wereendowed with greater numbers of votes; intuitively heyshould therefore have possessed more power. Equation 5provides he test we seek. As seen in that equation, therelationship etween he voting strength f a nation and tsShapley alue s indeed nonlinear; t is quadratic, uch thatas the number of votes controlled by a nation ncreases, tsShapley value rises more than proportionately p. 225).

This seems to be confusing two claims. First, largernations had more votes (see Bates 1997, 138, fordetails). Second, the relation between voting strengthand power is quadratic, such that when the number ofvotes rises, the power rises more than proportionately.Even if the Shapley value has been a linear function ofthe number of votes, the larger producers would havehad more power. Further puzzlement is provided bythe fact that Bates uses statistical estimation to deter-mine the Shapley value as a function of voting strength,whereas I would have thought that the relation obtainsby definition.

It is quite possible that most of these puzzles are just(severe) problems of exposition, not of reasoning. Itmay seem petty to dwell on them, but I believe theprogrammatic nature of the volume justifies my doingso.

SOME GENERAL ISSUESThe volume is also beset by some more general andperhaps more serious problems. The objections citedabove mainly concern second-decimal issues. Even if

691




the authors had gotten them right, there would still bea number of first-decimal questions to be asked.

The Assumption of Rational ActorsThe authors uniformly assume that the main actorsacted rationally. With the exception of Levi andRosenthal (who, albeit not in his formal mode, givessome place to the causal efficacy of religion), they alsoassume that the actors were motivated by their materialinterest. In general, of course, rational choice theorycan easily accommodate nonmaterial or nonselfishinterests. What matters is whether the actors pursuetheir goals in an instrumentally rational manner, notwhether these goals are defined in terms of materialself-interest. Yet, even when expanded to includebroader goals, rational choice theory is often inade-quate because people may not conform to the canonsof instrumental rationality.

Let me distinguish between two issues. First, we mayoppose rational choice theory as applied by the authorsto the paradigm of bounded rationality (e.g., Nelsonand Winter 1982) and to the research program vari-ously labeled behavioral economics or quasi-rationalchoice (Thaler 1991). In the chapters by Greif andRosenthal, actors (individual and collective) are en-dowed with the capacity to make very complicatedstrategic decisions. There is no reference to the factthat for most people the ability to engage in strategicreasoning is severely limited. Consider, for instance,French President Chirac's disastrous call for electionsin June 1997. I conjecture that his coalition lost be-

cause voters understood that if he wanted early elec-tions, it was because he knew something they did not,something which made him believe that he would loseif he waited. By calling early elections, he revealedwhat he knew, or at least that he knew something, andtherefore gave them a reason to vote against him.

The point is not that this was a failure of rationalityon Chirac's part, but that the failure is a very commonone. Experiments on the Winner's Curse (Thaler 1992)show that people are very bad at anticipating howbehavior may reveal beliefs. In Chirac's case, he did notanticipate that the voters would infer his beliefs fromhis behavior. In the Winner's Curse, subjects fail to

anticipate the beliefs that they themselves will be ableto infer from the behavior of others. In both cases,there seems to be a limit to how many inferential stepspeople can make ahead of time. It is not that it isparticularly difficult; it just does not seem to comenaturally.

Why, then, should I believe that Greif's Genoanswere capable of reasoning through nested games ofbackward induction, complete with discounting over aninfinite horizon? Also, why should I believe in Greif'sassumption of exponential discounting when there is"compelling" evidence (Barro 1999, 1126) that timediscounting is hyperbolic rather than exponential? It

will not do to say that the hypothesis of exponentialdiscounting is a useful first approximation, since itpredicts behavior that is qualitatively different fromwhat we would expect on the hypothesis of hyperbolic

692

discounting. By contrast, the hypothesis of quasi-hyper-bolic discounting (Laibson 1997) is a useful first ap-proximation.

Second, the departure from rationality may be muchmore radical if the actors are motivated by fairnessconcerns or by emotions. The Ultimatum Game (Roth1995) is a standard example of a situation in whichpeople are willing to take nothing rather than some-thing, either because they react to being unfairlytreated or because they are motivated by emotions ofenvy or anger. Weingast's chapter, for instance, ignorescompletely the emotional charge of the slavery issue.Commenting on the congressional debates over slaverybefore the Civil War, Miller (1996, 184, emphasis inoriginal), observes that "the slavery interest has trans-muted into the glorious Southern cause." As he docu-ments, the thinking of the slave owners and theirpolitical representatives was seriously distorted by pas-sion. (This may have made their blackmail more cred-ible than it otherwise would have been.) Levi does notignore these nonrational concerns, although she mis-describes them as rational. Although I know moreabout seventeenth-century French literature thanabout the politics of the period, even my scatteredknowledge of the latter confirms the overwhelmingimpression conveyed by the former, namely, that theruling elites were obsessed with glory and honor. Tounderstand, for instance, the deep pleasure they tookin humiliating their rivals, we must read Saint Simon'sMemoires rather than Games and Decisions.

The alternative to the rational actor is not, however,someone who is in the complete grip of passions and

completely ignores concerns of security and self-inter-est. It would be as absurd to deny that the Frenchabsolutist kings were preoccupied with revenue as toassert that this was their only concern. They were, in aword, reward-sensitive. That is not to say, however,they were reward-maximizing. Nonrational concernscould often distort their thinking or shorten their timehorizon so as to undermine the instrumental efficiencyof their behavior. Yet, once their material interestswere seriously threatened, they were capable of facingreality. The social sciences today, however, cannot offera formal model of the interaction between rational andnonrational concerns that would allow us to deduce

specific implications for behavior. As mentioned ear-lier, the idea of modeling emotions as psychic costs andbenefits is jejune and superficial. The fact that emotioncan cloud thinking to the detriment of an agent'sinterests is enough to refute this idea. Also, if (say)guilt were simply a psychic cost, then agents should bewilling to pay for a pill that would eliminate theemotion-but in reality they would, I believe, feelguilty about doing so.

High Levels of Aggregation

Some of the explananda in the volume are vast histor-

ical entities: the development of absolutism, the stabil-ity of antebellum politics, the secular expansion of theGenoan economy. Some of the actors that enter intothe explanations are huge collectives-clans, the elite,




North, South, and so on. When these are endowed withbeliefs and goals and are assumed to engage in complexstrategic calculations, credulity breaks down. Rationalchoice explanations divorced from methodological in-dividualism have a dubious value. Although there maybe collective actors who can for practical purposes be

treated as individuals, the ones who appear in AN arenot-or have not been shown to be-among them.The most fundamental question is, of course,

whether aggregates can coherently be said to havepreferences at all. Although social choice theoristshave argued that preference aggregation is a fragilebusiness, there may be special cases in which it can bejustified. We may be entitled, perhaps, to treat firms asindividuals to the extent that they are subject to hardmarket constraints. Even larger aggregates such asnations may be forced by their own statements tobehave consistently. Suppose there is a majority cycleof policies A, B, and C, and that procedural rules lead

to the adoption of A, which is then stated publicly asthe position of the country. If some of those whoformerly preferred C over A change their preferencebecause of the cost to the country of being perceived asinconsistent, even procedural rules that would origi-nally have led to the adoption of C may now simplyconfirm A. This particular story may not apply to any ofthe cases in AN. To be taken seriously, however, anyaccount that imputes goal-oriented behavior to aggre-gate entities has to explain why we should expectconsistency in their behavior.

Little Concern for Intentions and Beliefs

One would expect rational choice theorists to take theidea of choice seriously, by looking for evidence thatthe agents whose behavior they want to explain did infact have the goals and the beliefs they describe. Tosome extent, Bates and Levi provide direct evidenceabout mental states, drawing on interviews (Bates) andparliamentary proceedings (Levi). Weingast tries butfails. The other chapters neglect this aspect of social-scientific explanation. One might say, perhaps, thatthey are ruthlessly instrumentalist in the Milton Fried-man tradition. But to explain observed behavior byintentions and beliefs that are imputed rather than

documented is acceptable only if (1) the sources do notpermit us to establish intentions and beliefs directly,(2) the observed empirical fit is very good, (3) otherimplications of the imputed intentions and beliefs arededuced and verified, and (4) plausible alternativeexplanations are given a good run for their money andthen rejected. At the least, I do not feel confident thatthe chapters by Greif, Rosenthal, and Weingast satisfythese criteria.

Much of applied rational choice theory is a combi-nation of just-so stories and functionalist explanation.One constructs a model in which the observed behaviorof the agents maximizes their interests as suitably

defined, and one assumes that the fit between behaviorand interest explains the behavior. Suppose that highereducation tends to make people pay more attention tothe future, and paying more attention to the future

tends to make people better off. It is then a temptingstep to conclude that people choose higher educationin order to reduce their rate of time discounting(Becker and Mulligan 1998), or (in a more generalversion) that this particular benefit of education ex-plains why it is chosen. It is, however, a temptation that

should be firmly resisted, in either version (Elster 2000,26-9). Unless one can demonstrate an intention (firstversion) or a causal feedback loop from the conse-quences of the behavior to the behavior (generalversion), the coincidence of behavior and interest maybe only that-a coincidence.

No Concern for UncertaintyExcept for a minor feature of Weingast's chapter, allthe models in AN assume full information. In the realworld, of course, high-stake politics is permeated byuncertainty. No model of political behavior that ig-

nores this fact can be successful in predicting out-comes. There are at least five kinds of uncertainty. Thefirst is brute factual uncertainty (will there be a majorearthquake in greater Los Angeles over the nextdecade?). The second is higher-order uncertaintyabout the cost of resolving first-order uncertainty (do Ihave time to ascertain the enemy's position beforegoing into battle?). The third is strategic uncertaintydue to multiple equilibria (do cartel members playtit-for-tat or sudden death?). The fourth is uncertaintydue to asymmetric information (is my opponent irra-tional or only faking?). The fifth is uncertainty due toincomplete causal understanding (will tyrannical mea-

sures imposed by a dictator make the subjects morecompliant or less?). The compound effect of these (andperhaps other) forms of uncertainty will, in mostcomplex situations, tend to be overwhelming. At thesame time, existing models of decision making underuncertainty, of equilibrium selection, and of gameswith asymmetric information tend to be very artificial.One might say, perhaps, that it is to the credit of theAN authors that they stay away from these models.Yet, I think the proper conclusion would have been toeschew formal modeling altogether.

CONCLUSIONIt is not clear to me what an analytic narrative is.Although the authors deploy rational choice theory,they assert that analytic narratives also could derivefrom "the new institutionalism ... or from analyticMarxism" (p. 3). In the absence of specific examples, Ishall understand an analytic narrative as a deductiveexplanation of individual historical facts, with "fact"being taken in a broad sense that may include long-term developments. In practice, this means rationalchoice explanation, for the simple reason that rationalchoice theory is the only theory in the social sciencescapable of yielding sharp deductions and predictions.

The theory rests on maximization theory, the beauty ofwhich is that it typically yields a unique maximum.I have argued that the contributions in AN fail to

sustain this research program. I have also hinted that,

693




over and above flaws of execution, the failure may bedue to the intrinsic nature of the program. As in thecase of sociobiology (Kitcher 1985), deductive rationalchoice modeling of large-scale historical phenomenamay be a case of excessive ambition. But this analogymay be imperfect. In the case of sociobiology, we knowfrom first principles that patterns of animal behaviormust have an explanation in terms of biological fitness.The main objection to sociobiological just-so stories isthat they are premature, not that some story of thiskind will not eventually come to be found. In theseventeenth century Pascal ([1660] 1995, 22) made asimilar observation about the Cartesian attempt tocreate a mechanistic biology: "In general terms onemust say: 'That is the result of figure and motion,'because it is true, but to name them and assemble themachine is quite ridiculous-pointless, uncertain, andarduous."

Rational choice history is in a worse situation than

that of either mechanistic biology in the seventeenthcentury or of sociobiology today. The analogy would beappropriate if it were mainly a question of refining thetheories and of gathering more evidence. One couldrefine theory by incorporating bounded rationality andquasi-rational choice, so as to match more closely theway in which actual decisions are made. One couldgather more evidence by paying careful attention tosources that illuminate the beliefs and goals of theactors. When we look at the public justifications givenfor the votes in the first French constituent assembly,for instance, it appears that all deputies had loftypublic-spirited motivations. When we read the letters

they wrote to their wives, it is apparent that some votedagainst bicameralism or against an absolute veto forthe king because they feared for their life if they did not(Egret 1950). After the publication by Alain Peyrefitte(1994, 1997) of the notes of his conversations with deGaulle, we can distinguish between what the latter saidfor public consumption and what really animated hisbehavior. There is no reason to think that de Gaullewas also being strategic in these conversations, whosesubstance has been confirmed by his son, AdmiralPhilippe de Gaulle. By proceeding on these two fronts,one would approach the ideal of explaining behavior bydeductive models that rely on (1) realistic assumptions

about the way agents make their decisions, given theirmental states, and (2) assumptions about their mentalstates that are independently verified rather then stip-ulated.

Unfortunately, this will not take us very far. I do notwant to insist on the issue of uncertainty. If we canidentify the actual beliefs of the agents, we need notworry too much about indeterminacy in rational beliefformation. I also do not want to insist on the problemof aggregation. We can proceed from the bottom up toexplain the behavior of aggregates of agents, for exam-ple, by putting together individual-level explanations ofthe voting behavior of all members of an assembly to

explain the overall vote. In many cases, the votingprocedure, including the ways in which alternatives arehelp up against each other, can simply be taken asparametric. The main obstacle to analytic narratives,

694

understood as rational choice history, arises at the levelof motivations.

As suggested by my earlier discussion, I want tomake two claims. First, nonrational motivations areimportant and pervasive. Wars have been lost becausesoldiers were taught that it was dishonorable to takedefensive measures (Dixon 1976, 54-5). Analyses ofwhy some individuals harbored Jews in Germany orGerman-occupied countries during World War IIwhereas others did not suggest that a major factor wasthat the former were asked by someone to do so(Varese and Yaish 1999). On a rational choice account,this would be a matter of information: To harborsomeone, you first have to know about their existence.On an alternative and perhaps more plausible account,it is a matter of the emotional difficulty of refusing aface-to-face request. In war trials after World War II,individuals accused immediately after liberation weresentenced much more severely than those tried for

identical crimes two years later (Elster 1998a), whichmost plausibly can be explained by appeal to thedynamics of anger and hatred. These are scatteredexamples, which could be multiplied indefinitely.

Yet, if we embrace the most abstract characteriza-tion of analytic narrative as deductive history, ratherthan rational choice history, such facts are not neces-sarily fatal to the project. To the extent that emo-tions-their triggering and their dynamics-can bemodeled in a way that yields definite predictions, theycan be incorporated into an analytic narrative. Mysecond claim, however, is that we do not know how toconstruct such models. We do not know how to predict

the behavior that will occur when an individual isentirely in the grip of an emotion. Fear, for instance,can lead to fight, flight, or freezing, and we do not knowwhich will be triggered in a given situation. We may noteven be able to predict which emotion will be triggered.If A favors B at the expense of C, will C feel envytoward B or anger toward A? Also, we do not havegood models of the trade-offs at work when emotionand rational pursuit of a goal coexist as motivations.(On all these points, see Elster 1999.)

If pressed, I might have to choose between twoversions of the latter claim. Do I mean, a la Pascal, thatdeductive modeling of emotions is a premature

project? Or do I want to state, more strongly, that suchmodeling will forever remain beyond our reach? Al-though I incline toward the second answer, my reac-tions are tempered by previous attempts to legislate apriori what science cannot do-often followed, de-cades or centuries later, by scientists who carry out theallegedly impossible task. Nevertheless, two reasonsmake me believe that deductive history will foreverremain impossible. First, the micromechanisms, if andwhen we find them, are likely to be of very fine grain.Second, however assiduously we search the historicalrecord, the evidence is unlikely ever to match thefineness of grain of the mechanisms.

All this is not to say that rational choice theorycannot illuminate historical analysis, as long as itsclaims are suitably modest. Collective action theory haschanged forever the way social scientists and historians




think (or ought to think) about rebellion, revolution,and related phenomena. Hobbes, Tocqueville, andMarx may use language that reminds us of moderndiscussions of the free-rider problem, but formal anal-ysis is needed to bring out its relation to the subtlydifferent phenomena modeled in the game of Chicken

or the Assurance Game. Montaigne and Descartes mayhave understood at a qualitative level that iteratedinteractions differ importantly from one-shot interac-tions, but they did not and could not anticipate game-theoretic results about the conditions under whichbehavioral differences are likely to arise. Modern anal-yses of credibility and precommitment have revolution-ized our understanding of strategic behavior. The ideaof burning one's bridges has always been known, butonly after Schelling (1960) has the motivation for suchbehavior been fully understood. Again, examples couldbe multiplied indefinitely.

The need for modesty appears in two ways. First, as

I have been at some pain to emphasize, one shouldavoid the postulate of hyperrationality. Collective ac-tion, iterated games, and credibility are simple ideasthat can be and have been refined to yield rococo (orbaroque?) constructions that no longer bear any rela-tion to observable behavior. To be useful, they have tobe constrained by what we know about the limitationsof the human mind. Second, because formal analysishas nothing to say about the motivation of the agents,it cannot by itself yield robust predictions. Although itis extremely useful to know that the structure ofmaterial interests in a given case is that of a one-shotPrisoner's Dilemma, that fact does not by itself imply

anything about what the agents will do. If they havenonmaterial or even nonrational motivations, theymight behave very differently from the noncooperativebehavior we would expect if they were exclusivelyswayed by material interests. If they are in fact ob-served to cooperate, then we will have to search fornonmaterial or nonrational motivations. Rationalchoice theory tells us what to look for, not what we willfind.

In closing, to focus the discussion I might put sixquestions to the contributors to AN. (1) Do they agreewith my characterization of analytic narrative as deduc-tive history and with my statement that in practice this

tends to mean rational choice history? (2) Do theyagree that a plausible analytic narrative requires inde-pendent evidence for intentions and beliefs? (3) Dothey agree that a plausible analytic narrative must be atthe level of individual actors or, failing that, thatspecific reasons must be provided in a given case toexplain why aggregates can be treated as if they wereindividuals? (4) Do they agree that standard rationalchoice theory needs to be modified to take account ofthe findings of bounded rationality theory and ofbehavioral economics? (5) Do they agree that it also

needs to be modified to take into account nonrationalmotivations? (6) Do they agree that at present there isno way to model nonrational motivations and how theyinteract with rational motivations, at least not in a waythat yields determinate deductions?

REFERENCESBarro, Robert J. 1999. "Ramsey Meets Laibson in the Neoclassical

Growth Model." Quarterly ournal of Economics 114 (3): 1025-52.Bates, Robert H. 1997. Open-Economy Politics: The Political Econ-

omy of the World Coffee Trade. Princeton, NJ: Princeton UniversityPress.

Bates, Robert H., and Da Hsiang Donald Lien. 1985. "On theOperation of the International Coffee Agreement." InternationalOrganization 39 (Summer): 553-59.

Becker, Gary S., and Casey B. Mulligan. 1997. "The EndogenousDetermination of Time Preferences." Quarterly Journal of Eco-nomics 112 (3): 729-58.

B6renger, Jean. 1996. "Guerre." In Dictionnaire de lAncien Regime.Paris: Presses Universitaires de France.

Dixon, N. F. 1976. On the Psychology of Military Incompetence.London: Futura.Egret, Jon. 1950. La revolution des notables. Paris: Armand Colin.Elster, Jon. 1998a. "Coming to Terms with the Past." Archives

Europ~ennes de Sociologie 39 (1): 7-48.Elster, Jon. 1998b. "Emotions and Economic Theory." Journal of

Economic Literature 36 (March): 47-74.Elster, Jon. 1999. Alchemies of the Mind. Cambridge: Cambridge

University Press.Elster, Jon. 2000. Ulysses Unbound. Cambridge: Cambridge Univer-

sity Press.Fudenberg, Drew, and Jean Tirole. 1991. Game Theory. Cambridge,

MA: MIT Press.Green, Donald, and Ian Shapiro. 1994. Pathologies of Rational

Choice. New Haven, CT, and London: Yale University Press.Kitcher, Philip. 1985. Vaulting Ambition: Sociobiology and the Quest

for Human Nature. Cambridge, MA: MIT Press.Laibson, David I. 1997. "Golden Eggs and Hyperbolic Discounting."Quarterly ournal of Economics 112 (2): 443-77.

Levi, Margaret. 1997. Consent, Dissent, and Patriotism. Cambridge:Cambridge University Press.

Lien, Da Hsiang Donald, and Robert H. Bates. 1987. "PoliticalBehavior in the Coffee Agreement." Economic Development andCultural Change 35 (April): 629-36.

Miller, William Lee. 1996. Arguing about Slavery. New York: Knopf.Nelson, Richard, and Sidney G. Winter. 1982. An Evolutionaty

Theory of Economic Change. Cambridge, MA: Harvard UniversityPress.

Pascal, Blaise. [1660] 1995. Pens~es, trans. A. J. Krailsheimer.Harmondworth: Penguin Books.

Peyrefitte, Alain. 1994. C'&tait de Gaulle, vol. I. Paris: Editions deFallois.

Peyrefitte, Alain. 1997. C'&tait e Gaulle, vol. II. Paris: Editions deFallois.Roth, Alvin E. 1995. "Bargaining Experiments." In Handbook of

Experimental Economics, ed. John H. Kagel and Alvin E. Roth.Princeton, NJ: Princeton University Press. Pp. 253-348.

Schelling, Thomas. 1960. The Strategy of Conflict. Cambridge, MA:Harvard University Press.

Thaler, Richard H. 1991. Quasi-Rational Economics. New York:Russell Sage.

Thaler, Richard H. 1992. The Winner's Curse. New York: Free Press.Varese, Federico, and Meir Yaish. 2000. "The Importance of Being

Asked: The Rescue of Jews in Nazi Europe." Rationality andSociety 12 (3): 307-34.

695

Elster Analytic Narratives

Documents