Fearon, James D. 1991. "Counterfactuals and Hypothesis Testing in Political Science." World Politics 43 (2) - 169-95

Trustees of Princeton University
Counterfactuals and Hypothesis Testing in Political Science

Author(s): James D. Fearon
Source: World Politics, Vol. 43, No. 2 (Jan., 1991), pp. 169-195
Published by: Cambridge University Press
Stable URL: http://www.jstor.org/stable/2010470 .
Accessed: 29/10/2013 14:59
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
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 forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Cambridge University Press and Trustees of Princeton University are collaborating with JSTOR to digitize,
preserve and extend access to World Politics.
http://www.jstor.org
This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

All use subject to JSTOR Terms and Conditions
COUNTERFACTUALS AND
HYPOTHESIS TESTING IN
POLITICAL SCIENCE
By JAMES D. FEARON*
Without the prior democraticmodernization of England, the reac-

tionarymethodsadopted in Germanyand Japanwould scarcelyhave
been possible. Without both the capitalist and reactionaryexperi-
ences, the communistmethod would have been somethingentirely
different,if it had come into existenceat all.
-Barrington Moore
Nuclear weapons did not cause the condition of bipolarity.... Had

the atom never been split,[the U.S. and the Soviet Union] would far
surpass the othersin militarystrength.
-Kenneth Waltz
THE epigraphs,provideexamplesof counterfactualconditionals,or

propositionsthattake the genericform"If it had been thecase that
C (or not C), it would have been thecase thatE (or not E)." Counterfac-
tuals make claims about eventsthatdid not actuallyoccur. It is argued
in thispaper thatsuch propositionsplay a necessaryand fundamental,if
oftenimplicitand underdeveloped,role in the effortsof political scien-
tiststo assess theirhypothesesabout the causes of the phenomena they
study.Particularlyin small-N researchdesigns,scholarsin comparative
politicsand internationalrelationsroutinelyevaluate causal hypotheses
by discussingor simplyreferringto counterfactualcases in which a hy-
pothesized causal factoris supposed to have been absent. Though this
procedureis quite common,itsmethodologicalstatusand itsviabilityare
unclear and are worthexploring.How does the strategyof counterfac-
tual argumentrelate,if at all, to methodsof hypothesistestingbased on
the comparisonof actual cases, such as regressionanalysisor J.S. Mill's
Method of Difference?Are counterfactual"thoughtexperiments"a vi-
* Preparedfordeliveryat the Annual Meetingsof the AmericanPoliticalScience Associ-
ation,San Francisco,August30-September2, i990. For helpfulcommentsand advice I wish
to thank Chris Achen,RobertM. Adams, David Collier,Russ Faeges, Don Green, Marcus
Kurtz, JimMahon, MerrillShanks,Laura Stoker,Arun Swamy,and an anonymousreferee.
I am particularlyindebted to David Collier for his encouragementand close reading of
severaldrafts.None mentioned(or unmentioned)are responsibleformymistakes.
I Moore, Social Originsof Dictatorshipand Democracy(Boston: Beacon Press, i966), 414;
Waltz, TheoryofInternational Politics(New York: Random House, 1979), i8o.
WorldPolitics43 (Januaryi99i), i69-95

170 WORLD POLITICS
able means of assessinghypothesesabout nationaland internationalout-

comes, or are theymethodologicallyinvalid in principle?
The articlehas threeprincipalaims. First,I seek to supportthe claim
made above, thatcounterfactualpropositionsand argumentsplay a cen-
tral role in the effortsof politicalscientiststo assess theircausal hypoth-
eses. Support is drawn frommethodologicalargumentand fromexam-
ples showing the counterfactualstrategyat work in researchin the areas
of comparative politics and internationalrelations.I discuss examples
fromthe literatureon the causes of World War I, the nonoccurrenceof
World War III, social revolutions,thebreakdownof democraticregimes
in Latin America, and the originsof fascistand corporatistregimes in
interwarEurope.
Second, the paper examines how the strategyof counterfactualargu-
ment is related to but also differsfrom methods of hypothesistesting
based on the comparisonof actual cases. The two approaches are found
to be closelyrelated:analystswithfewcases and manyvariablesare com-
pelled to resortto counterfactualargumentby a statisticalprinciple;and
counterfactualsalso appear to play a key role in the assumptionsthat
justifylarge-N regressionanalysis,when the data employedis quasi-, or
nonexperimental.2The differencebetween regressionand the counter-
factualstrategyis not thatone relieson counterfactualswhile the other
does not. Rather,the strategiesdifferin theway thateach employscoun-
terfactualsand in the way that each evaluates supportfor a causal hy-
pothesis.
Finally, the paper addresses the question posed above: Is counterfac-
tual argumenta viable means of assessingcausal hypothesesin nonex-
perimentalresearchsettings?I give no firmanswer here but instead in-
troducesome of the problemsand issuesinvolved.My purpose is neither
to advocate the use of counterfactualargumentin preferenceto compar-
isons with otheractual cases nor to suggestthatthe counterfactualstrat-
egy is fundamentallyinvalid or has no value in principle.Since political
scientistsoftenuse counterfactualargumentwhen assessingor justifying
causal hypotheses,particularlyin small-N researchsettings,it seems im-
portantsimplyto understandwhat the strategyentails.If the paper car-
ries a methodologicalprescription,it is that researcherswho use coun-
terfactual argument to support causal hypotheses should be
methodologicallyaware of what theyare doing and should make their
counterfactualargumentsas explicitand defensibleas theycan.
These general aims are pursued in three sections.The firstdistin-
groupsis referredto
Data notgeneratedby randomassignmentto controland treatment
as quasi-, or nonexperimental.

COUNTERFACTUALS AND HYPOTHESIS TESTING 171
guishes betweentwo strategiesof hypothesistesting the comparisonof

actual cases and counterfactualargument and examines key methodo-
logical similaritiesand differencesbetween them. The second section
shows how thecounterfactualstrategyappears in practiceby considering
examples fromwork in internationalrelationsand comparativepolitics.
The examples make clear thatcounterfactuals matterboth when the re-
searcher is focusingon one actual case (for example, the outbreak of
World War I or the Brazilian militarytakeoverin i964) and when the
researcherconsidersseveral actual cases (forexample, social revolutions
or interwarEuropean regime types).The thirdsectionreturnsto some
theoreticalissues concerning the link between causal arguments and
counterfactualpropositions,issues thatbear on the question of whether
the counterfactualstrategyis a viable one. Two logical problemsrelated
to the use of counterfactualsare discussedbriefly:(I) Is any eventC that
appears to satisfy"if C had not occurred,E would not have occurred" to
be called a "cause" or E? And (2) Are some counterfactualcomparisons
more "legitimate"or appropriatethan others?
ACTUALCASE COMPARISONS,
COUNTERFACTUALS, ANDTHE LOGIC OF
INFERENCE
Suppose it is hypothesizedthatC was a cause of event E. I would argue

thatwhen experimentalcontroland replicationare not possible,analysts
have available a choice betweentwo and only two strategiesfor"empir-
ically"assessingthishypothesis.Eithertheycan imaginethatC had been
absent and ask whetherE would have (or mighthave) occurredin that
counterfactual case; or theycan searchforotheractual cases thatresemble
the case in question in significantrespects,3 except that in some of these
cases C is absent (or had a differentvalue). In the latterprocedure,the
analystthen checks the associationbetween the occurrenceof C and E
in the set of actual cases.4If successful(fromthe analyst'spointof view),
both strategieswould tend to supportthe hypothesisthat the proposed
cause in fact produces (or produced) the effect.
As an illustration, consider the hypothesis that international structural
rather than domestic political factors have been the principal causes of
3The sense of "significant

respects"is discussedbelow.
4 These summarystatementsof the two strategiesare not complete.Qualificationsand
elaborationsforeach are discussedin the restof the paper,with more attentionpaid to the
counterfactual case strategy.The potentialdifficulties
withthe methodof comparingactual
cases, which is formallyknown as regressionanalysisthoughinformallypracticedin such
works as Theda Skocpol, Statesand Social Revolutions(Cambridge: Cambridge University
Press, 1979), are extensivelydiscussedin theeconometricsand statisticsliteratures.

172 WORLD POLITICS
major aspectsof Soviet foreignpolicy.The analystapplyingthe counter-

factualstrategywould evaluate the hypothesisby examiningarguments
thatany regimein Russia, Soviet or not,wouldhave made essentiallythe
same foreignpolicy choices. The analystusing the actual case strategy
would search forcases of statesin both similarand dissimilarstructural
positionsas Soviet Russia and then would check the sample for a rela-
tionshipbetweenstructuralpositionand foreignpolicies.5
Both methodologicalstrategiesaim to solvethesamestatistical problem.
Our analystbegins with one case and at least one explanatoryvariable,
which means negative degrees of freedom.6Legitimate causal imputa-
tionscannot be made on the basis of negativedegreesof freedom,so the
analyst wishing to assess a causal hypothesisor to assess the relative
weightsof different causes has no choice but to add or createmore cases:
eithera counterfactualcase (or cases) thatneveractuallyexistedor actual
cases.
Put otherwise,the analyst,in explainingwhy some particularevent E
occurred,cannothelp but explain why E occurredratherthansome other
possible outcome or outcomes.These otherpossibleoutcomesdefinethe
range of variationthatthe analystaccountsfor,and thisrange is treated
differently in differentresearchtraditions.For example,much historical
analysis leaves implicitthe other thingsthat might have been had the
historian'sfavored causes varied. In more methodologicallyself-aware
small-N work, analyststend to be more explicitabout what mighthave
happened.7Finally, in the actual case strategy,analyststake their cues
about what mighthave happened fromotheractual cases. Thus, an elec-
tions specialistmay explain why a respondentvoted Republican rather
than Democratic (as did otheractual respondents);studentsof interna-
tional conflictmay explain why deterrencefailed in one actual case but
not in other cases; expertsin comparativepoliticsmay explain why in-
terwarGermany became a fascistdictatorshipratherthan a liberal de-
5 "Structuralposition"here would entail the numberof great powers and the basic geo-
political circumstancesof the Soviet Union. Waltz (fn. i); and idem, "Another Gap?" in
RobertOsgood et al., Containment, SovietBehavior,and GrandStrategy, Policy Papers in In-
ternationalAffairsNo. i6 (Berkeley:Instituteof InternationalStudies,Universityof Cali-
fornia,i98i). On structuralversusdomesticpoliticalor ideologicalexplanationsof Soviet
foreignpolicy,see also BarryR. Posen, "Competing Images of the Soviet Union," World
Politics39 (JulyI987), 579-97.
6 Degrees of freedomare the numberof cases minusthe numberof explanatory variables
minusone.
7For example,BarringtonMoore, The Social BasesofObedienceand Revolt(London: Mac-
millan, 1978); Stephen Van Evera, "The Cult of the Offensiveand the Originsof the First
World War," International Security9 (Summer I984), 58-I07. Of course,historianscan be
quite carefulabout theircounterfactualarguments.For examples, see McGeorge Bundy,
Danger and Survival(New York: Random House, i988); and George Kennan,Russiaand the
WestunderLenin and Stalin(Boston: LittleBrown,i960), 29-32.

mocracylike England or (taking a larger range of otheractual cases) a

social democracylike Sweden or a traditionaldictatorshiplike Austria
in the Dollfuss-Schuschniggperiod.8
Not only do counterfactualand actual case strategiesboth attemptto
solve the same statisticalproblem,but both also run importantmethod-
ological risks.Less obvious is thatin each of the strategies,the principal
risksare closelyconnectedto the role played by counterfactuals.
The main risk in the firststrategyis obvious and serious-how can
we know what would have happened with any degree of confidence?
Historians,when confrontedwiththesuggestionthatthevalidityof their
causal inferencesnecessarily depends on counterfactualargument,have
oftendismissedout of hand or ignoredthe idea in favorof the view that
theirjob is to deal with reality.9Political scientistsand sociologists,too,
with the exceptionof a neglectedmethodologicalpiece by Max Weber
and some recentwork by JonElster,have also tended to avoid explicit
discussionor open embrace of the counterfactualstrategy,probablybe-
cause it is felt that an empirical political science must deal only with
actual cases. This beliefwould seem to be reflectedin the titleof a recent
book of essaysby politicalscientistsworkingwith counterfactualprem-
ises: WhatIf?: Essaysin Social ScienceFiction.The play on "science fic-
tion" is no accident here.Io
The risks of the second methodologicalstrategy-that of increasing
the degrees of freedomby consideringotheractual cases are also well
known. Whereas in the counterfactualapproach one tries to imagine
another (not actual) case in which the presumed causal agent is absent
but everythingelse thatis relevantis identical,in the second strategythe
analystadding actual cases may not know if the additional cases are ap-
propriatelyidentical. If there are other causes of the phenomenon in
On deterrence,see Paul Huth, ExtendedDeterrence of War (New Ha-
and thePrevention
ven: Yale UniversityPress, I988); on fascismversusliberalismin Germany,see Moore (fn.
I); on fascismversus liberalism,corporatism,or traditionaldictatorship,
see GregoryM.
Luebbert,"Social Foundationsof PoliticalOrder in InterwarEurope," WorldPolitics39 (July
I987), 449-78.
9 For example,accordingto A. J.P. Taylor,"a historianshould neverdeal in speculations
about what did nothappen"; Taylor,The Strugglefor Masteryin Europe,i848-i9i8 (London:
OxfordUniversityPress, 1954), 513. Or, in M. M. Postan'swords,"The might-have-beens of
historyare not a profitablesubject of discussion"; quoted in J. D. Gould, "Hypothetical
History,"EconomicHistoryReview,2d ser.,22 (August i969), 195-207. See also David Hack-
ettFischer,Historians'Fallacies (New York: Harper Colophon Books, 1970), I5-2I; and ex-
amples given in Peter McClelland, Causal Explanationand Model-Buildingin History,Eco-
nomics,and theNew EconomicHistory(Ithaca,N.Y.: Cornell UniversityPress, 1975).
? Weber, "Objective Possibilityand Adequate Causation in HistoricalExplanation," in
The Methodology oftheSocial Sciences(New York: Free Press,I 949); Elster,Logic and Society:
Contradictions and PossibleWorlds(New York: Wiley, 1978); citationsbelow (fn. 56); and
Nelson Polsby,ed., WhatIf?: Essaysin Social ScienceFiction(Lexington,Mass.: Lewis Pub-
lishing,i982).

174 WORLD POLITICS
question that are not consideredexplicitlyin the analysis,and if any of

theseare in factsystematically relatedto the causes explicitlyconsidered,
then effectsof the othercauses will be wronglyattributedto thoseof the
causes that are being evaluated. Simply put, estimatesof the effectsof
the proposed causes will be biased. In statisticsthis is the familiarprob-
lem of whetherany independentvariablesare correlatedwith the con-
tentsof the errorterm(which containsthe effectof all unspecified,un-
measured "other causes"). Such correlationmay occur due to failureto
include relevantindependentvariables,errorsin measuringthe indepen-
dent variables,or unrecognizedreciprocalcausation. In the comparative
politicsliteratureit is oftenposed as the question of whethera research-
er's several cases are comparable,or if the ceterisparibus assumptionis
adequately satisfied.,IBecause of the severityof this risk,some analysts
tend to be skepticalof large-N or comparativehistoricalwork; theypre-
fercase studies in which the risks of (an oftenimplicit)counterfactual
strategymay seem intuitivelyless serious.
Less well understoodis the link betweenthis centralrisk run by the
actual case strategyand counterfactuals. While the paper focusesprimar-
ily on the role of counterfactualsin small-N research,a few words on
theirrole in quasi-experimentalregressionanalysisare useful as a pre-
lude to making clearerexactlyhow the two strategiesdiffer.
To supporta causal interpretation of estimatedregressioncoefficients,
the large-N analystusing nonexperimentaldata needs to make a number
of theory-drivenassumptions.12 As noted, chief among these is the as-
The notionof comparabilityplaysa major role in the methodologicaland applied writ-

ings of specialistsin comparativepolitics.My impressionis that nonethelessthe notion re-
mains a deeplyvague one. It seemsto include,at varioustimes,theidea thattheothercauses
should be uncorrelatedwith the independentvariables(E(X'e) = o); that everythingelse
should have as littleinfluenceas possible(E(e'e) should be close to zero); thatmeasureswill
not be as valid or reliableacross countriesand cultures;and othermeanings.(Throughout,
E(-) is the expectationsoperator;X is an n X k matrixof n observationson k independent
variables;e is an n x i vectorof errorterms.)
Posing the main risk for analysisacross sets of actual cases in termsof the regression
validityof theceterisparibusassumptionalso bearsqualification.For regressionestimatesto
be unbiased,we do not need the otherthingsto be literallyequal, thoughit is truethatthe
moreequal theyare, thegreatertheprecisionofour estimatedeffects. For unbiasedestimates
of causal effectswe need only requirethatthe otherthingsnot be systematically relatedto
the prospectivecauses and thedependentvariablethatwe are evaluating.This pointappears
notto have been fullyclear to Mill (workingbeforestatistics was well developed),who some-
timeswritesin hisSystemofLogic (London: JohnW. Parker,i85i) as thougheverything else
has to be literallyidenticalin orderforthe Methodof Differenceto work.The same confu-
sion seems to carryover today in the work of some specialistsin comparativepoliticswho
take Mill as a principalmethodologicalguide (e.g., Theda Skocpol and MargaretSomers,
"The Uses of ComparativeHistoryin MacrosocialInquiry,"ComparativeStudiesin Society
and History22 [April i980], I74-97). That said, I should also note thatthose who conduct
large-N researchoftendo referto thisassumptionas the"ceterisparibusassumption"simply
forconvenience,and I will followthisusage here.
12This is true as well of actual experimentsin which cases are assigned at random to

sumptionthatexplanatoryvariablesand the errors(the othercauses) are

uncorrelated.Formally,the argumentthatestimatedcoefficients are un-
biased depends on the assumption that E(X'e) = o. It is easy to show
thatthisassumptionis credibleifand onlyifa counterfactualproposition
is credible; namely,the proposition
(Pi) If the cases in the sample had assumed differentvalues on the inde-
pendent variables,the contentsof the errortermwould not have dif-
feredsystematically.
If Pi is false,then E(X'e) does not equal zero. If E(X'e) does not equal
zero, Pi cannot be true(Q.E.D.).
This argumentsays thatassumingthatE(X'e) -=o in a quasi experi-
ment is equivalentto assuming the truthof a counterfactualproposition
about what would have happened if we could have altered a variable's
value forany case in the sample. Althoughone may not thinkabout the
ceteris paribus assumption in termsof a counterfactualproposition,a
counterfactualpropositionis necessarilyinvolved nonetheless.In actual
experimentsrandom assignmentguaranteesthe truthof Pi (within the
limitsof sampling variance). In quasi experiments,a causal interpreta-
tionof estimatedcoefficients requiresbeliefin thecredibilityof thecoun-
terfactualPi for justification.If we believe the resultsof a regression
analysis,we must be willing to believe that,say, if Joe Respondenthad
been a Republican as opposed to a Democrat, he would have been
roughly"so much" more likelyto have voted forReagan in i984; or that
if Israel had not moved troopsquicklyto itsnortheasternborderin Sep-
tember I970, then Syria would have been much less likelyto have been
deterredfromadvancing on the Jordaniancapital.'3
If both strategiesof confirmation
are means of solvingthe same statis-
tical problem,and if both depend in some measure on counterfactuals,
then how do theydiffer?The answer is thateach strategyprovides its
"empirical" confirmationfora causal hypothesisin a differentway.
In the actual case strategysupportfora hypothesizedcausal connec-
tion comes principallyin the formof a frequencyor magnitudeof asso-
ciation across actual cases. Of course, theory-drivenassumptions-
among them a counterfactualone are needed to supportor justifyany
regressionresult.But the resultone looksfor in regressionanalysis is an
treatmentand controlgroups.See Leland Neuberg,ConceptualAnomaliesin Economicsand
Statistics:
Lessonsfrom theSocial Experiment(Cambridge:CambridgeUniversityPress, i988).
Among otherthings,Neuberg shows thata counterfactual assumptionis needed to justify
estimatesof samplingvariancein actual experiments.I suggestbelow, however,thatcoun-
terfactualsplay a key role in quasi-experimentalhypothesistestingthattheydo not play in
actual experiments.
i The laterexamplecomes fromHuth (fn.8), 97, who oftenuses counterfactual argument
about particularcases to make more plausiblethe resultsof his regressionanalysis.

176 WORLD POLITICS
estimatedcoefficient significantlydifferentfromthe null hypothesis,and

thisdifferencederivesfroma frequencyof associationin the sample.
In the counterfactualcase strategy,by contrast,frequenciesof associ-
ation cannot be meaningfullyassessed. They are arguably irrelevantin
any event,since the researcheris attemptingto performthe perfectex-
periment,in which everythingbut the testfactoris equal. Instead, sup-
port for a causal hypothesisin the counterfactualstrategycomes from
arguments about what would have happened.These argumentsare made
credible (i) by invokinggeneral principles,theories,laws, or regularities
distinctfromthe hypothesisbeing tested;and (2) by drawing on knowl-
edge of historicalfactsrelevantto a counterfactualscenario.'4
An example will help make thispointconcrete.It has been proposed
thata "cult of the offensive" the widespreadconvictionheld by Euro-
pean civilian and militaryleaders that there were enormous strategic
advantages to strikingfirst was an importantcause of World War I.'5
According to the analysis above, there are two means of empirically
checking this hypothesis.Following the actual case strategy,we could
assemble a set of internationaldisputes,some of which escalated to war
and some of which did not. We could thenconstructa measure of mili-
taryand civilianbeliefsabout theadvantagesof a firststrike,presumably
frommilitarywritingsand fromstatementsof politiciansand generals
about theirexpectationsforwar. Finally,afterthinkinghard about what
otherindependentvariablesrequiredstatisticalcontrol,we could testfor
the strengthof association between commitmentto offensivedoctrines
and escalation.To assess the contributionof thiscause to the likelihood
of World War I in particular,we would check the value of the several
independent variables for this case, comparing their various contribu-
tions with thatof beliefin firststrikeadvantages.'6
Alternatively,we might employ the counterfactualcase strategy,
14 I am relyinghereon whatDavid Lewis calls "metalinguistic" theoriesofcounterfactuals.

These hold that"a counterfactual is true,or assertable,ifand onlyif itsantecedent,together
with suitablefurtherpremises,impliesits consequent";Lewis, Counteifactuals (Cambridge:
Cambridge UniversityPress, I973), 65. The "furtherpremises"may include both factsand
causal laws, or "lawlike generalizations."For example,thecounterfactual "If thatmatchhad
been struck,it would have lit" is truegiven the existenceof certainlaws concerningsulfur,
oxygen,friction, and heat,plus certainfactualconditions,includinga drymatch,presenceof
oxygen,etc. A counterfactual is thus a "condensedor incompleteargument"(J.L. Mackie,
"Counterfactualsand Causal Laws," in R. J. Butler,ed., AnalyticalPhilosophy[Blackwell:
Oxford,i962], 68). There are otheraccountsof what makes a counterfactual true(or assert-
able), based on notionsof distancebetween"possibleworlds"; see Lewis.
'5 Van Evera (fn.7); JackSnyder,"Civil-Military Relationsand the Cult of the Offensive,
1914 and i984," International Security9 (Summer i984), 108-46.
6 Of course,each step of thisprocess fromidentifying a sample to interpreting relative
importance-is fraughtwithmethodologicalperil.Both strategies,it should be emphasized,
are risky.

which oftengoes under the name "case study."'7In thisinstance,careful

researcherswould make an expliciteffortto imagine the prewar world
withouta cult of the offensivebut otherwisesimilar.They would then
constructan argument showing that the outbreak of a general war
would have been much less likely in the counterfactualcase. Such an
argumentwould depend forits credibilityon the principlesand histori-
cal knowledge used to draw the pictureof what would have happened.
Stephen Van Evera adopts preciselythis strategyto supporthis cult of
the offensivehypothesis,relyingat bottomon generalprinciplesof ratio-
nality.,8He asks, in essence: How would statesmenhave behaved if they
had believed that defenseratherthan offensehad the advantage? A re-
constructionof what rational actions would have followed from these
beliefsyields the conclusion that escalation would have been much less
likelyin a crisislike thatof JulyI914 (thatis, in the counterfactualcase).
The differencebetween the two means of hypothesistestingwould
thusappear to be quite stark,and on one level it is. In the counterfactual
strategythe analystsupportsone causal hypothesisby invoking others-
laws,regularities,or principles
thatare takenas havingsomeindepen-
dentcredibility.In theactualcasestrategyno otherprinciples needto be
invokeddirectly to supportthecausalhypothesis: onlya strength ofas-
sociationacrossactualcasesmatters. Indeed,fromthisvantagepointthe
counterfactual strategyfor"empirically" checkinga causal hypothesis
seemsonlyindirectly empirical,sincetheconfirmation it providesde-
pends principally on othertheories, whichare presumably themselves
supported byempiricalevidencefromactualcasecomparisons.
On whatmaybe a deeperlevel,thisapparently centraldifferencebe-
tweenthetwo strategies seemslesssharplydrawn.As noted,whenthe
actualcasestrategy isemployedina nonexperimental thevalidity
setting,
of theresultsin contingent
of a causal interpretation on thetruthof a
counterfactual assumptionabout the otherunspecified, unmeasured
causes.We mustbe readyto accepttheproposition thathad variableX
takenvaluesdifferent fromthosein thesample,no suchothercausesof
thedependentvariablewouldhavebeensystematically different
as well.
Our confidence thattheothercauseswouldnotvarywiththeindepen-
dentvariablesdependson our confidence in our theoryaboutwhatthe
I7I want to suggestthatcounterfactual reasoningmust underlieeffortsto inferor assess

the relativeweightsof causes in case studieswhere the analyst'sdegreesof freedomin the
actual world are negative.In practice,thosewho use case studiesoftenresortas well to casual
comparisonswithotheractual cases (e.g., "Whereas in manyotherAfricancountries.. ., in
Kenya .. .") and testingmultipleimplicationsof a theory;see Donald Campbell, " 'Degrees
of Freedom' and the Case Study,"Comparative PoliticalStudies8 (July1975), 178-93.
i8 Van Evera (fn.7).

178 WORLD POLITICS
other causes are and about how theymight be related to the variables
being testedexplicitly.
Two othercontrastsbetween the counterfactualand actual case strat-
egies should be noted. The firstconcernsthe appraisal of relativecausal
weight.In the actual case strategy,such appraisals can be carried out in
several ways, essentiallyby contrastingour estimatesof the effectsof
differentindependentvariables.'9Ultimately,we can do thisbecause we
have a sample fromwhich relevantfrequenciesand magnitudescan be
extracted.In the counterfactualstrategy,by contrast,we have no con-
crete frequenciesor magnitudes,and the degrees of freedomproblem
will bite every time we introducea new variable that may have influ-
enced the particulareventto be explained.Explicitjustificationof claims
about relativeeffectswill requirea proliferationofcounterfactualcases.
Suppose, forexample, a historianor politicalscientistwishes to argue
thatbothA and B were causes ofeventE, but thatA was more important
than B. The above analysiswould suggestthatwe now need not one but
at least two counterfactualscenarios to support this claim. We would
need to contrasta counterfactualcase where A is presentbut B absent
with one where B is presentbut A absent,and theninvoke general prin-
ciples and relevantfactsto argue thatE would have been more likelyto
have occurredin the firstinstance.20
One might well object that such argumentsabout what would have
happened in multiplecounterfactualscenarioswill be veryimpreciseand
uncertain.The second contrastbetweenthe two strategiesrelatesto this
issue of "precision of estimates."In the actual case strategywhen N is
large, frequenciesand magnitudesallow the researcherto get an idea of
how much risk attachesto the beliefthatthe truecausal effectof a vari-
able is as distinctfrom the null hypothesisas the resultsshow. In the
counterfactualstrategythereis no such formalcriterionforgauging the
risk of error associated with some independentvariable. All depends
instead on the plausibilityof argumentsabout what would have hap-
pened. As will be seen in the example of the debate on the origins of
World War I, argumentsabout the relativeimportanceof possiblecauses
becomeargumentsabout the relativeplausibilityof different counterfactual
scenarios.
19There is, however,more than one meaningfulsense to the idea of causal importancein
a regressionmodel. See J.MerrillShanks,"The Importanceof Importance"(Workingpaper,
SurveyResearchCenter,Universityof California,Berkeley,i982); ChristopherAchen,In-
and UsingRegression
terpreting (BeverlyHills, Calif.: Sage Publications,i982).
20 Some philosophersof historyworkingon theproblemof how historianscan and should
attributecausal weightingshave proposed similarcriteria.See Raymond Martin,"Causes,
Conditions, and Causal Importance,"Historyand Theory21 (i982), 53-74, and citations
therein.

The analysisin this sectionbears on currentmethodologicalissues in

the fieldin at least two ways. First,scholarsin comparativepoliticsand
internationalrelationsoftenargue that because statisticalmethods are
inapplicablewhen we have few cases and manyvariables,othermethods
need to be developed to enable sound explanations;theseinclude,among
others,the comparativemethod,structured,focused comparisons,pro-
cess tracing,and what are sometimescalled "qualitative methods." Fol-
lowing the analysis here, we would emphasize that statisticalmethods
are inapplicable in these circumstancesfor a good reason, namely, the
lack of enough cases to supporta causal claim. Further,statisticallogic
implies thatassessinga causal claim would require the addition of coun-
terfactualor actual cases. Statisticalprinciplesdo not simplycease to op-
erate when the number of actual cases dips below twentyor fifteenor
ten,creatingroom foralternativeways of testingcausal hypotheses.
Second, researchersshould choose between the two strategiesof con-
firmationon the basis of the typesof riskstheyare willingto run. Some-
times the counterfactualclaims needed to support a causal inference
seem entirelyunproblematic.For example, we do not require a formal
survey and regressionanalysis to support the claim that a gunshot
throughthe heartcaused the death. Less trivially,a researchermightbe
skeptical of regressionanalysis showing no clear relation between do-
mesticpolitical troubleand the initiationof war if it seemed clear from
counterfactualreasoningthat in a number of cases, domestic problems
were a factorimpellingthe leadershipto starta war.21Where thereare
serious problems in identifyinga sample, operationalizingand measur-
ing variables,and conceivingof relevantcontrols,counterfactualargu-
mentabout one or severalcases may be more compellingthana statistical
effort.
Indeed, understandingthatone can tryto explain counterfactualvari-
ation in single cases, as well as actual variationacross actual cases, may
help resolvesome of the puzzle over how case studiesfunctionmethod-
ologically to assess theories and hypotheses.22 There is a substantial
amount of work in politicalsciencewheretheanalystdeclaresan interest
in explainingphenomenonX (forexample,war, revolution,democracy),
chooses a set of cases where X actuallyoccurred,and ends up drawing
conclusions about the causes of phenomenonX. Those who engage in
large-N analysistend to view thisprocedureas totallyinvalid. Such an-
alysts"sample on theirdependentvariables"; iftheyfail to include cases
21Cf. Jack Levy, "Domestic Politics and War," Journalof Interdisciplinary

History i8
(Spring 1988), 653-73.
22 Anothertack on thispuzzle is takenby Campbell (fn. 17).

180 WORLD POLITICS
where X does not occur, how can theyfindcauses thatdifferentiate be-

tween outcomes? But if we see thateach case studyproposescauses that
selected the actual outcome froma range of possiblecounterfactual out-
comes in that case, we see the source of the "not Xs" and the variance
such analystsaccount for.This does not justifythe approach particu-
larlysince it is usuallyperformedunconsciously-but it does make some
methodologicalsense of it.
COUNTERFACTUAL ARGUMENT IN PRACTICE
The most controversialpoint made above is probablythat concerning

the necessityof counterfactualargumentfor justifyingcausal claims in
small-N settings.The approach is not put forwardas simply another
option on the menu forsmall-N practitioners. Rather,I have argued, the
point is that when degrees of freedom in the actual world are negative,
a causal claim requiresargumentabout counterfactualcases forits justi-
fication(or addition of other actual cases). This section will consider
some examples of how thislogical constraintmakes itselffeltin practice.
One does not findcounterfactualsplayingcentralroles in all small-N
political science research.My impression,afterreviewingliteraturefor
examples and evidence,is thatcounterfactuals are mostlikelyto be found
performingconfirmatory work in case studies where the analystis ex-
plicitlyconcerned with giving a causal explanation for some event or
phenomenon.23Of course, case studiesmay be used forother purposes,
such as evaluating the performanceof rival theoriesor simply giving
informationrelevantto various theoreticalconcerns.24 In addition,even
in what is nominallya case studyanalystsoftenemploy both strategies
of confirmation.
I will firstdiscuss the use of counterfactualsin three examples of
N = I case studies.25 Here counterfactualscenariosmustbe developed to
support explicit causal claims and to support assertionsabout relative
causal weight.Second, I will considerthe role of counterfactualsin sev-
23 E.g., Youssef Cohen, "Democracy fromAbove: The PoliticalOrigins of MilitaryDic-
tatorshipin Brazil," WorldPolitics40 (October i987), 30-54; Hyug Baeg Im, "The Rise of
BureaucraticAuthoritarianismin South Korea," WorldPolitics39 (Januaryi987), 231-57;
JoanneGowa, "Hegemons, 1Os, and Markets:The Case of theSubstitutionAccount,"Inter-
nationalOrganization38 (Autumn I984), 66I-83.
24 For examples,see articlesin FredericDeyo, ed., The PoliticalEconomyoftheNew Asian
Industrialism(Ithaca, N.Y.: Cornell UniversityPress,i987).
25 As the precedingdiscussionshould suggest,an N = I case studyin which causal infer-
ences are drawn is, strictlyspeaking,impossible,since othercounterfactualcases must be

invoked to supportcausal claims. I use N here to referto the numberof cases in the actual
world. On the idea of actual versuspossibleworlds,see Michael Loux, ed., The Possibleand
Actual:Readingsin theMetaphysics ofModality(Ithaca,N.Y.: Cornell UniversityPress,i985).

eral N > i examplesfromcomparativepolitics.Here analystsoftenmake

primaryuse of the actual case strategywhen groundingcausal assertions,
but even thencounterfactualsmay be needed to justifyinferencesfully.
N = I EXAMPLES
Some of the clearestexamples of the importanceof counterfactualargu-

ment come fromresearchon the causes of World War I. Over the years
political scientistsand historianshave identifiedan enormous collection
of possible factors,26which are typicallyargued to be causes on the fol-
lowing grounds: If cause X had not been present,the war eitherwould
not have occurredor would have been much less likelyto have occurred.
Thus, in arguing the causal importanceof misperceptionsin 914, Rob-
ert Jerviswrites,"Had the participantsrealized not only that the first
offensivewould not end the war, but also thatthefightingwould last for
fourpunishingyears,theymightwell have held back."27Note thatJervis
relieson a rationalityprinciple(sensitivity
to war costs)to make credible
the causal inference drawn from the counterfactualproposition.
On similargrounds,Van Evera has developed the thesisthata cult of
the offensivewas a major cause of World War 1.28 Further,he argues
thatmilitaryand civiliantendenciesto glorifytheoffensivehad theeffect
of "feedingor magnifyinga wide rangeof secondarydangers" thatother
analyststhoughtwere independentor unrelatedcauses. To establishthis,
Van Evera discussesthe secondarydangersone by one, arguing in each
case that had the cult of the offensivenot been present,the secondary
cause would nothave operated with as much (or any) force.His conclu-
sion nicely summarizes these counterfactualarguments.Throughout,
Van Evera reliesprimarilyon implicitrationalityprinciples:he supposes
leaders had different beliefsand thendraws conclusionsabout appropri-
ate or rationalbehaviorgiven such beliefs.29
26
These include,but are not limitedto, nationalism,imperialism,capitalism,social Dar-
winism,a fatalisticintellectualmood, the balance of power system,populationgrowth,dif-
ferentialindustrialization,a power transition, mis-
long cycles,tightalliances,multipolarity,
perceptions,psychologicalpathologies,leader personalities,essentiallyaggressiveGerman
intent,militarydoctrine(i.e., the cult of the offensive),militaryorganization,diplomatic
errors,the Russian mobilization,the archduke'sassassination,and the outcomesof recent
crises.
27 Jervis,"War and Misperception,"Journalof Interdisciplinary Historyi8 (Spring i988),
684.
28
Van Evera (fn.7)-
principlesare nottheonlyones thatmightbe used to limn
29 I should note thatrationality
counterfactual scenarios.One mightargue,forexample,thathad some independentvariable

been different, a key actor would have ignored it due to cognitivedissonanceor wishful
thinking.
Even so, thefrequentuse of rationality principlesto sketchcounterfactualscenariosshould
notbe surprising.The counterfactual strategyis oftenused byanalystsexplainingan outcome

182 WORLD POLITICS
The consequences of the cult of the offensiveare illuminatedby imagining

the politics of I9I4 had European leaders recognized the actual power of
the defense.... All European states would have been less tempted to
mobilize first,and each could have toleratedmore preparationsby adver-
saries before mobilizing themselves,so the spiral of mobilization and
counter-mobilizationwould have operated more slowly,if at all. If armies
mobilized, theymighthave rushed to defend theirown trenchesand for-
tifications,instead of crossingfrontiers,
divorcingmobilization fromwar.
Mobilizations could moreeasilyhave been confinedto single frontiers,lo-
calizing the crisis.Britaincould moreeasilyhave warned the Germans and
restrainedthe Russians,and all statesmencould moreeasilyhave recovered
and reversed mistakes made in haste or on false information.Thus the
logic that led Germany to provoke the I9I4 crisiswould have been under-
mined, and the chain reactionby which the war spread outward fromthe
Balkans would have been very improbable. In all likelihood, the Austro-
Serbian conflictwould have been a minor and soon-forgottendisturbance
on the peripheryof European politics.30
The use of counterfactuals is explicit and clear in Van Evera's analysis

because he is methodologically self-conscious about providing a causal
explanation. This is less true of much historical scholarship on the causes
of World War I, where the key counterfactual propositions are often left
implicit or underdeveloped.
Explicit treatment of counterfactual cases may in turn have the advan-
tage of sharpening substantive debates. In the example at hand Scott
Sagan has offered some important qualifications to the arguments ad-
vanced by Van Evera and Jack Snyder. He argues, among other things,
that Van Evera and Snyder "have overlooked the negative consequences
that would have resultedif the great powers had adopted purely defensive
military doctrines."3' He takes issue, in other words, with Van Evera's
counterfactual scenario. Sagan holds that the offensive doctrines of the
major European powers were rationally chosen to provide extended de-
terrence to key strategic allies and were not simply or solely the result of
the biases of military organizations. He suggests that defensive doctrines
might have left states unable to offer credible threats on behalf of their
allies, possibly increasing the chances for smaller wars that would have
as the resultof human choices.This entailssayingwhyotherpossiblechoiceswere not seen

as desirableby the actors.In game-theoreticterms,analystsusingthe counterfactual strategy
are oftendescribingwhy some particularset of choices was an equilibrium (or, at least,
rationalizable)strategyin the "game" faced by the actors.On Nash equilibriumversusra-
tionalizabilityas game-theoreticsolutionconcepts,see B. Douglas Bernheim,"Rationalizable
StrategicBehavior,"Econometrica52 (i984), 1007-28.
3? Van Evera (fn.7), 105 (emphasisadded).
Security2 (Fall i986), 151-75, at 159 (em-
3' Scott Sagan, "1914 Revisited,"International
phasis added).

altered the balance of power against them (for-example,Germany loses

Austriato Russia,or Russia loses France to Germany).Sagan's argument
containsexcellentexamples of the use of general principlesand specific
historicalknowledge to supportcounterfactualscenarios,as well as some
cleveruses of the actual case strategyto refutecounterfactualclaims with
which he disagrees.32
Anothergood example of counterfactualanalysisin internationalre-
lations researchfocuseson the nonoccurrenceof an importantphenom-
enon, namely,that therehas not been a war among major powers since
I945. This outcome mightbe explained by any of the followingcauses,
accordingto differenttheories:bipolarity,the presenceof nuclear weap-
ons, successfulbalance of power politics,or the obsolescence of major
war due to the "Hollandization" of the greatpowers.33If eithernuclear
weapons or Hollandization were in factthe trueor major cause of post-
war militarystability,then we cannot hope to employ the actual case
strategyto check this,since neithervariable varied much before I945.34
With JohnMueller,who has recentlyargued the case forHollandization
against the more widely accepted nuclear weapons thesis,we would be
compelled to argue about what would have happened had nuclear weap-
ons not been inventedand amassed in thisperiod.35As Mueller puts it:
The postwar world mightwell have turnedout much the same even in the
absence of nuclear weapons. Without them, world war would have been
discouraged by the memoryof World War II, by superpowercontentment
with the postwar statusquo, by the nature of Soviet ideology,and by the
fearof escalation [to conventionalwar].36
Mueller proceeds to argue the counterfactualcase foreach of these "in-

dependent variables" favoringpostwar stability.Though he does not
denythatnuclear weapons may have had some damping effecton poten-
32 See also Snyder'sresponseto Sagan's critiqueand Sagan's reply, Security9

International
(Winteri986-87), i87-98. Their discussionis carriedout largelyin the realmof the counter-
factual(e.g., what was theprobabilitythatthe Schlieffenplan would work).
-3 The Hollandization thesisis developedby JohnMuellerin Retreatfrom Doomsday(New
York: Basic Books, i989), where he argues that gradual changes in the governmentand
societiesof advanced industrialstateshave made themmore peaceable in theirexternalaf-
fairs.For a reviewof argumentson thecauses of thelong peace, see JohnLewis Gaddis, The
Long Peace (Oxford: OxfordUniversityPress,i987), chap. 8.
34Depending on how one countsthe "poles," neitherdoes bipolarity;see Waltz (fn. i).
35 To assess the question of relativeimportance, we would also need to ask about what
would have happened if nuclearweapons existedbut Hollandization did not. Mueller does
not explore this second counterfactualscenarioexplicitly.To hold thatHollandization has
been the more importantcause, he would need to argue thatpostwarstateslacking the key
Hollandization attributesmightnot have been deterredfromfightinga major war, despite
nuclearweapons.
36 Mueller, "The Essential Irrelevanceof Nuclear Weapons: Stabilityin the Postwar
World," InternationalSecurityI3 (Fall i988), 55-79,at 56 (emphasisadded).

184 WORLD POLITICS
tial escalation,he holds thattheircausal effecthas been redundant,due

to the combinedimpactof theothervariables.The claim about the coun-
terfactualcase the postwarworld withno nuclear weapons and no ma-
jor war is supported by some specifichistoricaldetail (for example,
characteristicsof Soviet ideology) and by at least one general principle:
"Wars are not begun out of casual capriceor idle fancy,but because one
countryor anotherdecides that it can profitfrom(not simplywin) the
war the combination of risk, gain, and cost appears preferable to
peace."37Taking thisas eithera theoretically plausible or an empirically
confirmedregularity,Mueller suggeststhateven disregardingthe added
costs posed by nuclear weapons, the costs of conventionalwar in these
years would have been enough to deterthe U.S. and Soviets froma hot
war.38
A final example of the counterfactualstrategyas used in an N = I
case studycomes fromwork on the breakdownof democraticregimes.39
Alfred Stepan's explanation forthe i964 militarytakeover in Brazil il-
lustratesa fairlycommon way thatcounterfactuals are employedin com-
parative politics and internationalrelationscase studies.40Stepan pro-
poses thattheactual outcome themilitarycoup-was made possibleby
the operation of certainsocial, economic,and ideological "macropoliti-
cal" factorsbut that these did not make the coup "inevitable." "There
remained a small margin of maneuverabilitywithin which the process
of increasing democratizationand participationcould have been ex-
panded."41Stepan is here definingthe rangeof counterfactual variation
that he wishes to explain. Brazil in i964 could have seen a democratic
outcome but did not. The micropoliticalfactorsthatreduced the "mar-
gin of maneuverability"and selectedtheauthoritarianoutcome fromthe
range of possibilitiesare then assigned causal status above that of the
37 Ibid., 68-69.
38 The fortunateabsence of actual cases of nuclearconflicthas led a numberof historians
and politicalscientiststo reflecton the role of counterfactuals in nuclear history.See John
Lewis Gaddis, "Nuclear Weapons and InternationalSystemicStability,"AmericanAcademy
of Arts and Sciences Occasional Paper No. 2 (Cambridge: AAAS, i990). This paper was
preparedforan AAAS workshopentitled"Nuclear Historyand theUse ofCounterfactuals."
In a different vein,RichardNed Lebow and JaniceGross Stein("Beyond Deterrence,"Jour-
nal of Social Issues43 [Winter i987], 3-7i) have brieflydiscussedthe role of counterfactuals
in defininga sample of cases of successfuldeterrence.
39 Juan Linz and AlfredStepan, eds., The Breakdownof DemocraticRegimes(Baltimore,
Md.: JohnsHopkins UniversityPress,1978).
4 Stepan, "Political Leadership and Regime Breakdown: Brazil," ibid. For otherexam-
ples,see RichardSmoke, War:Controlling Escalation(Cambridge:Harvard UniversityPress,
1977), and citationsin fn.23.
4' Stepan (fn.40), 134, and see also 120.

macropolitical factors,which, in Stepan's view, were not "sufficient"

themselvesto determinethe result.42
In his historicaltreatmentof theeventsleading up to the coup, Stepan
identifiespolitical strategychoices by the incumbent president,Joao
Goulart, as the key micropoliticalcauses of the breakdown of the dem-
ocraticregime.In an atmosphereof politicalstalemate,Goulart lost im-
portantmilitaryand middle-class allies by proposing major economic
and constitutionalreformsand biddingforthesupportof theleftto back
them. But still, "as late as twelve days after[the declaration of these
reforms]no 'winning coalition' existedto overthrowGoulart." A naval
mutinyby lower-levelofficersand sailorsthenoccurred,forcingGoulart
to choose between alienating either the mutineersor the higher-level
officers,who saw the mutinyas a major "threatto the principleof mili-
tarydiscipline."43His decision to be lenientwith the mutineershad the
unforeseeneffectof galvanizing high-levelmilitarysupportfora coup.
These two key politicalchoices are posed as causes of the breakdown
of the regime on counterfactualgrounds: had Goulart chosen different
strategies,the analysis suggests,a coup might not have occurred. The
counterfactualcontrastingcase is justifiedby referenceto specifichistor-
ical detail (evidence thatthe militarywas divided and generallynot sup-
portiveof directmilitaryrule beforethechoices were made) and to gen-
eral principles (for example, the propositionthat plottersdo not act
unless they expect sufficientsupport,or lack of resistance,from other
key actors).
Though Stepan is much more carefuland explicitabout his counter-
factualcomparisonthan is oftenthe case, I would argue thathis analysis
only goes part of the way to justifyhis causal claims,essentiallybecause
he does not spell out the counterfactualscenarioin quite enough detail.
Goulart's reasons for choosing the left-oriented, constitutionalreform
strategy,and thus in a sense the deeper causes of the takeover,are left
unclear. Stepan seems to suggestthatGoulart's destabilizingmove left-
ward was more a functionof his personalityand aspirationsthan it was
of the untenabilityof otheralternatives.But we need more carefulspec-
ulation about what would have happened had he insteadstayedwith his
divided and indecisivecoalition. If in the longer run his position was
simplyimpossible-that is, if no civilian leader could govern given the
political stalemateunder existinginstitutionalarrangements-then the
42 The distinction is similarto thatbetweenunderlyingcauses and specificor proximate

causes-a frameworkoftenused by historians.
43 Stepan (fn.40), i29 and I30.

186 WORLD POLITICS
macropoliticalfactorswould seem to gain in importanceas causal fac-

tors.44
N> I BUT STILL SMALL-N EXAMPLES
Researcherswith more than one actual case are not logicallycompelled

to use the counterfactualstrategyto justifya causal claim,as long as they
do not have more independentvariablesthan cases (less one), or two or
more independentvariablesthatvarytogether(perfectmulticollinearity).
Roughly speaking,these conditionsensure thatregressionestimatescan
be derived,and theyare usuallymet withease in large-N researchproj-
ects. Quite often,however,researchersin comparativepoliticsand inter-
national relationswork with few cases and many variables.45There are
sometimesopportunitiesin thisintermediaterange to employthe actual
case strategy,but the applicationof statisticalmethodseitherwould fail
to yield estimatesof causal effectsor would yield wildly impreciseesti-
mates. In these circumstances,I would argue, one typicallyfindsa mix-
ing of the actual and counterfactualcase strategies,with each used to
make the othermore credible.In good large-N research,the credibility
of causal effectestimatesderivesin thefirstinstancefromample degrees
of freedom.Given the theoreticalassumptionssupportinga causal inter-
pretation,causal claims are empiricallysupportedby regularitiesof as-
sociation,in Humean fashion.With an N between 2 and (say) I5, how-
ever, the regularityjustificationis weaker and may need support from
more detailed treatmentsof individualcases. Readers may want to know
not only thatthe proposed causes correctlypartitionoutcomesacross the
few actual cases, but that in each case the proposed causes indeed pro-
ducedtheeffectsattributedto them.In such efforts one can findexamples
where researchersresort implicitlyor explicitlyto the counterfactual
strategyand examples where theywould have to use it to defend their
causal claims.
One common methodologicalpracticein comparativepoliticsand in-
ternationalrelationswork could be called the"loading up of explanatory
factors."The researcherlists several causes for the phenomenon being
explained,all of which were presentin the cases where the phenomenon
occurred. In formalterms,the researcherhas a multicollinearity prob-
lem. In such instances,counterfactualargumentswould be necessaryto
analysisto assess the impactof

44 In a currentproject,Stepan uses explicitcounterfactual
presidentialas opposed to parliamentary systemson democraticregimebreakdownin South

America and SouthernEurope.
45Arend Lijphart,"ComparativePoliticsand the ComparativeMethod,"AmericanPoliti-
cal ScienceReview65 (September1971), 682-93.

support the claim thatany one of the proposed conditionshas a causal

effect.
Consider, for example, one of Stepan's arguments in his work on
Peru.46His goal is to explain the successor failureof attemptsto install
corporatistpoliticalarrangementsin Latin Americanstates.He identifies
fiveindependentvariablesand gives general hypotheseslinkingeach to
the likelihood of success or failure.These fivevariablesare then shown
to discriminatebetween actual cases of successand failurein the follow-
ing sense: where theywere all basicallyfavorableto corporatistinstalla-
tion,installationsucceeded; where theywere basicallyunfavorable,it did
not.The difficulty here is thatwithoutcounterfactualargument,we can-
not decide which of these variables mattered-whether at all or how
much. It could be, forexample,thatonlyone or two of thesevariablesis
reallycriticaland thatthe restare totallyirrelevant.There are only two
ways to decide: (i) findnew actual cases where one explanatoryfactoris
presentbut othersare not,or (2) argue counterfactually thatthe removal
of any one of the variableswould have damaged chances forcorporatist
success in the actual cases we have. The tendencyto "load up" explana-
tory factorsis quite common. For instance,BarringtonMoore's land-
mark work containsmany examples of this practice,such as the list of
five"main conditionsthathave apparentlybeen most importantforthe
developmentof democracy.47
Counterfactualsmay also come into play in what is nominallyactual
case work when analystsuse historicaltreatmentsof particularcases to
make credible claims based on actual case associations.For example,
Theda Skocpol identifiesthreekey variablesthatdifferentiate her "pos-
itive cases" of social revolution(I 789 France, I 9 I 7 Russia, and I 9 I I-49
China) fromactual cases in which social revolutionsdid not occur (for
example,Meiji Japan,seventeenth-century England, i807 and i848 Prus-
sia, Russia afterthe Crimean War and in I905, early-eighteenth-century
France).48Ratherthan simplystatingthe values of the independentvar-
iables forthe different cases and showingthattheydifferentiate between
outcomes, Skocpol undertakes moderately extensive historical treat-
mentsof each positivecase, detailinghow the independentvariablesshe
identifiesproduced social revolutionin each one. Though Skocpol makes
46 Stepan,The Stateand Society: Peruin ComparativePerspective (Princeton:PrincetonUni-

versityPress, 1978).
Diplomacy
47 Moore (fn. I), 430. See also AlexanderL. George et al., The Limitsof Coercive
(Boston: LittleBrown, 1971), 227.
48 Skocpol (fn.4). Only fourof these"negativecases" are treatedexplicitly and at length,
thoughSkocpol is well aware thatothersmentionedare used in the same fashion.

188 WORLD POLITICS
frequentuse of the actual case strategywithinhistoricaltreatments,49 her

approach is broadly similar to that of Stepan in the Brazil example, in
that the independentvariablesare shown to selectout certainhistorical
actualitiesfroma range of possibilities.
Gregory Luebbert's exemplaryuse of the actual case approach in a
small-N settingand provides some finalexamples of how the counter-
factualstrategymay be employed in such analyses.Luebbert firstiden-
tifiestwo independentvariablesthatperfectlypartitionhis fourteenac-
tual cases of European interwarregime types."Pluralist democracies"
occurredonlyin countrieswhereliberalpartiesgained dominancebefore
World War I. Regime typeelsewhere was determinedby which party
successfullyformeda coalitionwiththerural"middle peasants": ifit was
socialists,then "corporatistdemocracy"resulted;if liberals,then "tradi-
tional dictatorship";ifneither,thenfascism.Luebbertdoes not dwell on
the causal links between coalitionmembershipand regimetype.In jus-
tifyingthe causal argumentthat the effectsof World War I made plu-
ralistdemocracyalong the lines of Britainand France improbableelse-
where, he does explore the following counterfactual: "In ... Italy,
Norway, and Sweden, anothergenerationof peace mighthave resultedin
pluralistdemocraticregimes."50But on the whole he is contentto let the
perfectassociationand theintuitively acceptableidea thatcoalitionmem-
bers determinethe policyregimesupportthe causal claim.
Instead,Luebbert turnshis analyticattentionto identifying"the con-
ditions that produced each of these coalitions."'51By implication,these
will be the finalor deeper causes of regime type.This effortis marked
by both actual case comparisonsand implicitcounterfactualarguments.
For examples of the latter,consider Luebbert's explanationof why so-
cialist partiesallied with middle peasants ratherthan with the agricul-
tural proletariatin Norway and Denmark (thusyieldingcorporatistde-
mocracy).In Norway, he notes,therewere few landless laborers,so they
were not a temptinggroup forsocialiststo mobilize. Ratherthan corre-
latingsize of ruralworkerpopulationswithsocialistmobilizationefforts
across severalactual cases, Luebbert simplyappeals to a rationalityprin-
ciple that would supportthe appropriatecounterfactualargument.52In
Denmark the socialistscould not mobilize what was a much larger ag-
49 Ibid., e.g., 63.

5?Luebbert(fn.8), 457-58 (emphasisadded).
5 Ibid., 452.
52 The rationality principleis: Partiesdesirousof electoralsuccesswill seek partnersthat
can carrymanyvoteswiththem.The implicitcounterfactual argumentis: If therehad been
manylandlesslaborersin Norway,the socialistsmighthave soughtto forma coalitionwith
them,and fascismmighthave resulted.

riculturallabor forcebecause "this population.had already been heavily

mobilized by anotherparty."53 Instead,theymobilized the middle peas-
ants, leading to the corporatistcoalition.The implicitcounterfactualis:
If the agriculturallabor forcehad not already been mobilized, then it
mighthave been mobilized by the socialists,and fascismratherthan cor-
poratismwould have resulted.Thus, a particularfactabout Danish pre-
war politicsbecomes an ultimatecause of corporatismratherthan dic-
tatorshipor fascismin thatcountry.
COUNTERFACTUALS AND CAUSATION: Two THEORETICAL ISSUES WITH

PRACTICAL IMPLICATIONS
The propositionthat a cause of a particularhistoricalevent may be es-

tablishedby imaginingthe effectof its(counterfactual)absence has been
made before.In what remainsone of the bestessayson the topic,Weber
argued vigorouslyforrecognitionof thelink betweencausal explanation
and counterfactualsin historicalresearch.
[The question ofl what might have happened if, for example, Bismarck
had not decided to make war [in i866] is by no means an "idle" one [con-
traryto theviewofhistorian
EduardMeyer].It doesindeedbearon some-
thingdecisivefor the historicalmouldingof reality,namely,on what
causal significance is properlyattributedto this individual decision in the
contextofthetotality
ofinfinitely . . .54
numerous"factors"
Since Weber has been a methodologicalguru forgenerationsof soci-
ologistsand politicalscientists,it is somewhatsurprisingthatthispartic-
ular essay has been so littlediscussed and explicitlyapplied. In recent
yearsthe onlyseriousand sustaineddebate on the role of counterfactuals
outside of philosophytook place among historians(and without refer-
ence to Weber), in theirdiscussionsof the use of counterfactualsby some
practitionersof the "new economic history."55 The onlypoliticalscientist
I know of who has examined the topicat lengthis Elster,particularlyin
his Logic and Society.56 He presentsa novel "branchingworlds" theory
53Luebbert (fn.8), 466.

54Weber (fn. io), i64 (emphasisin original).
55RobertFogel, Railroadsand AmericanEconomicGrowth(Baltimore,Md.: JohnsHopkins
UniversityPress, i964); McClelland (fn. 9); Gould (fn.9); Fritz Redlich,"'New' and Tra-
ditionalApproachesto Economic Historyand Their Interdependence," JournalofEconomic
History25 (i965), 480-95; and T. A. Climo and P. G. A Howells, "Possible Worlds in His-
toricalExplanation,"Historyand TheoryI5 (1976), I-20. Fischer (fn. 9) lists furtherrefer-
ences.
56 Elster (fn. io). See also Elster,ExplainingTechnicalChange (Cambridge: Cambridge
UniversityPress, 1983), chap. i; idem, "Reply to Comments,"Inquiry23 (June1980), 213-
32; Steven Lukes, "Elster on Counterfactuals," Inquiry23 (Junei980), 145-55; Brian Barry,

190 WORLD POLITICS
for assessing the validityof counterfactualpropositionsand uses it to

analyze some examples in economic history.Outside the social sciences,
of course,analyticphilosophershave been writingabout counterfactuals
and causationforyears.57 While much of thisliterature(forexample,that
concernedwith the metaphysicsof modality)58 would seem largelyirrel-
evant to social scientists,some recentwork on counterfactuals and expla-
nation by philosophersof historyhas practicalvalue and probablyde-
servesgreaterattention.59
Scholars who have dealt with counterfactualshave often expressed
dismay,doubt,and bewildermentat thesortsof logical and philosophical
problemssuch propositionsseem to entail.This sectionbrieflyintroduces
two problemsthat seem particularlybothersometo social scientistsand
historians.
The firstis sometimesreferredto as the"Cleopatra's Nose Problem."6o
According to Pascal, ifCleopatra's nose had been shorter,Antonymight
not have been so infatuated,and the course of Western historymight
have been different.Does thisimplythatthe gene controllingthe length
of Cleopatra's nose was a cause of World War I? More generally,if we
believe thatan eventA satisfies
B wouldnothaveoccurred,
(P2) If A had notoccurred,
then are we committedto sayingthatA was a cause of B?
This is not just an idle question.As we have seen,social scientistsoften
argue that A was a cause of B on preciselythese grounds-that had A
not occurred, B might not have occurred.How do we distinguishbe-
"Superfox,"PoliticalStudies28 (i980), i39-43. Politicalscientistshave broachedissues raised

by counterfactualsin a varietyof places. See, forexample,AlexanderGeorge and Timothy
McKeown, "Case Studiesand Theoriesof OrganizationalDecision Making,"in RobertCou-
lam and RichardSmith,eds.: Advancesin Information Processing in Organizations
(Greenwich,
Conn.: JAI Press,i985), 2:33-34; Charles Ragin,The Comparative Method(Berkeley:Univer-
sityof CaliforniaPress,i987),39; Donald Moon, "The Logic of PoliticalInquiry:A Synthesis
of Opposed Perspectives,"in Nelson Polsbyand Fred Greenstein,eds.,HandbookofPolitical
Science(Reading,Mass.: Addison-Wesley,1975), 1:13i-228.
57 Nelson Goodman, "The Problemof Counterfactual Conditionals,"JournalofPhilosophy
44 (1947), 113-38, reprinted in his Fact Fiction and Forecast (Cambridge: Harvard University
Press, i983); Lewis (fn. 14); Ernest Sosa, ed., Causationand Conditionals(Oxford: Oxford
arisesfromtheir
UniversityPress, 1975). Part of the philosophicalinterestin counterfactuals
bearingon key issues in the philosophyof science.See FrederickSuppes, "The Search for
PhilosophicUnderstandingof ScientificTheories,"in Suppes, ed., The Structure ofScientific
Theories (Urbana: University of Illinois Press, 1977), 3-232, at 36-45, and references cited
there;ErnestNagel, The Structure ofScience(New York: Harcourt,Brace,and World, i96i).
Nagel (chap. I5) also saw thatcounterfactuals playa keyrole in historicalexplanation.
58 Loux (fn.25).
59 Martin(fn.20). See also Martin,"Beyond Positivism:A ResearchProgramforPhiloso-
phyof History,"Philosophy ofScience48 (i98i), ii2-2i; and idem, "Singular Causal Expla-
nation,"Theoryand Decision2 (1972), 221-37.
60 Edward Hallet Carr, WhatIs History?(New York: Knopf, I962); Gaddis (fn.38).

tween the infinityof particularfactorsthat would have precluded or

reduced the likelihood of some interestingevent had they not taken
place?
One approach would be to hold that causalityshould not be defined
in termsof counterfactualslike P2, that"A satisfiesP2" does not imply
that A is a cause of B. Consider the view thata cause is somethingthat
producesits effectwhenever(or usually when) it occurs. The cult of the
offensivecan be understoodto have producedWorld War I in thissense,
but Cleopatra's nose reallycannot.This strategyamounts to acceptinga
regularitytheoryof causation.61 Accidentalhappeningsthathelp lead to
specificeventsare not "causes" but only"conditions";conditionsof par-
ticular events that generalize or could regularlyproduce the effectare
labeled causes. The distinctionbetweencauses and conditionscould con-
ceivablybe a usefulone forpoliticalscientistsengaged in small-N work,
and particularlyfor case studies.62The distinctioncan do violence to
common sense and ordinaryusage, however, as in the following: the
unlucky person's death was not "caused" by the fallingshingle,it was
caused by skull fracture;the shinglewas onlya "condition."
Anotherapproach would be more lenientwithcertainaccidentalhap-
penings. We could argue that the length of Cleopatra's nose did not
make World War I any more likelythan myriadother possible worlds
thatcould have followed,whereas the presenceof the cult of the offen-
sive did significantly"selectout" the particularoutcome thatwas World
War I. That is, the same theoreticalargumentholdingthatthe probabil-
ity of World War I conditionalon Cleopatra's nose being shorterwas
zero implies as well thatthe probabilityof World War I conditionalon
her nose being as it was must have been almostzero. On thisaccount,an
accidental (or "random") happening-say, a monkey bite leads to the
death of a king, whose replacementbegins a war63-could qualify as a
"cause' of a particularevent. The importantpoint is that in both ac-
counts eventsthat satisfyP2 are not necessarilycauses of the phenome-
non being explained. Though counterfactuals like P2 mightbe explored
to lend credence to a causal claim, a cause does more than just satisfy
P2.64 Both suggestionscould have practicalvalue for political scientists
arguing causalityon counterfactualgrounds.
6,
On these,see Tom Beauchamp and Alexander Rosenberg,Hume and the Problemof
Causation(Oxford: OxfordUniversityPress,i98i).
62
For example, Luebbert (fn. 8) might have distinguishedmore carefullybetween the
conditionsprevailingin particularcountriesthatallowed thecausesof regimetype-coalition
membership-to operate as theydid. On relatedphilosophicaldistinctionsbetweencauses
and conditions,see J.L. Mackie, "Causes and Conditions,"in Sosa (fn.57), 15-38; and Martin
(fn-59, 198I, 1972).
63 Carr (fn.6o), citingChurchill.
64 A thirdsuggestionforresolvingthisproblemwould be to add a conditionof temporal

192 WORLD POLITICS
The second vexingproblemforscholarsconcernswhat Elstercalls the

"legitimacy"of a counterfactualproposition.65 Elster argues thata coun-
terfactualthoughtexperimentundertakento assess a causal hypothesisis
not legitimateifwe have a theorysayingthatthecounterfactualanteced-
ent could not have happened. Suppose we wish to learn the effectof the
railroad on the growthof GNP in nineteenth-century America and that
we attemptto do so by imaginingthe last centurywithoutthe railroad.66
Elster thinksit nonsensicalto speculatewhetherthe internalcombustion
engine would have been inventedearlierthan it was (in the counterfac-
tual nineteenth-century America without railroads), since an answer
would require a theoryof technicalchange strongenough to make the
original counterfactualproposition implausible. If we could predict
whether the gas engine would have been invented earlier, surely we
would also have a theoryshowingthattherailroads"had to be" invented
when theywere. Elster calls this"the scissorsproblem,"67 or "the unim-
portance of inevitable":68the betterour theories,the more things we
know "had to occur" as theydid, and thusthe fewercounterfactualswe
can legitimatelyassert.
In social science practicethis problemoftenappears in the following
guise. On the basis of actual case comparisons,a comparativistclaims
that C caused E in countryX, suggestingthat if C had been different,
the outcome in countryX might have been more like the outcome in
countryY. A specialiston countryX criticizesthis as absurd, arguing
thatdue to a complexof historicaland culturalfactorsparticularto coun-
tryX, C could not have been different.
The insistencethat counterfactualpropositionsbe "legitimate" may
confound,or at least obscure,two distinctproblems.69The real issue is
not legitimacy,if thismeans thatin our counterfactuals we cannot legit-
imatelyvarycauses thathad to occur as theydid. Whetherevent C had
to occur has no directbearing on its causal statuswith respectto E. A
variable may help explain one outcome and still itselfbe explained by
the action of othervariables.In large-N work thispatternis commonly
found in structuralequations models, in which a dependentvariable in
one equation may be an independentvariablein anotherequation.
or causal proximityto P2; that is, A is a cause of B if P2 is true and A precedes B by a

relativelyshorttimeperiod,or ifthecausal chain is not too long. But thisraisesthe problem
of how long?
65 Elster (fn. io).
66 Fogel (fn-55)-
67 Elster(fn.56, I983), 38.
68 Elster(fn. io), i85.
69 For relatedcriticisms legitimacy,see Barry(fn.56);

of Elster'snotionof counterfactual
and Lukes (fn.56).

The key issue, it seems to me, is what philosopherNelson Goodman

called "cotenability."In Goodman's account,a counterfactualassertion
is judged trueif (i) the counterfactualantecedent,when joined with ap-
propriatetheoriesand facts,impliestheconsequent;and (2) the counter-
factual antecedent is "cotenable" with the factsor "initial conditions"
used to draw the inference,meaning thatif the antecedenthad actually
occurred,the initial conditionscould also have occurred.70Thus, in his
critique of the cult-of-the-offensive hypothesis,Sagan in effectargues
that supposing the absence of the cult is not cotenablewith supposing a
I9I4 otherwiseidentical to the actual I9I4. As noted, he suggeststhat
beliefsin defensivesuperioritywould have created a differentstrategic
problem forstateleaders,one thatcould also have produced war.
The question,then,is not whethera factorhad to occur but whether
varyingthe factorimplies changing other factorsthatalso would have
materiallyaffectedtheoutcome.7'It is notappropriateto criticizea coun-
terfactualargument by saying that the antecedentcould not have oc-
curred. Rather,we need an explicitargumentsayingthatif the antece-
dent had been the case, other changes would be required in the
counterfactualscenariothatwould have affectedtheoutcome in a differ-
ent way.
An obvious methodological prescriptionfollows: analysts using the
strategy of counterfactualargument should pay close attention to
whether their counterfactualsuppositionsare cotenable with the facts
and theoriesused to draw the causal inferencestheymake. This is per-
haps a more precisestatementof what it means to make a counterfactual
argument plausible. I expect that in practice,the cotenabilityrequire-
ment will be more plausibly satisfiedfor small causes, such as specific
policydecisions,than forbig causes, such as nationalism,imperialism,or
a cult of the offensive.Historyoftenprovidesevidence thatleaders con-
sidered several possible choices at certain junctures,and in some in-
stancesit may be feasibleto imaginea different choice withoutchanging
other major influenceson the outcome in question. The fewer the
7"Goodman (fn.57, i983), 15-17. See also references in fn. 14. Goodman pointsout thatit
is quite problematicto use a counterfactual to definegeneraltruthconditionsforcounterfac-
tuals. See Mackie (fn. 14) fora possibleway around thisproblem(which at any rate may be
of greaterinterestto philosophersthan to politicalscientists).
I should note thatElster(fn. Io) is well aware of the issue of cotenability,
which he refers
to as "compossibility"(p. 177) and also "compatibility"(p. i83). Indeed, his "branching
worlds" theoryfor assessing the truthof counterfactuals can be seen as a suggestionfor
assessingcotenability.
7' Note the similarity of the cotenabilityconditionto PI, the key assumptionjustifyinga
causal interpretation of regressioncoefficients derived fromquasi-experimentaldata. The
likenessunderscoresthe point thatquasi experimentsand the counterfactual strategyshare
relianceon counterfactual suppositions.

194 WORLD POLITICS
changes fromthe actual world required by a counterfactualsupposition,

the easier it will be to draw and supportcausal inferences,and the more
defensibletheywill be.72
CONCLUSION
Counterfactualsand the counterfactualstrategyof hypothesistesting

play an importantbut oftenunacknowledged and underdevelopedrole
in the effortsof politicalscientiststo assesscausal hypotheses.I have tried
to show thatany nonexperimentalresearchthatmakes causal claims, be
it of the large-N or small-N variety,mustconfrontcounterfactuals in the
formof key assumptionsor in the use of hypotheticalcomparisoncases.
Particularlyin small-N research,the common condition of too many
variablesand too few cases makes counterfactualthoughtexperimentsa
necessarymeans forseriousjustification ofcausal claims. I close withtwo
simple suggestionsforanalystsevaluatingcausal claims via counterfac-
tual argumentratherthan via regularitiesof associationin a sample of
actual cases.
First, small-N analysts could strengthen(or simply specify) their
causal arguments by being explicitabout the counterfactualscenarios
needed to support their hypotheses.Quite commonly,researchersin
comparativepoliticsand internationalrelationsassertthat theirdepen-
dent variable is X, where X is some particulareventor phenomenon.X
mightbe the failureof the U.S. to play the role of internationalhegemon
between the world wars, a change in the nuclear proliferationregime,
the dominance of the Liberal Democratic Partyin Japan,or the collapse
of communismin Eastern Europe. Analystsexplainingsuch eventsneed
to understand that none of these are variables. They become values of
variablesifalternative,counterfactualscenariosare identifiedor ifactual
cases, some of which differin outcome,are added to the analysis.73 If,for
whateverreason,one is reluctantto add actual cases, then it is essential
to make explicitwhat mighthave happened if a possiblecause had var-
ied. Counterfactual comparison cases need not be exhaustively de-
tailed-just specified-so the readerknows what variationthe theoryor
hypothesisproposesto explain.
The second key step is making inferencesdrawn froma counterfac-
tual comparisondefensible.Sometimesthe argumentimplied by a coun-
72 This suggestionis influenced byexamplesprovidedby McGeorge Bundy (fn.7), and by
Gaddis's discussionof them(fn.38).
73 The pointthata variableis distinctfromany particularrealizationof it should be obvi-
ous but is sometimesmissed.The pointthatthe varianceexplainedmightbe definedacross
actual or counterfactualcases is rarelyseen.

terfactualpropositionis quite obvious and requires littleor no unpack-

ing.74Other times,particularlywhen the hypothesisis evaluated against
otherhypotheses,analystsshould make clear what argumentssupportit
and how theydo so. The analystneeds to ask whetherthe causal infer-
ence does indeed follow from the theoriesand historicalfactsused to
sketchthe comparisoncase and then,whetherthe counterfactualprop-
osition is cotenable with the counterfactualscenario. Cotenabilityre-
quires that if the counterfactualassertionhad been true (forexample, if
there had been no cult of the offensive),nothingelse would also have
been different in a way thatwould have materiallyaffectedtheoutcome.
Of course, we can never be certainabout what else would have been
differentifC had been different. But perhapswe can venturearguments
thatcan be judged more credibleor less credible,depending on our use
of historicaldetail and theoriesabout the way people behave. I should
emphasize that I am not suggestingthattherecan be a special method-
ology for determiningpreciselywhat would have happened. Nor am I
advocating a new methodologyto rival establishedapproaches to hy-
pothesistesting.Rather,the intenthas been, first,to show that counter-
factualscannot be avoided in nonexperimentalhypothesistesting,then,
to explicate their roles, and finally,to recommend that when political
scientistsuse counterfactuals,theydo so explicitlyand carefully.
74For example, "if that match had been struck,it probablywould have lit" will not be
controversialin most circumstances.Neither are the counterfactualargumentsimplied by
Luebbert(fn.8) on whythe fascistcoalitiondid notdevelop in Norway and Denmark.


Fearon, James D. 1991. "Counterfactuals and Hypothesis Testing in Political Science." World Politics 43 (2) - 169-95

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fearon, James D. 1991. "Counterfactuals and Hypothesis Testing in Political Science." World Politics 43 (2) - 169-95

Uploaded by

Copyright:

Available Formats

Trustees of Princeton University

Counterfactuals and Hypothesis Testing in Political Science

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

Without the prior democraticmodernization of England, the reac-

Nuclear weapons did not cause the condition of bipolarity.... Had

THE epigraphs,provideexamplesof counterfactualconditionals,or

WorldPolitics43 (Januaryi99i), i69-95

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

able means of assessinghypothesesabout nationaland internationalout-

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

guishes betweentwo strategiesof hypothesistesting the comparisonof

Suppose it is hypothesizedthatC was a cause of event E. I would argue

3The sense of "significant

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

major aspectsof Soviet foreignpolicy.The analystapplyingthe counter-

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

mocracylike England or (taking a larger range of otheractual cases) a

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

question that are not consideredexplicitlyin the analysis,and if any of

The notionof comparabilityplaysa major role in the methodologicaland applied writ-

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

sumptionthatexplanatoryvariablesand the errors(the othercauses) are

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

estimatedcoefficient significantlydifferentfromthe null hypothesis,and

14 I am relyinghereon whatDavid Lewis calls "metalinguistic" theoriesofcounterfactuals.

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

which oftengoes under the name "case study."'7In thisinstance,careful

I7I want to suggestthatcounterfactual reasoningmust underlieeffortsto inferor assess

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

The analysisin this sectionbears on currentmethodologicalissues in

21Cf. Jack Levy, "Domestic Politics and War," Journalof Interdisciplinary

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

where X does not occur, how can theyfindcauses thatdifferentiate be-

COUNTERFACTUAL ARGUMENT IN PRACTICE

The most controversialpoint made above is probablythat concerning

ences are drawn is, strictlyspeaking,impossible,since othercounterfactualcases must be

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

eral N > i examplesfromcomparativepolitics.Here analystsoftenmake

Some of the clearestexamples of the importanceof counterfactualargu-

counterfactual scenarios.One mightargue,forexample,thathad some independentvariable

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

The consequences of the cult of the offensiveare illuminatedby imagining

The use of counterfactuals is explicit and clear in Van Evera's analysis

as the resultof human choices.This entailssayingwhyotherpossiblechoiceswere not seen

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

altered the balance of power against them (for-example,Germany loses

Mueller proceeds to argue the counterfactualcase foreach of these "in-

32 See also Snyder'sresponseto Sagan's critiqueand Sagan's reply, Security9

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

tial escalation,he holds thattheircausal effecthas been redundant,due

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

macropolitical factors,which, in Stepan's view, were not "sufficient"

42 The distinction is similarto thatbetweenunderlyingcauses and specificor proximate

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

macropoliticalfactorswould seem to gain in importanceas causal fac-

N> I BUT STILL SMALL-N EXAMPLES

Researcherswith more than one actual case are not logicallycompelled

analysisto assess the impactof

presidentialas opposed to parliamentary systemson democraticregimebreakdownin South

This content downloaded from 171.65.249.4 on Tue, 29 Oct 2013 14:59:35 PM

support the claim thatany one of the proposed conditionshas a causal

46 Stepan,The Stateand Society: Peruin ComparativePerspective (Princeton:PrincetonUni-