5

The Comparative Method

David Collier

Comparison is a fundamental tool of analysis. Itsharpens our power of description, and plays a centralrole in concept-formation by bringing into focus .suggestive similarities and contrasts among cases.Comparison is routinely used in testing hypotheses, and itcan contribute to the inductive discovery of newhypotheses and to theory-building.

The forros of comparison employed in thediscipline of political science vary widely and includethose contained in statistical analysis, experimentalresearch, and historical studies. At the same time, thelabel "compara tive method" has a standard meaningwithin the discipline and in the social sciences morebroadly: it refers to the methodological issues that arisein the systematic analysis of a small number of cases, ora "small N."I This chapter examines altemativeperspectives on the comparative method that haveemerged over roughly the past two decades. Althoughthe primary focus is on discussions located in the fields ofcomparative politics and intemational studies, theapplícation of the comparative method is by no meansrestricted to those fields.

The decision to analyze only a few cases isstrongly influenced by the types of polítical phenomenaunder study and how they are conceptualized. Topics forwhich it is productive to examine relatively few casesinclude revolutions, particular types of national politicalregimes (e.g.,post-communist regimes), or particularforros of urban political systems. This focus on a smallnumber of cases is adopted because there exist relativelyfew instances of the pbenomenon under consideration thatexhibit the attributes of interest to the analyst.Alternatively, some analysts believe that politicalpbenomena in general are best understood tbrough tbecareful examination of a small number of cases. In thefield of comparative and international studies, tbe practiceof focusing on few cases bas achieved greater legitimacyin recent years in conjunction with the rise of the schoolof "comparative historical analysis," in which smallnumbers of countries are studied over long periods. Thisdose scrutiny of each country limits the number ofnational cases a scbolar can consider."

Choosing to study few cases routinely poses theproblem of having more rival explanations to assess thancases to observe, or tbe quandary of "many variables,small N" (Lijphart 1971, 686). Elementary statisticsteaches us that as the number of explanatory factorsapproaches the number of cases, the capacity toadjudicate among the explanations tbrough statisticalcomparison rapidly diminishes. This problem has stimu-lated much discussion of bow most productively toanalyze a small N.

The late 1960s and earIy 1970s saw a boom inwriting on comparative metbod (e.g., Merritt and Rokkan1966; Kalleberg 1966; Verba 1967; Smelser 1968;Lasswell 1968; Przeworski and Teune 1970; Sartori1970; Merritt 1970; Etzioni and Dubow 1970; Lijphart1971; Vallier 1971; Zelditch 1971; Armer and Grimshaw1973). This literature established a set of norros andpractices for small-N research, proposed altemativestrategies for conducting such analyses, and created abase line of understanding that bas played an importantrole in the ongoing practice of small-N studies. Thiscbapter assesses tbe issues of comparative method thatbave been debated in the intervening years and considerstheir implications for ongoing researcb. The point ofdeparture is Arend Lijphart's (1971) article "ComparativePolitics and Comparative Method." Among the studiespublished in that period, Lijphart's piece stands out for itsimaginative synthesis of basic issues of comparison andof the relation between comparative method and otherbranches of methodology.' It therefore provides abelpful framework for examining, and building upon,new developments in the field.

A central theme that emerges in tbe discussionbelow is that refinements in methods of small-N analysishave substantially broadened the range of techniquesavailable to comparative researchers. The most fruitfulapproach is eclectic, one in whicb scholars are willingand able to draw upon these diverse techniques.

106 The Comparative Method

Synopsis of Lijphart

Lijphart defines the comparative method as theanalysis of a small number of cases, entailing at least twoobservations, yet too few to permit the application ofconventional statistical analysis. A central goal of hisarticle is to assess the comparative method in relation tothree other methods-experimental, statistical, and case-study-and to evaluate these different approaches by twocriteria: 1) how well they achieve the goal of testingtheory through adjudicating among rival explanations, and2) how difficult it is to acquire the data needed to employeach method (see Figure 1).

The experimental method has the merit ofproviding strong criteria for eliminating rival explanationsthrough experimental control, but unfortunately it isimpossible to generate appropriate experimental data formost topics relevant to political analysis. The statisticalmethod has the merit of assessing rival explanationsthrough the weak:er but still valuable procedure ofstatistical control, but it is often not feasible to collect asufficient1y large set of reliable data to do this form ofanalysis.

The case-study method has the merit ofproviding a framework in which a scholar with modesttime and resources can generate what may potentially beuseful data on a particular case. Unfortunate1y,opportunities for systematically testing hypotheses are farmore limited than with the other methods. Yet Lijphart(pp. 691-93) insists that case studies do make acontribution to testing hypotheses and building theory,and he offers a suggestive typology of case studies basedon the nature of this contribution. He distinguishesamong atheoretical case studies; interpretative casestudies (that self-consciously use a theory to illuminate aparticular case); hypothesis-generating case studies;theory-confirming case studies; theory-infirming casestudies (that, although they cannot by themselvesdisconfirm a theory, can raise doubts about it); anddeviant case analyses (that seek to elaborate and refinetheory through a close examination of a case that departsfrom the predictions of an established theory). Lijphartemphasizes that "certain types of case studies can even beconsidered implicit parts of the comparative method" (p.691), and to the extent that the assessment of hypothesesdoes occur in some case studies, it is often because thecase studies are placed in an implicit or explicitcomparative framework. Yet even within thisframework, he emphasizes that findings from a singlecase should not be given much weight in the evaluation ofhypotheses and theory (p. 691).

The comparative method, as defined hy Lijphart,has an intermediate status in terms of both bis criteria. Itprovides a weak:er basis than the experimental or

statistical method for evaluating hypotheses, due to thelack of experimental control and the problem of manyvariables, small N. Yet it does offer a stronger basis forevaluating hypotheses than do case studies. Despite theconstraint of addressing more variables than cases, thecomparative method allows systematic comparison that, ifappropriately utilized, can contribute to adjudicatingamong rival explanations.

Although the data requirements of thecomparative method may be much greater than for casestudies, Lijphart argues that they are less demanding thanfor experimental or statistical research. He thereforeviews the comparative method as most appropriate inresearch based on modest resources, and he suggests thatstudies using the comparative method might often serveas a first step toward statistical analysis.

If at all possible one should generally use thestatistical (or perhaps even the experimental)method instead of the weaker comparativemethod. But often, given the inevitablescarcity of time, energy, and financialresources, the intensive comparative analysisof a few cases may be more promising thana more superficial statistical analysis ofmany cases. In such a situation, the mostfruitful approach would be to regard thecomparative analysis as the first stage ofresearch, in which hypotheses are carefullyformulated, and the statistical analysis as thesecond stage, in which these hypotheses aretested in as large a sample as possible.(1971, 685)

Lijphart also proposes solutions to both sides ofthe problem of many variables, small N (1971, 686 ft).With regard to the small number of cases, even ifresearchers stop short of a statistical study, they cannonetheless try to increase the number of cases used inassessing hypotheses. With regard to the large number ofvariables, he suggests two approaches. First, analystscan foeus on "comparable cases," that is, on cases thata) are matched on many variables that are not central tothe study, thus in effect "controlling" for these variables;and b) differ in terms of the key variables tbat are thefoeus of analysis, thereby allowing a more adequateassessment of their influence. Hence, the selection ofcases acts as a partial substitute for statistical orexperimental control. Second, analysts can reduce thenumber of variables either by combining variables in asingle scale or through theoretical parsimony, that is,through developing a theory that focuses on a smallernumber of explanatory factors.

Thus, Lijphart provides a compact formulationof the relationship between the comparative method and

Figure 1. Situating the Comparative Method as af 1971: Lijphart's ScbemeCase Study Method Comparatlve Method Experimental Method

i Merlt: Permits inten-

Isive examination 01cases even with lim-ited resources

Inherent Problem:Contributes less tobuilding theory thanstudies with morecases

Types 01 Case Stu-

I dles:

1

1. Atheoretical2. Interpretive

1

3. Hypothesis-generating

14. Theory-conlirmingI 5. Theory-inlirmingI (i.e., case studiesI that weaken aI theory marginally)

i 6. Deviant case studies 1

Deflned as: Syste-matic analysis 01 smallnumber 01 cases("small-N" analysis)

Merit: "Given inevit-able scarcity 01 time,energy, and financialresources, the inten-sive analysis 01 a fewcases may be morepromising than thesuperficial st~tisticalanalysis 01 manycases" (Lijphart, p.685)

Inherent Problem:Weak capacity to 5Or!out rival explanations,specificaily, the prob-lem of "many vari-ables, few cases"

Potential Solutions:1. Increase number of

cases2. Focus on com par-

able cases3. Reduce number of

variablesa. Combine vari-

ablesb. Employ more par-

simonious theory

Merit: Eliminates rivalexplanations throughexperimental control

Inherent Problem:Experimental control isimpossible for many ormos! topics ofrelevance to field ofcomparative politics

Statistical Method

Merit: Assesses rivalexplanations throuahstatistical control

Inherent Probl.m:Difficult to collect ade-quate information in asufficient number ofcases, due to limitedtime and resources

108 The Comparative Method

other methodologies, and he offers solutions to thecharacteristic dilemmas of the comparative method.

Further Perspectives on SmalI-NAnalysis

The two decades following Lijphart's study haveseen the emergence of new perspectives on small-Nanalysis, as well as a renewed focus on methodologicalalternatives already available before he wrote his article.Though many of these innovations appear in workexplicitly concerned with the comparative method,conventionally understood, others appear in writing onthe experimental, statistical, and case-study methods.The result has been an intellectual cross-fertilization ofgreat benefit to the comparative method. Figure 2provides an overview of these innovations.

Innovations in the Comparative Method

Innovations in the comparative method can bediscussed in terms of the issues introduced above, encom-passing the goals of comparison, the justification forfocusing on few cases, and the problem of manyvariables, small N.

Goals of Comparison

A central and legitimate goal of comparativeanalysis is assessing rival explanations. However, asTheda Skocpol and Margaret Somers (1980) argue,comparative studies should be understood not merely interms of this single goal, but in terms of three distinct,yet ultimately connected, goals." The first is thatconsidered above: the systematic examination ofcovariation among cases for the purpose of causalanalysis. S The second is the examination of a number ofcases with the goal of showing that a particular model orset of concepts usefully illuminates these cases. No realtest of the theory occurs, but rather the goal is theparallel demonstratíon of theory, This use of comparisonplays an important role in the process through whichtheories are developed. The third type of comparison isthe examination of two or more cases in order tohighlight how different they are, thus establishing aframework for interpreting how parallel processes ofchange are played out in different ways within eachcontext. This contrast of contexts is central to the more"interpretive" side of the social sciences and reflects yetanother way that comparison is frequently used.

In addition to providing a more multifacetedaccount of the goals of comparison, Skocpol and Somerssuggest the intriguing idea of a research "cycle" among

these approaches (pp. 196-197). This cycle arises inresponse to the problems that emerge as scholars pusheach approach up to - or beyond -- the limits of itsusefulness. For example, a "parallel demonstration"scholar might introduce a new theory and show how itapplies to many cases. "Hypothesis-testing" scholars,wanting to specify the conditions under which the theorydoes not hold, could make further comparisons with thegoal of discovering these conditions. Hypothesis-testingstudies that too brashly compare cases that are profoundlydifferent might, in turn, stimulate "contrast of contexts"scholars to examine more carefully the meaning of thedifferences among the cases. It is thus useful to lookbeyond an exclusive focus on the role of comparison inbroad causal analysis, to an understanding thatencompasses the different elements in this research cycle.

This is not to say that assessing hypotheses doesnot remain a paramount goal of comparison, and manyscholars insist that it is the paramount goal. Yet thisbroader perspective offers a valuable account of howcomparative work proceeds within a larger researchcommunity, pointing usefully to the interaction amongdifferent goals of comparison.

Iusttfication for Small N

A second trend is toward a more elaboratejustification of a focus on relatively few cases. Lijphart'srationale seems in retrospect rather modest, in that itemphasizes only the problem of inadequate resources andtreats the small-N comparison as a way station on theroute to more sophisticated statistical analysis.

A very different defense of working either with asmall N or with case studies had previously beenavailable in arguments favoring a "configurative"approach (Heckscher 1957,46-51, 85-107), and thisperspective was elaborated a few years before thepublication of Lijphart's analysis in Sidney Verba's(1967) review essay advocating the "disciplinedconfigurative approach." In evaluating Robert A. Dahl'sPolitical Oppositions in Western Democracies (1966),Verba points both to the sophistication of the hypothesesentertained in the book and to the difficulty of assessingthem adequately, except through a close command of thecases, leading him to advocate this disciplinedconfigurative mode of research. Verba's formulation isappealing because he is concemed with systematichypothesis testing and theory building. At the same time,he link.s this priority with a more explicit appreciation ofthe difficulty of testing hypotheses adequately and thevalue of proper1y executed case studies in providingsubtle assessments of hypotheses.

It might be claimed that the difficulty ofadequately testing hypotheses ultimately derives from the

Agura 2. Innovations Relevant to the Comparatíve MetbodComparetlva Mathod Experimental Mathod

Broadened Understanding of Types of Dlffuslon of Oldar Ideas and Introduc-Comparativa Studlea tlon of New ldeas on Quaai-1. Emphasis on interpretive understanding Experimental Deslgn

2. Idea of a "research cycle" among the 1. Methodology of quasi-experiments andtypes (Skocpol and Somers) interrupted tíme-senes design becomes

Further Justlficatlons for Focua on a more widely known

Small-N 2. Analysis of the Connecticut crackdown

Case Study Method 1. To pursue "disciplined configurative on speeding as exemplar of interrupted

approach" (Verba. reinforced by ~ time-serias analysis (Campbell and

New Perspectives on Case Studles Almond and Genco) Ross)

1. Ne;;; dõ:ensc vi lll~\;Cl';~ :»lUUy rnethoo 2. To avoid premem of "conceptuat 3. Dit!usio~ cf ideae abcut quasi-

(Campbelll stretchinq" ~S!!!'!ori) experiments encouraged by codification

2. Refinements in lijphart's typology of 3. To facilitate "thick description" andof evaluation research

case studies (Eckstein, George)1,..;3'

olher forms of inlerprelive understand- 4. Proposed statistical solution to problem

3. Improvemenl of causal analysis in case ing (Geertz and many others) of selection bias in quasi-experiments

sludies through "process tracing" 4. To achieve analytic depth of case-(Achen)

(George and McKeown) oriented approach (Ragin) Statistlcal Method4. Crilique of value of case sludies in Debates on Solutions to Problem of

assessing rational choice theory (Achen Many Varlables, Small-N New Wamlngs and New SoIutlonaand Snidal) 1. Value of increasing number of cases 1. Critieism of standard statistical practiee

2. Comparable cases versus contrasting ~in the social scienees (Freedman)

cases (Lijphart versus Przeworski and 2. New statistical techniques relevant toTeune) small-N analysis

3. Reducing number of variables in con- 3. Effort to refine statistical analysis of ajunct~on with using stronger theory small-N in the debate on corporatism

and economie growth in WestemEuropa (Lang&-Garren.Jaekman-Hieks-Panerson)

~~--------------------------------~---------------------

110 The Comparative Method

problem of limited resources discussed by Lijphart. Ifenough talented researchers worked long and hard, theycould carry out a Political Oppositions study for manydozens of countries. Yet the problem bere is somewbatdifferent from that empbasized in Lijpbart's initial formu-lation. It is not so mucb that resources are limited, butthat constructing adequate comparisons has proved moredifficult than had often been thought in the 1960s andearly 1970s, in the initial days of enthusiasm forcomparative statistical researcb. Among thesedifficulties, that of the valid application of conceptsacross diverse contexts has been especially vexing.

Within the literature on comparative method, akey step in elucidating these problems of validity, andthereby strengthening the justification for a small N, isGiovanni Sartori's (1970) classic discussion of "ConceptMisformation in Comparative Politics, " the basic themesof which are elaborated in his later book Social ScienceConcepts (1984). Sartori suggests that the application ofa concept to a broader range of cases can lead toconceptual "stretching," as some of the meaningsassociated with the concept fail to fit the new cases. Theconcepts that can most easily be applied to a broad rangeof cases are often so general that they do not bring intofocus the similarities and contrasts among cases that areessential building blocks in wortbwhile comparativeanalysis. Consequently, a study focused on concepts thatare carefully adapted to this "finer slicing" of a given setof cases sbould be extended to otber cases only with greatcaution. From this perspective, it may be argued that themost interesting studies will often be those tbat focus on asmaller number of cases.

With regard to the problems of increasing thenumber of cases under study," Adam Przeworski andHenry Teune's The Logic of Comparative Social Inquiry(1970) is a major source of insight. Although they arguethat achieving a higb leveI of generality should be a basicgoal of social science, their framework is centrallyconcemed with the difficulties that can arise ingeneralizing beyond an initial set of cases. With regardto problems of validity, they advocate tbe use, whennecessary, of "system-specific" indicators that serve tooperationalize the same concept in distinct ways indifferent contexts (pp. 124-130). For the scholar seekingto move toward a larger set of cases, the potential needfor system-specific indicators necessitates the closeexamination of every new case.

Przeworski and Teune also address the problemthat as the analyst incorporates more cases into a study,distinct causal patterns may appear in the new cases. Todeal witb this problem, Przeworski and Teune advocate"replacing proper names" of social systems by identifyingthose systems in terms of the explanatory factors thataccount for why causal relations take a particular form

within each system (pp. 26-30).7 This approach makesthe invaluable contribution of providing a theoretical,rather than an idiosyncratic and case-specific, basis foranalyzing differences in causal patterns. However,extending an analysis to additional cases on the basis ofthis procedure again requires a painstaking assessment ofeach new contexto Thus, Przeworski and Teune providea valuable tool for adequately analyzing a larger numberof cases, but their approach again shows that this must bedone with caution.

Since 1970, the renewal of a Weberian concemwith interpretive understanding, i.e., with deciphering themeaning of behavior and institutions to the actorsinvolved, bas also strengthened the justificationforadvancing cautiously with one or very few cases.Clifford Geertz's (1973) label "thick description" iscommonly evoked to refer to this concern," and thisfocus has appeared in many guises relevant to politicalresearch, including Gabriel Almond and Stephen J.Genco's analysis of "Clouds and Clocks" (1977) andSkocpol and Somers' "contrast of contexts" approach,which encompasses studies that use comparison to richlycontextualize research findings. Charles C. Ragin's TheComparative Method (1987) explores another facet of thisconcem in his analysis of the "holistic" orientation ofwhat he calls "case-oriented" research and the complexproblems of "conjunctural causation" - that is, causalpatterns that vary according to the context -- to whichconfigurative scholars are typically far more sensitive.

Finally, the intellectual success in recent years ofthe school of comparative historical analysis has playedan important role in legitimating a focus on a small N.This approach was pioneered in works such as ReinhardBendix (1964), Barrington Moore (1966), and Lipset andRokkan (1967), and more recent works include Rokkan(1970), Tilly (1975), Paige (1975), Bendix (1978),Trimberger (1978), Skocpol (1979), Bergquist (1986),Luebbert (1991), Goldstone (1991), CoIlier and CoIlier(1991), and Rueschemeyer, Stephens, and Stephens(1992). Methodological statements focused on thistradition include Skocpol and Somers (1980), Skocpol(1984), Tilly (1984), and Ragin (1987). .

The particular form of analysis in these studiesvaries considerably, as suggested by Skocpol and Somers'typology noted above. In varying combinations, thesestudies employ both rigorous qualitative comparisons thatextend across a number of nations, and also historicalanalysis that often evaluates each national case over anumber of time periods.? This tradition of research thuscombines well-thought-out comparison with anappreciation of historical context, thereby contributing toan effort to "historicize" the social sciences.

Althougb the uses of comparison in this literatureare diverse, as Skocpol and Somers emphasized, it may

/;--------------------~--~~~--~~--------------~---------------------------------~

be argued that a major consequence of the growingimportance of comparative historical studies is to furtherlegitimate the approach that was Lijphart's originalconcem: the assessment of rival explanations, based onsystematic, qualitative comparison of a small number ofcases. In light of a spectrum of studies from BarringtonMoore's (1966) pioneering analysis of the emergence ofaltemative forms of modem regimes, to Skocpol's (1979)study of revolution, to Luebbert's (1991) analysis of theemergence of liberalism, fascism, and social democraeyin interwar Europe, it is evident that this literature hasgiven new legitimacy to the use of broad historicalcomparison for systematic causal analysis. Efforts tocodify procedures for assessing hypotheses in this type ofanalysis, such as that in Ragin's Comparative Method(1987), further reinforce the plausibility of insisting onthe viability of small-N analysis as a middle groundbetween case studies and statistical studies.

Solutions 10 lhe Problem of Many Variables, Small N

The evolving debates on comparative methodhave suggested further refinements in Lijphart's originalthree solutions for the problem of many variables, smallN, i.e.: 1) increasing the number of cases, 2) foeusingon matched cases, and 3) reducing the number ofvariables.

1. Increase lhe Number of Cases At the timeLijphart wrote, it was believed in some eircles thatcomparative social science would increasingly be orientedtoward large-N comparative studies, based on extensivequantitative data sets and rigorous statistical analysis.Today there can be no question that, for better or worse,quantitative cross-national research in the subfield ofcomparative politics, and quantitative intemational politicsin the subfield of intemational relations, have not come tooccupy as dominant a position as many had expected.Within these two subfields, they hold the status of oneapproach among many.

Various factors have placed limits on the successof large-N research based on quantitative data sets,among which is certainly the renewed concem withclosely contextualized analysis and interpretive studies.Broad quantitative eomparison may have been set back asmany scholars discovered how extraordinarilytime-consuming it is to construct appropriate data sets,often out of proportion to the professional rewards thatseem to be forthcoming. This is particularly a problemwhen the focus of analysis extends beyond the advancedindustrial countries to regions for which it is oftenextremely difficult to develop reliable data. In addition,the quantitative-comparative approaeh has probably beenhurt by the publication of too many studies in whiehconcepts are operationalized with dubious validity and

i

Collier 111

which employ causal tests that are weak, unconvincing,or inappropriate (Ragin 1987, chap. 4).

Yet the faet that broad quantitative comparisonhas not become a predominant approach should not leadscholars to overlook what has been accomplished. RobertJacleman (1985) insists that comparative statisticalresearch has had more success than is recognized, andLijphart's own recent work moves in this direction(1990). The failure to seize good opportunities to doquantitative research could certainly be viewed as beingas much of a mistake as premature quantification, and thefruitful debate on corporatism and economic growth inWestem Europe discussed below is one of manyexamples of how statistical methods can effectivelyaddress interesting analytic issues. Further, theavailability of new statistical techniques (also discussedbelow) has made it far more productive to do quantitativeanalyses with as fewas ten to fifteen cases. Conse-quently, the option of increasing the "N" at least to thatleveI is still worth pursuing, and it should probably bepursued more often.

2. Focus on Comparable Cases Therecommendation that analysts focus on carefully matchedcases has been both reinforced and challenged. In adiscussion published in the mid-1970s, Lijphart (1975)explores further the trade-off he noted in 1971 betweenthe goal of increasing the number of cases and the goal ofmatching cases as a substitute for statistical control.Obviously, if a researcher is to select cases that are reallysimilar, however that similarity is defined, the number ofappropriate cases is likely to become limited. In the faceof this trade-off, Lijphart opts in favor of the morecareful matching of fewer cases, and he goes so far as torestriet the application of the term "comparative method"to analyses that focus on a smalI number of earefullymatched cases. This emphasis parallels a much earlierperspective on the comparative method referred to as themethod of "controlled comparison" (Eggan 1954).Arthur Stinchcombe's (1978) advocacy of themethodology of "deep analogy," i.e., the comparativeanalysis of very few, extremely closely matched, casespushes this approach even further.

A contrasting strategy is advocated byPrzeworski and Teune (1970, 32-39) and Przeworski(1987, 38-41). They suggest that even with carefulmatching of cases in what they label a "most similar"systems design, there remains a problem of"overdetermination," in that this design fails to eliminatemany rival explanations, leaving the researcher with noeriteria for ehoosing among them. They prefer instead a"most different" systems design, based on a set of caseswhich are highly diverse and among which the analysttraces similar processes of change." Przeworskisuggests that the strength of this design is in part

112 The Comparative Method

responsible for important advances in the literature ondemocratization, such as the work of O'Donnell,Schmitter, and Whitehead (1986). Przeworski maintainsthat this literature addresses such a broad range of casesthat analysts are forced to distill out of that diversity a setof common elements that prove to have great explanatorypower.!'

This discussion can be placed in perspeetive byrecognizing that cases that are closely matched from onepoint of view may contrast sharply from another. Myown recent work (Collier and Collier 1991) combines thetwo strategies by starting with a set of eight LatinAmerican countries that are roughly matched on anumber of broad dimensions. Among the eight countries,the analysis focuses on pairs of countries that arenonetheless markedly different. The overall matchingassures that the contexts of analysis are analyticallyequivalent, at least to a significant degree, and the pairedcomparison places parallel processes of change in sharprelief because they are operating in settings that are verydifferent in many respects.

In conjunction with the debate over the merits ofmost similar and most different systems designs, it isimportant to reeognize that in many studies, theconclusions reached in the overall comparison of casesare also assessed -- implicitly and sometimes explicitly --through within-case analysis. In the section on casestudies below, the discussion of "pattem matching" and"process tracing" suggests some of the forms this takes,It is no coincidence that within the school of comparativehistorical analysis, findings are often reported in books,rather than articles. Part of the reason is that thepresentation of detailed information 00. each case servesto further validate the conclusions drawn fromcomparisons across cases.

These within-case comparisons are critical to theviability of small-N analysis. As Stanley Lieberson(1991, 312-315) has correetly insisted, taken bythemselves, comparisons across a small number of cases,using either a most similar or a most different systemsdesign, provide a weak basis for causal inference.However, if one considers the role of these intemalcomparisons, the "N" is substantiaUy increased, therebystrengthening causal analysis.P

This use of within-case comparison can also helpproteet the analyst from a problem that arises in the mostdifferent systems design, in which countries are matchedon the dependent variable and differ in terms of a seriesof background variables. Barbara Geddes (1990) hasshown that if cases are seleeted on the basis of scores onthe dependent variable, which is how most differentsystems designs are often carried out, the lack of varianceon the outcome to be explained introduces a "selectionbias" that can greatly weaken causal inference. One way

of mitigating this problem is to introduce greater varia-bility through intemal comparison.

The ongoing debate on most similar versus mostdifferent systems designs has implications for the statusof area studies. Dankwart Rustow (1968) argued sometime ago in favor of moving beyond an area studiesapproach, and many scholars agree that cases should beseleeted in response to the analytic requirements ofparticular research projeets, rather than 00 the basis of ageographic proximity that at best is often a poorsubstitute for the analytic matching of cases. Reeent"cross-area" studies 00 successful export-led growth and00 democracy suggest that this altemative perspective isgaining ground."

However, the area studies approach is a boomingbusiness today for a variety of reasons, including theimpressive funding of area studies by U.S. foundations inrecent decades, as well as institutiooal momentum. Infact, from the point of view of the theoretically orientedsmall-N comparativist, this is not a bad outcome. Thecountry case studies produced by area speeialists arecrucial building blocks in most comparative work, andwithout them cross-area studies would be on far weakerground. It is essential to recognize that these case studiesbenefit greatly from the intellectualleverage gained whenindividual scholars develop, over many years, acumulative and well-contextualized understanding of aparticular region. Particularly in light of currentconcerns that broad comparative studies should beattentive to the cootext of analysis, the contribution ofarea specialists is essential.

3. Reduce the Number of Variables: The thirdsolution to the small-N problem is to reduce the numberof explanatory factors, either through combiningvariables, sometimes referred to as "data reduction," orthrough employing a theoretical perspective tbat focuses00 a smaller set of explanatory factors. One of thepromising sources of parsimonious explanatory theory isthe "rational choice" approach that has gained increasingattention amoog political scientists. Rational choicemodeling offersa productive means of simplifyingarguments that contain a multitude of interestingvariables, but that may fail to specify ~e most criticalones. Within the field of comparative analysis, Geddes's(1991) study of administrative reform in Latin America,which models the impact of different eleetoral and partysystems 00 the incentives of legislators to adopt reform,provides ao excellent example of a productivesimplification of a complex topic. As such models gainincreasing acceptance in the comparative field, analystswill acquire a useful tool for addressing the small-Nproblem."

More work on concept formation is also needed,ootwithstanding the sustained contribution of Sartori

(1970, 1984, 1991, 1993, and Sartori, Riggs, and Teune1975); the work of authors such as McKinney (1966),Kalleberg (1966), and DeFelice (1980); and also Burger's(1976) invaluable synthesis of the Weberian approach toconcept formation. Comparativists do not devote enoughattention to thinking through how well or poorly conceptsare serving them and therefore may have insufficientground for knowing whether they are making appropriatechoices in the effort to achieve theoretical parsimony.

The field of cognitive science has recentlyprovided insights into categorization that may be useful inrefining the concepts employed in comparative studies.The application of these insights is illustrated by GeorgeLakoff's (1987) challenge to frameworks, such as that ofSartori, that employ what Lakoff calls "classicalcategorization, " in which the meaning of concepts isunderstood in terms of defining characteristics that areseen as giving the concepts well-defined boundaries.This understanding is crucial to Sartori's framework, inthat the problem of conceptual stretching which heanalyzes hinges on these boundaries. Cognitive scientistsargue that in ordinary language, the meaning of conceptsderives not from defining characteristics, but from animplicit "cognitive modelo that underlies the concept andfrom "exemplar" cases that serve to anchor the concept'smeaning and provide a point of reference for identifyingbetter and worse cases. This perspective provides adifferent view of the question of boundaries, and hence ofconceptual stretching. More work is needed to discovertbe degree to whicb tbese patterns in ordinary languageare also present in social science usage, and if so, theimplications for the use of concepts in"comparativeanalysis (see Collier and Mabon 1993).

Innovations Suggested by Work on OtherMethods

Experimental Method

Although the experimental metbod itself may beof little relevance to the topics addressed in mostcomparative research, ideas derived from theexperimental metbod can improve small-N studies.Donald Campbell and Julian Stanley's classicExperimental and Quasi-Experimental Designs forResearch (1963) sbows bow tbe logic of experimentaldesign can be applied to "quasi-experiments," that is, to"observational" studies that include some event or innova-tion tbat bas a forro analogous to an experimentalintervention, but that occurs in a "natural" setting. Anexample would be the initiation of a new public policywhose impact one wishes to assess.

Campbell and Stanley underline the great valuein quasi-experiments of the "interrupted time series"

Collier 113

designo In this design the analyst looks at a long seriesof observations over time, so that the values of theobserved variable are examined not only at two pointsimmediately before and after the policy cbange or otherinnovation (which "interrupts" the series), but also wellbefore and well after. To illustrate the risk of restrietingthe analysis to these two observations, the autbors presentseveral hypothetical configurations of data in whichrestricting the analysis to two observations leads to afinding of sbarp discontinuity, whereas the full timeseries reveals continuity. Causal inferences about theimpact of discrete events can be risky without anextended series of observations. Comparativistsemploying small-N analysis must heed this warning, sincethey routinely analyze the impact of discrete events,ranging from wars, revolutions, and military eoups tospecific public policies.

Donald Campbell and Laurence Ross's (1968) subse-quent analysis of the impact on traffie fatalities of theConnectieut craekdown on ~in in the 1950s providesa stunning "exemplar" of the imaginative application of aquasi-experimental design to public policy analysis.Indeed, Przeworski (1987, 31) bas argued thatmethodology is influenced far more by exemplars than byformal attempts to "legislate" correct methods, and theConnectieut crackdown article bas certainly played thatrole."

The case appears to be a simple one. When theState of Connectieut initiated striet enforeement of thevehieular speed limit in the 1950s and traffie deathsdropped sbarply, the cause and effect relationship seemedobvious. Yet in evaluating this causal link, Campbell andRoss do an impressive analysis of potential threats to its"internal validity" (was that really tbe cause inConnectieut?) and its "external validity" (ean tbe findingbe generalized?). No sensitive analyst can read thisarticle without aequiring a more sober view of tbeproblems of evaluating poliey impaets.

Ideas about quasi-experimental and interruptedtime series design bave also heen disseminated throughthe large body of writing on evaluation researeh. Thisincludes studies that apply these ideas to tbe analysis ofpolitical development (Hoole 1978), as well as excellenttreatments of experimental design and evaluation researchin introductory textbooks on social science metbodology,such as Babbie (1992).

Although much writing on quasi-experimentsappears to offer helpful guidance and practical advice tosmall-N analysts, Christopher H. Achen's The StatisticalAnalysis of Quasi-Experiments (1986) may leave themfeeling that the methodological challenges posed by thistype of design are overwhelming. In studies of theimpact of public poliey, the core problem is the laek of"randomization" in the application of the policy, which

114 The Comparative Method

may result in selection bias. For example, the benetits ofa policy are commonly received by some groups and notby others, on the basis of certain attributes possessed bythe groups, and it is possible that these prior attributeswill themselves reinforce the outcomes that the policyseeks to promote. In the absence of true experimentaldata, this poses the challenge of disentangling the impactof the policy from the impact of these prior attributes.This causal riddle can be addressed by constructing amodel of how citizens are selected to be recipients of thepolicy. This model then becomes a building block in theanalysis of the policy's impact, in that these priorconsiderations can be "factored out" in assessing thepolicy. Achen shows that solving the riddle requires acomplex form of "two-stage" statistical analysis.

The implications of Achen's book may be~ g{scouraging for analysts working with a small number ofI -----;;cases. An adequate solution to the lack of randomization

requires a form of statisticaI analysis which can beapplied to an elaborate quantitative data set, but thistechnique would be hard to apply in a small-N study. Amore hopeful view might be that the literature onexperiments and quasi-experiments at least providesuseful warnings about the perils of analyzing discreteevents as if they were true experimental interventions. Inthe absence of appropriate data sets, the researcher mustexercise caution in making causal claims.

Innovations in Statistics

Recent work on statistical analysis has providedboth new warnings about the risks of statistical studiesand new opportunities for doing meaningful statisticalwork with relatively modest case bases. The statisticianDavid Freedman has launched a major assault on the useof multivariate quantitative analysis in the social sciences(1987, 1991), which he claims fails because theunderlying research design is generally inadequate andbecause the data employed fail to meet the assumptions ofthe statistical techniques. Ris criticism may bringconsiderable satisfaction to those who have been skepticalabout statistics all along and who take comfort in thegreater "control" of the material they feel derives fromanalyzing relatively few cases through more qualitativetechniques. It is realistic to expect that we may gothrough a period of greater questioning of the use ofstatistics in the social sciences. However, as with therejection of quantitative cross-national research discussedabove, it would be unfortunate if a reaction againstquantitative studies went too faro

The emergence of new statistical techniques thatare helpful in the analysis of relatively few cases makessuch a blanket rejection unwarranted. One example is thedevelopment of "resampling strategies" such as the"bootstrap" and "jackknife" (Diaconis and Efron 1983,

Mooney and Duval 1992). These techniques usecomputer simulation to create, from an initial set of realdata, a large number of hypothetical replications of thestudy, which can then be used in statistical tests that arenot as vulnerable to violations of distributional assump-tions as are more conventional tests. These techniquesmay be especiaUy useful when there is greatheterogeneity among units, as may readily occur incross-national comparisons.

The development of "robust" and "resistant"statistical measures (Hampel et aI. 1987; Hartwig 1979;MosteUer and Tukey 1977) is promising in much thesame way. These measures are relatively unaffected byextreme or deviant values and can therefore helpovercome the problem in small-N analysis that findingsmay be seriously distorted by a single observation that isgreatly in error.

Another set of techniques concemed with thissame problem is "regression diagnostics" (Bollen andJackman 1985; Jackman 1987). These are tests used inconjunction with conventional regression analysis toassess whether unusual values on particular observations,called influential cases, have distorted the findings. Theadvantage of regressiori. diagnostics in comparison withrobust and resistant statistics is that one can employ themwith the more familiar coefficients associated withregression analysis.

The use of regression diagnostics is nicelyillustrated in the recent debate on the relationship betweencorporatism and economic growth in 15 WestemEuropean countries (Lange and Garrett 1985, 1987;Jackman 1987, 1989; Hicks 1988; Hicks and Patterson1989; Garrett and Lange 1989). The starting point ofthis debate is Peter Lange and Geoffrey Garrett's 1985article, which presents an interesting and complex idea ina simple formo They argue that the organizationalstrength of unions in the labor market and the politicalstrength of the left in the electoral and govemmentalarenas both have an impact on economic growth, but thatthis impact is shaped by a complex interplay betweenthese two factors, which they represent through an"interaction" term in their regression analysis of the 15cases.

In a reanalysis of their artic1e, Robert W.Jackman (1987) employs regression diagnostics toexamine certain influential cases that he believes distorttheir findings. In the ensuing discussions among thesetive authors, an expanded model with further controlvariables is proposed, this expanded model is bothchallenged and defended, and Lange and Garrettsubsequently defend their original model and caU for newdata and further tests.

This scholarly debate brings together animportant substantive problem, a high Ievel of areaexpertise and knowledge of specific cases, the inventive

i

use of a relatively straightforward statistical model, aconstroctive critique based on regression diagnostics, anda sustained process of cumulative knowledge generationbased on the scrotiny of a shared data set. Just as theCampbell and Ross artic1e on the Connecticut speedingcrackdown is an exemplar of a quasi-experimental design,this debate should stand as an exemplar of amethodologically sophisticated effort by several scholarsto solve an important problem within the framework ofsmall-N quantitative analysis. This debate also showsthat although an "N" of 15 might often be approachedthrough qualitative small-N comparison, it can likewisebe subjected to statistical analysis, with interestingresults.

Another area in which potential problems ofstatistical analysis are amenable to solution concems theissue of •average effects" in regression studies. Theresults of the simpler forms of regression analysis arebased on an average of the strength of causal relationsacross the cases being studied. For the coefficientsproduced by regression analysis to be meaningful, it isnecessary that these causal relations be homogeneousacross the cases. Yet Ragin (1987, chap. 4), amongothers, has forcefully argued that this assumptioncommonIy does not hold, given the complex forms of"multiple conjunctural causation" often encountered incomparative studies. In different contexts of analysis, theinteraction among causal factors may vary.

However, solutions to this problem are available.John E. Jackson (1992) shows how it can be addressedwith advanced statistical techniques, and the interactionterm in the Lange-Garrett regression analysis, discussedabove, deals with precisely this problem: that the effectof one explanatory factor varies depending on the valueof another explanatory factor. Finally, Przeworski andTeune's procedure of "replacing proper names," alsodiscussed above, takes this problem of causal complexityand tums it into an opportunity to deal more theoreticallywith the diversity of causal patterns.

Innovations tn lhe Case-Study Method

When Lijphart wrote his 1971 artic1e, heapparently felt some hesitation about including adiscussion of case studies in an assessment of thecomparative method." Yet the two topics are closelylinked, and his helpful typology of the uses of casestudies in hypothesis testing and theory building set thestage for refinements in case study analysis laterintroduced by other scholars.

One of the most suggestive discussions of thecase-study method is that of Campbell (1975). Hedramatically recants the bold assertion he made in hisearlier book with StanIey that "one-shot" case studies are

Collier 115

"of almost no scientific value" (1963, 7). He showsinstead that case studies are the basis of most comparativeresearch, that they offer many more opportunities than isoften recognized for falsifying the researcher's mainhypotheses, and that much can be leamed from makingexplicit the comparisons that are often implicitly built intocase studies. For example, any given hypothesis about acase has implications for many facets of the case.Campbell uses the label "pattem matching" to refer to theprocess of discovering whether these implications arerealized. The analyst can thereby increase the "N" bymultiplying the opportunities to test hypotheses withinwhat may initially have been viewed as a "single" case.

This procedure of pattem matching is helpful inaddressing the long-standing concem that case studies areuseful for generating hypotheses, but that the same casecannot then be used to test the hypothesis because itoffers no possibility of disconfirmation. This issometimes referred to as the problem of ex post factohypotheses.'? The procedure of pattem matching opensthe possibility that an hypothesis initially generated by aparticular case could subsequently fail to be supported bythe same case. Thus, the problem of ex post factohypotheses can be partially overcome."

Harry Eckstein (1975, 113-123) is likewiseconcemed with testing, as opposed to generating,hypotheses in case-study analysis, and he arguesforcefully that many analysts have greatly underestimatedthe value of case studies for hypothesis testing. Inparticular, the carefully constrocted analysis of a "criticalcase" -- for example, one about which the analyst hasparticularly strong expectations that it will fit thehypothesized causal pattem -- can provide an invaluableopportunity to falsify the relevant hypothesis.

Alexander George and Timothy McKeown(1985), building on George (1979), present a helpfulsynthesis of two key building blocks in the processthrough which hypotheses are tested in case studies. Thefirst corresponds to the conventional approach to placinga case in comparative perspective, which they call the"congroence procedure." The scholar examines thevalues of an hypothesized independent and dependentvariable for a given case and determines, in light ofexplicit or implicit comparison with other cases, whetherthese values are consistent with the predictions of thehypothesis under consideration (pp. 29-30). The secondis "process-tracing," through which the researcherengages in a dose processual analysis of the unfolding ofevents over time within the case (pp. 34-41). The goal isto assesses whether the dynamics of change within eachcase plausibly reflect the same causal pattem suggestedby the comparative appraisal of the case in relation toother cases. Process tracing may be seen as a specificinstance of Campbell's pattem matching, and as with

116 The Comparative Method

pattem matching the analyst makes a series of within-caseobservations against which the hypothesis can be furtherassessed.

Overall, these artic1es, along with works such asRobert K. Yin's Case Study Research (1984), offer asystematization of case-study procedures that provi de avaluable point of reference for scholars concemed withsmall-N analysis. At the same time, the debate continueson the proper role of case studies in assessing andbuilding theory. An interesting part of this debate,published as a special issue of the joumal World Politics(1989) focuses on the contribution of case studies toevaluating one application of rational choice analysis,i.e., rational deterrence theory in intemational relations.The opening artic1e by Achen and Snidal (1989) arguesthat the case studies employed by many intemationalrelations specialists do not adequately address the centralideas of this body of theory, thereby raising an issueperhaps not often enough considered in discussions of thecomparative method: How can the methodologicalconcem with executing good comparisons be linked to thekey analytic issues posed by the particular theories thatare to be evaluated? Achen and Snidal also note theproblem of selection bias in case studies of deterrencetheory, that is, the problem that case studies usualIy focuson deterrence failure, whereas much or most of the timedeterrence works. The issue of the joumal inc\udes aseries of artic\es by scholars c\ose to the case-studytradition who debate the issues raised by Achen andSnidal. These artic\es constitute a valuable effort to think-lhrough how case studies have functioned in relation tothe assessment of a particular body of theory, a line ofinquiry that should be taken up more often.

In this debate on deterrence theory, anintellectual tension emerges that has been a recurringtheme in this chapter: between analyses that seek toachieve a generic understanding, based on relatively fewvariables and encompassing many cases, as opposed toanalyses that seek to draw out the complexities ofparticular cases.

Conclusion

Among the diverse approaches discussed in thischapter, three major analytic altematives stand out.First, new perspectives on the case-study method havestrengthened the viability of that approach. Discussionsof opportunities for within-case comparisons have in factbegun to blur the distinction between case studies and thecomparative method, although the case-study approachdoes remain a distinct tradition. Interest in case studieshas been reinforced by several factors, inc\uding therenewed concem with interpretive social science, the

continuing intellectual and institutional strength of areastudies, and deep skepticism in some circ1es about thevalidity of broad comparison.

Second, it is evident that quantitative techniquesemploying a relatively small number of cases cansuccessfulIy address important substantive questions.This approach merits attention in light of the newstatistical tests suitable for small-N analysis. Theopportunity for cumulative scholarly leaming provided bystatistical studies is nicely illustrated by theLange-Garrett-Jackman-Hicks-Patterson debate. Thisdebate is also relevant to the issue of linking rivalresearch traditions, because it shows that insights derivedfrom case studies and from more qualitative comparativework can, after alI, serve as stepping-stones on the pathtoward statistical analysis.

The third altemative has been reinforced as well:the systematic comparison of a small number of cases,with the goal of causal analysis, which is the approachthat Lijphart originally advocated. In this perspective,broad qualitative comparison is seen as both possible andproductive. The growing influence of the school ofcomparative historical analysis has substantialIy enhancedthe credibility of this approach, and it plays an importantrole as an analytic middle ground between the case-studytradition and small-N statistical analysis.

AlI three of these approaches will persist, and akey question is how well they can be linked. Thetradition of research on Westem Europe provides anencouraging model, in that the findings of quantitativecomparative scholars play an important role in generaldebates in that field." In research on Latin America,by contrast, quantitative comparative work receivesconsiderably less attention from mainstream scholars.Yet the kind of cross-fertilization found in the WestEuropean field can make an important contribution tostrengthening research. With good communication,country specialists and experts in qualitative small-Ncomparison can push the comparative quantifiers towardmore carefully contextualized analysis. Likewise, thecomparative quantifiers can push the country specialistsand experts in qualitative comparison toward moresystematic measurement and hypothesis testing. A centralgoal must be to sustain such communication.

The implications for graduate training are c1ear.If Ph.D. candidates are to be prepared to address theseissues of comparison, they should have enough training instatistical methods to evaluate quantitative studies thatemploy old, and new, methods of statistical analysis andto use such methods when appropriate. Those moreoriented toward statistical analysis should have enoughbackground in qualitative small-N comparison and casestudy analysis to be able to build on the analyticcontribution of those approaches. Both groups shouldhave substantial exposure to basic writings on the

.Jhilosophy of science and logic of inquiry that canprovide the framework for more informed choices aboutthese methodological alternatives.

In this way, the foundation can be laid for aneclectic practice of small-N analysis that takes advantageof opportunities that present. themselves on both sides ofwhat could otherwise be a major intellectual divide.

Notes

This is a revised and expanded version of an article earlierpublished in Dankwart A. Rustow and Kenneth Paul Erickson, OOs.,Comparaiive Political Dynamics: Global Research Perspectives (NewYork: Harper Collins, 1991). Permission to reprint granted by HarperCollins. Ruth Berins Collier, Kenneth Paul Erickson, LeonardoMorlina, Elizabeth Busbee, and Carol A. Medlín rnade particularlyuseful suggestions on earlier drafts. I also acknowledge comments fromChristopher Achen, Stephen Collier, James Fearon, David Freedrnan,Deborah Norden, Robert Powell, Merrill Shanks, and Laura Stoker.Ada Finifter and two anonymous reviewers likewise rnade helpfulcomments. This research has been supported by a GuggenheimPeIlowship, the Social Science Research Council, and the Institute ofGovemmental Studies at Berkeley. FinaIly, I would like to note a verypromising manuscript (King, Verba, and Keohane 1992) thatunfortunately came to my attention too late to be discussed in thischapter.

1. "N" is used to refer to the nurnher of cases analyzed inany given study.

2. References to rcpresentalive works of comparativehistorieal analysis are presented below.

3. In his comparison of these methods, Lijphartacknowledges his debt to Smelser's (1968) excellent analysis thatemployed a parallel framework. See also Smelser (1976).

4. This perspective has been elaborated by Skocpol (1984,chap. l I), and a parallel formulation is found in Charles Tilly (1984,chap.4).

5. Skocpol and Somers (1980, 181-87) refer to Ihis as"rnaero-causal" analysis. Yet small-N studies that generate and testhypotheses ean have both a maero and a micro focus, and it does notseem productive to exclude from Ihis category those wilh a micro focus.Hence, this altemative label is used.

6. Allhough Przeworski and Teune are centrally concemedwilh issues that arise when additional cases are added to an analysis, lheproblerns Ihey discuss are also more Iikely to occur if one is dealingwith a larger N to begin wilh.

7. For example, instead of referring to "Venezuela," onewould refer to a country in which, due to lhe impact of rnassive oilrevenues, a particular causal rclationship assumes a distinct formo

8. "Ihick description" is sometirnes místakenly understoodto refer simply to "detailed description," which is not what Geertzintends,

9. Given that these studies often focus on long periods oftime within each case, it might be argued that the number of cases couldbe greatly increased Ihrough comparison over time, thereby makingthem something other than srnal1-N studies. However, since the goaI ofmany studies in this tradition is to explain overall configurations ofnational outcomes as they are manifested over long periods, theseoutcomes often cannot be disaggregated into a series of longitudinalobservations. Hence, lhe number of cases cannot realistical1y beincreased Ihrough the use of comparison over time.

10. The most similar and most differcnt systerns designs

Collier 117

correspond, respectively, to John Stuart MiII's (1974) method ofdifference and method of agreement. Whereas Przeworski and Teune'slabels of "similar" and "different" refer to whether the cases arematched, as opposed to contrasting, on a series of background variables,MiII's labels of "difference" and "agreement" refer to whether lhe casesare contrasting, as opposed to matched, on the dependem variable.

1l. Personal communication from Adam Przeworski.12. Christopher Achen, personal communication, haslong

insisted on this point.13. For example, Gereffi and Wyman (1990), Haggard

(1990), Przeworski (1991), and Rueschemeyer, Stephens, and Stephens(1992).

14. For a discussion of strategic choice modela (a closelyrelated type of model) that have been applied to lhe analysis of politicalreform, democratization, and democratic consolidation in LatinAmerica, and that likewise offer fruitful simplificalions of complexphenomena, see Collier and Norden (1992).

15. The reprinting of this article in a reader on socialscience methodology (Tufte 1970) rnade it widely available to politicalscientists, and its influence has been substantial.

16. Personal communication from Arend Lijphart.17. This problem is routinely discussed in introductory

melhodology texts, e.g., Babbie (1992, 24-25, 427).18. Although pattem rnatching wilhin lhe same case

introduces the possibility of falsifying the hypothesis, it does notovercome ali of lhe problerns of ex post facto hypolheses. Thus,pattem matching will probably not overcome a problem ofunrepresentativeness which may arise due to selection bias or to lhechance selection of an atypical case.

19. See, for example, lhe debate on interest mediation andcorporatism in Westem Europe, including Wilensky (1976), Hibbs(1978), Schmitter (1981), and Cameron (1984). The debate started byLange and Garrett (1985) is a continuation of this line of analysis.

Bibliography

Achen, Christopher H. 1986. The Statistical Analysis 01 Quasi-Experiments, Berkeley and Los Angeles: University ofCalifornia Press,

Achen, Christopher H., and DuncanSnidal. 1989. "RationalDeterrence Theory and Comparative Case Studies." WorldPolitics 41: 143-69.

Almond, Gabriel A., and StephenJ. Genco. 1977. "Clouds, Clocks,and lhe Study of Politics." World Polirics 29:489-522.

Armer, Michael, and Allen Grimshaw, eds. 1973. Comparative SocialResearch, New York: John Wiley.

Babbie, Earl. 1992. The Practice 01 Social Research, 6th edition.Belmont, CA: Wadsworth.

Bendix, Reinhard. 1964. Nation-Building and Cirizmship: Studies 01Our Changing Social Order. New York: lohn Wlley.

Bendix, Reinhard. 1978. Kings or People: Power and lhe MtrniUzle toRule, Berkeleyand Los Angeles: University of CaliforniaPress.

Bergquist, Charles. 1986. Labor in Latin America: ComparativeEssays on Oule, Argentina, Yenezuela, and Colombia,Stanford: Stanford University Press.

Bollen, Kenneth A., and Robert W. Jackman. 1985. "RegressionDiagnostics: Ao Expository Treatment of Outliera andInfIuential Cases." Sociological Methods and Researcn13:510-42.

Burger, Thornas. 1976. Max Weber's Theory 01 Concept Formation:History, Laws, and Ideal Types . Durham: Duke University.

)

-~--.U-'i,;_~_~ '" _ÊJ . ~--'''=l1x:ncy-

41»~.á~ ..L,~&2 . ,....------~~~~4'7........._4'~tfQFúflt'ffJf' riJ .lfufriJJ/!M [J.flJhJ/d ~P< * ..bL>~

~~~&-.rú'r <:rl7/T~f""'7~ry <:::ZrF'r#?:l~7'777 .:Ter. .roán.LI.

Goldthorpe. Ne'\VYork.: Ox-ford Univcrsity Press.

Campbell, Donald T. 1975. "'Degrees of Freedom' and lhe CaseStudy. • Comparative Political Studies 8: 178-93.

Campbell, Donald T., and H. Laurence Ross. 1968. "The ConnecticutCrackdown on Speeding: Time Series Data in Quasi-Experimental Analysis." Law and Society Review 3:33-53.

Campbeíl, Donald T., and Julian C. Stanley. 1963. Experimental andQuasi-Experimensal Designs for Research, Chicago: RandMcNally.

Collier, David, and James E. Mahon. 1993. "Conceptual 'Stretching'Revisited: Altemative Views of Categories in ComparativeAnalysis." American Political Science Review,

Collier, David, and Deborah L. Norden. 1992. "Strategic ChoiceModels of Political Change in Latin America," ComparativePolitics 24:229-243.

Collier, Ruth Berins, and David Collier. 1991. Shaping lhe PoliticalArena: Crirical Junctures, lhe Labor Movemeru, and RegimeDynamics in Latin America. Princeton: Princeton UniversityPress.

Dahl, Robert A., ed. 1966. Polirical Oppositions in WestemDemocracies. New Haven: Yale University Press,

DeFelice, E. Gene , 1980. "Comparison Misconceived: CommonNonsense in Comparative Politics." Comparative Politics\3:119-26.

Diaconis, Persi, and Bradley Efron, 1983. "Computer-IntensiveMethods in Statistics." Scientific American 248:116-30.

Eckstein, Harry. 1975. "Case Study and Theory in Political Science."10 Handbook of Political Science, vol, 7, ed. Fred Greensteinand Nelson W. Polsby. Reading, MA: Addison-Wesley.

Eggan, Fred. 1954. "Social Aothropology and lhe Melhod ofControlled Comparison." American Anthropologist 56:743-63.

Etzioni, Amitai, and Frederic L. Dubow, eds, 1970. ComparativePerspectives: Theories and Methods . Boston: Linle,Brown.

Freedman, David A. 1987. "As Others See Us: A Case Study in PathAnalysis." Joumal of Educational Statistics 12:101-28.

Freedman, David A. 1991. "Statistical Models and Shoe Leather." 10Sociological Methodology 1991, ed, Peter Marsden. SanFrancisco: Jossey-Bass, r

Garrett, Geoffrey, and Peter Lange. 1989. "Govemment Parlisanshipand Economic Perfonnance: When and How Does 'WhoGoverns' Matter?" Joumal of Politics 51:676-93.

Geddes, Bárbara. 1990. "How the Cases Vou Choose Affect lheAnswers Vou Get: Selection Bias in Comparative Politics."10 Polirical Analysis, vol. 2, ed. James A. Stimson. AnoArbor: University of Michigan Press.

Geddes, Barbara. 1991. "A Game Theoretic Model of Reform in LatinAmerican Democracies." American Polittcal Science Review85:371-392.

Geertz, Clifford. 1973. "Thick Description: Toward an 10terpretiveTheory of Culture." 10 lhe Interpretaüon of Cultures, 00.Clifford Geertz. New York:: Basic Books.

George, Alexander L. 1979. "Case Studies and Theory Development:The Method of Structured, Focused Comparison." 10Diplomacy: New Approaches in History, 1heory, and Policy,ed. Paul Gordon Lauren. New York: The Free Press.

George, Alexander L., and Timolhy I. McKeown. 1985. "CaseStudies and Theories of Organizationa1 Decision Making."Advances in Informaiion Processing in Organizations, vol. 2.Santa Barbara, CA: IA! Press.

Gereffi, Gary, and Donald L. Wyman, eds. 1990. Manufacturing

~-~~"9 ~~.-....,

~/~ /húcç/,on.' rTrÍ1Cewn (ím'versíty ~~,.?"'-~~.-.I'- .......•..~........:: ....,--..-~ ~~~~

ftroa:crn """"o~. açrló.c.lcy Ana Loa Angc::lc:a; Únfvcratt)

Califorrua Press.

Haggard, Stephan. 1990. Pathways from lhe Periphery: 1he Politic;ofGrowth in lhe Newly Industrializing Countries. Ithaca:Cornell University Press.

Hampel, Frank R., et al. 1987. Robust Statistics: 1he ApproachBased on Influence Functions. New York: John Wt1cy.

Hartwig, Frederick, wilh Brian E. Dearing. 1979. Exploratory DataAnalysis, Sage University Paper, Series No. 07-016.Beverly HilIs, CA: Sage Publications.

Heckscher, Gunnar. 1957. The Study of Comparative Govemment andPolitics. London: George Allen and Unwin.

Hibbs, Douglas A., Ir. 1978. "On lhe Politica1 Economy of Long-RunTrends in Strike Activity." British Joumal of PoliricalScience 8:153-175.

Hicks, Alexander. 1988. "Social Democratic Corporatism andEconomic Growth." Joumal of Politics 50:677-704.

Hicks, Alexander, and William David Patterson. 1989. "On lheRobustness of lhe Left Corporatist Model of EconomicGrowth." Joumal of Politics 51 :662-675.

Hoole, Francis W. 1978. Evaluation Research and DevelopmeruActivities, Beverly Hills, CA: Sage Publications,

Jackman, Robert W. 1985. "Crosa-National Statistical Research andlhe Study of Comparative Política." American Joumal ofPolitical Science 29: 161-82.

I ackrnan, Robert W. 1987. "The Politics of Economic Growth inIndustrial Democracies, 1974-80: Leftist Strength or NorthSea Oil?" Joumal of Politics 49:242-56.

Jackman, Robert W. 1989. "The Politics of Economic Growth, OnceAgain." Joumal of Politics 51 :646-661.

Jackson, John E. 1992. "Estimation of Models wilh VariableCoefficienta." 10 Political Analysis, vol, 3. ed, James A.Stimson. Ano Arbor: University of Michigan Press.

Kalleberg, Arlhur L. 1966. "The Logic of Comparison: AMelhodological Note on lhe Comparative Study of PoliticalSystems." World Politics 19:69-82.

King, Gary, Sidney Verba, and Robert O. Keohane. 1992. ScientificInference in Qualitarive Research, Unpublished manuacript,Department of Govemment, Harvard University.

Lakoff, George. 1987. Women, Fire, and Dangerous Things: WhatCasegories Reveal abous lhe Mint!. Chicago: University ofChicago Press.

Lange, Peter, and Geoffrey Garrett. 1985. "The Politics ofGrowth:Strategic Interaction and Economic Performance in AdvancedIndustrial Democracies, 1974-1980." Joumal of Politics47:792-827.

Lange, Peter, and Geoffrey Garrett. 1987. "The Politics of GrowthReconsidered." Joumal of Politics 49:257-74.

Lasswell, Harold D. 1968. "The Future oflhe Comparative Melhod."Comparative Politics 1:3-18.

Lieberson, Stanley. 1991. "Small N's and Big Conclusions: AoExamination of lhe Reasoning in Comparative Studies Basedon a Small Number of Cases." Social Forces 70:307-20. '

Lijphart, Arend. 1971. "Comparative Politics and ComparativeMelhod." American Political Science Review 65:682-93.

Lijphart, Arend . ..l?75. "Tbe Comparable Cases Strategy inComparative,Research." . Comparative Political Studies8:158-77.

Lijphart, Arend. 1990. "The Political Consequences of ElectoralLaws, 1945-1985." American Political Science Review84:481-96.

Lipset, Seymour Marlin, and Stein Rokkan. 1967. "Cleavages,Structures, Party Systems, and Voter Alignmenta: Ao Intro-

duction." In Party Systems and Voter Alignments, eds,Seymour Martin Lipset and Stein Rokkan. New York: FreePress.

Luebbert, Gregory M. 1991. Llberalism, Fascism, or SocialDemocracy: Social Classes and lhe Poiitical Origins ofRegimes in Iruerwar Europe, New York: Oxford UniversityPress.

McKinney,Iobn C. 1966. Constructive 1'ypology and Social Theory,New York: Meredilh Publishing Company.

Merritt, Richard. 1970. Systematic Approaches 10 ComparativePolitics, Chicago: Rand McNally.

Merritt, Richard, and Stein Rokkan, eds. 1966. Comparing Nations:lhe Use of Quanlilalive Data in Cross-National Research,New Haven: Yale University Press.

Mill, John Stuart. [1843]1974. A System of Logic. Toronto:University of Toronto Press.

Mooney, Christopher Z., and Robert D. Duva!. 1992. "BootstrapInference: A Prelirninary Monte Carlo Evaluation." Paperpresented at lhe annual meetings of lhe American PoliticalScience Association, Chicago.

Moore, Barrington. 1966. Social Origins of Dicuuorship andDemocracy: Lord and Peasaru in lhe Maldng of the ModemWorld. Boston: Beacon Press.

Mosteller, Frederick, and Iobn W. Tukey. 1977. DaUJ Analysis andRegression, Reading, MA: Addison-Wesley.

O'Donnell, Guillermo, Philippe C. Schmitter, and Lawrence Whitehead,eds. 1986. Transitions from AurhorilOrian Rule, Baltimore:John Hopkins University Press.

Paige, Jeffrey. 1975. Agrarian Revoluiion: Social Movements andExport Agriculture in lhe Underdeveloped World. NewYork: Free Press.

Przeworski, Adam. 1987. "Melhods of Cross-National Research,1970-1983: An Overview." MeinolfDierkes, Hans N.Weiler, and Ariane Berthoin Antal, eds. Comparative PolicyResearch: Leaming from Experience, Brookfield, VT:Gower.

Przeworski, Adam. 1991. Democracy and the Market: Polirical andEconomic Reforms in Eastem Europe and Latin America,

Cambridge: Cambridge University Press.Przeworski, Adam, and Henry Teune. 1970. lhe Logic of

Comparative Social Inquiry, New York: Iohn Wiley.Ragin, Charles C. 1987. lhe Comparative Method: Moving beyond

Qualitarive and Quantilalive Strategies, Berkeley and Los, Angeles: University of California Press.

Rokkan, Stein. 1970. Citizens, Elections, Parties: Approaches 10 lheComparative Study of Processes of Development, New York:David McKay.

Rueschemeyer, Dietrich, Evelyne Huber Stephens, and John D.Stephens. 1992. Capitalist Developmeni and Democracy,Chicago: University of Chicago Press.

Rustow, Dankwart. 1968. "Modernization and Comparative Politics:Prospects in Research and Theory." Comparative Politics1:37-51.

Sartorí, Giovanni. 1970. "Concept Misformation in ComparativePolitics." American Polirical Science Review 64:1033-53.

Sartori, Giovanni, ed. 1984. Social Science Concepts: A SystematicAnalysis. Beverly Hills, CA: Sage.

Sartori, Giovanni. 1991. "Comparing and Miscomparing." JoumalofTheoretical Politics 3 :243-57.

Sartori, Giovanni. 1993. "Totalitarianism, Model Mania, and Learningfrom Error." Joumal of Theoretical Polirics.

Sartori, Giovanni, Fred W. Riggs, and Henry Teune. 1975. Tower ofBabei: On lhe Definition and Analysis of Concepts in lheSocial Sciences. International Studies Association,Occasional Paper No. 6, University of Pittsburgh.

Collier 119

Schrnitter, Philippe C. 1981. "Interest Intermediation and RegimeGovernability in Contemporary Westem Europe and NorthAmerica." In Organizing Interests in Westem Europe, ed.Suzanne D. Berger. Cambridge; Cambridge UnivenityPreIS.

Skocpol, Theda. 1979. SIOleS and Social Revo/ulions: A Comparaüv«Analysis of France, Russia, and China. Cambridge:Cambridge University Press.

Skocpol, Theda. 1984. V/SÍon and Merhod in Historical Sociology.Cambridge and New York: Cambrídge Univenity Press,

Skocpol, Theda, and Margaret Somers. 1980. "The Uses ofComparative History in Macrosocial Inquiry." ComparaiiveStudies in Society and History 22:174-97.

Smelser, Neil. 1968. "The Methodology of Comparative Analysis ofEconomic Activity." In Essays in SociologicalInterpretaiion, ed. Neil Smelser. Englewood Cliffs, NI:Prentice-Hall,

Smelser, Neil. 1976. Comparative Merhods in lhe Social Sciences,Englewood Cliffs, NI: Prentice Hall,

Stinchcombe, Arthur L. 1978. Theoretical Methods in Social History,New York: Academic Press.

Tilly, Charles. 1975. lhe Formation of National SIOleS in WestemEurope. Princeton; Princeton University Presa.

Tilly, Chades. 1984. Big Structures, Large Processes, HugeComparisons. New York: Russell Sage.

Trimberger, Ellen Kay. 1978. Revolution from Above: MilitaryBureaucrats and Development in Japan, Turkey, Egypt, andPeru, New Brunswick, NJ; Transaction Books.

Tufte, Edward R. 1970. lhe Quanlilalive Analysis of Social Problems.Reading, MA: Addison-Wesley.

Vallier, Ivan, ed. 1971. Comparative Merhods in Sociology: Essayson Trends and Applicaüons, Berkeley and Los Angeles:University of California Press.

Verba, Sidney. 1967. 'Some Dilemmas in Comparative Research."World Politics 20: 111-27.

Wllensky, Hacold L. 1976. 1he 'New Corporatism,' Centralization,and lhe Welfare State . Professional Papers in ContemporaryPolitical Sociology Séries. Beverly HiIIs: Sage Publications.

Yin, Robert K. 1984. Cose Srudy Research: Design and Methods.Applied Social Research Melhods Series, vol. 5. BeverlyHills: Sage Publications.

Zelditch, Jr., Morris. 1971. "lntelligible Comparisons." InCompararive Merhods in Sociology: Essays on Trends andApplications, ed. Ivan Vallier. Berkeley and Los Angeles:University of California Press.

Political Science:The State of the Discipline

II

Edited by Ada W. Finifter

1993

A publication supported by the Evron M. Kirkpatrick Fundof the American Political Science Association

*The American Political Science Association

1527 New Hampshire Avenue, N.W.Washington, De 20036