Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.



Published on

  • Be the first to comment

  • Be the first to like this


  1. 1. American Political Science Review Vol. 98, No. 2 May 2004What Is a Case Study and What Is It Good for?JOHN GERRING Boston University T his paper aims to clarify the meaning, and explain the utility, of the case study method, a method often practiced but little understood. A “case study,” I argue, is best defined as an intensive study of a single unit with an aim to generalize across a larger set of units. Case studies rely on the same sort of covariational evidence utilized in non-case study research. Thus, the case study method is correctly understood as a particular way of defining cases, not a way of analyzing cases or a way of modeling causal relations. I show that this understanding of the subject illuminates some of the persistent ambiguities of case study work, ambiguities that are, to some extent, intrinsic to the enterprise. The travails of the case study within the discipline of political science are also rooted in an insufficient appreciation of the methodological tradeoffs that this method calls forth. This paper presents the familiar contrast between case study and non-case study work as a series of characteristic strengths and weaknesses—affinities— rather than as antagonistic approaches to the empirical world. In the end, the perceived hostility between case study and non-case study research is largely unjustified and, perhaps, deserves to be regarded as a misconception. Indeed, the strongest conclusion to arise from this methodological examination concerns the complementarity of single-unit and cross-unit research designs. he case study occupies a vexed position in the tion of work generated by the discipline, the case studyT discipline of political science. On the one hand, methodologists generally view the case studymethod with extreme circumspection (Achen and method is held in low regard or is simply ignored. Even among its defenders there is confusion over the virtues and vices of this ambiguous research design. Practi-Snidal 1989; King, Keohane, and Verba 1994; Lieberson tioners continue to ply their trade but have difficulty[1991] 1992, 1994; Njolstad 1990). A work that focuses articulating what it is that they are doing, methodolog-its attention on a single example of a broader phe- ically speaking. The case study survives in a curiousnomenon is apt to be described as a “mere” case study. methodological limbo. At the same time, the discipline continues to pro- How can we understand the profound disjunctureduce a vast number of case studies, many of which that exists between the case study’s acknowledged con-have entered the pantheon of classic works (Allen 1965; tributions to political science and its maligned statusAllison 1971; Dahl 1960; Johnson 1983; Kaufman 1960; within the discipline? If case studies are methodologi-Lazarsfeld, Berelson, and Gaudet 1948; Lijphart 1968; cally flawed, why do they persist?Pressman and Wildavsky 1973). Judging by recent The paper is divided into two parts. The first part fo-scholarly output, the case study method retains con- cuses on matters of definition. I argue that for method-siderable appeal, even among scholars in research ological purposes a case study is best defined as ancommunities not traditionally associated with this in-depth study of a single unit (a relatively boundedstyle of research—e.g., among political economists and phenomenon) where the scholar’s aim is to elucidatequantitatively inclined political scientists (Acemoglu, features of a larger class of similar phenomena. It isJohnson, and Robinson 2003; Bates et al. 1998; Rodrik demonstrated that case studies rely on the same sort2003). By the standard of praxis, therefore, it would of covariational evidence utilized in non-case study re-appear that the method of the case study is solidly en- search. Thus, the case study method is correctly un-sconced and, perhaps, even thriving. derstood as a particular way of defining cases, not a Thus, a paradox: Although much of what we know way of analyzing cases or a way of modeling causalabout the empirical world is drawn from case studies relations. I show, finally, that this understanding of theand case studies continue to constitute a large propor- subject illuminates some of the persistent ambiguities of case study work, ambiguities that are, to some extent, intrinsic to the enterprise.John Gerring is Associate Professor, Department of Political Sci- In the second part of the paper I proceed to examineence, Boston University, 232 Bay State Road, Boston, MA 02215 the contrast between case study and non-case study( Helpful comments were received from Robert Adcock, Andrew work. The central argument here is that the differencesBennett, Henry Brady, Bear Braumoeller, David Collier, Michael between these two genres are best understood as char-Coppedge, Colin Elman, Peter Hall, Alan Jacobs, Evan Lieberman, acteristic strengths and weaknesses—affinities—ratherJim Mahoney, Jason Seawright, David Waldner, and three anony- than antagonistic approaches to the empirical world.mous reviewers for the journal. William Barndt and JoshuaYesnowitz provided crucial assistance at the final stages of manuscript Tradeoffs, rather than dichotomies, characterize the on-preparation. George and Bennett 2004 reached me in manuscript going case study/non-case study debate.form as I was developing the ideas for this paper and had a substantialimpact on my own understanding of the case study—although somedifferences of opinion remain. Funding for this research was gener-ously provided by the Frederick S.Pardee Center for the Study of the WHAT IS A CASE STUDY?Longer-Range Future. Work was completed while the author wasin residence at the School of Social Science, Institute for Advanced What is a case study, and how is it differentiated fromStudy. other styles of research? Regretfully, the term “case 341
  2. 2. What Is a Case Study? May 2004study” is a definitional morass. To refer to a work as a For those familiar with the rectangular form of a datasetcase study might mean (a) that its method is qualita- it may be helpful to conceptualize observations as cells,tive, small-N (Yin 1994); (b) that the research is ethno- variables as columns, cases as rows, and units as eithergraphic, clinical, participant-observation, or otherwise groups of cases or individual cases (depending upon“in the field” (Yin 1994); (c) that the research is charac- the proposition and the analysis).terized by process-tracing (George and Bennett 2004); The most important point is that all these terms are(d) that the research investigates the properties of a sin- definable only by reference to a particular propositiongle case (Campbell and Stanley 1963, 7; Eckstein [1975] and a corresponding research design. A country may1992); or (e) that the research investigates a single phe- function as a case, a unit, a population, or a case study.nomenon, instance, or example (the most common us- It all depends upon what one is arguing. In a typi-age). Evidently, researchers have many things in mind cal cross-country time-series regression analysis (e.g.,when they talk about case study research.1 As a result of Przeworski et al. 2000), units are countries, cases arethis profusion of meanings, proponents and opponents country-years, and observations are collected for eachof the case study marshal a wide range of arguments case on a range of variables. However, shifts in thebut do not seem any closer to agreement than when unit of analysis of a proposition change the referen-this debate was first broached several decades ago. tial meaning of all terms in the semantic field. If one How, then, should the case study be understood? The moves down one level of analysis the new populationfirst three options enumerated above (a–c) seem inap- lies within the old population, the new sample withinpropriate as general definitions of the topic since each the old sample, and so forth, such that an observationimplies a substantial shift in meaning relative to estab- in the original proposition now becomes a case. Pop-lished usage. One cannot substitute case study for qual- ulation, unit, case, and observation are nested withinitative, ethnographic, or process-tracing without feeling each other. Since most social science research occursthat something has been lost in translation. These defi- at several levels of analysis these terms are generally innitions are best understood as describing certain kinds flux. Nonetheless, they have distinct meanings within(subtypes) of case studies, rather than the general phe- the context of a single proposition, which defines thenomenon itself. The fourth option (d) equates the case principal unit of with the study of a single case, the N = 1 research I do not issue this somewhat novel definition of casedesign. This is simply wrong, as argued at length below; study (an intensive study of a single unit for the purposecase studies always employ more than one case. The of understanding a larger class of units) with any hopesfifth option (e), centering on phenomenon, instance, or of displacing common usage. Indeed, there is no harm inexample as the key term, is correct as far as it goes continuing to refer to a case study in the various waysbut also ambiguous. Imagine asking someone, “What listed above (options a–e). What is important is thatis your instance?” or “What is your phenomenon?” A we have recourse to a narrower and clearer definitioncase study presupposes a relatively bounded phenome- when methodological confusions arise so that we have anon, an implication that none of these terms captures. way to arbitrate such confusions. The definition chosen As a substitute for these flawed definitions, I propose here is useful in this regard. Moreover, it captures theto define the case study as an intensive study of a sin- essential features of other extant definitions; it is reso-gle unit for the purpose of understanding a larger class nant (Gerring 2001, chap. 3). Finally, as the succeedingof (similar) units. A unit connotes a spatially bounded portions of this paper show, it clarifies the distinctivephenomenon—e.g., a nation-state, revolution, political features of a broad class of work in the discipline ofparty, election, or person—observed at a single point in political science and in neighboring fields of the socialtime or over some delimited period of time. (Although sciences. It is theoretically useful.the temporal boundaries of a unit are not always ex-plicit, they are at least implicit.)2 To clarify this definition we must establish the rela- The Case Study Method Consideredtionship of the case study, so defined, to other terms in as an Empirical Endeavorthis crowded semantic field. Following is a set of nesteddefinitions, which should be read carefully. A “pop- The distinctiveness of the case study is most clearly un-ulation” is comprised of a “sample” (studied cases), derstood when placed within a broader set of method-as well as unstudied cases. A sample is comprised of ological options. To understand what a case study is,several “units,” and each unit is observed at discrete one must comprehend what it is not.points in time, comprising “cases.” A case is comprised All empirical evidence of causal relationships is co-of several relevant dimensions (“variables”), each of variational in nature. A purported cause and effect mustwhich is built upon an “observation” or observations. be found to covary. They must appear and disappear, wax and wane, or perform some other transformation1 In addition to sources cited above, see Brady and Collier 2004, in tandem or at some regular, more or less predictable,Campbell (1975) 1988, Davidson and Costello 1969, Feagin, Orum, intervals. Even where this covariation is imagined, asand Sjoberg 1991, George 1979, McKeown 1999, Ragin 1987, 1997, in a counterfactual thought experiment, the evidenceRagin and Becker 1992, and the symposium, “The Case Study we imagine is of a covariational sort. Conversely, theMethod in Sociology,” in Current Sociology, Volume 40, Number1 (Spring 1992). absence of such covariation is taken as disconfirm-2 Similar understandings of the term “unit” can be found elsewhere ing evidence. If the appearance and disappearance(e.g., King, Keohane, and Verba 1994, 76–77). (waxing/waning et al.) of X and Y are not associated342
  3. 3. American Political Science Review Vol. 98, No. 2 TABLE 1. Research Designs: A Covariational Typology Temporal Variation No Yes None (1 unit) [Logically impossible] (a) Case study I Within-unit (b) Case study II (c) Case study III Spatial Variation Across-unit (d) Cross-sectional (e) Time-series cross-sectional Across- and within-unit (f) Hierarchical (g) Hierarchical time-series; Comparative-historicalin any way that can be rationally explained, and hence designs the case study is probably closest to the latter,predicted (or postdicted), then the empirical evidence where levels of analysis move up and down more or lesssuggests that a causal relationship does not exist.3 simultaneously and where a small number of units are This provides a useful way of typologizing various subjected to intensive study. Indeed, the comparative-research designs. Covariation may be observed (a) in historical study may be looked upon as a series ofa single unit diachronically, (b) within a single unit case studies combined with explicit cross-unit analysissynchronically, (c) within a single unit diachronically, (Mahoney and Rueschemeyer 2003).(d) across units synchronically, (e) across units syn- Having placed these standard cross-unit research de-chronically and diachronically, (f) across and within signs within a covariational typology one must also takeunits synchronically, or (g) across and within units syn- note that each of these methods might also be em-chronically and diachronically, as depicted in Table 1. ployed as a case study. A case study may employ cross- It will be seen that the case study occupies one of sectional, TSCS, hierarchical, hierarchical time-series,three possible cells. Type I case studies examine vari- and perhaps even comparative-historical models. It allation in a single unit over time, thus preserving the depends upon the proposition in question. Specifically,primary unit of analysis. Other case studies break down it is the purposes to which these analyses are put, andthis primary unit into subunits, which are then sub- hence the definition of a unit, that determines whetherjected to covariational analysis—either synchronically or not they are appropriately referred to as case studies.(type II) or synchronically and diachronically (type III). This will become clearer as we proceed.These are the three logically conceivable approachesto the intensive study of a single unit where that unit isviewed as an instance of some broader phenomenon. The N QuestionConsequently, when one refers to the case study method I have argued that what distinguishes the case studyone is in fact referring to three possible methods, each method from all other methods is its reliance on covari-with a different menu of covariational evidence. ation demonstrated by a single unit and its attempt, at The bottom half of Table 1 lays out various across- the same time, to illuminate features of a broader set ofunit research designs (where some important element units. It follows from this that the number of cases (N)of the empirical analysis involves comparisons across employed by a case study may be either small or largeunits). Here I have listed the methods most commonly and, consequently, may be evaluated in a qualitative oridentified with these research designs. Across-unit anal- quantitative fashion.4ysis without any explicit temporal component (d) is To see why this must be so let us consider how a caseusually classified as “cross-sectional” (even though a study of a single event—say, the French Revolution—temporal component is usually simulated with indepen- works. Intuitively, such a study provides an N of onedent variables that are assumed to precede the depen- (France). If one were to broaden the analysis to includedent variable under investigation). When a temporal a second revolution (e.g., the American Revolution), itcomponent is included we often refer to the analysis as would be common to describe the study as comprising“time–series cross-sectional” (TSCS) or pooled time– two cases. Yet, as I have argued preliminarily, this isseries (e). When one examines variation across- and a gross distortion of what is really going on. It wouldwithin units in the same research design one is said to be more correct to describe such a study as comprisingbe employing a “hierarchical” model (f). Finally, when two units, rather than two cases, for a case study of aall forms of covariation are enlisted in a single research single event generally examines that event over the resulting method is described as “hierar- France is observed before, during, and after the eventchical time–series” (if quantitative) or “comparative- to see what changed and what remained the same afterhistorical” (if qualitative) (g). Of all cross-unit research this cataclysmic event. These patterns of covariation offer the empirical clues one needs to reach conclusions3 Note that covariation (or correlation) refers to the mutual relation- about causation. They also create multiple cases out of that individual unit. N = 2, at the very least (e.g., beforeship between X and Y; variation, to the behavior of a single variable.These words are often used interchangeably. Hume’s word for this and after a revolution), in a case study of type I.was constant conjunction, and others have been employed as well. Ishould clarify that although the empirical component of a causal argu-ment is covariational in nature, successful causal arguments depend 4 This section explains and elaborates on a theme first articulated byupon more than just covariation. Among other things, a convincing Campbell (1975) 1988, itself a revision of Campbell’s earlier perspec-causal account must identify a causal mechanism (see below). tive (Campbell and Stanley 1963). 343
  4. 4. What Is a Case Study? May 2004 If, instead, there is no temporal variation—if, for ex- not type I, involve a change in level of analysis, sinceample, the French Revolution is examined at a single cases are drawn from phenomena within the primarypoint in time—then the object of investigation will be unit (as defined by the proposition of interest). Thus,covariational patterns within that unit, a case study of some case studies—but not all—involve a change in thetype II. Within-unit cases consist of all cases that lie at a primary unit of analysis. To complicate matters further,lower level of analysis relative to the inference under in- case studies often combine observations of the primaryvestigation. If the primary unit of analysis is the nation- unit over time (type I) with synchronic and diachronicstate, then within-unit cases might be constructed from observations of within-unit covariation (types II andprovinces, localities, groups, or individuals. The pos- III). Many case studies are thus hybrids of all threesibilities for within-unit analysis are, in principle, in- research designs. A final complication is introduced byfinite. Indeed, within-unit N often swamps across-unit the fact that it is often difficult to figure out which sort ofN, particularly where individuals comprise the relevant covariational evidence is being mobilized at a particularwithin-unit case. A single national survey will produce a juncture. The difficulty owes something to the complex-larger sample than any conceivable cross-country anal- ities of within-unit analysis. Although the primary unitysis. Thus, in many circumstances case studies of type II of analysis is usually clear, within-unit cases are oftencomprise a larger N than cross-sectional analyses or multiple and ambiguous.TSCS analyses. A second source of ambiguity concerns the blurry Evidently, if a case study combines both temporal and line between a unit that is intensively studied—the casewithin-unit variation, as in case studies of type III, then study—and other adjacent units that may be broughtits potential N increases accordingly. This is probably into the analysis in a less structured manner. Recall thatthe most common genre of case study analysis. because a case study refers to a set of units broader than These covariational facts hold true regardless of the one immediately under study, a writer must havewhether the method is experimental or nonexperimen- some knowledge of these additional units (a) to choosetal. It is also true of counterfactual reasoning, which a unit for special treatment and (b) identify plausibletypically consists of four cases—the actual (as it hap- causal hypotheses. Case studies are not immaculatelypened) before and after cases and the before and after conceived; additional units always loom in the back-cases as reconstructed through counterfactual reason- (i.e., with an imagined intervention). In short, the To speak of a case study at all it is helpful to introducecase study does not preclude high-N; it simply precludes a distinction between formal and informal units. Theacross-unit N (by definition). formal unit is the unit chosen for intensive analysis— What, then, of the classic N = 1 research design, the person, group, organization, county, region, coun-which haunts the imaginations of social scientists ev- try, or other bounded phenomenon of which the writererywhere? This hypothetical research design occupies has in-depth knowledge. Informal units consist of allthe empty cell in Table 1. Its cell is empty because it other units that are brought into the analysis in a pe-represents a research design that is not logically fea- ripheral way, typically in an introductory or concludingsible. A single unit observed at a single point in time chapter. Often, these informal units are studied onlywithout the addition of within-unit cases offers no evi- through secondary literature; they are always more su-dence whatsoever of a causal proposition. In trying to perficially surveyed than the formal unit under a causal relationship from this snapshot—a sin- Sometimes, the status of informal units is left case without within-unit covariation—we would be This may be warranted in circumstances where the rel-engaging in a truly random operation, since an infinite evant comparison or contrast between the formal unitnumber of lines might be drawn through that one data and other units is obvious or generally accepted. In anypoint. case, the distinction between a formal and an informal unit is always a matter of degrees. The more equality of treatment granted to peripheral units, the more a studyAmbiguities–Necessary and Unnecessary leans toward a cross-unit style of analysis. The greater the predominance of a single unit, the more it meritsThe effort in this section has been to clarify what it the appellation case study.means to conduct a case study. I have argued that a case A third ambiguity occurs whenever a single workstudy is most usefully defined as the intensive study combines single-unit and across-unit analysis in a for-of a single unit wherever the aim is to shed light on mal manner. This would be true of comparative-a question pertaining to a broader class of units. Al- historical work as well as any work in which an inten-though this definitional exercise does not settle all the sively studied unit is “nested” within a broader researchambiguities besetting the case study research design, it design (Coppedge 2002; Lieberman 2003). Indeed, thedoes provide a way of understanding ambiguities that only thing that distinguishes the single-unit study fromremain. Six issues deserve emphasis. a sample (which is of course also designed to elucidate The first ambiguity concerns the problem of distin- the features of some larger phenomenon) is that theguishing different types of covariational evidence. We latter is generally understood as composed of morehave pointed out that case studies may observe a sin- than one unit. Case studies, like samples, seek to rep-gle unit through time (type I), synchronic within-unit resent, in all ways relevant to the proposition at hand,variance (type II), or synchronic and diachronic within- a population of cases. A series of case studies mightunit variance (type III). Notice that types II and III, but therefore be referred to as a sample; it is a matter of344
  5. 5. American Political Science Review Vol. 98, No. 2emphasis and of degree. The more case studies one systems in ethnically divided societies. Reilly argues, onhas, the less intensively each one is studied, and the the basis of several case studies, that single–transferablemore confident one is in their representativeness (of vote (STV) electoral systems have a moderating effectsome broader population), the more likely one is to on group conflict relative to first-past-the-post (FPP)describe them as a sample rather than a series of case electoral systems. Here is a good example of a casestudies. study that is more than simply suggestive (for other A fourth ambiguity afflicting case studies is that such examples see Eaton 2003, Elman 1997, Lijphart 1968,works generally partake of two empirical worlds. They and Stratmann and Baur 2002). For present purposes,are both studies tout court and case studies of something what is significant is that both styles of argumentation—more general. As a study, the population is restricted to the suggestive and the falsifiable—are legitimately re-the unit under investigation. As a case study, the pop- ferred to as case studies. Evidently, they have very dif-ulation includes adjacent units—perhaps quite a large ferent methodological implications. But these implica-number of them. This tension is evident in Graham tions should not be confused with the case study format,Allison’s (1971) renowned work, whose subtitle, Ex- which can be implemented in interpretivist as well asplaining the Cuban Missile Crisis, invokes a narrow positivist modes.topic, whereas the title, Essence of Decision, suggests Having flagged these six ambiguities of the casea much larger topic (government decision-making). study, the question is begged: Are they necessary? AreTo complicate matters further, different propositions they intrinsic to the research design, or might they bewithin the same work commonly apply to different pop- avoided?ulations. Some may be restricted to the unit under study, In many instances, ambiguities can be removed sim-whereas others have a wider ambit. This is clearly the ply by more careful attention to the task of specifi-case in Allison’s study and is noted explicitly in the cation (Gerring 2001, 90–99). Writers should be clearintroduction. about which propositions are intended to describe the To complicate matters further, the status of a work unit under study and which are intended to apply to amay change as it is digested and appropriated by a com- broader set of units. Regrettably, many studies focusedmunity of scholars. “Meta-analyses” are systematic at- on some element of politics in the United States frametempts to integrate the results of individual studies into their analysis as a study of politics—by implication, pol-a single quantitative analysis, pooling individual cases itics in general (everywhere and always). One is left todrawn from each study into a single dataset (with vari- wonder whether the study pertains only to Americanous weightings and restrictions). The ubiquitous “liter- politics, to all contemporary polities, or, in varying de-ature review” often aims at the same objective, albeit grees, to both. Indeed, the slippage between study andin a less synoptic way. Both statistical meta-analyses case study accounts for much of the confusion that weand narrative literature reviews assimilate a series of encounter when reading single-unit analyses. To the ex-studies, treating each of them as case studies in some tent that propositions—and their attendant cases, units,larger project—whether or not this was the intention and populations—are stated clearly and explicitly, theof the original authors. author avoids confusion and the work attains a higher A final ambiguity concerns the sort of argument that degree of falsifiability. This may involve some sacrificea case study is intended to prove or demonstrate. One in narrative flow, but it is rightly regarded as the entryspecies of case study examines a loosely defined general price of social science.topic—war, revolution, gender relations—in a particu- However, it hardly seems plausible that the six am-lar setting but offers no specific proposition that might biguities noted above arise solely from the sloppy orbe applied across a larger set of units. E. P. Thompson’s unduly belletristic habits of case study practitioners.The Making of the English Working Class (1963) is usu- Indeed, a certain degree of ambiguity is inherent in theally construed as a case study of class formation. This enterprise of the case study. This pertains, most of all,suggests a very general purview, perhaps applicable to to the study/case study distinction.all countries in the modern era. Yet Thompson does It would be difficult to write a study of a single unitnot proffer a theory of class formation, aside from the that does not also function as a case study, and vicerather fuzzy notion of a working class participating in versa, for reasons already explored. Indeed, it may beits own development. Thus, his work is probably cor- difficult to neatly separate the study and case studyrectly understood as a study of how a more general components of a work (e.g., into different chapters orphenomenon occurred in one country setting. Virtu- differently labeled propositions). The reason for thisally any intensive study of a relatively bounded topic structural ambiguity is that the utility of the single-unitqualifies as a case study in this minimal sense, so long as study rests partly on its double functions. One wishesit can be linked with some larger topic via a key word to know both what is particular to that unit and what(e.g., class formation). Indeed, the narrowest terrains is general about it. It should be kept in mind that casesometimes claim the broadest extensions. Studies of a studies often tackle subjects about which little is pre-war are studies of war, studies of a farming commu- viously known or about which existing knowledge isnity are studies of farming communities everywhere, fundamentally flawed. The case study typically presentsstudies of individuals are studies of humanity, and so original research of some sort. Indeed, it is the opportu-forth. nity to study a single unit in great depth that constitutes A very different style of argumentation informs one of the primary virtues of the case study methodBenjamin Reilly’s (2001) study of the role of electoral (see below). If a writer were to restrict herself only to 345
  6. 6. What Is a Case Study? May 2004 TABLE 2. Single-Unit Versus Cross-Unit Research Designs: Tradeoffs and Affinities Affinity Case Study Cross-Unit Study 1. Type of inference (a) Descriptive + (b) Causal + 2. Scope of proposition (a) Depth + (b) Breadth + (c) Boundedness + 3. Unit homogeneity (a) Case comparability (internal) + (b) Representativeness (external) + 4. Causal insight (a) Causal mechanisms + (b) Causal effect + 5. Causal relationship (a) Invariant + (b) Probabilistic + 6. Strategy of research (a) Exploratory (theory generation) + (b) Confirmatory (theory testing) + 7. Useful variance (a) For only a single unit + (b) For many units + 8. Ontology Indeterminateelements of the unit that were generalizable (i.e., if she WHAT IS A CASE STUDY GOOD FOR?rigorously maintains the “case study” mode of anal-ysis), a reader might justifiably complain. Such rigor It has been demonstrated that the difference betweenwould clarify the population of the primary infer- a case study and a study (tout court) is rarely clear-cut.ence, but it would also constitute a considerable waste Indeed, the case study is probably best understood asof scholarly resources. Imagine a study of economic an ideal-type rather than a method with hard-and-fastgrowth that focuses on Mauritius as a case study yet re- rules. Yet the fact that the case study is fuzzy aroundfuses to engage causal questions unless they are clearly the edges does not mean that it is lacking in distinctiveapplicable to other countries (since this is a case study characteristics. When considered as an ideal type theof a more general phenomenon, growth). No mention case study research design, like all research designs, ex-of factors specific to the Mauritian case is allowable; hibits characteristic strengths and weaknesses relativeall proper nouns are converted into common nouns to its across-unit cousin. These pertain to the type of(Przeworski and Teune 1970). Such a study seems un- inference under consideration (descriptive or causal),duly narrow; its conclusions may mislead. the scope of the proposition (its depth, breadth, and Indeed, it is often difficult to tell which of the many boundedness), the degree of unit homogeneity foundfeatures of a given unit are typical of a larger set of among cases and between the sample and the popula-units (and hence fodder for generalizable inferences) tion, the sort of causal insight desired (causal effectand which are particular to the unit under study. The ap- or causal mechanism), the strategy of research (ex-propriate response to such ambiguity is for the writer to ploratory or confirmatory), and the kind of empiricalreport all facts and hypotheses that might be relevant— evidence available. Tradeoffs along these seven dimen-in short, to overreport. Much of the detail provided by sions are summarized in Table 2. Ontological presuppo-the typical case study may be regarded as “field notes” sitions are also important but of indeterminate importof possible utility for future researchers—perhaps with (as indicated in Table 2).a rather different set of inferences in mind. Again, it It should be underlined that these tradeoffs rep-seems justifiable for case studies to work on two levels resent methodological affinities, not invariant laws.simultaneously, the unit itself and some broader class Exceptions can be found to each of the general tenden-of (perhaps difficult to specify) units. cies identified here. Even so, the strengths and weak- As a general observation we might say that methods, nesses often noted in case study research, reproducedstrictly defined, tend to lose their shape as one looks in many subfields and disciplines over the course ofcloser at their innards. A study merges into a case study, many decades, are not the product of a purely stochas-a single-unit study merges into a study of a sample, a tic process. General patterns suggest general inter-longitudinal study merges into a latitudinal study, infor- pretations.mal cases merge into formal cases, and so forth. Meth- I should also emphasize that each of these trade-ods that seem quite dissimilar in design bleed into one offs carries a ceteris paribus caveat. Case studies areanother when put into practice. There are few “pure” more useful for forming descriptive inferences, all othermethods. And this is probably a good thing. Chastity is things being equal. Since ceteris is not always paribusnot necessarily an attribute to be cherished in research the reader should not jump to any conclusions about thedesign. research design appropriate to a given setting without346
  7. 7. American Political Science Review Vol. 98, No. 2considering the single-unit/cross-unit options available tually studying the phenomenon in question each timewithin that research context. one sallies forth with a new pronouncement about the Finally, readers should note that although many of world.5my examples are drawn from the subfield of compara- It should be clear that the affinity between case studytive politics, with nation-states as the principal unit of research and descriptive inference does not denigrateconcern, these examples could be replicated with other the possibility of causal analysis through case study re-units and in other research settings. The problem of the search, of which one might cite many illustrious exam-case study is not limited to a single subfield. ples. Indeed, the discussion that follows is primarily concerned with propositions of a causal nature. My point is simply that it is easier to conduct descrip-Type of Inference: Descriptive Versus Causal tive work than to investigate causal propositions while working in a case study mode.Descriptive inference remains an important, if under-valued, trope within the social sciences (Gerring 2001,chap. 6; King, Keohane, and Verba 1994, chap. 2). Thus, Scope of Proposition: Breadth andit is not at all pejorative to observe that there is a Boundedness Versus Depthmethodological affinity between descriptive inference The variable utility of the case study is also partly aand case study work. When one is examining correla- product of the scope of the causal argument that ative relationships or proximate causal relationships the writer wishes to prove or demonstrate. Arguments thatcase study format seems less problematic and is often strive for great breadth and boundedness are in greaterhighly informative. Indeed, many of the most famous need of cross-unit cases; causal arguments restricted tocase studies in anthropology, political science, and so- a small set of units more plausibly subsist on the basisciology are primarily descriptive in orientation (e.g., of a single-unit study. The extensive/intensive tradeoffFenno 1978, Hartz 1955, Lynd and Lynd [1929] 1956, is commonsensical. Insofar as one wishes to make anMalinowski [1922] 1984, and Whyte [1943] 1955). How argument about the universe of nation-states, one iscan we understand this affinity? on surer ground if one has studied more than one of What? and How? questions are easier to answer with- those nation-states. A case study of France probablyout recourse to cross-unit analysis than Why? questions. offers better evidence for an argument about EuropeThe simplest genre of descriptive case study asserts that than for an argument about the world. Propositionalthe unit under study (A) is like, or unlike, other similar breadth and evidentiary breadth generally go hand inunits (B and C). A more complicated descriptive case hand. The evidence should be commensurate with thestudy might assert a classificatory relationship among scope of the proposition.A, B, and C, such that A falls into a certain typological This statement, like all others, has a ceteris paribusrelationship with B and C. The latter, of course, is more caveat. There is a variety of ways in which single-complicated and is more likely to require some explicit unit studies can credibly claim to provide evidencecross-unit examination. However, a descriptive infer- for causal propositions of broad reach—e.g., by choos-ence does not make any assertions about causal rela- ing cases that are especially representative of the phe-tionships (beyond the most proximal) occurring within nomenon under study or by choosing “crucial” casesA, B, and C. In this sense, descriptive inference is sim- (Eckstein [1975] 1992). Even so, a proposition withpler, methodologically speaking. a narrow scope will be more conducive to case study To be sure, descriptive case study propositions are analysis than a proposition with a broad purview. Theimplicitly comparative and these comparisons must breadth of an inference thus constitutes one factor,have a cross-unit reference point. To say “green” is among many, in determining the utility of the case studyto imply “not blue.” However, it is usually fairly easy mode of make such comparisons without conducting a study Similarly, the boundedness of an inference is oftenof the presumed variation. One knows what blue is related to the degree to which it exploits cross-unitwithout going in search of blue cases. This illustrates variance. Precisely because their focus is so tight, casesomething important about the structure of descrip- studies often produce inferences with poorly definedtive propositions in social science. They are held to- boundaries. It is clear that an inference extends be-gether by language—by ordinary or technical terms yond the unit under study, but it is often unclear howand their definitions. When describing a phenomenon far the inference extends. Cross-unit research may alsoone is usually comparing it to an ideal-type definition. suffer from poorly bounded inferences; however, it isAmerican political culture is “liberal” or “republican” less likely to do so since the research design allows oneinsofar as it conforms to standard definitions of these to test the limits of an inference in an explicit fashion.two concepts (Smith 1997). To describe is to categorize,and to categorize is to rely on language to divide upthe world into identifiable entities. Language, in this 5 Granted, cross-unit examination may be useful, particularly whensense, provides the “laws” that allow for consistent in- the terms in question are broad and/or ambiguous. If, for exam-terpretations of the phenomenal world (Sartori 1984). ple, one is examining American political culture as an example of a broader class of political cultures deemed “liberal,” it will be difficultChairs are different from tables in predictable ways; to reach firm conclusions without a larger sample of nation-states.labeling entities as one or the other thus allows the use Even so, it will be easier to describe the subject without cross-unitof nomothetic descriptive generalizations without ac- reference-points than to explain it. 347
  8. 8. What Is a Case Study? May 2004The sharpness of the boundaries of the population— Causal Insight: Causal Effect Versuswhat is and is not “covered” by an inference—is easier Causal Mechanismsto establish when units in the population also constitutecases under study. A fourth tradeoff concerns the sort of insight into cau- By the same token, one of the primary virtues of the sation that one is able to achieve by examining empir-case study method is the depth of analysis that it offers. ical evidence of a particular X:Y relationship. Tradi-One may think of depth as referring to the detail, rich- tionally, quantitative researchers have been concernedness, completeness, wholeness, or degree of variance primarily with the estimation of a causal effect—the ef-that is accounted for by an explanation. The case study fect on Y of a given change in X, taking all backgroundresearcher’s complaint about the thinness of cross-unit circumstances into account (King, Keohane, and Verbaanalysis is well taken; such studies often have little to 1994, 81–82)—and, equally important, an estimate ofsay about individual cases. Otherwise stated, cross-unit the probability of that effect, captured statistically instudies are likely to explain only a small portion of the the error term. Assuming that the causal relationshipvariance with respect to a given outcome or to approach is probabilistic in nature (see discussion below), thethat outcome at a very general level. A cross-unit study researcher must examine several instances of this phe-might be satisfied to explain the occurrence of a revo- nomenon to gauge the average causal effect of X on Ylution in Country A, while a case study of Country A and the random element of that variation. The calcu-might also strive to explain specific features of that lation of a causal effect presumes the investigation ofevent—why it occurred when it did and in the way that cross-unit variation precisely because, for a probabilis-it did. tic cause, one cannot assume that the behavior of one I shall return to the advantages of taking an in-depth, unit will be indicative of the behavior of other units.Gestalt-oriented look, at a single unit. For present pur- Units may behave differently. Thus, the example of aposes, the conclusion is simple. Research designs invari- single unit, even if subjected to iterated testing, is notably face a choice between knowing more about less a good way to estimate causal effects and is certainlyand knowing less about more. The case study method inadequate to the task of estimating probabilities.may be defended, as well as criticized, along these lines But causal arguments depend not only on measur-(Ragin 2000, 22). ing causal effects. They also presuppose the identifi- cation of a causal mechanism (Gerring, forthcoming; Hedstrom and Swedberg 1998). X must be connectedUnit Homogeneity: Case Comparability with Y in a plausible fashion; otherwise, it is unclearVersus Representativeness whether a pattern of covariation is truly causal in nature. The identification of causal mechanisms hap-Single-unit studies provide cases that are likely to be pens when one puts together general knowledge of thecomparable to one another. After all, they are all drawn world with empirical knowledge of how X and Y inter-from the same unit (by definition). Cases drawn from relate. It is in the latter task that case studies enjoy adifferent units, in contrast, often force the researcher comparative make heroic assumptions about the comparability Case studies, if well constructed, allow one to peerof concepts and causal relationships across the chosen into the box of causality to the intermediate causes lyingcases. After all, they are different units. between some cause and its purported effect. Ideally, Yet the strength of the case study also suggests a cor- they allow one to “see” X and Y interact—Hume’sresponding weakness. Single-unit research designs of- billiard ball crossing the table and hitting a second ball.ten fall short in their representativeness—the degree to Clayton Roberts (1996, 66) describes process-tracingwhich causal relationships evidenced by that single unit as “the minute tracing of the explanatory narrative tomay be assumed to be true for a larger set of (unstud- the point where the events to be explained are mi-ied) units. Are the men chosen by Robert Lane (1962) croscopic and the covering laws correspondingly moretypical of the American male, white, immigrant, work- certain” (see also George and Bennett 2004). Often,ing class? Is Middletown representative of America the connections between a putative cause and its ef-(Lynd and Lynd [1929] 1956)? fect are rendered visible once one has examined the The tradeoff between comparability and representa- motivations of the actors involved. Intentionality is antiveness is a general feature of cross-unit sample size. integral part of causal analysis, as interpretivists haveNaturally, there are many ways to overcome the cor- been claiming for some time (Taylor 1970). Similarly,responding problems of comparability and representa- the investigation of a single unit may allow one to testtiveness in both case study and non-case study research. the causal implications of a theory, thus providing cor-Even so, it should be pointed out that the addition of roborating evidence for a causal argument. This is of-units to a research design can hardly increase the case ten referred to as pattern-matching. Here, the theorycomparability of a sample. Similarly, it is unlikely that of primary interest “generates predictions or expecta-the addition of units will decrease the representative- tions on dozens of other aspects of the culture, and [theness of a sample (though it is certainly possible). Thus, it writer] does not retain the theory unless most of theseseems appropriate to regard the tradeoff between com- are also confirmed. In some sense, he has tested the the-parability and representativeness, like other tradeoffs, ory with degrees of freedom coming from the multipleas intrinsic to the study/case study choice of research implications of any one theory” (Campbell [1975] 1988,design. 380).348
  9. 9. American Political Science Review Vol. 98, No. 2 One can readily see why the investigation of fashion; a cause increases the likelihood of an outcomecausal mechanisms (including both process-tracing and and/or the magnitude of a (scalar) outcome.7pattern-matching) is commonly associated with the Ceteris paribus, case study research designs havecase study research design. The in-depth analysis of a an easier time addressing invariant causes. Considersingle unit is useful in elucidating causal mechanisms that a necessary or sufficient causal proposition can bebecause its characteristic style of evidence-gathering— disproved with a single case study (Dion 1998). Provingover-time and within-unit variation—is likely to pro- an invariant causal argument generally requires morevide clues into what connects a purported X to a par- cross-unit cases. However, it is not nearly as compli-ticular Y. Cross-unit variation, in contrast, is often mute cated as proving a probabilistic argument for the simplewith respect to causal mechanisms. The Xs and Ys may reason that one assumes invariant relationships; conse-be at a considerable remove from one another; one quently, the single unit under study carries more weight.does not know, or must simply intuit, what connects Where the causal relationship is assumed to be proba-the dots. bilistic, on the other hand, case study evidence is easier The caveat here is that cross-unit evidence is not al- to dismiss; it is, after all, just one example of a generalways or necessarily mute with respect to causal mech- phenomenon assumed to have a stochastic component.anisms, and single-unit analysis is not always revela- Recall that an error term may be incorporated into casetory on this score. A cross-unit analysis is opaque if, study work since the N of a case study is indeterminate;or insofar as, (a) there is great causal distance be- however, this error term remains a property of single-tween the X and the Y variables, (b) the interven- unit analysis. In contrast, the error term in a large-Ning variables thought to lie between X and Y cannot cross-unit analysis represents the stochastic nature ofbe tested empirically in a cross-unit research design, an (assumed) probabilistic relationship.and (c) the X:Y relationship cannot be intuited fromcommon sense or deductive reasoning. There is a gen-eral perception—common at least among protagonists Strategy of Research: Exploratoryof the case study—that cross-unit studies entail large Versus Confirmatoryblack boxes with no peepholes and that, therefore, they Social science research involves a quest for new theo-must be supplemented by the in-depth analysis of key ries as well as a testing of existing theories, a series ofunits. This is not always the case. One can think of “conjectures and refutations” (Popper 1969). Regret-plenty of studies in which the relationship is quite clear, tably, social science methodology has focused almostand where a case study would be superfluous. And one exclusively on the latter. The former is quickly dis-can point out any number of studies in which interven- missed as a matter of guesswork, inspiration, or luck—aing variables are in fact investigated with a cross-unit leap of faith, in any case, and hence a poor subject forresearch design.6 Again, it is important to stress that we methodological reflection. Yet it will readily be grantedare examining typical, not definitional, characteristics that many works of social science, including most gen-of the case study. Ceteris paribus, case studies are more erally acknowledged classics, are seminal works. Theirlikely to shed light on causal mechanisms and less likely classic status derives from a new idea, a new perspec-to shed light on true causal effects. tive, that is subsequently subjected to more rigorous analysis. Indeed, it is difficult to devise a program ofCausal Relationship: Invariant falsification the first time a new theory is proposed.Versus Probabilistic Path-breaking research is, by definition, exploratory. Subsequent research on that topic is confirmationistCausal arguments may be either invariant (“determin- insofar as its primary task is to verify or falsify a preex-istic”) or probabilistic. Invariant causal relationships isting hypothesis or a set of hypotheses. Thus, the worldare asserted to be always true, given some set of back- of social science may be usefully divided accordingground circumstances. They take the form of necessary, to the predominant strategy of research undertaken,sufficient, or necessary and sufficient arguments. Proba- exploratory or confirmatory/disconfirmatory (Gerringbilistic arguments, in contrast, are true in a probabilistic 2001, chap. 10). These constitute two moments of em- pirical research, a generative moment and a skeptical6 moment, each of which is essential to the progress of a For example, a vast literature addresses the putative causal con-nections between trade openness and the welfare state. The usual discipline.empirical finding is that more open economies are associated with Case studies enjoy a natural advantage in researchhigher social welfare spending. The question then becomes why such of an exploratory nature. These same advantages,a robust correlation exists. What are the plausible interconnectionsbetween trade openness and social welfare spending? One possiblecausal path, suggested by David Cameron (1978), is that increasedtrade openness leads to greater domestic economic vulnerability to 7 I avoid the term “determinism” since it has multiple meanings, onlyexternal shocks (due, for instance, to changing terms of trade). In one of which—invariance—is relevant here. I assume that to makesubsequent work, writers have attempted to test this hypothesis by an invariant causal argument does not commit one to a view that allexamining the correlation between terms of trade and social welfare causes are perfectly determined; some causes may be invariant andspending. If Cameron’s causal mechanism is operative, one should others probabilistic. Useful discussions of invariance/determinismfind a robust correlation between these two variables in a cross- can be found in Adcock 2002, Dion 1998, 141, Goertz and Starrnational regression. As it happens, results are equivocal (Alesina, 2003, and Waldner 2002. Goertz (2003, 76–94) includes a sample ofGlaeser, and Sacerdote 2001). The point is, writers can and do exploit 150 necessary condition hypotheses deployed in various fields of thecross-unit variation to test assumptions about causal mechanisms. social sciences (Goertz 2003, 76–94). 349
  10. 10. What Is a Case Study? May 2004however, often serve as impediments in work of a con- I do not mean to suggest that case studies neverfirmatory nature. Let us explore why this might be so.8 serve a confirmatory role. As discussed, evidence drawn Traditionally, scientific methodology has been identi- from a single unit may disconfirm a necessary or suffi-fied with the segregation of conjecture and refutation; cient hypothesis. Case studies are also often useful inone should not be allowed to contaminate the other. conjunction with a cross-unit study for the purpose ofYet in the real world of social science, inspiration arises elucidating causal mechanisms, as discussed. However,from perspiration. “Lightbulb” moments build on a general theories rarely offer the kind of detailed andclose engagement with the particular facts of a partic- determinate predictions on within-unit variation thatular case (unit). Ragin (1997) notes that case study re- would allow one to reject a hypothesis through pattern-search is all about “casing”—defining the topic, includ- matching (without additional cross-unit evidence). Theing the hypothesis(es) of primary interest, the outcome, point is, theory confirmation/disconfirmation is not theand the set of cases that offer relevant information vis- case study’s strong suit. The selection of “crucial” casesa-vis the hypothesis. It is a highly circular process. A` cannot overcome the fact that cross-unit N is of the French Revolution may be conceptualized We are unlikely to reject a hypothesis, or to consider itas a study of revolution, of social revolution, of revolt, definitively proved, on the basis of the study of a singleof political violence, and so forth. Each of these top- unit, particularly if the hypothesis has behind it a corpusics entails a different population and a different set of of scholarly work. Eckstein himself acknowledges thatcausal factors. A good deal of authorial intervention is his argument for case studies as a form of theory con-necessary in the course of defining a case study topic, firmation is largely conjectural. At the time of writing,for there is a great deal of evidentiary leeway. Yet the several decades ago, he could not point to any existingvery “subjectivity” of case study research allows for the study where a crucial case study had performed thegeneration of a great number of hypotheses, insights heroic role assigned to it (Eckstein [1975] 1992, 172).that might not be apparent to the cross-unit researcher I suspect that this is still more or less true. Indeed, itwho works with a thinner set of empirical data across is true even of experimental case studies in the naturala large number of units and with a more determinate sciences (Campbell and Stanley 1963, 3). A single case(fixed) definition of cases, variables, and outcomes. It study is still a single-shot, a single piece of evidenceis the very fuzziness of case studies that grant them lying at the same level of analysis as the propositiona strong advantage in research at exploratory stages, itself.for the single-unit study allows one to test a multitude The tradeoff between exploratory and confirmatoryof hypotheses in a rough-and-ready way. Nor is this an research helps us to reconcile the enthusiasm of caseentirely conjectural process. The covariational relation- study researchers and the skepticism of case study crit-ships discovered among different elements of a single ics. They are both right, for the looseness of case studyunit have a prima facie causal connection: They are all research is a boon to new conceptualizations just as itat the scene of the crime. This is revelatory when one is a bane to falsification. The problem is that work ofis at an early stage of analysis, for there is no identi- an exploratory nature, although it may receive praisefiable suspect and the crime itself may be difficult to from the discipline, is unappreciated, and greatly un-discern. The fact that A, B, and C are present at the dertheorized, by methodologists.expected times and places (relative to some outcomeof interest) is sufficient to establish them as indepen-dent variables. Proximal evidence is all that is required. Useful Variance: Single- Versus Multiple-UnitHence, the common identification of case studies as“plausibility probes,” “pilot studies,” “heuristic stud- The analysis of any causal relationship hinges on theies,” and “theory-building” exercises (Eckstein [1975] counterfactual assumption—that without X (or with1992; Ragin 1992, 1997; Rueschemeyer and Stephens more or less of X), Y would be different. In investi-1997). gating this assumption the preferred research designs A multiple-unit study, in contrast, generally allows are, in order of preference: (a) laboratory or field ex-for the testing of only a few hypotheses but does so periments, (b) “natural” experiments (where a singlewith a somewhat greater degree of confidence, as is unit undergoes unmanipulated change through timeappropriate to work of a confirmatory nature. There is that approximates a true experiment), (c) thought ex-less room for authorial intervention because evidence periments (counterfactuals), or (d) statistical controlsgathered from a cross-unit research design can only be (a quasi-experimental method of neutralizing irrele-interpreted in a limited number of ways. Another way vant variables so as to isolate the true causal effectsof stating the point is to say that whereas case studies of one or a few factors of theoretical interest). Thelean toward Type 1 errors (falsely rejecting the null hy- implication of this hierarchy of research designs is thatpothesis), cross-unit studies lean toward Type 2 errors a laboratory/field experiment, natural experiment, or(failing to reject the false null hypothesis). Perhaps this thought experiment involving a single unit may be moreexplains why case studies are more likely to be theory- useful than multiple units that attempt to mimic thegenerating, whereas cross-unit studies toil in the prosaic virtues of the experimental method with purely “statis-but highly structured field of normal science. tical” evidence. For example, in investigating the relationship be-8 For discussion of this tradeoff in the context of economic growth tween campaign efforts and voter turnout one mighttheory see Temple 1999, 120. be more convinced by a field experiment conducted in350
  11. 11. American Political Science Review Vol. 98, No. 2a single community than by multiple cross-community tries (the comparative-historical method), than by astudies or individual poll data that rely on a host of cross-country TSCS research design. The same prob-more or less unsatisfactory quasi-experimental con- lem affects the study of many social science questionstrols (Gerber and Green 2000). Similarly, in investigat- where only one unit, or a small number of units, un-ing the role of electoral systems in conditioning pub- dergoes the event that is to be explained: early in-lic policy outcomes one might be more convinced by dustrialization (England and the Netherlands), fascisma single natural experiment—a change in a country’s (Germany, Italy), the use of nuclear weapons (Unitedelectoral system—than by a cross-country study em- States), world war (WWI, WWII), democratization inploying statistical controls to examine these complex Africa (Botswana, Mauritius, South Africa), single-causal relationships (Horowitz 1985, 602). Even where nontransferable vote (SNTV) electoral systems (prere-one is forced to rely solely on counterfactual thought- form Japan, Jordan, Taiwan, Vanuatu), and settler soci-experiments to evaluate causal claims, one might still eties (Australia, Canada, New Zealand, South Africa,prefer the case study mode if useful variance is not United States).available across units. The time-honored question ofwhether early democratization leads to a quiescentworking class and lower levels of social welfare devel- Ontological Considerationsopment (Lipset 1963) is difficult to investigate cross-nationally for the simple reason that only one coun- Thus far, I have looked upon the choice between casetry granted suffrage to the (male) working class prior study and non-case study methods as a matter of logicto industrialization. In this circumstance, a historical (the logic of causal inquiry) and empirics (the inves-study focused on the United States (i.e., a case study) tigation of the empirical world). However, this choicemay provide the most compelling evidence of a general rests also on ontological presuppositions. An ontologyproposition. is a vision of the world as it really is, a more or less co- To clarify, the issue is not whether natural herent set of assumptions about how the world works,experiments—or real experiments for that matter—are a research Weltanschauung analogous to a Kuhniandesirable. They are always desirable. The issue is how paradigm (Hall 2003; Kuhn [1962] 1970). As a resultmany experiments are available and how (truly) exper- of its all-embracing character and ambiguous claims,imental are they? If many experiments are possible, an ontology cannot be proved or disproved, at leastas they typically are in natural science settings, then not in the usual (verificationist or falsificationist) sense.a cross-unit research design is probably justified. But Although it seems odd to bring ontological issues intothis is uncommon in social science settings. Similarly, a discussion of social science methodology, it may bereturning to our previous examples, if nation-states granted that social science research is not a purely em-exhibit substantial variance on the independent vari- pirical endeavor. What one finds is contingent uponables of interest—early democratization and electoral what one looks for, and what one looks for is to somechange—then one would probably choose a cross-unit extent contingent upon what one expects to find.research design (with a time-series dimension). This The significance of ontological presuppositions be-would not preclude a supplementary case study, but comes apparent whenever questions of breadth, casesuch a case study would undoubtedly carry less weight. comparability, and representativeness are vetted. Con-In the event, however, one is thrown back on more sider the possibility that a sample of units will becomeprimitive expedients. Much of the utility of the case more and more dissimilar (less comparable) as thatstudy is “contextual” in this special sense. The sort sample is enlarged, but with no obvious break-points.of natural or experimental variation that would make Here it evidently is the choice of the researcher how tocross-unit analysis fruitful is lacking; such units simply define the population of a given inference and, hence,do not exist and cannot be generated. what the relevant units of analysis will be. Where do This helps to explain why case studies often focus like cases end and unlike cases begin?on rare (“historical”) events. Let me define an event If adjacent units are thought to be entirely noncom-provisionally as an instance of substantial and rela- parable, the case study method is impossible. The per-tively quick change in an independent or dependent fectly ideographic universe displays such uniquenessvariable of theoretical interest. Now imagine a uni- among units that absolutely nothing can be learnedverse of empirical data in which such events occur in about one unit by studying another. The notion ofonly 13 instances. This approximates the situation of a “case study” is nonsensical. At the other extreme,scholars whose work focuses on the phenomenon of where all units of a given type are perfectly comparable,social revolution (e.g., Skocpol 1979, 287). In princi- the case study is equally nonsensical. Why focus on aple, the field of empirical evidence is virtually bound- single unit when other units will do just as well? This isless, including all nation-states over the past three the nomothet’s way of looking at things.centuries. (Let us suppose that this constitutes the do- Case study researchers are situated between thesemain of the inference.) Yet useful variation is severely two extremes. They are dubious about the viabil-limited. The world does not provide many units that ity of comparisons drawn across many apparently di-have experienced revolutionary upheaval. Arguably, verse units. “Variable-oriented” research involves “ho-one learns more about this phenomenon through an mogenizing assumptions” (Ragin 2000, 35). Yet theyintensive case study of France, or several case stud- are equally suspicious of the claim, implicit in muchies focused on the 13 or so social-revolutionary coun- historical and anthropological work, that each unit is 351
  12. 12. What Is a Case Study? May 2004 upon other units brings in its train six methodologi- FIGURE 1. The Ontology of Case Study cal ambiguities that are, to some extent, ineradicable. Research Design First, case studies may build upon a variety of covari- ational evidence; there is no single type of case study evidence, but rather three (see Table 1). Second, case studies assume a distinction between formally and in- formally studied units that is never entirely clear since the latter must be brought into the analysis in some fashion but would compromise the notion of a case study if fully integrated. Third, individual case studies are often grouped together in a single study, thus con- founding the distinction between single- and cross-unit analysis. Fourth, case studies usually perform a double function; they are studies (of the unit itself) as well as case studies (of a broader class of units). Finally, the inference(s) pursued by a case study may be either illustrative or falsifiable. These methodological issues bedevil the case study research design. And they shall continue to do so, for they are inherent in the enterprise. The travails of the case study within the discipline of political science are rooted, finally, in an insuffi- cient appreciation of the methodological tradeoffs that this method calls forth. I have argued that at least seven characteristic strengths and weaknesses must beunique. The case study thus occupies a tenuous onto- considered. Ceteris paribus, case studies are generallylogical ground midway between ideographic and nomo- more useful (1) when inferences are descriptive ratherthetic extremes, as depicted in Figure 1 (see Collier and than causal, (2) when propositional depth is prizedCollier [1991] 2002, 13–14). over breadth and boundedness, (3) when (internal) To be sure, there is no profit in dwelling on ontologi- case comparability is given precedence over (external)cal differences between case study and non-case study case representativeness, (4) when insight into causalresearchers. Ontological debates are, by definition, ir- mechanisms is more important than insight into causalresolvable. Once one has defended one’s position as effects, (5) when the causal proposition at issue is invari-a matter of ontology, further discussion is superfluous ant rather than probabilistic, (6) when the strategy ofexcept as it might bear upon matters of logic and co- research is exploratory, rather than confirmatory, andherence. If social science is understood as an evidence- (7) when useful variance is available for only a singlebased form of inquiry then matters of ontology are unit or a small number of units. (Ontological consider-simply not relevant or are only tangentially relevant. ations also come into play when one chooses betweenNonetheless, insofar as our ontological presuppositions a single-unit and a cross-unit research design, thoughinfluence our construction of cases, we had best be cog- the methodological implications of these assumptionsnizant of this fact. Indeed, the middle-range position are equivocal.)of case study research on this crucial question may Of these seven considerations, the last is perhaps thehelp to account for its ambiguous position in the so- most important. There is little point in pursuing cross-cial sciences. It is neither fish nor fowl, ontologically unit analysis if the units in question do not exhibit vari-speaking. ation on the dimensions of theoretical interest and/or the researcher cannot manage to hold other, potentiallyCONCLUSION confounding, factors constant. Of course, a preliminary canvassing of these units is necessary to perceive theseThis paper has attempted to shed light on the apparent facts about the sample. But beyond this, cross-unit anal-disjuncture between an often-maligned methodology ysis may be of little consequence. By the same token,and a heavily practiced method, both of which go by individual units may offer useful variance or they maythe name of case study. The torment of the case study not. In any case, the most important single questionbegins with its definitional penumbra. Frequently, this researchers should ask themselves as they contemplatekey term is conflated with a set of disparate method- various research designs is which one of these optionsological traits that are not definitionally entailed. The most closely approximates the experimental ideal.first task of the essay, therefore, was to craft a narrower Throughout this discussion I have avoided the discus-and more useful concept for purposes of methodologi- sion of methodological considerations that are practicalcal discussion. The case study, I argued, is best defined in nature or rooted in specific research contexts. Myas an intensive study of a single unit with an aim to concern has been with general methodological issues.generalize across a larger set of units. Before concluding this essay, however, it is important In other respects, the predicament of the case study is to acknowledge that practical and contextual consid-not merely definitional but rather inheres in the method erations are often paramount in the choice between aitself. To study a single unit with intent to shed light case study and a non-case study research format.352
  13. 13. American Political Science Review Vol. 98, No. 2 The collection of original data is typically more dif- or its method of modeling causal relations. The caseficult in cross-unit analysis than in case study analysis, study research design constructs cases from a single unitinvolving greater expense, greater difficulties in iden- while remaining attentive to inferences that span simi-tifying and coding cases, learning foreign languages, lar units outside the formal scope of investigation. Non-traveling, and so forth.9 Whatever can be done for a case study research designs construct cases across unitsset of units can usually be done more easily for a sin- exemplifying the principal causal inference. This dif-gle unit. Similarly, case studies commonly afford multi- ference in “casing” has important consequences, as weple observations of a single case, thus providing firmer have shown. However, it does not render the case studyevidence of the factual accuracy of a given proposi- epistemologically distinct from the cross-unit analysis.tion than would be possible in the analogous cross-unit Indeed, the two modes of analysis are interdependent,study. and this is as it should be. One is at pains to imagine A second practical consideration concerns the state cross-unit research that does not draw upon case studyof research on a given topic. Here one is concerned with work or case study work that disregards adjacent units.the “triangulation” of evidence rather than the ease The strongest conclusion to arise from this methodolog-of evidence-gathering. Social scientists are accustomed ical reflection concerns the complementarity of single-to the idea that research occurs within the context of unit and cross-unit research ongoing tradition. All work is dependent, for theidentification of topic, argument, and evidence, on thisresearch tradition. What we need to know, and hence REFERENCESought to study, is to some extent contingent upon what Acemoglu, Daron, Simon Johnson, and James A. Robinson. already known. It follows from this that the utility of “An African Success Story: Botswana.” In In Search of Prosper-case study research relative to non-case study research ity: Analytic Narratives on Economic Growth, ed. Dani to some extent the product of the state of research Princeton, NJ: Princeton University Press, 80–122. Achen, Christopher H., and Duncan Snidal. 1989. “Rational Deter-within a given field. A field dominated by case studies rence Theory and Comparative Case Studies.” World Politics 41may have little need for another case study. A field (January): 143–69.where cross-unit studies are hegemonic may be des- Adcock, Robert. 2002. “Determinism and Comparative-Historicalperately in need of in-depth studies focused on single Analysis: Clarifying Concepts and Retrieving Past Insights.” Pre-units. sented at the Annual Meeting of the American Political Science Association, Boston. Indeed, much of the debate over the utility of the case Alesina, Alberto, Edward Glaeser, and Bruce Sacerdote. 2001. “Whystudy method has little to do with the method itself and Doesn’t the US Have a European-Style Welfare State?” Brookingsmore to do with the state of current research in that Papers on Economic Activity 2: 187–277.field. If both case study and cross-unit methods have Allen, William Sheridan. 1965. The Nazi Seizure of Power: The Ex- perience of a Single German Town, 1930–1935. New York: Watts.much to recommend them (the implicit argument of Allison, Graham T. 1971. Essence of Decision: Explaining the Cubanthis paper), then both ought to be pursued—perhaps Missile Crisis. Boston: Little, Brown.not in equal measure but at least with equal diligence Bates, Robert H., Avner Greif, Margaret Levi, Jean-Laurentand respect. There is no virtue, and potentially great Rosenthal, and Barry Weingast. 1998. Analytic Narratives. Prince-harm, in pursuing one approach to the exclusion of the ton, NJ: Princeton University Press. Brady, Henry E., and David Collier, eds. 2004. Rethinking Socialother or in ghettoizing the practitioners of the minority Inquiry: Diverse Tools, Shared Standards. Lanham, MD: Rowmanapproach. The triangulation essential to social scien- & Littlefield.tific advance demands the employment of a variety of Cameron, David. 1978. “The Expansion of the Public Economy:(viable) methods, including the case study. A Comparative Analysis.” American Political Science Review 72 (December): 1243–61. This paper is manifestly not a brief for the case study. Campbell, Donald T. [1975] 1988. “‘Degrees of Freedom’ and theRather, it is a brief for the better understanding of the Case Study.” In Methodology and Epistemology for Social Science,case study. We may or may not need more case stud- ed. E. Samuel Overman. Chicago: University of Chicago Press.ies in political science. It is hoped, however, that the Campbell, Donald T., and Julian Stanley. 1963. Experimental andforegoing discussion will encourage better case studies Quasi-Experimental Designs for Research. Boston: Houghton Mifflin.and a greater appreciation of their utility within the Collier, David, and James Mahoney. 1996. “Insights and Pitfalls: Se-discipline. lection Bias in Qualitative Research.” World Politics 49 (October): It should also be apparent that the perceived hos- 56–91.tility between case study and non-case study research Collier, Ruth Berins, and David Collier. [1991] 2002. Shaping the Po- litical Arena: Critical Junctures, the Labor Movement, and Regimeis largely unjustified and, perhaps, deserves to be re- Dynamics in Latin America. Notre Dame, IN: University of Notregarded as a misconception. Case studies may be small- Dame Press.or large-N, qualitative or quantitative, experimental or Coppedge, Michael J. 2002. “Nested Inference: How to Combine theobservational, synchronic or diachronic. The case study Benefits of Large-Sample Comparisons and Case Studies.” Pre-research design comports with any social-scientific the- sented at the Annual Meeting of the American Political Science Association, Boston.oretical framework including behavioralism, rational Dahl, Robert A. 1961. Who Governs?: Democracy and Power in anchoice, institutionalism, and interpretivism. What dif- American City. New Haven, CT: Yale University Press.ferentiates the case study from the cross-unit study is Davidson, P. O., and C. G. Costello, eds. 1969. N = 1: Experimentalits way of defining cases, not its analysis of those cases Studies of Single Cases. New York: Van Nostrand Reinhold. Dion, Douglas. 1998. “Evidence and Inference in the Comparative Case Study.” Comparative Politics 30 (January): 127–45.9Granted, a good deal of cross-unit work involves the reanalysis of Eaton, Kent. 2002. Politicians and Economic Reform in New Democ-existing datasets; in this situation the barriers to entry are not so high. racies. University Park: Pennsylvania State University Press. 353