STOR
Comparative Politics and the Comparative Method
Arend Lijphart
The American Political Science Review, Vol. 65, No. 3. (Sep., 1971), pp. 682-693. Stable URL:
http://linksjstor.org/sici?sici=0003-0554%28197109%2965%3A3%3C682%3ACPATCM%3E2.0.CO%3B2-I
The American Political Science Review is currently published by American Political Science Association.
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/apsa.html.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.
JSTOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact support@jstor.org.
http://www.j stor.org/ Sun Apr 30 15:45:09 2006
Comparative Politics and the Comparative Method*
Arend Lijphart
University of Leiden
Among the several fields or subdisciplines into which the discipline of political science is usually divided, comparative politics is the only one that carries a methodological instead of a substantive label. The term "comparative politics" indicates the how but does not specify the what of the analysis. The label is somewhat misleading because both explicit methodological concern and implicit methodological awareness among students of comparative politics have generally not been very high.1 Indeed, too many students of the field have been what Giovanni Sartori calls "unconscious thinkers" —unaware of and not guided by the logic and methods of empirical science, although perhaps well versed in quantitative research techniques. One reason for this unconscious thinking is undoubtedly that the comparative method is such a basic, and basically simple, approach, that a methodology of comparative political analysis does not really exist. As Sartori points out, the other extreme—that of the "overconscious thinkers," whose "standards of method and theory are drawn from the physical paradigmatic sciences"—is equally unsound.2 The purpose of this paper is to contribute to "conscious thinking" in comparative politics by focusing on comparison as a method of political inquiry. The paper will attempt to analyze not only the inevitable weaknesses and limitations of the comparative method but also its great strengths and potentialities.
* This article is a revised version of a paper presented to the Round Table Conference on Comparative Politics of the International Political Science Association, held in Turin, Italy, September 10-14, 1969. I am very grateful to David E. Apter,- Donald T. Campbell, Robert A. Dahl, Giuseppe Di Palma, Harry Eckstein, Lewis J. Edinger, Samuel E. Finer, Galen A. Irwin, Jean Laponce, Juan J. Linz, Stefano Passigli, Austin Ranney, Stein Rokkan, Dankwart A. Rustow, and Kurt Sontheimer for their comments and suggestions on earlier drafts of the paper, which were very helpful in the preparation of the revision.
'The reverse applies to the relatively new field of "political behavior": its name indicates a substantive field of inquiry, but especially the derivative "behaviorism" has come to stand for a general approach or set of methods. See Robert A. Dahl, "The Behavioral Approach in Political Science: Epitaph for a Monument to a Successful Protest," American Political Science Review, 55 (December, 1961), pp. 763-72.
a Giovanni Sartori, "Concept Misformation in Comparative Politics," American Political Science Review, 64 (December, 1970), p. 1033.
In the literature of comparative politics, a wide variety of meanings is attached to the terms "comparison" and "comparative method." The comparative method is defined here as one of the basic methods—the others being the experimental, statistical, and case study methods—of establishing general empirical propositions. It is, in the first place, definitely a method, not just "a convenient term vaguely symbolizing the focus of one's research interests."3 Nor is it a special set of substantive concerns in the sense of Shmuel N. Eisenstadt's definition of the comparative approach in social research; he states that the term does not "properly designate a specific method . .., but rather a special focus on cross-societal, institutional, or macrosocietal aspects of societies and social analysis."4
Second, the comparative method is here defined as one of the basic scientific methods, not the scientific method. It is, therefore, narrower in scope than what Harold D. Lasswell has in mind when he argues that "for anyone with a scientific approach to political phenomena the idea of an independent comparative method seems redundant," because the scientific approach is "unavoidably comparative."5 Likewise, the definition used here differs from the very similar broad interpretation given by Gabriel A. Almond, who also equates the comparative with the scientific method: "It makes no sense to speak of a comparative politics in political science since if it is a science, it goes without saying that it is comparative in its approach."8
'Arthur L. Kalleberg, "The Logic of Comparison: A Methodological Note on the Comparative Study of Political Systems," World Politics, 19 (October 1966), p. 72.
* Shmuel N. Eisenstadt, "Social Institutions: Comparative Study," in David L. Sills, ed., International Encyclopedia of the Social Sciences (New York: Macmillan & Free Press, 1968), Vol. 14, p. 423. See also Eisenstadt, "Problems in the Comparative Analysis of Total Societies," Transactions of the Sixth World Congress of Sociology (Evian: International Sociological Association, 1966), Vol. 1, esp. p. 188.
»Harold D. Lasswell, "The Future of the Comparative Method," Comparative Politics, 1 (October, 1968), p. 3.
«Gabriel A. Almond, "Political Theory and Political Science," American Political Science Review, 60 (December, 1966), pp. 877-78. Almond also argues that comparative politics is a "movement" in political science rather than a subdiscipline. See his
682
1971
Comparative Politics and the Comparative Method
683
Third, the comparative method is here regarded as a method of discovering empirical relationships among variables, not as a method of measurement. These two kinds of methods should be clearly distinguished. It is the latter that Kalleberg has in mind when he discusses the "logic of comparison." He defines the comparative method as "a form of measurement"; comparison means "nonmetrical ordering," or in other words, ordinal measurement.7 Similarly, Sartori is thinking in terms of measurement on nominal, ordinal (or comparative), and cardinal scales when he describes the conscious thinker as "the man that realizes the limitations of not having a thermometer and still manages to say a great deal simply by saying hot and cold, warmer and cooler."8 This important step of measuring variables is logically prior to the step of finding relationships among them. It is the second of these steps to which the term "comparative method" refers in this paper.
Finally, a clear distinction should be made between method and technique. The comparative method is a broad-gauge, general method, not a narrow, specialized technique. In this vein, Gunnar Heckscher cautiously refers to "the method (or at least the procedure) of comparison,"9 and Walter Goldschmidt prefers the term comparative approach, because "it lacks the preciseness to call it a method."10 The comparative method may also be thought of as a basic research strategy, in contrast with a mere tactical aid to research. This will become clear in the discussion that follows.
The Experimental, Statistical, and Comparative Methods
The nature of the comparative method can be understood best if it is compared and con-
"Comparative Politics," in International Encyclopedia of the Social Sciences, Vol. 12, pp. 331-36. 'Kalleberg, op. cit., pp. 72-73; see also pp. 75-78.
8 Sartori, op. cit., p. 1033. See also Paul F. Lazars-feld and Allen H. Barton, "Qualitative Measurement in the Social Sciences: Classification, Typologies, and Indices," in Daniel Lerner and Harold D. Lasswell, eds., The Policy Sciences: Recent Developments in Scope and Method (Stanford: Stanford University Press, 1951), pp. 155-92.
9 Gunnar Heckscher, The Study of Comparative Government and Politics (London: Allen and Un-win, 1957), p. 68 (italics added).
10 Walter Goldschmidt, Comparative Functionalism: An Essay in Anthropological Theory (Berkeley: University of California Press, 1966), p. 4. Oscar Lewis argues that "there is no distinctive 'comparative method' in anthropology," and that he therefore prefers to discuss "comparisons in anthropology rather than the comparative method." See his "Comparisons in Cultural Anthropology" in William L. Thomas, Jr., ed., Current Anthropology (Chicago: University of Chicago Press, 1956), p. 259.
trasted with the two other fundamental strategies of research; these will be referred to, following Neil J. Smelser's example, as the experimental and the statistical methods.11 All three methods (as well as certain forms of the case study method12) aim at scientific explanation, which consists of two basic elements: (1) the establishment of general empirical relationships among two or more variables,13 while (2) all other variables are controlled, that is, held constant. These two elements are inseparable: one cannot be sure that a relationship is a true one unless the influence of other variables is controlled. The ceteris paribus condition is vital to empirical generalizations.
The experimental method, in its simplest form, uses two equivalent groups, one of which (the experimental group) is exposed to a stimulus while the other (the control group) is not. The two groups are then compared, and any difference can be attributed to the stimulus. Thus one knows the relationship between two variables—with the important assurance that no other variables were involved, because in all respects but one the two groups were alike. Equivalence—that is, the condition that the cetera are indeed paria—can be achieved by a process of deliberate randomization. The experimental method is the most nearly ideal method for scientific explanation, but unfortunately it
u For the idea of discussing the comparative method in relation to these other basic methods, I am indebted to Neil J. Smelser's outstanding and most enlightening article "Notes on the Methodology of Comparative Analysis of Economic Activity," Transactions of the Sixth World Congress of Sociology (Evian: International Sociological Association, 1966), Vol. 2, pp. 101-17. For other general discussions of the comparative method, see L€o Moulin, "La Meth-ode comparative en Science Politique," Revue Internationale d'Histoire Politique et Constitutionelle, 7 (January-June, 1957), pp. 57-71; S. F. Nadel, The Foundations of Social Anthropology (London: Cohen and West, 1951), pp. 222-55; Maurice Duverger, Methodes des Sciences Sociales (3rd ed., Paris: Presses Universitaires de France, 1964), pp. 375-99; John W. M. Whiting, "The Cross-Cultural Method," in Gardner Lindzey, ed., Handbook of Social Psychology (Reading, Mass.: Addison-Wesley, 1954), Vol. 1, pp. 523-31; Frank W. Moore, ed., Readings in Cross-Cultural Methodology (New Haven, Conn.: HRAF Press, 1961); Adam Przeworski and Henry Teune, The Logic of Comparative Social Inquiry (New York: Wiley-Interscience, 1970); and Robert T. Holt and John E. Turner, "The Methodology of Comparative Research," in Holt and Turner, eds., The Methodology of Comparative Research (New York: Free Press, 1970), pp. 1-20.
"The case study method will be discussed below.
13 Eugene J. Meehan, The Theory and Method of Political Analysis (Homewood, 111.: Dorsey Press, 1965). He expresses this idea in three short sentences: "Science seeks to establish relationships" (p. 35); "Science ... is empirical" (p. 37); "Science is a generalizing activity" (p. 43).
684
The American Political Science Review
Vol. 65
can only rarely be used in political science because of practical and ethical impediments.
An alternative to the experimental method is the statistical method. It entails the conceptual (mathematical) manipulation of empiricially observed data—which cannot be manipulated sit-uationally as in experimental design—in order to discover controlled relationships among variables. It handles the problem of control by means of partial correlations. For instance, when one wants to inquire into the relationship between political participation and level of education attained, one should control for the influence of age because younger generations have received more education than older generations. This can be done by partialing—dividing the sample into a number of different age groups and looking at the correlations between participation and education within each separate age group. Paul F. Lazarsfeld states that this is such a basic research procedure that it "is applied almost automatically in empirical research. Whenever an investigator finds himself faced with the relationship between two variables, he immediately starts to 'cross-tabulate,' i.e., to consider the role of further variables."1*
The statistical method can be regarded, therefore, as an approximation of the experimental method. As Ernest Nagel emphasizes, "every branch of inquiry aiming at reliable general laws concerning empirical subject matter must employ a procedure that, if it is not strictly controlled experimentation, has the essential logical functions of experiment in inquiry."15 the statistical method does have these essential logical functions, but it is not as strong a method as experimentation because it cannot handle the problem of control as well. It cannot control for all other variables, merely for the other key variables that are known or suspected to exert influence. Strictly speaking,
"Paul F. Lazarsfeld, "Interpretation of Statistical Relations as a Research Operation," in Lazarsfeld and Morris Rosenberg, eds., The Language of Social Research: A Reader in the Methodology of Social Research (Glencoe, 111.: Free Press, 1955), p. 115. However, control by means of partial correlations does not allow for the effects of measurement error or unique factor components; see Marilynn B. Brewer, William D. Crano and Donald T. Campbell, "Testing a Single-Factor Model as an Alternative to the Misuse of Partial Correlations in Hypothesis-Testing Research, Soci-ometry, 33 (March, 1970), pp. 1-11. Moreover, partial correlations do not resolve the problem of the codiffusion of characteristics, known in anthropology as "Galton's problem"; see Raoul Naroll, "Two Solutions to Galton's Problem," Philosophy of Science, 28 (January, 1961), pp. 15-39, and Przeworski and Teune, op. cit., pp. 51-53.
"Ernest Nagel, The Structure of Science (New York: Harcourt, Brace, and World, 1961), pp. 452f.
even the experimental method does not handle the problem of control perfectly, because the investigator can never be completely sure that his groups are actually alike in every respect.16 But experimental design provides the closest approximation to this ideal. The statistical method, in turn, is an approximation—not the equivalent—of the experimental method. Conversely, one can also argue, as Lazarsfeld does, that the experimental method constitutes a special form of the statistical method, but only if one adds that it is an especially potent form.17
The logic of the comparative method is, in accordance with the general standard expounded by Nagel, also the same as the logic of the experimental method. The comparative method resembles the statistical method in all respects except one. The crucial difference is that the number of cases it deals with is too small to permit systematic control by means of partial correlations. This problem occurs in statistical operations, too; especially when one wants to control simultaneously for many variables, one quickly "runs out of cases." The comparative method should be resorted to when the number of cases available for analysis is so small that cross-tabulating them further in order to establish credible controls is not feasible. There is, consequently, no clear dividing line between the statistical and comparative methods; the difference depends entirely on the number of cases.18 It follows that in many re-
18 For instance, if the groups are made equivalent by means of deliberate randomization, the investigator knows that they are alike with a very high degree of probability, but not with absolute certainty. Moreover, as Hubert M. Blalock, Jr., states, so-called "forcing variables" cannot be controlled by randomization. See his Causal Inferences in Nonexperimental Research (Chapel Hill: University of North Carolina Press, 1964), pp. 23-26. In general, Blalock emphasizes "the underlying similarity between the logic of making causal inferences on the basis of experimental and nonexperimental designs" (p. 26).
"Lazarsfeld, "Interpretation of Statistical Relations as a Research Operation," p. 119. Talcott Parsons makes a similar statement with regard to the comparative method: "Experiment is . . . nothing but the comparative method where the cases to be compared are produced to order and under controlled conditions." See his The Structure of Social Action (2nd ed., New York: Free Press, 1949), p. 743. Another advantage of the experimental method is that the time variable is controlled, which is especially important if one seeks to establish causal relationships. In statistical design, this control can be approximated by means of the panel method.
18 In order to highlight the special problems arising from the availability of only a small number of cases, the comparative method is discussed as a distinct method. Of course, it can be argued with equal justice that the comparative and statistical methods should be regarded as two aspects of a single method. Many authors use the term "comparative method" in the
1971
Comparative Politics and the Comparative Method
685
search situations, with an intermediate number of cases, a combination of the statistical and comparative methods is appropriate. Where the cases are national political systems, as they often are in the field of comparative politics, the number of cases is necessarily so restricted that the comparative method has to be used.
From the vantage point of the general aims and the alternative methods of scientific inquiry, one can consider the comparative method in proper perspective and answer such questions as the following, raised by Samuel H. Beer and by Harry Eckstein: Can comparison be regarded as "the social scientist's equivalent of the natural scientist's laboratory?"19 and: "Is the comparative method in the social sciences .. . really an adequate substitute for experimentation in the natural sciences, as has sometimes been claimed?"20 The answer is that the comparative method is not the equivalent of the experimental method but only a very imperfect substitute. A clear awareness of the limitations of the comparative method is necessary but need not be disabling, because, as we shall see, these weaknesses can be minimized. The "conscious thinker" in comparative politics should realize the limitations of the comparative method, but he should also recognize and take advantage of its possibilities.
broad sense of the method of multivariate empirical, but nonexperimental, analysis, i.e., including both the comparative and statistical methods as defined in this paper. This is how A. R. Radcliffe-Brown uses the term when he argues that "only the comparative method can give us general propositions." (Brown, "The Comparative Method in Social Anthropology," Journal of the Royal Anthropological Institute of Great Britain and Ireland, 81 [1951], p. 22.) fimile Dürkheim also follows this usage when he declares that "comparative sociology is not a particular branch of sociology; it is sociology itself, in so far as it ceases to be purely descriptive and aspires to account for facts." (Dürkheim, The Rules of Sociological Method, translated by Sarah A. Solovay and John H. Mueller, [8th ed., Glencoe, 111.: Free Press, 1938], p. 139.) See also the statements by Lasswell and Almond cited above. Rodney Needham combines the two terms, and speaks of "large-scale statistical comparison," i.e., the statistical method. (Needham, "Notes on Comparative Method and Prescriptive Alliance," Bijdragen tot de Taal-, Land- en Volkenkunde, 118 [1962], pp. 160-82.) On the other hand, E. E. Evans-Pritchard uses exactly the same terminology as used by Smelser and as adopted in this paper, when he makes a distinction between "small-scale comparative studies" and "large-scale statistical ones." See his The Comparative Method in Social Anthropology (London: Athlone Press, 1963), p. 22.
"Samuel H. Beer, "The Comparative Method and the Study of British Politics," Comparative Politics, 1 (October, 1968), p. 19.
20 Harry Eckstein, "A Perspective on Comparative Politics, Past and Present," in Eckstein and David E. Apter, eds., Comparative Politics: A Reader (New York: Free Press of Glencoe, 1963), p. 3.
The Comparative Method: Weaknesses and Strengths
The principal problems facing the comparative method can be succinctly stated as: many variables, small number of cases. These two problems are closely interrelated. The former is common to virtually all social science research regardless of the particular method applied to it; the latter is peculiar to the comparative method and renders the problem of handling many variables more difficult to solve.
Before turning to a discussion of specific suggestions for minimizing these problems, two general comments are in order. First, if at all possible one should generally use the statistical (or perhaps even the experimental) method instead of the weaker comparative method. But often, given the inevitable scarcity of time, energy, and financial resources, the intensive comparative analysis of a few cases may be more promising than a more superficial statistical analysis of many cases. In such a situation, the most fruitful approach would be to regard the comparative analysis as the first stage of research, in which hypotheses are carefully formulated, and the statistical analysis as the second stage, in which these hypotheses are tested in as large a sample as possible.
In one type of comparative cross-national research, it is logically possible and may be advantageous to shift from the comparative to the statistical method. Stein Rokkan distinguishes two aims of cross-national analysis. One is the testing of "macro hypotheses" concerning the "interrelations of structural elements of total systems"; here the number of cases tends to be limited, and one has to rely on the comparative method. The other is "micro replications," designed "to test out in other national and cultural settings a proposition already validated in one setting."21 Here, too, one can use the comparative method, but if the proposition in question focuses on individuals as units of analysis, one can also use the statistical method; as Mer-ritt and Rokkan point out, instead of the "one-nation, one-case" approach, nationality can simply be treated as an additional variable on a par with other individual attributes such as occupation, age, sex, type of neighborhood, etc.22
21 Stein Rokkan, "Comparative Cross-National Research: The Context of Current Efforts," in Richard L. Merritt and Rokkan, eds., Comparing Nations: The Use of Quantitative Data in Cross-National Research (New Haven: Yale University Press, 1966), pp. 19-20. Rokkan specifically recommends the use of "paired comparisons" for this purpose; see his "Methods and Models in the Comparative Study of Nation-Building," in Citizens, Elections, Parties: Approaches to the Comparative Study of the Processes of Development (Oslo: Universitetsforlaget, 1970), p. 52.
"Merritt and Rokkan, op. ext., p. 193.
686
The American Political Science Review
Vol. 65
Terence K. Hopkins and Immanuel Wallerstein make a similar distinction between truly "cross-national studies" in which total systems are the units of analysis, and "multi-national but cross-individual research."23
The second general comment concerns a dangerous but tempting fallacy in the application of the comparative method: the fallacy of attaching too much significance to negative findings. The comparative method should not lapse into what Johan Galtung calls "the traditional quotation/illustration methodology, where cases are picked that are in accordance with the hypothesis—and hypotheses are rejected if one deviant case is found."24 All cases should, of course, be selected systematically, and the scientific search should be aimed at probabilistic, not universal, generalizations. The erroneous tendency to reject a hypothesis on the basis of a single deviant case is rare when the statistical method is used to analyze a large sample, but in the comparative analysis of a small number of cases even a single deviant finding tends to loom large. One or two deviant cases obviously constitute a much less serious problem in a statistical analysis of very many cases than in a comparative study of only a few —perhaps less than ten—cases. But it is nevertheless a mistake to reject a hypothesis "because one can think pretty quickly of a contrary case."25 Deviant cases weaken a probabilistic hypothesis, but they can only invalidate it if they turn up in sufficient numbers to make the hypothesized relationship disappear altogether.26
"Terence K. Hopkins and Immanuel Wallerstein, "The Comparative Study of National Societies," Social Science Information, 6 (October, 1967), pp. 27-33 (italics added). See also Przeworski and Teune, op. cit., pp. 34-43.
24 He adds: "This is a very naive conception of social science propositions; if only perfect correlations should be permitted social science would not have come very far." Johan Galtung, Theory and Methods of Social Research (Oslo: Universitetsforlaget, 1967), p. 505. The functions of deviant case analysis will be discussed below.
25 W. J. M. Mackenzie, Politics and Social Science (Harmondsworth: Penguin Books, 1967), p. 52. I have been guilty of committing this fallacy myself. In my critique of Giovanni Sartori's proposition relating political instability to extreme multipartism (systems with six or more significant parties), one of my arguments consists of the deviance of a single historical case: the stable six-party system of the Netherlands during the interwar years. See Arend Lijphart, "Typologies of Democratic Systems," Comparative Political Studies, 1 (April, 1968), pp. 32-35.
M It is clearly incorrect, therefore, to argue that on logical grounds a probabilistic generalization can never be invalidated; cf. Guenter Lewy's statement: "To be sure, a finding of a very large number of . . . [deviant cases] would cast doubt upon the value of the
After these introductory observations, let us turn to a discussion of specific ways and means of minimizing the "many variables, small N" problem of the comparative method. These may be divided into four categories:
(1) Increase the number of cases as much as possible. Even though in most situations it is impossible to augment the number of cases sufficiently to shift to the statistical method, any enlargement of the sample, however small, improves the chances of instituting at least some control.27 Modern comparative politics has made great progress in this respect as a result of the efforts of the field's innovators to fashion universally applicable vocabularies of basic politically relevant concepts, notably the approaches based on Parsonian theory and Gabriel A. Almond's functional approach.28 Such a restatement of variables in comparable terms makes many previously inaccessible cases available for comparative analysis. In addition to extending the analysis geographically, one should also consider the possibilities of "longitudinal" (cross-historical) extension by including as many historical cases as possible.29
It was the promise of discovering universal laws through global and longitudinal comparisons that made Edward A. Freeman enthusiastically espouse the comparative method almost
proposition, but logically such evidence would not compel its withdrawal. The test of the hypothesis by way of a confrontation with empirical or historical data remains inconclusive." Lewy, "Historical Data in Comparative Political Analysis: A Note on Some Problems of Theory," Comparative Politics, 1 (October, 1968), p. 109.
27 Furthermore, unless one investigates all available cases, one is faced with the problem of how representative one's limited sample is of the universe of cases.
28 On the necessity of establishing general concepts not tied to particular cultures, see Smelser, op. cit., pp. 104-09; Nadel, op. cit., pp. 237-38; Douglas Oliver and Walter B. Miller, "Suggestions for a More Systematic Method of Comparing Political Units," American Anthropologist, 57 (February, 1955), pp. 118-21; and Nico Frijda and Gustav Jahoda, "On the Scope and Methods of Cross-Cultural Research," International Journal of Psychology, 1 (1966), pp. 114— 16. For critiques of recent attempts at terminological innovation in comparative politics, see Sartori, "Concept Misformation in Comparative Politics"; Robert T. Holt and John M. Richardson, Jr., The State of Theory in Comparative Politics (Minneapolis: Center for Comparative Studies in Technological Development and Social Change, 1968); Robert E. Dowse, "A Functionalist's Logic," World Politics, 18 (July, 1966), pp. 607-23; and Samuel E. Finer, "Almond's Concept of "The Political System': A Textual Critique," Government and Opposition, 5 (Winter, 1969-70), pp. 3-21.
29 Michael Haas, "Comparative Analysis," Western Political Quarterly, 15 (June, 1962), p. 298n. See also Lewy, op. cit., pp. 103-10.
1971
Comparative Politics and the Comparative Method
687
a century ago. In his Comparative Politics, published in 1873, he called the comparative method "the greatest intellectual achievement" of his time, and stated that it could lead to the formulation of "analogies . . . between the political institutions of times and countries most remote from one another." Comparative politics could thus discover "a world in which times and tongues and nations which before seemed parted poles asunder, now find each one its own place, its own relation to every other."30 The field of comparative politics has not yet achieved—and may never achieve—the goals that Freeman set for it with such optimism. But his words can remind us of the frequent utility of extending comparative analyses both geographically and historically. (The value of this suggestion is somewhat diminished, of course, because of the serious lack of information concerning most political systems; for historical cases in particular this problem is often irremediable.)
(2) Reduce the "property-space" of the analysis. If the sample of cases cannot be increased, it may be possible to combine two or more variables that express an essentially similar underlying characteristic into a single variable. Thus the number of cells in the matrix representing the relationship is reduced, and the number of cases in each cell increased correspondingly. Factor analysis can often be a useful technique to achieve this objective. Such a reduction of what Lazarsfeld calls the "property-space" increases the possibilities of further cross-tabulation and control without increasing the sample itself.31 It may also be advisable in certain instances to reduce the number of classes into which the variables are divided (for instance, by simplifying a set of several categories into a dichotomy), and thus to achieve the same objective of increasing the average number of cases per cell. The latter procedure, however, has the disadvantage of sacrificing a part of the information at the investigator's disposal, and should not be used lightly.
(3) Focus the comparative analysis on "comparable" cases. In this context, "comparable" means: similar in a large number of important characteristics (variables) which one wants to treat as constants, but dissimilar as far
80 Edward A. Freeman, Comparative Politics (London: Macmillan, 1873), pp. 1, 19, 302. See also Gideon Sjoberg's argument in favor of global comparative research: "The Comparative Method in the Social Sciences," Philosophy of Science, 22 (April, 1955), pp. 106-17.
31 Lazarsfeld and Barton, op. cit., pp. 172-75; Barton, "The Concept of Property-Space in Social Research," in Lazarsfeld and Rosenberg, op. cit., pp. 45-50.
as those variables are concerned which one wants to relate to each other. If such comparable cases can be found, they offer particularly good opportunities for the application of the comparative method because they allow the establishment of relationships among a few variables while many other variables are controlled.32 As Ralph Braibanti states, "the movement from hypothesis to theory is contingent upon analysis of the total range of political systems,"33 but it is often more practical to accord priority to the focus on a limited number of comparable cases and the discovery of partial generalizations.
Whereas the first two ways of strengthening the comparative method were mainly concerned with the problem of "small N," this third approach focuses on the problem of "many variables." While the total number of variables cannot be reduced, by using comparable cases in which many variables are constant, one can reduce considerably the number of operative variables and study their relationships under controlled conditions without the problem of running out of cases. The focus on comparable cases differs from the first recommendation not only in its preoccupation with the problem of "many variables" rather than with "small N," but also in the fact that as a by-product of the search for comparable cases, the number of cases subject to analysis will usually be decreased. The two recommendations thus point in fundamentally different directions, although both are compatible with the second (and also the fourth) recommendation.
This form of the comparative method is what John Stuart Mill described as the "method of difference" and as the "method of concomitant variations." The method of difference consists of "comparing instances in which [a] phenomenon does occur, with instances in other respects similar in which it does not." The
S2Smelser, op. cit., p. 113. Holt and Turner refer to this strategy as the process of "specification" {op. cit., pp. 11-13). It is probably also what Eisenstadt has in mind when he mentions the possibility of constructing "special intensive comparisons of a quasi-experimental nature" {op. cit., p. 424). See also Erwin K. Scheuch, "Society as Context in Cross-Cul-tural Comparison," Social Science Information, 6 (October, 1967), esp. pp. 20-23; Mackenzie, op. cit., p. 151; Fred Eggan, "Social Arithopology and the Method of Controlled Comparison," American Anthropologist, 56 (October, 1954), pp. 743-63; and Erwin Ackerknecht, "On the Comparative Method in Anthropology," in Robert F. Spencer, ed., Method and Perspective in Anthropology (Minneapolis: University of Minnesota Press, 1954), pp. 117-25.
83 Ralph Braibanti, "Comparative Political Analytics Reconsidered," Journal of Politics, 30 (February, 1968), p. 36.
688
The American Political Science Review
Vol. 65
method of concomitant variations is a more sophisticated version of the method of difference: instead of observing merely the presence or absence of the operative variables, it observes and measures the quantitative variations of the operative variables and relates these to each other. As in the case of the method of difference, all other factors must be kept constant; in Mill's words, "that we may be warranted in inferring causation from concomitance of variations, the concomitance itself must be proved by the Method of Difference."34
Mill's method of concomitant variations is often claimed to be the first systematic formulation of the modern comparative method.35 It should be pointed out, however, that Mill himself thought that the methods of difference and of concomitant variations could not be applied in the social sciences because sufficiently similar cases could not be found. He stated that their application in political science was "completely out of the question" and branded any attempt to do so as a "gross misconception of the mode of investigation proper to political phenomena."36 Dürkheim agreed with Mill's negative judgment: "The absolute elimination of adventitious elements is an ideal which cannot really be attained; . . . one can never be even approximately certain that two societies agree or differ in all respects save one."37 These objections are founded on a too exacting scientific standard—what Sartori calls "over-conscious thinking." It is important to remember, however, that in looking for comparable cases, this standard should be approximated as closely as possible.
The area approach appears to lend itself quite well to this way of applying the comparative method because of the cluster of characteristics that areas tend to have in common and that can, therefore be used as controls.38 But opinions on the utility of the area approach
"John Stuart Mill, A System of Logic (8th ed., London: Longmans, Green, Reader, and Dyer, 1872), Book III, chapter 8.
85 Nadel, op. cit., pp. 222-23; Kenneth E. Bock, "The Comparative Method of Anthropology" Comparative Studies in Society and History, 8 (April, 1966), p. 272.
"Mill, op. cit., Book VI, chapter 7; see also Book III, chapter 10.
37 Dürkheim, op. cit., pp. 129-30. But he hailed the method of concomitant variations, which he evidently interpreted to mean a combination of the statistical and comparative methods, as "the instrument par excellence of sociological research" (p. 132). See also Francois Bourricaud, "Science Politique et So-ciologie: Reflexions d'un Sociologue," Revue Franchise de Science Politique, 8 (June, 1958), pp. 251-63.
** If the area approach is often preferable to research efforts with a global range in order to maximize comparability, the era approach may be preferable to longitudinal analysis for the same reason. Cf. the
differ sharply: Gunnar Heckscher states that "area studies are of the very essence of comparative government," and points out that "the number of variables, while frequently still very large, is at least reduced in the case of a happy choice of area."39 Roy C. Macridis and Richard Cox also argue that if areas are characterized by political as well as non-political uniformities, "the area concept will be of great value, since certain political processes will be compared between units within the area against a common background of similar trait configuration"; they cite Latin America as an example of an area offering the prospect of "fruitful intra-area comparison."*0 On the other hand, Dank-wart A. Rustow declares in a recent article that area study is "almost obsolete," and he shows little faith in it as a setting for "manageable comparative study." He argues that "mere geographic proximity does not necessarily furnish the best basis of comparison," and furthermore that "comparability is a quality that is not inherent in any given set of objects; rather it is a quality imparted to them by the observer's perspective."41 This is a compelling argument that should be carefully considered.
It is not true that areas reflect merely geographic proximity; they tend to be similar in many other basic respects. By means of an inductive process—a factor analysis of 54 social and cultural variables on 82 countries—Bruce M. Russett discovered socio-culturally similar groupings of countries, which correspond closely to areas or regions of the world as usually de-
following statement by C. E. Black: "There is much greater value in comparing contemporary events and institutions than those that are widely separated in time. The comparison of societies or smaller groups that are concerned with reasonably similar problems is more likely to lead to satisfactory conclusions than comparisons between societies existing many centuries apart." Black, The Dynamics of Modernization: A Study in Comparative History (New York: Harper and Row, 1966), p. 39. 39 Heckscher, op. cit., p. 88.
"Roy C. Macridis and Richard Cox, "Research in Comparative Politics," American Political Science Review, 47 (September, 1953), p. 654. See also John D. Martz, "The Place of Latin America in the Study of Comparative Politics," Journal of Politics 28 (February, 1966), pp. 57-80.
"Dankwart A. Rustow, "Modernization and Comparative Politics: Prospects in Research and Theory," Comparative Politics, 1 (October, 1968), pp. 45-47. Area study may also be criticized on the ground that, in the words of Dell G. Hitchner and Carol Levine, in Comparative Government and Politics (New York: Dodd, Mead, 1967): "Its very method of delimitation puts emphasis on what may be particular to a limited group of states, as opposed to the universal generalizations which fully comparative study must seek" (pp. 7-8). This argument has been answered above in terms of the need for partial generalizations as a first step. See also Braibanti, op. cit., pp. 54-55.
1971
Comparative Politics and the Comparative Method
689
fined.42 Comparability is indeed not inherent in any given area, but it is more likely within an area than in a randomly selected set of countries. It seems unwise, therefore, to give up the area approach in comparative politics. But two important provisos should be attached to this conclusion. First, the area approach can contribute to comparative politics if it is an aid to the comparative method, not if it becomes an end in itself. Otherwise, area study may indeed become "a form of imprisonment."43 It is against this danger that the thrust of Rustow's argument is directed. Second, the area approach should not be used indiscriminately, but only where it offers the possibility of establishing crucial controls. In this respect, some of the smaller areas may offer more advantages than the larger ones—Scandinavia, for example, which has barely been exploited in this manner, or the Anglo-American countries, which have received greater comparative attention (but which do not constitute an area in the literal sense).44
An alternative way of maximizing comparability is to analyze a single country diachroni-cally. Such comparison of the same unit at different times generally offers a better solution to the control problem than comparison of two or more different but similar units (e.g., within the same area) at the same time, although the control can never be perfect; the same country is not really the same at different times. A good example of diachronic comparative analysis is Charles E. Frye's study of the empirical relationships among the party system, the interest group system, and political stability in Germany under the Weimar and Bonn Republics. Frye argues that "for the study of these relationships, Weimar and Bonn make a particu-
42 Bruce M. Russett, "Delineating International Regions," in J. David Singer, ed., Quantitative International Politics: Insights and Evidence (New York: Free Press, 1968), pp. 317-52. See also Russett, International Regions and the International System (Chicago: Rand McNally, 1967).
"George I. Blanksten, "Political Groups in Latin America," American Political Science Review, 53 (March, 1959), p. 126. See also Sigmund Neumann, "The Comparative Study of Politics," Comparative Studies in Society and History, 1 (January, 1959), pp. 107-10; and I. Schapera, "Some Comments on the Comparative Method in Social Anthropology," American Anthropologist, 55 (August, 1953), pp. 353-361, esp. p. 360.
44 See Seymour Martin Lipset, "The Value Patterns of Democracy: A Case Study in Comparative Analysis," American Sociological Review, 28 (August, 1963), pp. 515-31; Robert R. Alford, Party and Society: The Anglo-American Democracies (Chicago: Rand McNally, 1963); Leslie Lipson, "Party Systems in the United Kingdom and the Older Commonwealth: Causes, Resemblances, and Variations," Political Studies, 7 (February, 1959), pp. 12-31.
larly good case [strictly speaking, two cases] because there are more constants and relatively fewer variables than in many cross-national studies. Yet the differences could hardly be sharper."45
Unless the national political system itself constitutes the unit of analysis, comparability can also be enhanced by focusing on intrana-tion instead of internation comparisons. The reason is again the same: comparative intrana-tion analysis can take advantage of the many similar national characteristics serving as controls.46 Smelser illustrates the utility of this strategy with the example of a hypothetical research project on industrialization in Germany and Italy: "For many purposes it would be more fruitful to compare northern Italy with southern Italy, and the Ruhr with Bavaria, than it would be to compare Germany as a whole with Italy as a whole. These two countries differ not only in level of industrialization, but also in cultural traditions, type of governmental structure, and so on." The advantage of intra-unit comparison is that inter-unit differences can be held constant. "Then, having located what appear to be operative factors in the in-tra-unit comparisons, it is possible to move to the inter-unit comparisons to see if the same differences hold in the large."47
As Juan J. Linz and Amando de Miguel point out, a particularly promising approach may be the combination of intranation and internation comparisons: "The comparison of those sectors of two societies that have a greater number of characteristics in common while differing on some crucial ones may be more fruitful than overall national comparisons."48 An illustrative example of this approach in the political realm is suggested by Raoul Naroll: "If one wishes to test theories about the difference between the cabinet and
43 Charles E. Frye, "Parties and Pressure Groups in Weimar and Bonn," World Politics, 17 (July, 1965), pp. 635-55. (The quotation is from page 637.) The postwar division of Germany also offers the opportunity of analyzing the effects of democratic versus totalitarian development against a similar cultural and historical background. See Ralf Dahrendorf, "The New Germanies: Restoration, Revolution, Reconstruction," Encounter, 22 (April, 1964), pp. 50-58. See also Sylvia L. Thrupp, "Diachronic Methods in Comparative Politics," in Holt and Turner, eds., The Methodology of Comparative Research, pp. 343-58.
"Heckscher, p. 69; Heinz Eulau, "Comparative Political Analysis: A Methodological Note," Midwest Journal of Political Science, 6 (November, 1962), pp. 397-407. Rokkan, too, warns against the "whole-nation" bias of comparative research ("Methods and Models," p. 49).
"Smelser, op. cit, p. 115.
48 Juan J. Linz and Amando de Miguel, "Within-
690
The American Political Science Review
Vol. 65
the presidential systems of government. . . one is better advised to compare Manitoba and North Dakota than to compare Great Britain and the United States, since with respect to all other variables Manitoba and North Dakota are very much alike, while Great Britain and the United States have many other differences."*9
(4) Focus the comparative analysis on the "key" variables. Finally, the problem of "many variables" may be alleviated not only by some of the specific approaches suggested above but also by a general commitment to theoretical parsimony. Comparative analysis must avoid the danger of being overwhelmed by large numbers of variables and, as a result, losing the possibility of discovering controlled relationships, and it must therefore judiciously restrict itself to the really key variables, omitting those of only marginal importance. The nature of the comparative method and its special limitations constitute a strong argument against what Lass-well and Braibanti call "configurative" or "contextual" analysis: "the identification and interpretation of factors in the whole social order which appear to affect whatever political functions and their institutional manifestations have been identified and listed for comparison" (Braibanti's definition).60 Lasswell argues that the comparative method as usually applied has been insufficiently configurative, and calls for the exploration of more variables: the entire context—past, present, and future—"must be continually scanned."51
Scanning all variables is not the same as including all variables, of course, as long as one is on one's guard against an unrealistic and eventually self-defeating perfectionism. Comparative politics should avoid the trap into which the decision-making approach to the study of international politics fell, of specifying and calling for the analysis of an exhaustive list of all variables that have any possible influence on the decision-making process.52 Parsimony
Nation Differences and Comparisons: The Eight Spains," in Merritt and Rokkan, op. cit., p. 268.
"Naroll, "Scientific Comparative Politics and International Relations," in R. Barry Farrell, ed., Approaches to Comparative and International Politics (Evanston, 111.: Northwestern University Press, 1966), pp. 336-37.
"Braibanti, op. cit., p. 49. In this context, "configurative" analysis is not synonymous with the traditional single-country approach, as in Eckstein's definition of the term: "the analysis of particular political systems, treated either explicitly or implicitly as unique entities" ("A Perspective on Comparative Politics," p. ID-
™ Lasswell, op. cit., p. 6.
52 See Richard C. Snyder, H. W. Bruck, and Burton
suggests that Joseph LaPalombara's call for a "segmented approach" aiming at the formulation of middle-range propositions concerning partial systems makes a great deal of sense.53 Similarly, Eckstein's urgent call for greater manageability of the field should be carefully heeded: "The most obvious need in the field at present is simplification—and simplification on a rather grand scale—for human intelligence and scientific method can scarcely cope with the large numbers of variables, the heaps of concepts, and the mountains of data that seem at present to be required, and indeed to exist, in the field."5*
It is no accident that the most fruitful applications of the comparative method have been in anthropological research. In primitive societies, the number of variables is not as bewilderingly large as in more advanced societies. All relevant factors can therefore be more easily surveyed and analyzed. In this respect, anthropology can be said to provide "almost a laboratory for the quasi-experimental approach to social phenomena."55 Political science lacks this advantage, but can approximate it by focusing attention on the key variables in comparative studies.
A final comment is in order about the relationship of comparative politics as a substantive field and comparison as a method. The two are clearly not coterminous. In comparative politics, other methods can often also be employed, and the comparative method is also applicable in other fields and disciplines. A particularly instructive example is James N. Rosenau's study of the relative influence of individual variables (personal policy beliefs and "personalizing tendencies") and role variables (party role and committee role) on the behavior of United States senators during two similar periods: the "Acheson era," 1949-1952, and the "Dulles era," 1953-1956. Rosenau argues that these two eras were characterized by a generally similar international environment and that the two secretaries of state conducted similar foreign
Sapin, eds., Foreign Policy Decision-Making (New York: Free Press of Glencoe, 1962).
"Joseph LaPalombara, "Macrotheories and Micro-applications in Comparative Politics," Comparative Politics, 1 (October, 1968), pp. 60-77. As an example he cites Robert A. Dahl, ed., Political Oppositions in Western Democracies (New Haven: Yale University Press, 1966), esp. chapters 11-13. See also LaPalombara, "Parsimony and Empiricism in Comparative Politics: An Anti-Scholastic View," in Holt and Turner, eds., The Methodology of Comparative Research, pp. 123-49.
"Eckstein, "A Perspective on Comparative Politics," p. 30. M Nadel, op. cit., p. 228.
1971
Comparative Politics and the Comparative Method
691
policies and also resembled each other in personal qualities. He terms the method that he uses in his analysis the method of "quantitative historical comparison." One of its basic characteristics is the testing of hypotheses by comparing two eras (cases) that are "essentially comparable ... in all respects except for the . . . variables being examined." The method is called "quantitative" because the variables are operationally defined in quantitative terms, and "historical" because the two cases compared are historical eras.56 The method is, therefore, a special form of the comparative method. It illustrates one of very many ways in which an imaginative investigator can devise fruitful applications of the comparative method.57
The Comparative Method and the Case Study Method
The discussion of the comparative method is not complete without a consideration of the case study method. The statistical method can be applied to many cases, the comparative method to relatively few (but at least two) cases, and the case study method to one case. But the case study method can and should be closely connected with the comparative method (and sometimes also with the statistical method); certain types of case studies can even be considered implicit parts of the comparative method.
The great advantage of the case study is that by focusing on a single case, that case can be intensively examined even when the research resources at the investigator's disposal are relatively limited. The scientific status of the case study method is somewhat ambiguous, however, because science is a generalizing activity. A single case can constitute neither the basis for a valid generalization nor the ground for disproving an established generalization.
Indirectly, however, case studies can make
M James N. Rosenau, "Private Preferences and Political Responsibilities: The Relative Potency of Individual and Role Variables in the Behavior of U.S. Senators," in Singer, ed., Quantitative International Politics, pp. 17-50, esp. p. 19. Rosenau adds that if "the findings are not so clear as to confirm or negate the hypotheses unmistakably, then of course the analyst moves on to a third comparable period" (p. 19). If such a third or even more periods can be found—which seems unlikely in the case of Rosenau's particular research problem—they should be included regardless of the outcome of the analysis of the first two eras (if the available resources permit it, of course).
87 See also the proposed use of "multiple comparison groups," as an approximation of the experimental method, by Barney G. Glazer and Anselm L. Strauss, "Discovery of Substantive Theory: A Basic Strategy Underlying Qualitative Research," American Behavioral Scientist, 8 (February, 1965), pp. 5-12.
an important contribution to the establishment of general propositions and thus to theory-building in political science. Six types of case studies may be distinguished. These are ideal types, and any particular study of a single case may fit more than one of the following categories:
(1) Atheoretical case studies;
(2) Interpretative case studies;
(3) Hypothesis-generating case studies;
(4) Theory-confirming case studies;
(5) Theory-infirming case studies;
(6) Deviant case studies.
Cases may be selected for analysis because of an interest in the case per se or because of an interest in theory-building. The first two types of cases belong to the former category. Atheoretical case studies are the traditional single-country or single-case analyses. They are entirely descriptive and move in a theoretical vacuum: they are neither guided by established or hypothesized generalizations nor motivated by a desire to formulate general hypotheses. Therefore, the direct theoretical value of these case studies is nil, but this does not mean that they are altogether useless. As LaPalombara emphasizes, the development of comparative politics is hampered by an appalling lack of information about almost all of the world's political systems.58 Purely descriptive case studies do have great utility as basic data-gathering operations, and can thus contribute indirectly to theory-building. It can even be claimed that "the cumulative effect of such studies will lead to fruitful generalization," but only if it is recognized that this depends on a theoretically oriented secondary analysis of the data collected in atheoretical case studies.59
As indicated earlier, the atheoretical case study and the other types of case studies are ideal types. An actual instance of an atheoretical case study probably does not exist, because almost any analysis of a single case is guided by at least some vague theoretical notions and some anecdotal knowledge of other cases, and usually results in some vague hypotheses or conclusions that have a wider applicability. Such actual case studies fit the first type to a large extent, but they also fit one or more of the other types (particularly the third, fourth, and fifth types) at least to some extent.
** LaPalombara, "Macrotheories and Microapplica-tions," pp. 60-65.
" See Michael Curtis, Comparative Government and Politics: An Introductory Essay in Political Science (New York: Harper and Row, 1968), p. 7. See also Macridis, The Study of Comparative Government (New York: Random House, 1955).
692
The American Political Science Review
Vol. 65
Interpretative case studies resemble atheoret-ical case studies in one respect: they, too, are selected for analysis because of an interest in the case rather than an interest in the formulation of general theory. They differ, however, in that they make explicit use of established theoretical propositions. In these studies, a generalization is applied to a specific case with the aim of throwing light on the case rather than of improving the generalization in any way. Hence they are studies in "applied science." Since they do not aim to contribute to empirical generalizations, their value in terms of theory-building is nil. On the other hand, it is precisely the purpose of empirical theory to make such interpretative case studies possible.60 Because of the still very limited degree of theoretical development in political science, such case studies are rare. One interesting example is Michael C. Hudson's imaginative and insightful case study of Lebanon in the light of existing development theories, in which he discovers a serious discrepancy between the country's socio-economic and political development.61
The remaining four types of case studies are all selected for the purpose of theory-building. Hypothesis-generating case studies start out with a more or less vague notion of possible hypotheses, and attempt to formulate definite hypotheses to be tested subsequently among a larger number of cases. Their objective is to develop theoretical generalizations in areas where no theory exists yet. Such case studies are of great theoretical value. They may be particularly valuable if the case selected for analysis provides what Naroll calls a sort of "crucial experiment" in which certain variables of interest happen to be present in a special way.62
Theory-confirming and theory-infirming case studies are analyses of single cases within the framework of established generalizations. Prior knowledge of the case is limited to a single
MAs Przeworski and Teune state: "The main role of a theory is to provide explanations of specific events. These explanations consist of inferring, with a high degree of probability, statements about particular events from general statements concerning classes of events" (p. 86).
61 Michael C. Hudson, "A Case of Political Underdevelopment," Journal of Politics, 29 (November, 1967), pp. 821-37. See also Beer, "The Comparative Method and the Study of British Politics," pp. 19-36.
62 Naroll, "Scientific Comparative Politics and International Relations," p. 336. An example of such a case study is my analysis of the determinants of Dutch colonialism in West Irian. In most cases, both objective (especially economic) and subjective factors can be discerned, but the case of West Irian is unique because of the complete absence of objective Dutch interests in the colony. See Lijphart, The Trauma of Decolonization: The Dutch and West New Guinea (New Haven: Yale University Press, 1966).
variable or to none of the variables that the proposition relates. The case study is a test of the proposition, which may turn out to be confirmed or infirmed by it. If the case study is of the theory-confirming type, it strengthens the proposition in question. But, assuming that the proposition is solidly based on a large number of cases, the demonstration that one more case fits does not strengthen it a great deal. Likewise, theory-infirming case studies merely weaken the generalizations marginally. The theoretical value of both types of case studies is enhanced, however, if the cases are, or turn out to be, extreme on one of the variables: such studies can also be labeled "crucial experiments" or crucial tests of the propositions.
Deviant case analyses are studies of single cases that are known to deviate from established generalizations. They are selected in order to reveal why the cases are deviant—that is, to uncover relevant additional variables that were not considered previously, or to refine the (operational) definitions of some or all of the variables.63 In this way, deviant case studies can have great theoretical value. They weaken the original proposition, but suggest a modified proposition that may be stronger. The validity of the proposition in its modified form must be established by further comparative analysis.64
Of the six types of case studies, the hypothesis-generating and the deviant case studies have the greatest value in terms of their contribution to theory. Each of these two types, however, has quite different functions in respect to theory-building: The hypothesis-generating case study serves to generate new hypotheses, while the deviant case study refines and sharpens existing hypotheses. The deviant case study—as
63 See Patricia L. Kendall and Katherine M. Wolf, "The Analysis of Deviant Cases in Communications Research," in Lazarsfeld and Frank Stanton, eds., Communications Research: 1948-49 (New York: Harper, 1949), pp. 152-57; Sjoberg, op. cit., pp. 114-15; and Lijphart, The Politics of Accommodation: Pluralism and Democracy in the Netherlands (Berkeley: University of California Press, 1968), chapter 10.
"This process of refining generalizations through deviant case analysis is what Robert M. Marsh calls "specification." See his article "The Bearing of Comparative Analysis on Sociological Theory," Social Forces, 43 (December, 1964), pp. 191-96. Specification should therefore definitely not be regarded as "the garbage bin" of comparative research; see Conrad Phillip Kottak, "Towards a Comparative Science of Society," Comparative Studies in Society and History, 12 (January, 1970), p. 102. See also Milton M. Gordon, "Sociological Law and the Deviant Case," Soci-ometry, 10 (August, 1947), pp. 250-58; and Andre J. F. Kobben, "The Logic of Cross-Cultural Analysis: Why Exceptions?", in Rokkan, ed., Comparative Research Across Cultures and Nations (Paris: Mouton, 1968), pp. 17-53.
1971
Comparative Politics and the Comparative Method
693
well as the theory-confirming and theory-in-firming case studies—are implicitly comparative analyses. They focus on a particular case which is singled out for analysis from a relatively large number of cases and which is analyzed within the theoretical and empirical context of this set of cases. The deviant case may be likened to the "experimental group" with the remainder of the cases constituting the "control group." Just as the analytical power of the comparative method increases the closer it approximates the statistical and experimental methods, so the analytical power of the case study method increases the more it approximates the comparative method in the form of deviant case analysis. Such case analysis requires, of course, that the position of the deviant case on the variables under consideration, and consequently also its position relative to the other cases, are clearly defined.
The different types of cases and their unequal potential contributions to theory-building should be kept in mind in selecting and analyzing a single case. Some of the shortcomings in Eckstein's otherwise insightful and thought-provoking case study of Norway may serve as instructive examples.85 Eckstein argues that the Norwegian case deviates from David B. Truman's proposition concerning "overlapping memberships,"66 because Norway is a stable democracy in spite of the country's deep and non-overlapping geographic, economic, and cultural cleavages. But he fails to place the case of Norway in relation to other cases. In fact, although he describes Norway's divisions as "astonishingly great, sharp, and persistent," he explicitly rules out any comparison with the cleavages in other countries. This exclusion seriously weakens the case study. Furthermore, instead of trying to refine Truman's proposition with the help of the deviant findings, Eckstein simply drops it. In terms of the sixfold typology of case studies discussed above, his analysis of the Norwegian case is only a theory-infirming one and is not made into a deviant case study.
From then on, the case study becomes a the-
85 Eckstein, Division and Cohesion in Democracy: A Study of Norway (Princeton, N.J.: Princeton University Press, 1966), esp. pp. 60-77, 177-201. Part of the critique which follows is included in my review of this book in the Journal of Modern History, 41 (March, 1969), pp. 83-87.
66 David B. Truman, The Governmental Process: Political Interests and Public Opinion (New York: Knopf, 1951).
ory-confirming one. Eckstein finds that the Norwegian case strikingly bears out his own "congruence" theory, which states that governments tend to be stable if there is considerable resemblance (congruence) between governmental authority patterns and the authority patterns in society.67 He demonstrates persuasively that both governmental and social patterns of authority are strongly democratic in Norway and thus highly congruent. The problem here is not that the Norwegian facts do not fit the theory, but that they fit the theory too perfectly. The perfect fit strengthens the theory marginally, but does not contribute to its refinement. The theory does not hold that complete congruence of authority patterns is required for stable democracy. In his original statement of the congruence theory, Eckstein himself points out the necessity of further work on the important questions of how much disparity can be tolerated and how degrees of congruence and disparity can be measured.68 Because the Norwegian case turns out to be a perfect theory-confirming one, it cannot be used to refine the theory in any of these respects. Therefore, Eckstein was unlucky in his selection of this case as far as the development of his congruence theory is concerned, and he fails to take full advantage of the case study method in analyzing the case in terms of Truman's theory of overlapping memberships.
*   * *
The comparative method and the case study method have major drawbacks. But precisely because of the inevitable limitations of these methods, it is the challenging task of the investigator in the field of comparative politics to apply these methods in such a way as to minimize their weaknesses and to capitalize on their inherent strengths. Thus, they can be highly useful instruments in scientific political inquiry.
07 In one respect, it is not altogether correct to call the Norwegian case study a theory-confirming study. Because the congruence theory has a rather narrow empirical basis, consisting chiefly of only two cases (Britain and Germany), it is a hypothesis rather than an established theory. The case study of Norway is, of course, not a hypothesis-generating study either. Perhaps it should be called a "hypothesis-strengthening" case study or, as Eckstein himself suggests, a "plausibility probe" (oral comment at the IPSA Round Table Conference in Turin, September 1969).
68 Eckstein, A Theory of Stable Democracy, Research Monograph No. 10 (Princeton, N.J.: Center of International Studies, 1961).