■ CHAPTER ONE Comparative Research and Social Science Theory Explanation and Theory in Social Science. Theory and Spatiotemporal Parameters: The Postulate of Substitittability. The Status of Proper Names of Systems. Summary. While a precise definition of comparative research will not be presented immediately, we assume in this book that the goal of social science is to explain social phenomena. We further postulate that the generality and parsimony of theories should be given primacy over their accuracy. In other words social science theories, rather than explaining pTicnoh1ferTa'as accurately as possible in terms relative to specific historical circumstances, should attempt to explain phenomena wherever and whenever they occur. Although this position is not new, this chapter is written with the belief that the implications of this preference for the conduct of comparative research will become clear when the assumptions underlying this choice are understood. We shall first discuss the assumptions underlying the construe? tion of general and parsimonious theories and, second, recommend son\i| procedures of data-gathering and data analysis in comparative research"? We will argue that the bridge between historically anchored observations and theoretical statements can be found in comparative studies and that general theories cannot be constructed in the social sciences without explicit reference to- factors operating at the level of systems. Explanation and Theory in the Social Sciences In 1954 in Bennington, Vermont, small businessmen supported the radical right more often than salaried employees with the same education. In 1963 in Poland, women who had had premarital sexual experience perceived their marriages as more successful than those who had had no pre- 17 18 Comparative Research and Social Science Theory Explanation and Theory in the Social Sciences 19 marital experiences. In 1958 in Finland, persons with congruent status (corresponding levels of income, education, and occupation) voted for the leftist parties more often than persons with incongruent status. These deliberately selected and disparate findings of social scientists can one day find their way into history books and become a part of the human heritage in such documents as the book relating the eating habits of Frenchmen during the second half of the eighteenth century or some future work reporting on the daily life in Vermont in the middle of the twentieth century. The way in which these findings are formulated makes us think of the future historian: most social scientists are more interested in finding out why social phenomena occur than where and when. But all observations of the sociopolitical realm are anchored in time and space. It is in Vermont, Poland, and Finland that these observations are made. They are made at a certain time and in a certain place, and if we were more concerned with historical veracity than with theoretical generality, we would never extend the findings beyond the particular spatiotemporal parameters within which the observations were made. The concern with building general theories of human behavior constitutes such an overriding goal of social scientists, however, that they are willing to risk the error of false generalizations rather than give up that concern.1 Thus we learn that "frustration brings aggression," that "outgroup hostility breeds ingroup solidarity," that "the disparity of culturally legitimate goals and means results in anomie," and that "a high level of economic development is necessary for a stable democratic political system." The goal of science is to explain and predict why certain events occur when and where they do. Why did small businessmen in Vermont support McCarthy in 1954? Why was the Kowalski marriage not successful? Why did Smith commit a crime? Why did Napoleon attack Russia? Science is concerned with the explanation of specific events by means of statements that are invariantly true from one set of circumstances to another. But what does it mean to "explain" or "predict" a concrete, specific event? Since a discussion of "explanation" exceeds the limits of this work, we 1A discussion of the types of historical generalizations and errors associated with those generalisations can be found in Stephan Nowak, "General Laws and Historical Generalizations in the Social Sciences," Polish Sociological Bulletin, 1, 1961, pp. 21-30. Nowak defines the problem in the following way: "If the sociologist is cautious, he is also alert to the fact that the more the limits of the validity of his theory exceed the investigated reality, the greater is the danger of his statements being false. . . If in addition he [the sociologist] is acquainted with the postulates of the methodology of science, he usually wants his propositions to be universal, free from limitations of time and space, so that they become scientific laws, since he is aware that statements of this type have many particularly valuable theoretical properties." shall base the subsequent analysis on the paradigm of explanation proposed by C. G. Hempel.2 Even Hempel's views, however, will not be presented here in their entirety, but will only be used as the basis for the discussion of comparative explanations of social phenomena. To explain a specific event is to state the conditions under which it always or usually takes place, that is, to cite general statements (laws) from which other statements concerning properties of specific events can be inferred with some reasonable certainty. In the social sciences such an explanation will most often be ofa" statistical nature. In order to understand why an individual behaved in a certain way in a given situation, we invoke general probabilistic statements that say that, for an individual of a particular type, it is likely that he will behave in this way, given this type of a situation. For example, why does Monsieur Rouget, age 24, blond hair, brown eyes, a worker in a large factory, vote Communist? To explain the vote of M. Rouget, one must rely upon general probabilistic statements that are relevant for voting behavior and have been sufficiently confirmed against various sets , of evidence. The particular features of M. Rouget must be used as the first premise of the explanation: M.'\3.ouget is a worker and works in a large factory and is young (24 years old). The second premise consists of a conjunction of general statements describing with a high likelihood the behavior of skilled workers, employees of large factories, and young persons. (No interaction is assumed.) One out of every two workers votes Communist; and employees of large organizations vote Communist more often than employees of small organizations; and young people vote Communist more often than older people. Therefore, it is likely that - ■ M. Rouget votes Communist. This explanation is incomplete. The probability of a French worker, 24, employed in a large factory, and voting Communist is still far from 1.00. Several other factors, such as place of residence, marital status, father's occupation, religiosity, and so forth, might have to be considered if the explanation (prediction) of M. Rouget's behavior were to approach certainty. Most explanations in the social sciences are incomplete in the ' C. G. Hempel, Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, Free Press, New York, 1965. See particularly the title essay, pp. 229-497. 20 Comparative Research and Social Science Theory sense that the probability, .of. the explained phenomenon taking place does not approach 1.00 (or zero). Since the rules of inference are probabilistic, we cannot expect that, even if the premises are true, the conclusions will invariably follow. As the probability of inferential rules increases, however, the probability of predicting a property also increases—it moves away from what could be expected randomly. The general statements thai serve as premises in an explanation constitute a theory. Usually more than one general statement is necessary to provide a relatively complete explanation. These statements must have certain logical properties: they must be interconnected, and none of their implications can contradict any other implication. Some rules must also be available to determine whether a specific event is an element of the class covered by the theory. In other words, these statements must be empirically interpretable. Finally, the set of such general statements should include a formal deductive framework, such that the inferred consequence is not an intuitively obvious result of the premises. The logical structure of theories, however, is not our central concern. The accumulation of knowledge consists of the process of gradual cot-firmation and/or modification of the theories that serve as the genial premises in the explanatory scheme. If we are to understand "what "happened in the last election in Denmark," "why was there a drop in suicides in Southern Italy," or "what made Joe drop out of school," we must have available a set of general theories sufficiently confirmed to provide a reasonable certainty that when these general statements are applicable, the expected consequences will follow. At any stage in the development of science it is likely, however, that more than one theory will explain the same class of events. Therefore additional goals are postulated that provide criteria for the evaluation of i theories: accuracy, generality, parsimony, and causality. i First, we expect a theory to be accurate, to explain as completely as possible, and to predict as much of the variation as possible. This criterion can be expressed in terms of the amount of variance accounted'for by the independent variables—the more variance accounted for by a. theory, the :. smaller the error of prediction. For comparative research this criterion implies that the goal of social science theory is to explain a given phenomenon as accurately as possible in each social system. If we wanted to explain the incidence of divorce, for example, we would construct theories that would minimize the error of prediction in each social system. We might find that in one society 99.9 percent of the variance of divorces can be explained by the education of spouses, their religion, and the degree of rigidity. In another social system, we may again be able to account for 99.9 percent of Explanation and Theory in the Social Sciences 21 the variance of divorces, but with the use of different factors, for example, mutual perceptions of the spouses, extent of premarital sexual experience, and sexual satisfaction in marriage. The two theories will be maximally accurate. They will provide a nearly complete explanation of divorce in each society. But they will not meet other requirements imposed on theories: accuracy, generality, parsimony, and causality. When the accuracy of theories is maximized, their generality and parsimony will often be low. Generality of a theory refers to the range of social phenomena to which it is applicable. The greater the generality of a theory, the greater the range of phenomena that can be explained by the theory. For example, one theoretical proposition may state that "education is related to political participation." This theory provides an explanation of one type of political activity of individuals, but only one type of activity. But what are we saying when we identify education as a determinant of behavior? What is "education*? Is it the number of questions concerning different branches of knowledge one is able to answer? The duration of protection of an individual by his family or by s ciety? Reinforcement of certain behavioral patterns by appropriate rewards? Chemical changes in the composition of: brain cells? A number of more general theories can be formulated if any of these definitions replaces the definition of education in terms of number of school years completed. For example, a more general theory may state that individuals who have been protected by society over an extended period of time are more likely to participate in social activities. Since attending school usually provides such protection and since political participation is a type of social activity, the original, less general theory can be deduced from the more general one. The development of natural science consists of more general theories superseding less general theories. As, Hem pel points out: "When a scientific theory is superseded by another in the sense in which classical mechanics and electrodynamics were superseded by the special theory of relativity, then the succeeding theory will generally have a wider explanatory range, including phenomena the earlier theory could not account for; and it will as a rule provide approximative explanations for the empirical laws implied by its predecessor."8 In social science, however, it is not always apparent that a less general theory can be deduced from a more general theory even if both are avail- 'Ibid., p. 345. 22 Comparative R esearch and Social Science Theory able. Both reward theory and a theory of political participation may potentially explain why an individual votes in an election; but it is not apparent that the explanatory role of education can be deduced from the reward theory or any other psychological theory. In the example discussed above, 99.9 percent of variance of divorce is predicted in each social system, but not a single statement can be made that would be true for both systems. Such an explanation will not only lack generality but also will not be parsimonious. The smaller the number of factors providing for a complete explanation of a given class of events, the more parsimonious the theory.4 It would be an interesting experiment to compare the interpretations given to findings derived from particular countries with the interpretations of similar data from various countries. Why were third parties never spectacularly successful in the United States? Because of the bipartisan tradition. Why were third parties never successful in Great Britain? Because of the shift in the composition of the labor force that took the working-class support away from the Liberals and gave it to Labour. But when we confront the two questions simultaneously—why were third parties never particularly successful in either country—the answer will tend to be formulated in terms of factors common to both such as the electoral system. To the extent that different theories—each involving a different set of independent variables—are used for different social systems, the formulation and testing of general theories in the social sciences is not possible. The criteria of generality and parsimony imply that the same theories must be evaluated in different systemic settings and that social science theories can gain" confirmation only ij theories formulated in terms of the common factors constitute the point of departure for comparative research. We recognize, however, that in some situations accuracy in a particular social system may be the most important value. For example, if we want to predict election results in the United States, we probably will not be concerned with the same factors that will predict election results in Great Britain. But if the goal is to provide understanding as to why people identify with political parties, then generality and parsimony will be more important than system-specific accuracy. If the role of a theory is to provide immediate 'The number of factors is only one of many aspects of parsimony. For a full discussion of this concept and its many definitions see the summary of the writings of Janina Kotarbinska in Henryk Skolimowski, Polish Analytical Philosophy, Routledge, Kegan and Paul. London, 1967. Explanation and Theory in the Social Sciences 23 guidelines for social practice, then accuracy in a specific social system may be the most important value.8 The fourth criterion imposed on theory concerns causality. We can think of causality in a twofold perspective. Causality is a property of a system of variables. A system of variables is said to be causal to the extent that (1) the dependent variable is not "overdetermincd"—no two variables within the system explain the same part of the variation of the dependent phenomenon and (2) the system of variables is isolated—the explanatory pattern"does not change when new variables are added." The extent to which a theory is causal, that is, the extent to which general premises are invariant, increases as the number of factors incorporated into the theory increases, la terms of comparative research, the postulate of causality implies that factors operating at different levels of analysis—groups, communities, region, nations, etc.—should be incorporated into theories and that their interaction with the factors operating within each of these systems should be examined. This particular model of theory as a general, parsimonious, and causal set of statements is assumed throughout the rest of this book. We are not arguing that this particular model constitutes the only or even the best model of theory, but that if this model is accepted, then cross-systemic studies must become an integral part of theory-building and theory-testing. 0 As W. E, Moore, has stated, generalization involves abstraction and abstraction involves a loss of information. "No [general] theory will yield a specific prediction, or yield a specific guide to policy . . . except by reversing the process and adding information to the general proposition." W. E. Moore, "The Social Framework of Economic Development," in R, J. Braibanti and J. J. Spengler, eds,, Tradition, Values-, and Socio-Economsc Development, Duke University Press, Durham, N.C., 1961, p. 5fo: "The first aspect of causality—the problem of overdetermination—has receOTHi become fashionable among social scientists, following Simon's article on "Caus^T Ordering and Identifiability" in W, C. Hood and T. C. Koopmans. eds.. Studies in Econometric Method, John Wiley & Sons, New York, 1953. Overdetermination can be tested, if strong assumptions are accepted, through the analysis of partial correlations or partial path coefficients. The second aspect of causality has so far received only lip service. The assumption of uncorrected errors is usually stated but not tested. It seems that Hempel came closest to suggesting an empirical test of this assumption in postulating the criterion of "maximal specificity" (op. cit., p. 402). This criterion implies that causal explanations should be accepted tentatively and then subjugated to the tests of invariance under the addition of new variables. If and only if the path coefficients do not change substantively when new variables are introduced, the assumption of relative isolation of the system of variables can be maintained. 24 Comparative Research and Social Science Theory Theory and Spatiotemporal Parameters: The Postulate of Substitutability In the Introduction we defined some general issues underlying the alleged contradiction between historically based observations and abstractly formulated statements. A question often discussed by students of society is whether historically anchored observations should be treated as specific to particular social systems or whether general theories, free of spatiotemporal parameters, can be developed and tested. As indicated earlier, the issue no longer appears in this extreme form. The problem now is to define the conditions under which general theories can be developed and the procedures that are appropriate for the development and testing of general theories. The extreme version of the relativistic argument allows no way of bridging historical and theoretical statements. Once a proposition is stated in historical terms, using such proper names as Ghana, Hitler, or British workers during the 1950s, it could be incorporated only into a theory consisting of more general historical propositions, containing such names as Africa, German leaders, or the British population during the 1950s. If, however, a proposition is stated in terms free of proper names, it can be incorporated only into theories consisting exclusively of such propositions. For example, the proposition that "economic crises give rise to charismatic leadership" can be generalized into a statement that "all crises give rise to charismatic leadership." This distinction between historical and theoretical generalization clarifies the alternative modes of theory construction. Any set of observations can be generalized in one of two ways, depending upon whether historical or theoretical generality is sought. But this distinction is only analytical. Actually spatiotemporal propositions (e.g., measurement statements) can be generalized theoretically, and general statements can be specified historically. On the one hand, the statement that "Hitler was a charismatic leader-who came to power in Germany as a result of a crisis" can be generalized into a theoretical proposition relating crises and charismatic leadership. On the other hand, the statement that "crises give rise to charismatic leaders" may have been observed to be true only in Africa. Thus the observed historical situation is not unique, but neither is the general theoretical statement universal. Historical statements are implicitly theoretical. They subsume under the proper names of the social systems a broad range of factors that might be used in theoretical explanation. But theoretical statements will generally include a historical component. As long as Africa differs from the other parts of the world, theoretical analysis is no longer possible, and therefore the name of a social system will have to be used in explanation. Thus "unique" factors can neither be the only ones nor can they be » . Theory and Spatiotemporal Parameters 25 , totally discarded in theoretical analysis-.. They are redefined, rather, as the j residuum of theoretical explanation. ? The ~b~ridge between historical observations and general theory is the , [ substitution of variables for proper names of social systems in the course of Veomparative research. The theoretical importance of this statement is best""" understood in terms of Hempel's requirement that classes of events referred to in theoretical ("lawlike") statements be essentially generalizable. Hempel . argued: "Surely a lawlike sentence must not be logically limited to a finite number of instances: it must not be logically equivalent to a finite conjunction of singular sentences, or, briefly, it must be of essentially generalized form."7 For example, a statement that "all Mexicans are taller than all Americans" is not a lawlike sentence that can be used for the explanation of the height of Mexicans or Americans. This sentence is logically equivalent to a conjunction of statements giving all asymmetrical relations between individual Mexicans and Americans, and it cannot sustain counterfactual and subjunctive conditional statements such as "if Mr. X, who is an American were a Mexican, he would have been taller." The sentence, "Persons living in warmer climates are invariably taller than persons living in colder climates," however, is not a conjunction of any finite number of statements concerning individuals; it can, at least logically, be extended ad infinitum. In other words, lawlike statements are possible in the social sciences if and only if spatiotemporal parameters are treated as residua of variables potentially contributing to the explanation. The postulate of substitutability concerns the ontological status of such concepts as "group," "organization," "culture," "nation," and "political i _ ■ system"—the ontological status of systems enclosed within some specific spatial and temporal parameters. Before proceeding to the discussion of-this postulate and its implications, we will analyze more closely the notionj;! of "historically located social systems," or "spatiotemporal parameters." A concept of "all historical social systems" or "all spatiotemporal para- i are selected on the basis of their position on some variable, such as income or education. The common and obvious procedure in cross-systemic re- 31