Science, Language, and Human Rights
(1952)

CONTENTS SYMPOSIUM: PHENOMENALISM I. Roderick Firth II. Max Black SYMPOSIUM: PROBLEMS OF CONCEPT AND THEORY FORMATION IN THE SOCIAL SCIENCES I. Ernest Nagel II. Carl G. Hempel SYMPOSIUM: WHAT IS A RULE OF LANGUAGE? I. N.L. Wilson II. R.M. Martin SYMPOSIUM: THE CONCEPT OF EXPRESSION IN ART I. Vincent A. Tomas II. Douglas N. Morgan SYMPOSIUM: THE CONCEPT OF UNIVERSAL HUMAN RIGHTS I. A.I. Melden II. W.K. Frankena PROBLEMS OF CONCEPT AND THEORY FORMATION IN THE SOCIAL SCIENCES ERNEST NAGEL AND CARL G. HEMPEL I. ERNEST NAGEL The study of human institutions and human behavior in institutional settings has been under cultivation for at least as long as has inquiry into physical and biological phenomena. But few will question the assertion that in no area of social study have a body of knowledge and systems of explanatory theory been achieved, which compare in scope, precision, and reliability with what the natural sciences have to offer. To be sure, systems of "social physics" have been proposed in the past as adequate for all human cultures and for all periods of history. Most of these explanatory systems are so-called "single-factor" theories. They identify some one "factor"-such as geography, race, economic structure, to mention but a few-as the "basic determinant" in terms of which the organization and development of societies are to be understood. However, none of these ambitious hypotheses has been able to survive critical scrutiny; and even in restricted domains of social study, such as economics, the empirical worth of the theories currently in development is still a seriously debated matter. In consequence, the right of any existing department of social study to the title of a "genuine science" has been repeatedly challenged. It is idle to debate an issue so stated, for it is little more than a dispute over the use of an honorific label. In any case, it is preposterous to deny that social scientists have produced valuable descriptive studies of many social phenomena, that they have thrown much light on the interdependence of a variety of social processes, or that 44 CONCEPT AND THEORY IN SOCIAL SCIENCE have devised techniques of measurement and empirical analysis possessing varying degrees of actual and potential use-fulness. On the other hand, many professional students of human affairs believe it is premature to attempt comprehensive theories for their subject matters; and most of them are indeed occupied with less ambitious though doubtless also less impressive tasks. Moreover, there is an influential group of thinkers who claim it is a fundamental error to measure the achievements of the social sciences by standards borrowed from the natural sciences; and they maintain that the construction of "abstract" theories which require to be warranted by objective evidence, like the theories of physics, is not the appropriate goal of social inquiry. The social sciences thus present a scene not only of widely diverse undertakings, but also of acute methodological controversy between different schools of inquiry. Anyone who proposes to discuss the social sciences, with the intent of comparing their theoretical structures and their logical methods with those of the natural sciences, is thus in a quandary. There is little general agreement as to what social theory ought to be, and as to what social theory ought to accomplish; and in any case, whatever sense is attached to the word "theory," there is no theoretical treatise in these disciplines comparable in scope and authority with those current in physics, chemistry, or biology. Whatever material 1s selected for analysis is likely to be judged by many students as unrepresentative, and the analysis itself as irrelevant to the central problems of social inquiry. But this risk is un-avoidable. In this paper I wish to examine some issues that seem to me perennial in the social sciences, and to discuss them in the context of an approach in them that is extensively exploited at present. Many outstanding investigators of human affairs believe that the best hope for their disciplines lies in the systematic cultivation of "functional" analyses of social processes and institutions; and some of them even proclaim functionalism to be a full-fledged "theory" of social phenomena. It is to a few broad problems raised by functionalism that the following somewhat desultory comments are addressed. ERNEST NAGEL 45 1 Since functional analysis in the social sciences is often a self-conscious attempt to adopt a type of explanation common in biology, and especially physiology, let me begin with a brief consideration of functional explanations in biological science. In biology, a functional account is an analysis of the structure and operations of various parts of an organism (or a system that includes living bodies) with the objective of exhibiting the manner in which certain characteristic activities or features of the organism (or system) are main-tained, despite variations in its external and internal en-vironment. Now it is demonstrable that functional analyses in biology do not differ in kind or method from the so-called "causal" explanations in the physical sciences with which they are often contrasted. For in the first place, a functional account in biology traces the objective effects which follow from the operation of various components of an organic system upon that system as a whole or upon some of its other parts. In consequence, the formulations of the relations of dependence that are thus discovered do not differ in content from formulations which simply specify the conditions (whether necessary or sufficient) under which events of a specified type occur. In fact, statements in biology that have a teleological form are translatable without loss of meaning into statements of a nonteleological kind. To say, for example, that the function of the human kidney is to maintain a certain chemical balance in the blood is to assert nothing different from saying that in the absence of kidneys in the normal human body the chemical composition of the blood does not remain approximately constant. In the second place, though physiologists conduct their experiments upon individual organic systems, their conclusions are intended to be valid not only for the particular systems actually investigated, but for all systems of a stated type. In short, physiological analyses terminate in general statements or laws. Moreover, in the process of ascertaining the functions of various parts of an organic system, explicit use is made of generalized knowled e dished by fre phini scene in the thirt 46 CONCEPT AND THEORY IN SOCIAL SCIENCE place, though the systems studied in biology are "self-main-taining" or "goal-directed," while the systems explored by the physical sciences are usually not of this nature, biologists no longer find it necessary to postulate purposes or special vital agents to account for the self-regulative character of the systems they investigate. Indeed, as the current literature on servo-mechanisms makes evident, it is now possible to define what distinguishes self-maintaining systems from those which are not, in terms that are neutral to the distinction between the living and the nonliving. In any event, biologists have succeeded in a number of cases in explaining the self-maintenance of organisms entirely in terms of the organization of their parts, and of compensatory physio-chemical changes in those parts. On the other hand, a critical biology recognizes that a teleological analysis is always relative to a specified system and a designated set of structures and activities. For the operation of a given organ will in general have a variety of consequences, so that to ask for the function of an organ is to ask for the consequences of its operation for one among several systems of which it may be a component. But this relativity of functional explanations does not make them illegitimate, and does not preclude the possibility of there being objective warrant for them. And finally, successful functional analysis in biology is not contingent upon the prior acceptance of any particular theory of organic processes. In particular, it does not rest upon the far-going assumption that the continued existence and operation of every part is indispensable to the organism, or that the actual behavior of each distinguishable component of an organic system is dependent on the character and mode of behavior of every other component. If by "theory" one understands an explicit formulation of determinate relations between a set of variables, in terms of which a fairly extensive class of empirically ascertainable regularities (or laws) can be explained (always provided that suitable boundary conditions for the application of the theory are supplied), and if the pursuit of functional analyses in biology is called "functionalism," then functionalism is not a biological theory. It is at best but a precept for orienting inquiry to the study of conditions and mechanisms of self-maintenance. ERNEST NAGEL 47 2 These brief reminders of the character of teleological analyses in biology will prove to be pertinent to the examination of functionalism in the social sciences, to which I now turn. However, despite their conviction of its great promise, proponents of functionalism in the social sciences are not in general formal agreement as to its content. I shall nevertheless take for my point of departure a statement of it which is by now classic, even though it is not tully representative of current functionalists' views and even if not all professed functionalists subscribe to it in its entirety. In an account of functionalism in anthropology, Malinowski explained that a functionalist analysis of culture aims at the explanation of anthropological facts at all levels of development by their function, by the part which they play within the integral system of culture, by the manner in which they are related to each other within the system, and by the manner in which this system is related to the physical sur-roundings.... The functional view of culture insists therefore upon the principle that in every type of civilization, every custom, material object, idea and belief fulfills some vital function, has some, task to accomplish, represents an indispensable part within a working whole.I But what is functionalism on this statement? Is it a hypothesis concerning the interrelations of social phenomena, is it simply a program for inquiry, is it both? The explicit insistence that every item discoverable in a culture is indispensable to it as a "working whole," inclines one to believe that it falls under the first alternative; but the remainder of the statement suggests that it is to be subsumed under the second. However, if functionalism is a "theory" about civili-zations, it must agree with available empirical evidence; and there is ample evidence to show that societies are not in general the tightly knit "organic" systems which this theory proclaims them to be, so that funtionalism in this version must be judged as false. On the other hand, if functionalism is construed as a methodological precept, then the above statement appears to be fully conveyed by the imperative: 48 CONCEPT AND THEORY IN SOCIAL SCIENCE "There are relations of dependence between cultural objects, and between cultural objects and their physical environment. Look for them!" But if this translation is not a complete caricature, it is difficult to understand why functionalism should be so widely hailed as a fresh and promising approach to the study of social phenomena. In point of fact, the content of functionalism (whether it is construed to be a theory or a program) is meager to the vanishing point, unless formulations of it such as the above are supplemented by a variety of material assumptions about the organization and the operation of societies. At the same time, the ambiguity just noted pervades not only the writings of Malinowski, but also the literature of functionalism in other areas of social study. 3 But I hasten to more substantial issues. Although it is usually quite clear in biology what are the activities of an organic system which must be maintained if the system is to endure, the corresponding question in the social sciences does not in general admit of a ready answer. For in biology there are certain familiar and generally acknowledged inclusive functions, such as respiration or reproduction, without whose continued maintenance the organism or the species will perish. In the social sciences, on the other hand, there is much less explicit agreement as to what are the indispensable activities and structures of social systems except when specifically biological needs of human beings are in question. The problem obviously involves complicated matters of empirical fact and analysis, though it is often settled by a tacit use of moral principles. For example, though Hayek's recent polemics against deliberate social planning employ the arguments of an implicit functionalism, his arguments have force only if one accepts the moral assumptions underlying them? In the absence of explicit agreement on what are the "essential" requirements of society, the functionalist's injunction to study the manner in which "the integral system of culture" is maintained, is frequently a concealed endorsement of some particular social ideal. But in any event, the characterization of a process or set of conditions as "functional" or "dysfunctional," according as it con- ERNEST NAGEL 49 tributes to or detracts from the "integration and effective-ness" of a social system, is empirically empty unless there is an antecedent specification of what is to be integrated and for what a process is effective. For every occurrence is functional for some things and at the same time dysfunctional for others, as indeed a critical biology makes plain. Some recent functionalists, if I understand them, seek to outflank these difficulties by frankly recognizing the relativity of teleological analyses of social systems. They propose systematically articulated "structural" categories allegedly so general as to comprehend all types of institutions. (For example, one proposal distinguishes between configurations of institutions according as they stress norms of conduct binding upon all members of a society indiscriminately, or norms which involve reference to special relations between individuals; and the proposal further distinguishes between configurations according as they promote social stratification on the basis of individual achievement, or on the basis of possession of special attributes. At least four types of social structure are thus recognized.) A functional analysis will therefore show just what, and in what manner, various social processes contribute to the maintenance of designated structural features of a social system. If I am correct in my reading of this version of functionalism, it makes central the possibility of alternate patterns of social organization. It thereby not only makes explicit the relativity of functional analyses, but also keeps distinct questions of actual fact from questions of social policy. The present and eventual worth of the categorical schemas thus far proposed is by no means clear, and is certainly disputable. But it is safe to say that without some such "structural" categories functionalism cannot intelligibly maintain this distinction, and cannot consistently escape basic confusion. 4 It is therefore a mistaken claim which asserts that functional analyses (or for that matter any responsible inquiry into human affairs) are necessarily committed to some pattern of cultural values, and which concludes that "objectivity" in any relevant sense of this difficult word is impossible in the 50 CONCEPT AND THEORY IN SOCIAL SCIENCE social sciences. I will begin with a quotation from Max Weber, whose views on this issue continue to exercise enormous influence. Weber maintained that it is the distinctive task of the social sciences to analyze phenomena in terms of their "cultural significance," and in this respect he was a functionalist despite his vigorous dissent from certain functionalist tendencies. But he also declared that The significance of a configuration of cultural phenomena and the basis of this significance cannot . . . be derived and rendered intelligible by a system of analytical laws, however perfect it may be, since the significance of cultural events presupposes a value-orientation toward these events. The concept of culture is a value-concept. Empirical reality becomes "culture" to us because and insofar as we relate it to value ideas. It includes those segments and only those segments of reality which have become significant to us because of this value-relevance. Only a small portion of existing concrete reality is colored by our value-conditioned interest and it alone is significant to us. It is significant because it reveals relationships which are important to us due to their connection with our values. ... The focus of attention on reality under the guidance of values which lend it significance and the selection and ordering of the phenomena which are thus affected in the light of their cultural significance is entirely different from the analysis of reality in terms of laws and general concepts. Weber finally concluded that an "objective" analysis of cultural events, which proceeds according to the thesis that the ideal of science is the reduction of empirical reality to "laws" is meaningless. . .. The transcendental presupposition of every cultural science lies not in our finding a certain culture or any "culture" in general to be valuable but rather in the fact that we are cultural beings, endowed with the capacity and the will to take a deliberate attitude towards the world and to lend it significance.3 ERNEST NAGEL 51 It would take more pages than are at my disposal to unravel the ambiguities in these pronouncements, or to separate what seems to me sound in them from what is false. I must limit myself to but one point. Why is the fact that an inquirer selects his material for study in the light of the problems that interest him, of greater moment for the logic of social inquiry than it is for the logic of natural science? If a social scientist discovers that the outcome-and in this sense the "significance"-of certain forms of individual activity is the perpetuation of a free economic market, in what way is his concentration upon processes that maintain this particular institution rather than something else, of greater relevance in evaluating the adequacy of his explanation than is, for the corresponding question, the concern of the physiologist with processes that maintain a relatively constant internal temperature of the human body? Were we not human beings, but were nevertheless capable of engaging in inquiry, we might perhaps show no interest either in the conditions that maintain a free market or in the conditions that make possible the homeostasis of human temperature. But there is no ascertainable difference on this score between the natural and the social sciences. The traits selected for study, with a view to discovering their conditions and consequences, may indeed be dependent on the fact that the investigator is a "cultural being"; but there is no reason whatever for believing that the validity of his conclusions is dependent on that fact in the same way. It may be said, however, that some point of view 1s required in selecting materials in the conduct of social inquiry, that the "point of view determines what is important and what is unimportant in the confusing maze of human events," and that since questions of importance involve problems of valuation, it is impossible to eliminate bias from the study of human affairs. But this argument is a non-sequitur, and gains its apparent plausibility from confounding several senses in which valuations may enter into inquiry. It should be noted in the first place that if the argument is sound, it applies equally to inquiries in both the natural and social sciences. In the second place, it is undoubtedly correct that the scientist brings certain "values" 52 CONCEPT AND THEORY IN SOCIAL SCIENCE to his investigations, and imposes certain demands upon the course of his inquiry. But the only values which are relevant in this connection are the values and standards implicit in the enterprise of obtaining warranted knowledge. Doubtless such standards change with the development of science, and doubtless also these changes are frequently associated with other cultural variations. But evidence is completely lacking to support the currently fashionable belief that even when the specific problem initiating an inquiry is carefully formulated and understood, individuals may be so disparate in their "cultural values" as to preclude in principle the possibility of agreement between them concerning the validity of a proposed solution for it. And in the third place, though judgments of importance are unavoidable in all inquiries, in the natural as well as the social sciences, such judgments are not necessarily "subjective" in any pejorative sense. In par-ticular, they need not be made in terms of some "private" set of values, since standards of relative importance can be explicitly formulated, and the question whether these standards are satisfied in a given case can be settled, at least in principle and when sufficient evidence is available, by customary methods of empirical investigation. To be sure, men's interests may be discordant, and the standards accepted by some in ascribing degrees of relative importance may be different from the standards accepted by others. It does not follow, however, that two apparently contradictory judgments of relative importance are necessarily incom-patible; for the appearance of contradiction may merely indicate the use of different criteria by which the meaning of "relative importance" is tacitly specified. Interests may clash, they may not be arbitrable by rational methods, and they may eventuate in the adoption of alternate principles for selecting and organizing empirical material. But once the meanings of terms have been decided upon and a principle of selection adopted, the clash of interests plays no logical role in establishing the cognitive worth of a proposed explanation of social phenomena. It is also frequently argued, to quote one fairly typical statement of the point, that a social scientist ERNEST NAGEL 53 cannot wholly detach the unifying social structure that, as a scientist's theory, guides his detailed investigations of human behavior, from the unifying structure which, as a citizen's ideal, he thinks ought to prevail in human affairs and hopes may sometime be more fully realized. His social theory is thus essentially a program of action along two lines which are kept in some measure of harmony with each other by that theory-action in assimilating social facts for purposes of systematic understand-ing, and action aiming at progressively molding the social pattern, so far as he can influence it, into what he thinks it ought to be. Now it is undoubtedly difficult in many inquiries to prevent our hopes and wishes from coloring the conclusions we draw; and it has taken centuries of devoted effort even in the natural sciences to develop habits and techniques of investigation which safeguard the course of inquiry against the intrusion of irrelevant personal factors. But to say this is to assume the possibility of distinguishing between fact and hope, for otherwise the statement becomes unintelligible. Even if it is invariably the case, as the above quotation maintains, that social scientists pursue a double line of activity, the claim that they do so makes clear sense only if it is possible to adjudicate between, on the one hand, contributions to theoretical understanding whose validity does not depend on allegiance to any social ideal, and on the other hand contributions toward the realization or perpetuation of some such ideal. Accordingly, the difficulty noted is a remedi-able practical difficulty. It can be overcome, as is often recognized, not by futile resolutions to remain unbiased, but, through the self-correcting processes of science as a social enterprise. For the tradition of modern science encourages the free exchange and criticism of ideas; and it permits and welcomes competition in the quest for knowledge on the part of independent investigators, whatever may be their prior doctrinal commitments. It would be ignorance to claim that these self-corrective processes have operated or are likely to operate in social inquiry as readily as they have in the natural sciences. But it would be a confusion in analysis 54 CONCEPT AND THEORY IN SOCIAL SCIENCE to conclude that therefore there is a basic difference in the character of warranted knowledge in these two areas of inquiry. 5 I must now turn to issues of a different order. In the study of biological functions the imputation of motives, atti-tudes, and purposes to organic systems or their parts is strictly irrelevant. In the study of social phenomena such imputation is highly pertinent. What is the significance of this fact for the objectives and the methods of the social sciences? According to an influential school of functionalists, all socially significant human behavior is an expression of motivated psychic states, so that the "dynamism" of social processes is identified with the "value-oriented" behavior of human individuals. An inquiry that is properly a social study has been therefore said to begin only with the ques-tion: "What motives determine and lead the individual members and participants in [a given] community to behave in such a way that the community comes into being in the first place and that it continues to exist?"6 In consequence, the social scientist cannot be satisfied with viewing social processes simply as the sequential concatenations of "ex-ternally related" events; and the establishment of correla-tions, or even of universal relations of concomitance, cannot be his ultimate goal. For he must not ignore the fact that every social change involves the assessment and readjustment of human activities relating means to ends (or "values"). On the contrary, he must construct "ideal types" or "models of motivation," in terms of which he seeks to "understand" overt social behavior by imputing springs of action to the actors involved in it. But these springs of action are not accessible to sensory observation; and the social scientist who wishes to understand social phenomena must imaginatively identify himself with its participants, and view the situation which they face as the actors themselves view it. Social phenomena are indeed not generally the intended resultants of individual actions; nevertheless the central task of social science is the explanation of phenomena as the unintended outcome of springs of action—of psychic ERNEST NAGEL 55 states which are familiar to us solely from our own "subjec-tive" experiences as volitional agents. In consequence, there is said to be a radical difference between explanations in the social and in the natural sciences. In the latter we allegedly understand the "causal nexus" of events only in an external manner; in the former we can grasp the peculiar unity of social processes, since these involve a dynamic synthesis of subjective urges, values, and goals, on the one hand, and the external environment on the other. A purely "objective" or "behavioristic" social science is thus declared to be a vain hope. For in the words of one recent writer, proponents of behaviorism in social science fail to perceive the essential difference from the standpoint of causation, between a paper flying before the wind and a man flying from a pursuing crowd. The paper knows no fear and the wind no hate, but without fear and hate the man would not fly nor the crowd pursue. If we try to reduce it to its bodily concomitants we merely substitute the concomitants for the reality expressed as fear. We denude the world of meanings for the sake of a theory, itself a false meaning which deprives us of all the rest. We can interpret experience only on the level of experience.? In short, since social science seeks to establish "meaningful" connections and not merely relations of concomitance, its goal and method are fundamentally different from those of natural science. I will not take time to comment here at length on the psychological preconceptions underlying this rejection of behaviorism, nor on the adequacy with which behaviorism is portrayed. Only one point requires brief mention in this connection. It is surely not the case that we must ourselves undergo (whether actually or in imagination) other men's psychic experiences in order to know that they have them, or in order to predict their overt behaviors. But if this is so, the alleged "privacy" or "subjectivity" of mental states has no bearing on the acquisition of knowledge concerning the character, the determinants, and the consequences of other not quite correct; the man can be Plown by the wind it tis strong enoyeh! 56 CONCEPT AND THEORY IN SOCIAL SCIENCE men's dispositions and actions. A historian does not have to be Hitler or even be capable of reenacting in imagination Hitler's frenzied hatreds, to write competently of Hitler's career and historical significance. For knowledge is not a matter of having images, whether faint or vivid; it is not a reduplication of, or a substitute for, what is claimed to be known. Knowledge involves the discovery through processes of controlled inference that something is a sign of something else; it is statable in propositional form; and it is capable of being verified through sensory observation by anyone who is prepared to make the effort to do so. It is therefore just as possible to know that a man is in a state of fear or that a crowd is animated by hatred, without recreating in imagination such fears and hatreds, as it is to know that a man is running away or that a crowd is pursuing him without an imaginary exercise of one's legs. It is possible to discover and know these things on the evidence supplied by the overt behaviors of men and crowds, just as it is possible to discover and know the atomic constitution of water on the evidence supplied by the physical and chemical behavior of that substance. But I must consider at greater length the claim that since the social sciences seek to "understand" social phenomena in terms of "meaningful" categories of human experience, the "causal-functional" approach of the natural sciences is not applicable in social inquiry. The abstract pattern of such "meaningful" explanations appears to be as follows. Let A be some complex set of conditions (e.g., membership in certain religious groups) under which a phenomenon B occurs (e.g., the development of modern forms of capitalistic enterprise). The social agents involved in A and B are then assumed to possess certain feelings, beliefs, etc. : 41 (e.g., belief in the sacredness of a wordly calling) and B1 (especial of ho atty, erines, and altemians abor); related, because of our familiarity with motivational patterns in our own experience; and the relations between A and A1, as well as between B and B", are also of the same alleged kind. Accordingly, the "external" connection between A and B is "meaningfully" explained, when each is "interpreted" ERNEST NAGEL 57 as an expression of certain "motivational" states A1 and Bi respectively, where the connection between the latter is "understood" in a peculiarly intimate way. But do such explanations require a special kind of logic, distinctive of the social sciences? At the risk of belaboring the obvious, I must state the grounds tor maintaining that the answer 1s negative. The imputation of emotions, attitudes, and purposes as an explanation of overt behavior is a twofold hypothesis; it is not a self-certifying one, and evidence for it must be supplied in accordance with customary canons of empirical inquiry. The hypothesis is twofold: for on the one hand, it assumes that the agents participating in some social phenomenon are in certain psychological states; and on the other hand, it also assumes definite relations of con-comitance between such states, and between such states and certain overt behaviors. But as the more responsible exponents of "meaningful" explanations themselves emphasize, it is in general not easy to obtain competent evidence for either assumption. We may identify ourselves in imagination with a trader in grain, and conjecture what course of action we would take were we confronted with the problems of a fluctuating market. But conjecture is not fact, however necessary conjecture may be as part of the process of discovering what is fact. None of the psychological states which we imagine the subjects of our study to possess may in reality be theirs; and even if our imputations should be correct, none of the overt actions which allegedly issue from those states may appear to us as "understandable" or "reasonable" in the light of our own experiences. If the history of anthropological research proves anything, it surely testifies to the errors students commit when they interpret the actions of men in unfamiliar cultures in terms of categories drawn uncritically from their limited personal lives. Moreover, do we "understand" the nature and operation of human motives and their issuance in overt behavior more adequately and with greater certitude, than we do the occurrences studied in the natural sciences? Do we understand more clearly and know with greater certainty why an insult tends to produce anger, than why a rainbow is produced when the sun's rays strike raindrops at a certain angle? The 58 CONCEPT AND THEORY IN SOCIAL SCIENCE question is rhetorical, for the obvious answer is "no." We may feel assured that if an illiterate and impoverished people revolts against its masters, it does so not because of adherence to some political doctrine but because of economic ills. But this assurance may only be the product of familiarity and a limited imagination; and the sense of penetrating comprehension that we may associate with the assertion, instead of guaranteeing its universal truth, may be only a sign of our provincialism. The contrast that is drawn between "understanding" in terms of "meaningful" categories and the merely "external" knowledge of causal relations which the natural sciences are alleged to provide is indeed far from clear. When we "meaningfully" explain the flight of a man from an angry crowd by imputing to him a fear of physical violence, we surely do not postulate in him a special agency called "fear" which impels him to run; nor do we intend by such an imputation that a certain immediate quality of the man's experience is the determinant of his action. What I think we are asserting is that his action is an instance of a pattern of behavior which human beings exhibit under a variety of circumstances, and that since some of the relevant circumstances are realized in the given situation the person can be expected to manifest a certain particular form of that pattern. But if something like this is the content of explanations in terms of "meaningful" categories, there is no sharp gult separating them trom explanations that involve merely "external" knowledge of causal connections. One final comment in this connection. While the "dyna-mism" of social processes is in general identified by many functionalists with the "motivational" character of human action, some of them have maintained that all social change must be understood in terms of those particular variations in motivation that are called "values." Thus, it has been argued that if we wish to ascertain why there has been a pronounced increase in the divorce rate in the U. S. during the past fifty years, a satisfactory explanation must be of the type which stipulates a change in valuation affecting the status of the family. The general indication is that divorce is more prevalent in those areas where the continuity ERNEST NAGEL of the family through several generations has less significance in the schema of cultural values than formerly or elsewhere. More generally, the claim has been made that "In so far as we are able to discover the changes of the evaluative schema of social groups we can attain, and thus only, a unified explanation of social change."8 It is not my concern to take sides in an issue of material fact, and in any case I am not qualified to do so in this particular one. But it is relevant to ask how an assumed alteration in a schema of cultural values can be established. The obvious procedure would be to take as evidence for such a change the explicit statements of the persons involved, whether these statements take the form of personal confes-sions, public speeches, or the like. However, explicit statements of the kind required are not generally available; and even when they are, they cannot always be taken at their face value. For there is often a great disparity between what men verbally profess and what they actually practice, and individuals may continue their verbal allegiance to a set of ideals even though their mode of living has been radically transformed. Students of human affairs are thus compelled to base their conclusions as to what are the operative evaluative schemas in a given society on evidence that is largely drawn from the overt behavior of men-from their conduct of business affairs, their mode of recreation, their domestic arrangements, and so on. There is in fact a risk that explanations of social change in terms of alterations in value-schemas collapse into tauto-logies, which simply restate in different language what is presumably explained. For example, if a change in the divorce rate is the sole evidence for the assumption that there has been a shift in the value associated with the continuity of the family, a proposed explanation of the former change by the latter is a spurious one. It does not follow that all explanations in terms of variations in evaluative schemas are necessarily sterile. It does follow, however, that if such sterility is to be avoided, the concept of a value-schema must be construed as a highly compact formulation of various regularities (most of which are perhaps never explicitly 59 pet remain totally ugeess 60 CONCEPT AND THEORY IN SOCIAL SCIENCE codified, or codified only in vague terms) between types of human behavior. The point I wish to make is that in imputing a certain schema of values to a community, one is imputing to its members certain attitudes. But an attitude is not something that can be established by introspection, whether in the case of our own persons or of others. An attitude is a disposi-tional or latent trait; and it is comparable in its theoretical status with viscosity or electrical resistance in physics, even if, unlike the latter, it can be usefully defined for socio-psychological purposes only in statistical terms. In any event, the concept is cognitively valuable only in so far as it effects a systematic organization of manifest data obtained from overt human responses to a variety of conditions, and only in so far as it makes possible the formulation of regularities in such responses. Whether, in point of fact, explanations of social changes in terms of variations in attitudes have a greater systematizing and predictive power than explanations employing different substantive concepts, is not the point at issue, and it cannot be settled by dogmatic a priori claims. But if these comments are well taken, there is nothing in such explanations which differentiates them in principle from explanations in the natural sciences, or which requires for their validation a distinctive logic of inquiry. 6 I will conclude, as I have begun, with some general reflections on the social sciences. Many commentators in recent as well as in earlier years have been in despair over the state of these disciplines, and have doubted the possibility of developing comprehensive and reliable theories of social processely deaned to is be co means the into one, character of social phenomena that is, the variation of modes and conditions of social processes with different societies and periods. It is therefore frequently asserted that unlike a law in the natural sciences, an assumed regularity in social matters is at best valid only within a specified institutional setting and only for a limited duration. Snell's law is presumably true for the refraction of light throughout the universe; but the variation of fertility rate with social ERNEST NAGEL 61 status which obtains in one community at one time is generally different from what is found in the same or a different community at another time. It would be futile to dismiss such scepticism as without foundation, or to deny that human actions are modified by their cultural environment which itself is in constant change. It must be admitted, at least as a possibility, that all social "laws" may have only a narrowly restricted universality. Nevertheless, the scepticism when it is wholesale often has its source in two misconceptions upon which I wish to comment briefly. One of them is the tacit assumption that celestial mechanics is the paradigm of any science worthy of the name; the other is the failure to distinguish between instability or variability in the specific materials and organization of systems, and an abstract uniformity which may nevertheless be pervasive in all the systems. It has been maintained, for example, that if a science of society were a true science, like that of astronomy, it would enable us to predict the essential movements of human affairs for the immediate and the indefinite future, to give pictures of society in the year 2000 or the year 2500 just as astronomers can map the appearance of the heavens at hixed points of time in the tuture. Such a social science would tell us exactly what is going to happen in the years to some and we should be powerless to change it by an effort of will.® But since "owing to the development of human experience, men ... are always growing and changing," so that "closed schemes cannot be made out of the data of the social sci-ences," it is concluded that there cannot be a "social science in any valid sense of the term as employed in real science."10 It needs little argument to show that the circumstances which permit long-range predictions in astronomy do not prevail in other branches of natural sciences, and that in this respect celestial mechanics is not a typical physical science. Such predictions are possible because for all practical purposes the solar system is an isolated system which will remain isolated, so there is reason to believe, during an 62 CONCEPT AND THEORY IN SOCIAL SCIENCE indefinitely long future. In most other domains of physical inquiry, however, this condition is not in general satisfied. Moreover, in many cases we are ignorant of the appropriate initial and boundary conditions, and cannot make precise forecasts even though available theory is adequate for that purpose. We can predict with reasonable accuracy the motion of a pendulum, because both the theory and the specific factual data for the system are known; but because we have no reason to assume that the system will continue to be immune indefinitely from external perturbations, the predictions can be made safely only for a limited period. On the other hand, we cannot predict with great accuracy where a fallen leaf will be carried by the wind in ten minutes, because though physical theory is in principle capable of resolving the problem, we do not have the requisite knowledge of the relevant initial conditions. It is clear, therefore, that inability to forecast the indefinite future is not unique to the study of human affairs. It is, moreover, an obvious error to believe that theoretical knowledge is possible only in those domains in which effective human control is lacking. Crude ores can be transformed into refined products not because no theory for such changes is available, but in large measure just because there is. And conversely, a domain does not cease to be a field for theoretical knowledge if, when suitable techniques have been developed, changes in that domain for which hitherto there has been no control become controllable. Will the principles of meteorology stop being valid, should we discover some day how to manufacture the weather? Men are ible to a lor various aspects de their scial ganizations; science of human affairs. The fact that social processes vary with their institutional settings, and that the specific uniformities found to hold in one culture are not pervasive in all societies, does not therefore preclude the possibility that these specific uniformities are specializations of relational structures invariant for all cultures. For the admitted differences between the ways in which different societies are organized may be the consequences simply of differences in the specific elements which ERNEST NAGEL 63 enter into an invariant pattern of relations. Consider the following purely physical phenomena: a lightning storm, the behavior of a mariner's compass, and the formation of an optical image on the ground glass of a camera. These are quite heterogeneous occurrences, and there is no antecedent reason for suspecting that they illustrate a common set of principles. But as is well known, they can all be understood 11 terms of modern electromagnetic theory; and though different special laws hold for them, the theory can explain them all when suitable initial and boundary conditions are supplied for each instance. Despite the variability and instability of social phenomena, they may nevertheless be subsumable under a common theory in an analogous way-though whether this is more than a fancy is at present any man's guess. But some things are fairly clear. If a comprehensive social theory is ever achieved, it will not be a theory of historical development, according to which societies and institutions succeed one another in a series of inevitable changes. Those who are seeking a comprehensive social theory by charting the rise and decline of civilizations, are looking for it in the wrong place. The theory will undoubtedly have to be highly abstract, if it is to cut across the actual cultural differences in human behavior. Its concepts will have to be apparently remote from the familiar and obvious traits found in any one society; its articulation will involve the use of novel algorithmic techniques; and its application to concrete materials will require special training of high order. But above all, it will have to be a theory for which a method of evaluating evidence must be available which does not depend on the vagaries of special insights and private intuitions. It will have to be a theory which, in its method of articulating its concepts and evaluating its evidence, will be continuous with the theories of the natural sciences. NOTES 1 Bronislaw Malinowski, "Anthropology," The Encyclopadia Britannica, Supplementary Volume I, New York and London, 1936, pp. 132-33. PROBLEMS OF CONCEPT AND THEORY FORMATION IN THE SOCIAL SCIENCES ERNEST NAGEL AND CARL G. HEMPEL II. CARL G. HEMPEL 1. Introduction. The concept of type has played a significant role in various phases of the development of empirical science. Many of its uses are by now of historical interest only; but some branches of research, especially psychology and the social sciences, have continued up to the present to employ typological concepts for descriptive and for theoretical purposes. In particular, various typologies of character and physique have been propounded as providing fruitful approaches to the study of personality; the investigation of "extreme" or "pure" types of physical and mental constitution has been advocated as a source of insight into the functioning of "normal" individuals; and as for social science, the use of ideal types has been declared one of the methodological characteristics which distinguish it essentially from natural science. Considering these recent uses of typological concepts and the various claims concerning their peculiar significance, it appears to be a matter of some interest and importance to have a reasonably clear understanding of their logical status and their methodological function. Now, there exists a voluminous literature on the subject, but a large part of it suffers from a definite inadequacy of the logical apparatus used for the analysis of the issues at hand. In particular, many of the studies devoted to the logic of typological concepts use only the concepts and principles of classical logic, which is essentially a logic of properties or classes, and thus is incapable of dealing adequately with 66 CONCEPT AND THEORY IN SOCIAL SCIENCE quantitative concepts. In a manner illustrative of this situa-tion, Max Weber, who so impressively champions the method of ideal types in the social sciences, makes a clear negative statement about their logical status: they cannot be defined by genus proximum and differentia specifica, and concrete cases cannot be subsumed under them as instancest i.e., they are not simply class, or property, concepts; but when it comes to a positive characterization, he resorts to much less precise, and often metaphorical, language. An ideal type, according to Weber, is a mental construct formed by the synthesis of many diffuse, more or less present and occasionally absent, concrete individual phenomena, which are arranged, according to certain one-sidedly accentuated points of view, into a unified analytical construct, which in its conceptual purity cannot be found in reality; it is a utopia, a limiting concept, with which concrete phenomena can only be compared for the purpose of explicating some of their significant components.? This characterization, and many similar accounts which Weber and others have given of the nature of ideal types, are certainly suggestive, but they lack clarity and rigor and thus call for further logical analysis. In addition to the logical status of typological concepts, some of the methodological claims which have been made for them appear to me to warrant reexamination. The present paper, then, is an attempt to explicate in outline the logical and methodological character of typological concepts, and to appraise their potential significance for the purposes they are intended to serve. The proposed investigation will naturally have to use some of the concepts and principles of contemporary logic; but it will not employ any symbolic devices. Our explicatory efforts will repeatedly invite comparative glances at concept formation in the natural sciences. By thus undertaking a comparative examination of certain aspects of the methodology of natural and social science, I hope this study will justify its inclusion in the present symposium on the concept of theory in the social and the physical sciences. It is a familiar fact that the term "type" has been used in several quite different senses. I propose to distinguish here three main kinds of type concepts, which for brief reference, CARL G. HEMPEL 67 and pending further clarification, will be called classificatory, extreme, and ideal types. These will now be considered in turn. 2. Classificatory types. The classificatory use of type concepts is illustrated by Ernst Kretschmer's rather influential typological theory of character and physique, in which types are construed as classes. In this case, the logic of typological procedure is the familiar logic of classification, which requires no discussion here. Methodologically, classificatory type formation, like any other kind of classification in empirical science, is subject to the requirement of systematic fruitfulness: The characteristics which serve to define the different types should not merely provide neat pigeonholes to accommodate all the individual cases in the domain of inquiry, but should lend themselves to sound generalization and thus offer a basis for prediction. Thus, e.g., constitutional typologies often aim at defining their types by reference to certain physical properties which are empirically associated with a variety of psychological traits, so that every type represents a cluster of concomitant characteristics. This objective is the methodological kernel of the search for "natural" as distinguished from "artificial" classes or types. In connection with classifcatory types, briet reterence should be made to the use of the term "typical" in the sense of average," for that usage evidently presupposes a classi-fication. Thus, the statement that the typical American college undergraduate is, say, 18.9 years old, purports to state the average value of a certain magnitude for a specified class. But since there are different kinds of average, and since none of these provides much information without an added measure of dispersion, it is clear that for any serious scientific purpose this use of the term "typical" has to be supplanted by a more precise formulation in statistical terms. 3. Extreme types. Attempts at typological classification in empirical science are orten trustrated, however, by the realization that those characteristics of the subject matter which are to provide the defining basis of the classification cannot fruitfully be construed as simple property concepts determining, as their 68 CONCEPT AND THEORY IN SOCIAL SCIENCE extensions, classes with neatly demarcated boundaries. Thus, e.g., if we try to formulate explicit and precise criteria for the distinction of extravert and introvert personalities it soon becomes clear that the adoption of classificatory criteria drawing a precise boundary line between the two categories would prove an "artificial," theoretically sterile procedure: it appears much more natural, much more promising syste-matically, to construe the two concepts as capable of grada-tions, so that a given individual will not be qualified either as extravert or as introvert, but as exhibiting each of the two traits to a certain extent. The purely extravert and the purely introvert personalities thus come to be conceived as "extreme" or "pure" types, of which concrete instances are rarely if ever found, but which may serve as conceptual points of reference or "poles," between which all actual occurrences can be ordered in a serial array. This general conception underlies several of the recent and contemporary systems of psychological and physical types, such as, e.g., Sheldon's theory of physique and temperament. 4 What is the logical form of these "extreme" or "pure" type concepts? Clearly, they cannot be construed as class concepts: individual cases cannot be subsumed under them as instances, but can only be characterized as to the extent to which they approximate them. In other words, if the term "T" represents an extreme type, an individual a cannot be said either to be I or to be non-T; rather, a may be, so to speak, "more or less T." But exactly how is this "more or less" to be objectively defined? A description, however vivid, of an extreme type with which concrete cases are to be compared does not by itself provide standards for such compari-son; at best, it may suggest a program of research, focusing attention upon certain empirical phenomena and regularities and stimulating efforts toward the development of a precise conceptual apparatus suited for their description and theoretical interpretation. But if an extreme type is to function as a legitimate scientific concept in scientific statements with clear objective meaning, then explicit criteria for the "more or less" of comparison must be provided. These criteria may take a nonnumerical, "purely comparative" form, or they may CARL G. HEMPEL 69 be based on quantitative devices such as rating scales or measurement. The formally simplest, purely comparative form of an extreme-type concept I can be specified by laying down criteria which determine, for any two individual cases a, b in the domain under investigation, whether (1) a is more I thas is b. For become than or er i sat as treme type, for example, this would require objective criteria determining, for any two individuals a, b whether they are equally introverted and, if not, which of them is the more introverted. Thus, an extreme type I of the purely comparative or ordering kind is defined, not by genus and differentia in the manner of a class concept, but by specifying two dyadic relations, "more T than" and "as much T as." Now, if the criteria defining those relations are to yield an ordering of all particular cases in a linear array reflecting increasing T-ness, then they must meet certain formal requirements: "more T than" must be an asymmetrical and transitive relation, "as much T as" must be symmetrical and transitive, and the two together must satisfy a trichotomy law to the effect that any two particular cases a, b meet the defining conditions for exactly one of the three alternatives (i), (ii), (iii) mentioned above. The kind of ordering concept here characterized is well illustrated by the definition, in mineralogy, of a purely comparative concept of hardness by reference to the scratch test: A mineral a is said to be harder than another, b, if a sharp point of a sample of a will scratch the surface of a sample of b, but not conversely. If neither of the materials is harder than the other, they are said to be of the same hardness. The two relations thus defined might be said to determine a purely comparative extreme type of hardness; but this terminology would tend to obscure rather than clarify the logic of the procedure, and it is not actually used. In psychology and the social sciences, it is difficult, to say the least, to find fruitful objective criteria, analogous to those based on the scratch test, which will determine a purely comparative typological order. We find therefore that proponents of extreme-type concepts, insofar as they provide 70 CONCEPT AND THEORY IN SOCIAL SCIENCE precise criteria and not merely suggestive programmatic characterizations, either end by construing their types as classes after all or else specify their typological orders by reference to rating scales or measuring procedures, which define a numerical "degree of T-ness," as it were. The first course is illustrated by Kretschmer's typology of physique and character: it uses the parlance of pure types for an intuitive characterization of the material to be investigated, while for exact formulations, it construes each of the main types as a class and accommodates the intermediate cases in some additional classes, designated as "mixed types." The second course is exemplified by Sheldon's typology of physique, which assigns to each individual a specific position on each of three seven-point scales representing the basic type traits of the theory: endomorphy, mesomorphy, and ectomorphy. But once suitable "operational" criteria of a strictly comparative or of a quantitative kind have been specified, the pure types lose their special importance: they simply represent extreme places in the ranges defined by the given criteria, and from a systematic point of view, the typological terminology is no more significant than it would be to say that the specific electric conductivity of a given material indicated how close it came to the extreme, or pure, type of a perfect conductor. The use of extreme-type concepts of the kind here considered reflects an attempt to proceed from the classificatory, qualitative level of concept formation to the quantitative one; ordering concepts of the purely comparative kind representing an intermediate stage. As long as explicit criteria for their use are lacking, they have, as we noted, essentially a programmatic but no systematic status; and once suitable criteria have been specified, the parlance of extreme types becomes unnecessary, for there are no logical peculiarities which differentiate extreme-type concepts from the other comparative quantitative concepts of empirical science; their logic is the logic of ordering relations and of measurement; henceforth, we will therefore refer to them also as ordering types. Methodologically, ordering as well as classificatory typolo- CARL G. HEMPEL 71 gies belong, as a rule, to an early stage in the growth of a scientific discipline, a stage which is concerned with the development of a largely "empirical" concept system and with its use for description and for low-grade generalization. Systematic fruitfulness, which is an essential requirement for all stages of concept formation, here consists, in the simplest case, in a high correlation between the criteria which "operationally define" a typological order (such as certain anthropometric indices, say) and a variety of other graded traits (such as further anatomical and physiological indices or psychological characteristics). For quantitative scales, such correlations may assume, in favorable cases, the form of a proportionality of several variables (analogous to the proportionality, at constant temperature, of the specific electric and thermic conductivities of metals), or they may consist in other invariant relationships expressible in terms of mathematical functions. 4. Ideal types and explanation in the social sciences. As was mentioned in the first section ideal types, too, are usually presented as the results of isolating and exaggerating certain aspects of concrete empirical phenomena, as limiting concepts which are not fully exemplified but at best approximated in reality.? Despite the suggestion conveyed by this description, I think that an adequate logical reconstruction has to assign to ideal types a status different from that of the extreme or pure types discussed above. For ideal types— or, as Howard Becker aptly calls them, constructed types--are usually introduced without even an attempt at specifying appropriate criteria of order, and they are not used for the kind of generalization characteristic of ordering types; rather, they are invoked as a specific device for the explanation of social and historical phenomena. I shall try to argue now that this conception reflects an attempt to advance concept formation in sociology from the stage of description and "empirical generalization," which is exemplified by most classificatory and ordering types, to the construction of theoretical systems or models. In order to amplify and substantiate this view, it will be necessary to examine more closely the character and function of ideal types as conceived by its proponents. 72 CONCEPT AND THEORY IN SOCIAL SCIENCE According to Max Weber and some writers holding similar views, the use of ideal types makes it possible to explain concrete social or historical phenomena, such as the caste system in India or the development of modern capital-ism, in their individuality and uniqueness. Such understanding is held to consist in grasping the particular causal relationships which interconnect the relevant elements of the total occurrence under examination. If such relationships are to afford a sociologically significant explanation they must be, according to this view, not only "causally adequate" but also meaningful, i.e., they must refer to aspects of human behavior which are intelligently actuated by valuation or other motivating factors. Weber characterizes the principles expressing those connections as "general empirical rules" concerning the ways in which human beings are prone to react in given situations; the "nomological knowledge" conveyed by them is said to be derived from our own experience and from our knowledge of the conduct of others. Weber mentions Gresham's law as a generalization of this kind: it is empirically well substantiated by the pertinent information available, and it is "a rationally clear interpretation of human action under certain conditions and under the assumption that it will follow a purely rational course."8 As for specific ways of discovering meaningful explanatory principles, Weber mentions the method of empathic understanding but adds the reminder that it is neither universally applicable nor always dependable. And indeed, as is made clear by the fuller argument presented in Professor Nagel's paper, the subjective experience of empathic identification with a historical figure, and of an immediate almost self-evidently certain-insight into his motivations, constitutes no knowledge, no scientific understanding at all, though it may be a guide in the search for explicit general hypotheses of the kind required for a systematic explanation. In fact, the occurrence of an empathic state in the interpreter is neither a necessary nor a sufficient condition of sound interpretation or understanding in the scientific sense: not necessary, for, as Professor Nagel's illustration shows, an appropriate theory of psychopathic behavior may provide the historian with an explanation of some phases of Hitler's actions even in the CARL G. HEMPEL 73 absence of empathic identification; not sufficient, for the motivational hypotheses suggested by the empathic experience may be factually unsound. Weber himself stresses that verification of subjective interpretation is always indispensable; he adds that in the absence of adequate experimental or observational data, "there is available only the dangerous and uncertain procedure of the 'imaginary experiment' which consists in thinking away certain elements of a chain of motivation and working out the course of action which would then probably ensue, thus arriving at a causal judgment."" By thus establishing what would have happened if certain specified constituents of the situation had been different, this method yields "judgments of objective possibility," which form the basis of causal imputation in the social sciences. Those judgments evidently have the form of contrary-to-fact condi-tionals, and students of the currently much discussed logic of counterfactuals might be interested in Weber's fascinating illustration of the proposed method by reference to interpretive problems of historiography, among them the question of the significance of the Persian Wars for the development of Western culture;10 Weber's discussion of these topics shows how well he was aware of the close connection between contrary-to-fact conditionals and general laws. An ideal type, then, is meant to serve as an interpretive or explanatory schema embodying a set of "general empirical rules" which establish "subjectively meaningful" connections between different aspects of some kind of phenomenon, such as purely rational economic behavior, a capitalistic society, a handicraft economy, a religious sect, or the like. But then, in intent at least, ideal types represent not concepts properly speaking, but rather theories; and the idea naturally suggests itself that if those theories are to serve their purpose, they must have a character quite similar to that of the theory of ideal gases, say.11 To elaborate and substantiate this con-ception, I will first try to show that the alleged differences between the explanatory use of ideal types and the method of explanation in natural science are spurious; then (in section 5) I will attempt a brief comparative analysis of the 74 CONCEPT AND THEORY IN SOCIAL SCIENCE status of "idealized" concepts, and the corresponding theories, in natural and social science. In natural science, to explain a concrete event means to explain the occurrence of some repeatable characteristic (a rise in temperature, the presence of corrosion, a drop in blood pressure, etc.) in a particular, i.e., at a specified place or in a specified object at a given moment or during a certain period of time (the air in New Haven during the morning hours of September 5, 1952, the hull of a specified ship, patient John Doe at a given time). Explanation of a concrete event does not and cannot reasonably mean an account of all the repeatable characteristics of a given particular, say b. For the latter include the fact that in such and such directions and at such and such spatiotemporal distances from b, there are particulars having such and such repeatable properties; as a consequence, to explain all the repeatable aspects of b is tantamount to explaining every concrete fact in the universe past, present, and future. Evidently this kind of explaining a concrete occurrence "in its uniqueness" is no more accessible to sociology than it is to physics; in fact, even its precise meaning is quite problematic. Thus, all that can be signincantly sought 1s the explanation of the occurrence of some repeatable characteristic U (which may be quite complex, of course) in a given particular b. The task of explaining Western capitalism in its uniqueness, for ex-ample, has to be construed in this fashion if it is to be at all significant; and it is then strictly analogous to the problem of explaining the solar eclipse of March 18, 1950. In either case, there are certain characteristics their combination is referred to as U above-for whose occurrence an explanation is sought (in the case of the eclipse, e.g., those characteristics might include the fact that the eclipse was annular, not visible in the United States, of a duration of 4 hours and 42 minutes, etc.), but there are innumerable other characteristics for which no account is intended (such as, say, the number of newspapers in which the event was described). It is worth noting here that the event thus to be explained, U(b) for short, is still unique because the particular b is unrepeatable. CARL G. HEMPEL 75 In natural science, to explain a unique concrete event, say U(b), amounts to showing that it had to be expected in view of certain other concrete events which are prior to or contemporaneous with it, and by virtue of specifiable general laws or theories. Formally, such explanation consists in the deduction of "U(b)" from those general principles and from the "boundary conditions" describing the antecedent and contemporaneous concrete occurrences. As Max Weber's writings clearly show, an adequate explanation of a concrete event in sociology or historiography has to be of essentially the same character. Reliance on emphatic insight and subjective "understanding" provides no warrant of objective validity, no basis for the systematic prediction or postdiction of specific phenomena; the latter procedures have to be based on general empirical rules, on nomological knowledge. Weber's limitation of the explanatory principles of sociology to "meaningful" rules of intelligible behavior, on the other hand, is untenable: many, if not all, occurrences of interest to the social scientist require, for their explanation, reference to factors which are "devoid of subjective meaning," and accordingly also to "non-under-standable uniformities," to use Weber's terminology. Weber acknowledges that the sociologist must accept such facts as causally significant data, but he insists that this does "not in the least alter the specific task of sociological analysis ..., which is the interpretation of action in terms of its subjective meaning."12 But this conception bars from the field of sociology any theory of behavior which foregoes the use of "subjectively meaningful" motivational concepts. This either means a capricious limitation of the concept of sociology— which, as a result, might eventually become inapplicable to any phase of scientific research—or else it amounts to an a priori judgment on the character of any set of concepts which can possibly yield explanatory sociological theories. Clearly, such an a priori verdict is indefensible, and indeed, the more recent development of psychological and social theory shows that it is possible to formulate explanatory principles for purposive action in purely behavioristic, non-introspective terms. In discussing, next, the role of experiments-in-imagination, 76 CONCEPT AND THEORY IN SOCIAL SCIENCE which are, of course, well known also in the natural sciences, it will be useful to distinguish two kinds of imaginary experi-ment: the intuitive and the theoretical. An intuitive experi-ment-in-imagination is aimed at anticipating the outcome of an experimental procedure which is just imagined, but which may well be capable of being actually performed. Prediction is guided here by past experience concerning particular phenomena and their regularities, and occasionally by belief in certain general principles which are accepted as if they were a priori truths; thus, e.g., in "explaining" the equi-distribution of results obtained in rolling a regular die, or in anticipating similar results for a game with a regular homogeneous dodecahedron, certain rules of symmetry, such as the principle of insufficient reason, are often invoked; and similar principles are sometimes adduced in imaginary experiments involving levers and other physical systems with certain symmetry features. Imaginary experiments of this kind are intuitive in the sense that the assumptions and data underlying the prediction are not made explicit and indeed may not even enter into the conscious process of anticipation at all: past experience and the-possibly unconscious-belief in certain general principles function here as suggestive guides for imaginative anticipation rather than as a theoretical basis for systematic prediction. The theoretical kind of imaginary experiment, on the other hand, presupposes a set of explicitly stated general principles-such as laws of nature and it anticipates the outcome of the experiment by strict deduction from those principles in combination with suitable boundary conditions representing the relevant aspects of the experimental situa-tion. Sometimes, the latter is not actually realizable, as when the laws for an ideal mathematical pendulum or for perfectly elastic impact are deduced from more general principles of theoretical mechanics. The question what would happen if, say, the thread of a pendulum were infinitely thin and perfectly rigid and if the mass of the pendulum were concentrated in the free end point of the thread is answered here, not by "thinking away" those aspects of a physical pendulum that are at variance with this assumption and then trying to envisage the outcome, but by rigorous deduction from CARL G. HEMPEL 77 available theoretical principles. Imagination does not enter here; the experiment is imaginary only in the sense that the situation it refers to is not actually realized and may indeed be technically incapable of realization. The two types of experiment-in-imagination here distinguished constitute extreme types, as it were, which are rarely realized in their pure form: in many cases, the empirical assumptions and the reasoning underlying an imaginary experiment are made highly, but not fully, explicit. Galileo's dialogues contain excellent examples of this procedure, which show how fruitful the method can be in suggesting general theoretical insights. But, of course, intuitive experiments-in-imagination are no substitute for the collection of empirical data by actual experimental or observational procedures. This is well illustrated by the numerous, intuitively quite plausible, imaginary experiments which have been adduced in an effort to refute the special theory of relativity; and as for imaginary experimentation in the social sciences, its customary reliance on empathy underscores its fallibility. Professor Nagel's example of an attempt to anticipate the behavior of a trader in grain provides a good illustration of this kind of mental experimentation. Thus, the results of intuitive experiments-in-imagination cannot strictly constitute evidence pertinent to the test of sociological hypotheses; rather, the method has an essentially heuristic function: it serves to suggest hypotheses, which must then be subjected, however, to appropriate objective testing procedures. The imaginary experiments mentioned by such writers as Max Weber and Howard Becker as a method of sociological inquiry are obviously of the intuitive variety; their heuristic function is to aid in the discovery of regular connections between various constituents of some social structure or process. These connections can then be incorporated into an ideal type and thus provide the basis for the explanatory use of the latter. 5. Ideal types and theoretical models. We have argued that since ideal types are intended to provide explanation, they must be construed as theoretical systems embodying testable general hypotheses. To what extent is this conception reconcilable with the frequent 78 CONCEPT AND THEORY IN SOCIAL SCIENCE insistence, on the part of proponents of the method, that ideal types are not meant to be hypotheses to be verified by empirical evidence, that deviation from concrete fact is of their very essence? As a point of departure in dealing with this question, let us consider more closely how those who hold such views conceive of the application of ideal-type concepts to concrete phenomena. There are few precise statements on this subject; perhaps the most explicit formulation has been given by Howard Becker, in an effort to develop what he terms "a logical formula for typology." Becker suggests that ideal, or constructed, types function in hypotheses of the form "If P then Q," where P is the type invoked, and l is some more or less complex characteristic.13 Concerning the application of such hypotheses to empirical data, Becker says: "In the very nature of type construction, however, the consequent seldom if ever follows empirically, and the antecedent is then empirically 'false.' If Q' then P' "14 By this deviation from empirical fact, by the occurrence of l' rather than 2, a constructed type acquires what Becker calls "negative utility": it initiates a search for factors other than those embodied iin P to account for the discrepancy.15 In this manner, according to Becker, "constructive typology makes planned use of the proviso 'All other conditions being equal or irrelevant' for the purpose of determining the 'inequality' or 'relevance' of the 'other conditions.' "16 This view calls for closer analysis, for it suggests-perhaps unintentionally-the use of the ceteris paribus clause for a conventionalistic defense of typological hypotheses against any conceivable disconfirming evidence.17 To illustrate this point, let us imagine, by way of analogy, a physicist propounding the hypothesis that under ideal conditions, namely in a vacuum near the surface of the Earth, a body falling freely for t seconds will cover a distance of exactly 16 t feet. Suppose now that careful experiment yields results differing from those required by the hypothesis. Then clearly the physicist cannot be content simply to infer that the requisite ideal conditions were not realized: in addition to this possibility, he has to allow for the alternative that the hypothesis under test is not correct. To state the point now in terms of Becker's general schema: we could infer that P CARL G. HEMPEL 79 is not realized only if, in addition to the observational finding ', we could take the truth of the hypothesis "If P then Q" for granted; but for this assumption, we surely have no warrant; in fact, it would make the entire test pointless. Thus, from the occurrence of l', we can infer only that either P was not realized or the hypothesis, "If P then Q," is false. Now, it might seem that we may with assurance assert our typological hypothesis it only we quality it by an appropriate ceteris paribus clause and thus give it the form: "All other factors being equal or irrelevant, l will be realized whenever P is realized." Evidently, no empirical evidence can ever disconfirm a hypothesis of this form since an apparently unfavorable finding can always be attributed to a violation of the ceteris paribus clause by the interference of factors other than those specifically included in P. In other words, the qualified hypothesis can be made unexceptionable by the convention to plead violation of the ceteris paribus clause whenever an occurrence of P is not accompanied by an occurrence of Q. But the very convention that renders the hypothesis irrefutable also drains it of all empirical content and thus of explanatory power: since the protective clause does not specify what factors other than P have to be equal (1.e., constant) or irrelevant if the prediction of l is to be warranted, the hypothesis is not capable of predictive application to concrete phenomena. Similarly, the idea of testing the given hypothesis becomes pointless. It is significant to note here by contrast that in the formulation of physical hypotheses, the ceteris paribus clause is never used: all the factors considered relevant are explicitly stated (as in Newton's law of gravitation or in Maxwell's laws) or are clearly understood (as in the familiar formulation of Galileo's law, which is understood to refer to free fall in a vacuum near the surface of the Earth) ; all other factors are asserted, by implication, to be irrelevant. Empirical test is therefore significant, and the discovery of discordant evidence requires appropriate revisions either by modifying the presumed functional connections between the variables singled out as relevant, or by explicitly introducing new relevant variables. Ideal-type hypotheses will have to follow the same pattern this land muc else here ) suggests "'the Frame problem 80 CONCEPT AND THEORY IN SOCIAL SCIENCE if they are to afford a theoretical explanation of concrete historical and social phenomena rather than an empirically vacuous conceptual schematism. But it is not true, after all, that in physics as well,, there are theories, such as those of ideal gases, of perfectly elastic impact, of the mathematical pendulum, of the statistical aspects of a game played with perfect dice, etc., which are not held to be invalidated by the fact that they possess no precise exemplification in the empirical world? And could not ideal types claim the same status as the central concepts of those "idealized" theories? Those concepts refer to physical systems satisfying certain extreme conditions which cannot be fully, but only approximately, met by concrete empirical phenomena. Their scientific significance lies, I think, in the following points: (a) The laws governing the behavior of the ideal physical systems are deducible from more comprehensive theoretical principles, which are well confirmed by empirical evidence; the deduction usually takes the form of assigning certain extreme values to some of the parameters of the comprehensive theory. Thus, e.g., the laws for an ideal gas are obtainable from more inclusive principles of the kinetic theory of gases by "assuming" that the volumes of the gas molecules vanish and that there are no forces of attraction among the molecules—i.e., by setting the appropriate parameters equal to zero. (b) The extreme conditions characterizing the "ideal" case can at least be approximated empirically, and whenever this is the case in a concrete instance, the ideal laws in question are empirically confirmed. Thus, e.g., the Boyle-Charles law for ideal gases is rather closely satisfied by a large variety of gases within wide, specifiable ranges of pressure and temperature (for a fixed mass of gas), and it is for this reason that the law can be significantly invoked tor explanatory purposes. The preceding analysis suggests the following observations on the "ideal" and the empirical aspects of ideal-type concepts in the social sciences: (i) "Ideal" constructs have the character not of concepts in the narrower sense, but of theoretical systems. The introduction of such a construct into a theoretical context requires, therefore, not definition by genus and differentia, but the CARL G. HEMPEL 81 specification of a set of characteristics (such as pressure, temperature, and volume in the case of an ideal gas) and of a set of general hypotheses connecting those characteristics. (ii) An idealized concept P does not, therefore, function in hypotheses of the simple form "If P then Q." Thus, e.g., the hypothesis "If a substance is an ideal gas then it satisfies Boyle's law," which is of that form, is an analytic statement entailed by the definition of an ideal gas; it cannot serve explanatory purposes. Rather, the hypotheses characterizing the concept of ideal gas connect certain quantitative characteristics of a gas, and when they are applied to concrete physical systems, they make specific empirical predictions. Thus, to put the point in a somewhat oversimplified form, what enters into physical theory is not the concept of ideal gas at all, but rather the concepts representing the various characteristics dealt with in the theory of ideal gases; only they are mentioned in the principles of thermodynamics. (iii) In the natural sciences at least, a set of hypotheses is considered as characterizing an ideal system only if they represent what might be called theoretical, rather than intuitive, idealizations; i.e., if they are obtainable, within the framework of a given theory, as special cases of more inclusive principles. Thus, e.g., the formula for the mathematical pendulum as empirically discovered by Galileo did not constitute a theoretical idealization until after the establishment of more comprehensive hypotheses which (a) have independent empirical confirmation, (b) entail the pendulum formula as a special case, (c) enable us to judge the degree of idealization involved in the latter by giving an account of additional factors which are relevant for the motion of a physical pendulum, but whose influence is fairly small in the case of those physical systems to which the formula is customarily applied. No theory, of course, however inclusive, can claim to give a completely accurate account of any class of empirical phenomena; it is always possible that even a very comprehensive and well-confirmed theory may be improved in the future by the inclusion of further parameters and appropriate laws: the most comprehensive theory of today may be but 82 CONCEPT AND THEORY IN SOCIAL SCIENCE a systematic idealization within the broader theoretical framework of tomorrow. Among the ideal-type concepts of social theory, those used in analytical economics approximate most closely the status of idealizations in natural science: the concepts of perfectly free competition, of monopoly, of economically rational behavior on the part of an individual or a firm, etc., all represent schemata for the interpretation of certain aspects of human behavior and involve the idealizing assumption that noneconomic factors of the sort that do in fact influence human actions may be neglected for the purposes at hand. In the context of rigorous theory construction, those ideal constructs are given a precise meaning in the form of hypotheses which "postulate" specified mathematical connections between certain economic variables; frequently, such postulates characterize the ideal type of behavior as maximizing a given function of those variables (say, profit). In two important respects, however, idealizations in economics seem to me to differ from those of the natural sciences: first of all, they are intuitive rather than theoretical idealizations in the sense that the corresponding "postulates" are not deduced, as special cases, from a broader theory which covers also the nonrational and noneconomic factors affecting human conduct. No suitable more general theory is available at present, and thus there is no theoretical basis for an appraisal of the idealization involved in applying the economic constructs to concrete situations. This takes us to the second point of difference: the class of concrete behavioral phenomena for which the "idealized" principles of economic theory are meant to constitute at least approximately correct generalizations is not always clearly specified. This of course hampers the significant explanatory use of those principles: an ideal theoretical system, as indeed any theoretical system at all, can assume the status of an explanatory and predictive apparatus only if its area of application has been specified; in other words, if its constituent concepts have been given an empirical interpretation which, directly or at least mediately, links them to observable phenomena. Thus, e.g., the area of application for the theory of ideal gases might be indicated, roughly speaking, by interpreting CARL G. HEMPEL 83 the theoretical parameters "P," "V," "T" in terms of the "operationally defined" magnitudes of pressure, volume, and temperature of gases at moderate or low pressures and at moderate or high temperatures. Similarly, the empirical applicability of the principles of an ideal economic system requires an interpretation in empirical terms whch does not render those principles analytic; hence the interpretation must not amount to the statement that the propositions of the theory hold in all cases of economically rational behavior-that would be simply a tautology; rather, it has to characterize, by criteria logically independent of the theory, those kinds of individual or group behavior to which the theory is claimed to be applicable. In reference to these, it has then to attach a reasonably definite "operational meaning" "cost proft, utity ce, in this tashion, the proposi tions of the theory acquire empirical import: they become capable of test and thus susceptible to disconfirmation and this is an essential characteristic of all potential explanatory systems. The results of the preceding comparison between the ideal constructs of economics with those of physics should not be considered, however, as indicating an essential methodological difference between the two fields. For in regard to the first of our two points of comparison, it need only be remembered that much effort in sociological theorizing at present is directed toward the development of a comprehensive theory of social action, relatively to which the ideal constructs of economics, in so far as they permit of empirical application, might then have the status of theoretical rather than intuitive idealizations. And quite apart from the attain-ability of that ambitious goal, it is clear that an interpretation is required for any theoretical system which is to have empirical import—in the social no less than in the natural sciences. The ideal types invoked in other fields of social science lack the clarity and precision of the constructions used in theoretical economics. The behavioral regularities which are meant to define a given ideal type are usually stated only in more or less intuitive terms, and the parameters they are 84 CONCEPT AND THEORY IN SOCIAL SCIENCE meant to connect are not explicitly specified; finally, there is no clear indication of the area of empirical applicability and consequent testability claimed for the typological system. In fact, the demand for such testability is often rejected in a sweeping manner which, I think, the preceding discussion has shown to be inconsistent with the claim that ideal types provide an understanding of certain empirical phenomena. If the analysis here outlined is essentially sound, then surely ideal types can serve their purpose only if they are introduced as interpreted theoretical systems, i.e., by (a) specitying a list of characteristics with which the theory is to deal, b) formulating a set of hypotheses in terms of those characteristics, (c) giving those characteristics an empirical interpretation, which assigns to the theory a specific domain of application, and (d), as a long-range objective, incorporating the theoretical system, as a "special case," into a more comprehensive theory. To what extent these objectives can be attained cannot be decided by logical analysis; but it would be self-deception to believe that any conceptual procedure essentially lacking in the first three respects can give theoretical understanding in any field of scientic inquiry. And to the extent that the program here outlined can actually be carried through, the use of "ideal types" is at best an unimportant terminological aspect, rather than a distinctive methodological characteristic, of the social sciences: the method of ideal types becomes indistinguishable from the methods used by other scientific disciplines in the formation and application of explanatory concepts and theories. 6. Conclusion. In sum, then, the various uses of type concepts in psychology and the social sciences, when freed from certain misleading connotations, prove to be of exactly the same character as the methods of classification, ordering, measure-ment, empirical correlation, and finally theory formation used in the natural sciences. In leading to this result, the analysis of typological procedures exhibits, in a characteristic ex-ample, the methodological unity of empirical science.