[p.v]

	“We may have many concepts but few confirmed theories; many points of view, but few theorems; many ‘approaches’ but few arrivals. Perhaps a shift in emphasis would be all to the good.”
	Robert K. Merton

Preface to the Second Edition

More than half of the material in this little book is new in the sense that it was not included in its first edition. Deletions from the first edition are equally extensive; also, everything relating to definitions, taxonomy, and descriptive studies has been reserved for a later, separate treatment.

In making these revisions I have benefited greatly from comments on the first edition. I would like to thank particularly Professors B. Andersson, G. Boalt, E. Dahlström, T. Hopkins, G. Karlsson, and P. F. Lazarsfeld for their helpful reviews. Parts of the material added to this edition have appeared in German, Italian, and Polish. I am especially grateful for the detailed comments on the German version given by Professors P. F. Lazarsfeld and H. Wold.

It is a coincidence that looks like a forethought that I was in Sweden during both writings of this work. The initial writing was done in 1952 at Professor Segerstedt’s Institute at Uppsala University, and the present revision [p.vi] was done in 1963 at Professor Dahlström’s Institute at Gothenburg University. The hospitality and adventurous spirit of reflective inquiry at these Institutes will always remain among my fondest memories. In both places the basic question “How is sociology possible?” was asked in earnest, and in both places I have been pleased to retort that sociology is possible, or at least easier, if it is theoretical. I was rather young and ignorant when I first said it, and I have appreciated this opportunity to say it again, and perhaps a little better.

The intellectual trends of thought and experiences that have shaped my emphasis on theoretical sociology might be briefly sketched. In Sweden my teacher, Torgny T. Segerstedt, like many others in many countries — allowed emphasis on theory to make up for soft methodology. My interest in this issue was aroused when I was called upon to defend this stand in a debate with psychometricians but found nothing written about it. Later I became acquainted with a somewhat parallel American debate of older standing at Columbia University. The issue here was whether sociology would advance more by concentrating on theory — a position taken by Robert K. Merton — or by concentrating or methodology — a position taken by Paul F. Lazarsfeld. I learned much from this discussion, and here again I sided with the theorists. As is evident from my book Social Theory and Social Practice, I have even come to trust applied theory as much as applied research. [p.vii]

This bet on theoretical sociology, however, has not emerged because I have rejected the arguments by the methodologists but because I have fully accepted them. As a graduate student at the University of Minnesota, Professors F. Stuart Chapin and Neal E. Gross introduced me to a rigid methodology based on the dictums of logical empiricism, and here I met the strict methodological ideas of George A. Lundberg, not merely as the ideals for research I had known from my first acquaintance with sociology, but as part of ongoing research practice. Using these strict standards, I eventually came to feel — apparently with several colleagues in America and Europe — that not only my own but even the most celebrated research projects in our field left something to be desired from a methodological point of view. And the whole research enterprise conducted in this fashion, which too often rendered trivial conclusions with efforts towards maximum precision, forced me to question sociology as a worthwhile occupation.

In this situation, the call for theory was neither an escape nor just a call for additional requirements to be met. It was simply a call for intellectual salvation. The saving quality of a theory is to coordinate many methodologically imperfect findings into a rather trustworthy whole in the form of a small number of information-packed sentences or equations. Moreover, some of the bits and pieces coordinated into this trustworthy whole can be the challenging insights of the classics of sociol-[p.viii]ogy and the celebrated writers of literature: in short, far from trivial propositions.

Contrary to the prevailing emphasis on taxonomical “social theory,” it also became clear that only propositional “theoretical sociology” contained such potentialities. This was epitomized in the motto from Robert K. Merton that opened the first edition of this book. The same motto remains for good reasons in the second edition, because the kind of theory it advocates has been drowned by louder taxonomizing voices in the past decade. All signs now are that the next decade will understand it better, and that the theoretical enterprise in sociology will see not only definitions, but more and more propositions, and thus will become theory rather than terminology.

I am enough convinced that this trend represents the future so that the new edition of this small book is conceived, not only as a pamphlet with a polemical cut, but also as a supplementary text which some teachers might find helpful in training future sociologists in courses on sociological research and in courses on sociological theory. As a science, sociology has already bridged the gulf between theory and research; this is true both in principle and in the work of several gifted scholars. The question now is to teach students to run back and forth across this bridge. Our compartmentalized instruction in theory and research might obscure the connection between the two for the students, and we need to establish [p.ix] a better pedagogical tradition at this critical juncture.

I am well aware that this text does not take into account all, or even most, of the niceties elaborated by various philosophies of science, and also that illustrations from the physical and biological sciences would often have conveyed with greater clarity the methodological principles involved. However, in a text for sociology students, the details of philosophy of science are out of place, and many of the points made in works on the logic and philosophy of science have little or no relevance or consequence for sociology as it is currently practiced. And it is an essential pedagogical requirement that our examples and illustrations should be taken from sociology. Actually, the time has passed when sociology students learn scientific method by examples from physics and other so called established sciences. By now, sociology itself is established, and it has become varied and sophisticated enough to provide the illustrations we need for the study of principles of theory construction and theory verification. Gothenburg in April 1963

Hans L. Zetterberg
Columbia University

[p.x]

Preface to the Third Edition

To this edition has been added one chapter on sociology as a humanistic discipline and one chapter on definitions in sociology. One may take this as sign that propositional theory in sociology now is so well entrenched that it can afford to take generous cognizance of competing approaches. Several corrections and additions have also been made to the text.

New York in October 1964
H.L.Z.

[p.xi-p.xiii]

Preface to the second edition
Preface to the third edition

1. On sociology as a Humanistic Discipline 1

The humanistic content of sociology 1Humanistic aspects of the education of sociologists 3

2. On sociology as a Scientific Discipline 9

The model of physics and biology 9Partial and grand theories 14Approaches emphasizing definitions and propositions 21

3. On Definitions in Sociology 30

Terms 32Varieties of conceptions 34The ordering of definitions 43Minimum terms: The uniqueness of our subject matter 49A consideration in selecting primitive terms for sociology 52The generation of derived terms 54Descriptive schemas 57

4. On Propositions in Sociology 63

Variates: Determinants and results 64The varieties of linkage between determinants and results 69Functional propositions 74Ordinary and theoretical propositions 79

5. On the Ordering of Sociological Propositions 87

Inventory of determinants 88Inventory of results 89Chain patterns of propositions 90Matrixes of propositions 93Axiomatic format with definitional reduction 94Axiomatic formats with propositional reduction 96

6. On the Confirmation of a Proposition 101

An overview of steps in confirming a proposition 104The separation of definitions and indicators 111

7. On the Decisions in Verificational Studies 114

Internal validity 114External validity 120Reliability 123Scope 126Representatives 128Designs 130The composite judgment of acceptance or rejection 150

8. The Confirmation of Complex Theories 157

Axiomatic theories and research 159Testing total theories through their gross preditions 166

Concluding Remarks 175

[p.1]

1. On Sociology as a Humanistic Discipline

The Humanistic Content of Sociology

Symbols are the stuff out of which cultures and societies are made. This assumption is basic to much recent work in sociology. ¹ For example, a sequence of conception, birth, nursing and weaning represents the biological reality of parenthood. But in analyzing human parenthood we find, in addition to the biological reality, a complex of symbols dealing with the license to have children, responsibilities for their care and schooling, rights to make some decisions on their behalf, obligations to launch them by certain rituals (such as [p.2] coming-out parties), privileges to enjoy their respect and to receive support to welfare in old age. Our language thus contains codifications of what parents are and what they shall do and what shall be done to them, and all these sentences in our language represent the social reality of parenthood. Social reality, in this as in all other cases, consists of symbols. ²

Academic disciplines such as foreign languages, literature, philosophy, arts, whose subject matter is symbols, are usually called humanistic ones. By this classification, sociology is a humanistic discipline, since social reality consists of symbols. It is not surprising, then, that the vocabulary used most comfortably by today’s sociologists has come from the world of letters. It is essentially the language of drama. ³ Sociologists talk about roles, publics, actors, decisions, choice, charisma, achievement, domination, and so forth. ⁴ The humanistic content of sociology and its affinity to literary and dramatic analysis and criticism is very obvious. Some prominent sociologists, e.g. Hugh D. Duncan, effec-[p.3]tively use even such terms as hero, villain, victim, tragedy and comedy in sociological discourse. ⁵

Humanistic Aspects of the Education of Sociologists

The process of learning sociology also has much in common with the learning of other humanistic disciplines. In spite of some extravagant claims to the contrary, one cannot at present adequately learn sociology by reading the latest textbook. The sociology student must also read the classical works of this field. In this sense he resembles the student of literature, philosophy and other humanistic subjects. The classics represent turning points, occasions when past formulations were superseded in giant steps by more far reaching and inclusive formulations. In this way the classics highlight the history of the field. Furthermore, they were written by men of foresight, men with a sense for the essentials, men who had a rare gift of feeling the crucial problems of their topic. Therefore, contemporary scholars return to them over and over again, not only to learn about the history of their discipline, but in search for new cues and insights. Not many books qualify as classical; the criticism that accumulates with the passing of time rele-[p.4]gates many books from the shelf of classics to the shelf of intellectual history.

A classic must stand at last alone: without apology, exegesis, or alibi. It must speak for itself to strangers; it must be intelligible, and seem true, after all its special friends are dead. It must have the minimum of weakness, vagueness, vanity, wind. It must be well made at the seams, to stand the long voyage it hopes to make, and to endure the waves either of contempt or of competition. It must have been made, in other words, by one who knew how to make such things, and nothing else about him will matter — who he was, how he looked, or what he thought about other things than the things he treated.⁶

The message of a classic is rarely straightforward and is not readily caught in a formula. Consider, for example, Lorenz von Stein’s Geschichte der soziale Bewegung in Frankreich. ⁷ Like most classical works it is like a nest of Chinese boxes and can be appreciated on many levels. The outside box is simply the history of the French Revolution. It is a famous and early guide to the Revolution for the German speaking world, written with intensity and insight, by a brilliant young Dane who went to study in Paris in the 1840’s. The second box is [ethical] socialism. Stein’s book became a standard source of knowledge about socialism before Marxism developed its full-fledged force. It continued for more than half a century to be an inspiration for reformist movements. Indeed, many of its theses remain very much alive today. For example, von Stein’s delineation of the limited importance of con-[p.5]stitutional reform as compared to changes in property distribution, and his contentions that voting rights are ineffectual as long as chances to acquire property and education are restricted, help us to understand puzzling problems of today as well as yesterday. The third doll is social theory. Stein developed a theory of society, coined terms such as “the proletariat” and used them objectively, and formulated verifiable propositions about class struggles and social change. His book is essential to the history of sociology. Clearly some of Marx’s ideas were developed under his influence. The alternative outcomes of the class struggle, according to von Stein, are “revolution or reform.” This position was rejected by Marx in favor of “the inevitable revolution,” but the evidence we have accumulated since then seems to favor von Stein’s theory. The innermost box in von Stein’s work is Hegelianism. We have here a readable Hegelian conception of society. Bursts of sudden changes due to accumulation of dissonances are seen as the key to history. Sociologists later abandoned this mechanism of change in favor of the equilibrium mechanism. However, in recent years disenchantment has grown with the equilibrium models and interest again turns to the older model. Thus, von Stein’s work is not only of historical importance, great as that may be; it meets also the test of a classical work because we return to it many years later to find cues and inspirations for the best ways of dealing with our current problems. The [p.6] richness of this and any classic is likely to escape us in any one reading. Classics cannot readily be summarized and closed; they feed continuous conversation and debate.

An education based on the classics remains even after we have forgotten the details of our classics. The panorama of major organizations and markets in civilized societies analyzed in Weber’s Wirtschaft und Gesellschaft ⁸ makes the reader a learned man if he remembers its details. However, even after he has forgotten its details, he remains an educated man, immune to the easy and sweeping generalizations about human society that plague us, yet somehow aware of the relatively simple forces that shape human society. Sense and good judgment thus are bred by exposure to the classics.

To sum up, sociology is a humanistic subject because its subject matter is symbols. Sociological education also follows a humanistic model to the extent it relies on the reading of classics of social thought as stimulation for contemporary thinking about society and as the primary means to convey sociological wisdom.

The guardians of the humanistic tradition in sociology often call themselves “social theorists.” Two interrelated conceptions of “theory” are found here. First, there is a habit of designating all of the better sociologi-[p.7]cal writings of older vintage as “social theory.” Statistical studies of suicide, historical studies of the effect of religion on the economy, informal observations on the role of secrets in social life, and anything else written at least a generation ago is likely to be called “social theory,” if the work is good enough to live in the memory of contemporary sociologists and to be read and cited by them. An alternative and better term than “theory” for this material would be “sociological classics.” Thus an anthology entitled Theories of Society⁹ contains mostly classical passages of sociological literature.

A second conception of “social theory,” also common in the humanistic tradition of sociology, equates it with a commentary on sociological writing, usually from an historical perspective. “Theorists” of this variety trace continuities in the accumulation of sociological knowledge; they discover the occasions when old wine has been poured into new bottles and new wine into old bottles. “Theory” here means essentially “sociological criticism.” An anthology containing sociological criticism is entitled Modern Sociological Theory in Continuity and Change.¹⁰

[p.8] But many sociologists do not accept classics and criticism as “theory” and they promote other conceptions of what constitutes “theory.” In turning now to these more “scientific” conceptions of social theory which are the topic of this book, let us agree from the onset that they will be poor in content if they are not informed by the humanistic tradition of sociology.

Notes to Chapter 1

1. The emergence of this view among prominent sociologists of the past is reviewed in Hugh D. Duncan, Communication and Social Order, Totowa, N.J.: The Bedminster Press, 1962. I have consciously adopted this basic tenet in my book Sociology in a New Key (Totowa, N.J.: The Bedminster Press, 1965) not only in reviewing problems of culture where this view long has been commonplace, but also in dealing with problems of social structure, an area in which this view is rarely made explicit. Moreover, I attempt to use this view as a basis for scientific rather than humanistic sociology.

2. Cf. Torgny T. Segerstedt, Verklighet och värde, Lund, Sweden: Gleerup, 1937.

3. This has not always been the case. See the accounts of the “mechanistic” and the “bio-organismic” schools of sociology in P. A. Sorokin, Contemporary Sociological Theories, New York: Harpers, 1928, chs. 1 and 4.

4. It should be superfluous to point out that use of statistics does not make the sociologists’ discipline less humanistic; statistics provide means of condensing information, of saying “more” and “less” in a precise way, and of discerning complex relations between events, all very useful skills in humanistic pursuits.

5. Hugh D. Duncan, The Symbolic Act, forthcoming.

6. Mark van Doren, The Happy Critic and Other Essays, New York: Hill & Wang, 1961, p. 27. [The quote in the text appeared originally in this footnote.]

7. Lorenz von Stein, History of the Social Movement in France, translated, edited, and introduced by Kaethe Mengelberg, Totowa, N.J.: The Bedminster Press, 1964.

8. Max Weber, Wirtschaft und Gesellschaft, 4th ed., Tübingen J. C. B. Mohr, 1956. English version, Economy and Society, edited by Guenther Roth [and Claus Wittich], Totowa, N.J.: The Bedminster Press, 1966.

9. Talcott Parsons, et al., Theories of Society: Foundations of Modern Sociological Theory, 2 volumes, New York: The Free Press of Glencoe, 1961.

10. Howard Becker and Alvin Boskoff (eds.) Modern Sociological Theory: In Continuity and Change, New York: The Dryden Press, 1957.

[p.9]

2. On Sociology as a Scientific Discipline

The Model of Physics and Biology

Despite their humanistic training and despite the humanistic content of their field, sociologists never seem to tire in telling their listeners and readers that their discipline also is a scientific one. This involves a series of assumptions to which rash answers are unwise. Basically, it presupposes that the same general mode of reflective inquiry that is used by scientists when the subject matter is biological and physical may be used when the subject matter consists of symbols. This is a bold assumption. From Dilthey’s celebrated elaboration of the distinction between Geisteswisserschaft and Naturwissenshaft to C. P. Snow’s analysis of the deep rift between the “two cultures” we have been told how very different humanistic study is from scientific body. The question how readily they combine in the practice of sociology is very interesting.

There is actually a rather smooth transfer of the mode [p.10] of inquiry developed in the study of biological and physical events to the study of symbol events. Perhaps the reason for this is less the physical nature of symbol events than the symbol nature of physical events. For the data of physics and biology are not accessible and intelligible to us in the raw; they come to us already couched in a language. The physical and biological sciences thus have, in some measure, symbols as their subject matter since they deal with the ways in which man perceives and interprets living and dead matter through his symbolizations. ¹

To grasp some of the implications of conceiving of sociology as a science, consider for a moment our education in the physical sciences. In our childhood many of us enjoyed reading some popular book in physics containing chapters called, “Automobiles,” “Aeroplanes,” “Radios,” “Guns,” etc. In high school, however, our physics texts did not have these titles. Now the chapter headings were “Mechanics,” “Optics,” “Thermodynamics,” etc., and the cars, planes, radios, and guns occurred only as illustrations of the principles valid in these various branches of physics. The scientists, we learned, had proceeded on the assumption that there was an underlying order behind the varied manifestations of physical events. Through observations and experiments they [p.11] learned about this order by finding regularities in the behavior of physical phenomena. They formulated these regularities in as compact form as possible and obtained their scientific laws. They combined and related their laws to each other and obtained the theories of physics. These theories then became the basis for calculations by engineers who, in response to practical needs, created the technological wonders of our age. We learned these theories, and understood the operation of planes, radios, guns, and many other things.

We may also say that these theories became the explanation of the technical wonders of our childhood. As is well known, we explain something by demonstrating that it follows the laws of other phenomena. To ask for an explanation in science is thus to ask for a theory. A scientific theory, then, is a sword that cuts two ways. On the one hand, it is a system of information packed descriptions of what we know; on the other hand, it is a system of general explanations. No further justification has to be given for interest in theories: the quest for informative description is the quest for theory, and the quest for explanation is a quest for theory. Of course, we may also ask for the explanation of a theory. This desire can be answered only to the extent that we know of a more inclusive theory — sometimes called “grand theory” — of which the former is a special case. In physics, the theory of relativity and the quantum theory are inclusive theories in terms of which most laws of physics can [p.12] be explained. Since they explain most laws they can also explain most phenomena. The final goal of the scientific enterprise is to know such a theory. It is interesting to note, however, that the quantum theory and the relativity theory cannot be derived from one another. Physics still awaits the grand theory that combines the two. The grand theory, however, cannot be explained. In the face of such a theory our curiosity would have to rest.

The case for sociology as a science now breaks down into affirmative answers to questions such as these: Is there an underlying order behind social reality? If so, have any sociological laws been discovered? If so, have these laws been combined into theories explaining social reality? If so, can these theories be used to calculate solutions to practical problems?

The critical question is the one about the existence of sociological laws. Sociologists, of course, know a large number of facts about their society — how many Negroes there are, how many people belong to voluntary associations, how many persons have advanced into high ranking jobs, and other facts reported in A Sociological Almanac for the United States and similar publications. But, apart from such facts, are there in the body of sociological knowledge any laws? The answer is undoubtedly “Yes.” However, the actual number of sociological laws is subject to debate, because different sociologists cannot agree on how stiff to make criteria [p.13] for calling a general statement about societal life a sociological law. Furthermore, there is a lack of agreement about the precise language and formulation of these laws. Any inventory of the laws of sociology becomes, therefore, subject to some convictions and preferences not shared by colleagues in all details. But such an inventory could nevertheless be made.

An inventory of knowledge gleaned from research on human behavior has been compiled by Berelson and Steiner. ² It contains 1045 numbered propositions. These propositions are not laws but research findings. Nor do they constitute theories; they are simply listed and no attempt is made to relate them to each other. Here are some examples:

	“Prolonged unemployment typically leads to a deterioration of personality: passivity, apathy, anomie, listlessness, dissociation, lack of interest and of caring” (p.403).
	“A person’s self evaluation is strongly influenced by the ranking of his class (that is, by the society’s evaluation of the group to which he belongs)” (p.489).
	“Television viewing by children is heaviest among the duller and the emotionally insecure” (p.535).
	“The more people associate with one another under conditions of equality, the more they come to share values [p.14] and norms and the more they come to like each other (p.327).
	“Even the simplest experiences are organized by the perceiver; and the perceived characteristics of any part are a function of the whole to which it appears to belong (p.104).
	“The leaders of major social changes in a society are unlikely to come from the group traditionally in control; they are more apt to come from deviant, marginal, disaffected groups (p.618).
	“People prejudiced against one ethnic group tend to be prejudiced against others (p.502).

As we shall see later (Chapter 6) the distinction between findings and laws is one of degree of generality and degree of empirical support. In reviewing the Burleson Steiner thousand plus propositions one finds that anywhere from five to fifty of them are general enough to qualify as laws, depending on how strict we make our criteria. At any rate, there is no doubt that sociology now has a number of lawlike propositions that can be called confirmed or trustworthy.

Partial and Grand Theories

We have several works in sociology that combine lawlike propositions into systems, that is, theories. A good [p.15] example in Hopkins’ book The Exercise of Influence in Small Groups.³ It summarizes a number of research studies into fifteen propositions. Each is explicitly stated and evaluated for its consistency with existing findings and consistency with other accepted propositions. The propositions are then ordered so that they each came to represent a part of a process which brings into balance the ranks held and the influences exercised by the members of any group. It is demonstrated, both by theoretical deduction and empirical investigations, how this process accounts for characteristics of group leaders, opinion changes among members, and group stability and instability. By all reasonable criteria, this is an acceptable scientific theory.

The history of sociology shows that we have been very eager to reach not only such theories of modest scope but to reach for grand theories. The last notable achievement here was made by Pitirim Sorokin. His theory on social and cultural change⁴ starts with a basic scale of mentality, to classify the symbols that constitute social reality. Symbols may be either ‘sensate,’ that is, they refer directly to sense data, or they may be ‘ideational,’ that is, their referents are many steps removed from sense data. Sorokin reviews the flow of events in art, sci-[p.16]ence, law, religion, and ethics from the beginnings of civilization to the present and tabulates the symbol products of culture in terms of the sensate or ideational mentality they exhibit. He establishes two basic lawlike generalizations. First, he shows that in each realm of society taken by itself, the mentality shows secular swings from sensate to ideational and back. Second, he shows that different realms tend to move together in their secular fluctuations between sensate and ideational mentality. Since no particular realm consistently seems to lead or initiate these changes, Sorokin rejects the claims of writers who attempt to locate the moving force of history in any particular realm of society such as religion, politics, art, or economy. Instead, the super rhythm, he speculates, has causes of its own operative in any realm; adaptation and efficiency is hampered by an excessive sensate mentality as well as by an excessive ideational mentality, and the movement toward a consistent mentality in the end always defeats itself.

Sorokin then proceeds to explore, by theoretical deduction and by research, the correlates of this super rhythm. His ambition is to show that if we can locate the phase of a civilization on the master curve of the super rhythm we should also be able to tell a good deal about it: the content of its dramatic works, the topics of its pictures, the organization of its government, the number of inventions made, the likelihood of wars and revolutions, etc. Most important among these correlates [p.17] are the nature of relations between men and his fellowmen; whether they are familistic, contractual, or compulsive. It is the large number of correlates to sensate and ideational mentality that makes Sorokin’s theory the grandest that so far has appeared in sociology. However, often it is not entirely clear how these correlates are derived from the two basic generalizations, and the empirical demonstrations of the correlatives are not always as convincing as one might wish. Hence, the theory is still controversial. Partly as a reaction against Sorokin’s effort to write grand theory, one of his students, Robert K. Merton, formulated a strategy that has become widely accepted by contemporary sociologists. Merton entered a plea for “theories of the middle range.”⁵ These are miniature theories, not grand theories; or, better expressed, partial theories rather than inclusive theories. When we call a theory partial (or middle range, or miniature) we admit that there are other accepted theories which are not contradicted by, or synonymous with, the one we call partial. Optics and thermodynamics are examples of partial theories in physics. As mentioned, in social psychology, anthropology, and sociology we also have a few such theories: e.g., Festinger’s theory of cognitive dissonance, Homans’ theory of elementary social behavior, [p.18] Hopkins’ theory of influence, Murdock’s theory of kinship structures, Pareto’s theory of elites. Sociology is believed to advance most rapidly by developing a large array of such partial theories.

At the horizon of sociological thought looms, however, the challenging issue of integrating these partial theories into a more inclusive whole. The plea for theories of the middle range would be misguided if it implied a condemnation of all efforts toward an inclusive theory. If inclusiveness is considered a matter of degree, it becomes again a manageable goal. (Inclusiveness is, indeed, a matter of degree: one theory can be more inclusive than another and yet be a partial theory in regard to a third theory). Steps toward inclusiveness comprising integration of two or more partial theories should be encouraged; herein lays one of the greatest challenges of the theoretical undertaking.

Thus there are two types of specialists in theoretical sociology: the man who develops new partial theories out of his own or other people’s research, and the man who takes a number of partial theories developed by others and integrates them into a more inclusive theory.

In spite of the fact that we have a few brilliant attempts toward grand theory and several examples of partial theories and recognize that the integration of the latter into a more inclusive theory is a possibility, the dominant impression in looking at the sociology of today is one of theoretical paucity. We have read, in [p.19] college or privately, some more or less popular texts of sociology. These texts contained chapter headings like “The Family,” “Social Class,” “Public Opinion,” “Race Relations,” etc. They dealt with the rather interesting, but theoretically unconnected, topics traditionally assigned to departments of sociology. They were more like our childhood popular physics books than the systematic physics texts of our later schooling. Present sociological thinking has actually rather little to offer the student who wants to go beyond this topical study to explain family structure, social class, public opinion, race relations, and other topics in terms of a few laws of sociology.

There is, of course, an enormous amount of research done which gives information about all these sociological topics. This flow of research findings, in fact, has become so great that it is now a losing game to try to keep abreast of all the findings. Monographs, journal articles, research proposals, mimeographed reports overtake man. To be a social theorist rather than a social researcher is no refuge from the flow of this research. The days are gone when “theory” and “speculation” meant the same thing and the theorist did not have to know anything except the location of the space bar on his typewriter. Generally speaking, the modern theorist, as we visualize him, has to know more empirical findings than the most down to earth researcher, since he is, among other things, concerned with the systematization [p.20] of the knowledge researchers have acquired: one outcome of his labor is in the form of documents that summarize the past discoveries and events in lawlike statements. We note with gratification that we now have obtained a few such research grounded theories in the field of sociology. But we also note that most work in theoretical sociology remains to be done.

The difficulty of the sociologists’ struggle with theory lies in part in the dilemma of sociology being both a humanistic and scientific discipline. We never escape the humanistic content of sociology but we try to treat it scientifically. Of the elders in the field, hardly anyone has devoted himself wholly to the task of theoretical sociology. It is understandable that one of the first sociological theoreticians in the modern sense, Vilfredo Pareto, did not begin his work in theoretical sociology until he was over fifty years old. ⁶ It is a sadder commentary that, fifty years later, a representative theoretician, George C. Homans, who received his training and has his career at one of the world’s best universities, confesses not only other pursuits prior to his endeavor in theoretical sociology, but also to a long process of unlearning of fettering traditional approaches to social [p.21] thought. ⁷ Some fortunate members of the generation now being trained in sociology will be the first ever to orient themselves from the very start of their careers toward actual theory construction.

Approaches Emphasizing Definitions and Propositions

Within the humanistic tradition of sociology we found that “social theory” could mean two related but different things: classical works and criticism. Within the scientific tradition of sociology, “social theory” also stands for two different but related enterprises. One is represented by an anthology such as Toward A General Theory of Action. ⁸ The task of the writers of this book is to develop an orderly schema defining anything to which sociologists (and other social scientists) should pay attention. Names are assigned to these subjects, and the reader is encouraged to go out and discover their concrete manifestations in all parts of society. A more specific term than “theory” for this system of definitions is “sociological taxonomy.” The anthology mentioned contains mostly suggestions for a general taxonomy of the social sciences.

[p.22] Concern with sociological classics, sociological criticism, and sociological taxonomy are all to the good. I, for one, enjoy pursuing these interests in my teaching and writing. However, in this book, I primarily want to pursue sociological theory in a fourth sense: systematically organized, lawlike propositions about society that can be supported by evidence: This is “theory” in the sense this word is used in other sciences. Taxonomy enters this enterprise only to the extent that we sometimes need to define the terms used in our propositions. Propositions are the central elements; definitions are auxiliary. As a reminder that this is a different breed of animal I shall speak of it as “theoretical sociology” rather than “social theory.” This usage is meaningless in itself, but I believe it will serve a good purpose. Let us spell out in some further detail how this propositional approach differs from the taxonomical one by considering the dilemma that both want to tackle.

An initial difficulty for the sociological theorist is, as mentioned, the great variety and complexity of phenomena with which his discipline customarily deals. As we noted, the topics have a wide range: family discord, social mobility, labor management relations, propaganda, public opinion, crime, housing, rural urban migration, race relations, and a series of technical subjects related to the organizations and institutions of government, industry, business, education, art, religion, welfare, civic affairs, mass media, and others. It is easy to [p.23] argue that no man knows enough or is wise enough to deal with all these phenomena.

There have been times when sociology was imperialistic enough to claim all aspects of all societal phenomena as its proper realm. But the expanding scientific knowledge about society can never be the monopoly of any one academic discipline. It is a joint enterprise of historians, economists, political scientists, demographers, sociologists, anthropologists, geographers, and others. The sociologists hold only a few of the pieces to the picture puzzle of society. Specialization is necessary. Sociology, too, is a specialized science.

These two statements — that sociology deals with just about every social phenomenon, and that sociology is but a specialized one of the many social sciences — do not complement each other. How can sociology deal with everything and yet be a specialized science of the social world? Or how can sociology be a specialized social science and yet deal with all societal phenomena? The diversity of subject matter and the necessity for specialization pose a dilemma. ⁹ In principle, the resolution of the dilemma does not appear difficult. No science seems to deal with all aspects of what common sense considers one phenomenon. In a recent text in sociology, beginning students are given a clear demonstration of this: [p.24]

Consider your instructor’s chair. If a specialist in the branch of physics called mechanics were to study it, he would see it as a combination of weights and balances; a biologist specializing in anatomy would see it as a receptacle for the human form and might assess its effect on the spinal column; an economist might see it as a product of mass production, a unit of cost and price; the psychologist might see it as a part of the perceptual frame of the student; and the sociologist might see in the chair a symbol of status. Like any field of inquiry, sociology is selective in its approach.¹⁰

Thus the specialization of sociology lies in its concentration on certain aspects of any social problem or any social institution, not in an inclusive study of one or two institutions or social problems.

A first resolution of our dilemma is to specify a small number of definitions which delineate the few aspects of reality with which sociology deals. These definitions, broadly speaking, tell the sociologist what is important for him to pay attention to when he views a human relationship, a group, or a society. The geographer, armed with definitions such as “latitude” and “longitude,” looks upon a given area of the earth in these terms, but can leave such problems as the age of the crust of the earth in a given area to a geologist. Likewise, a sociologist looking at a group in terms such as “rank” and [p.25] “norms,” which are among his key definitions, can leave problems of the members’ “personality traits” to the psychologist, who has a series of definitions to delineate them. Much work in sociology has concentrated on the development of definitions of the descriptive categories that a sociologist is to use.

This is what we call taxonomy. The goal is an orderly schema for the classification and description of anything social. Thus, when faced with any subject of research, the sociologist can immediately identify its crucial aspects or variables by using his taxonomy as a kind of “shopping list.” To “test” his taxonomy, he takes a fresh look at subject X and shows that the general terms defining his dimensions have identifiable counterparts in X. For example, Parsons assigns certain abstract attributes to a social system, and then turns to economy, for instance, and finds that economic thinking takes these dimensions into account. He concludes that the economy is a social system. This is occasionally called to “derive” X, or “explain” X — speech habits which are rather misleading; a better term is to “diagnose” X. To make a sociological diagnosis of the subject matter or problem X is to describe X in terms of a sociological taxonomy. For example, when Parsons and Smelser suggest that the distinction between short term supply and demand in the economy is a special case of the distinction between performance and sanction in a [p.26] social system ¹¹ this is not a sociological derivation or explanation of supply and demand; it is a sociological diagnosis.

Taxonomies summarize and inspire descriptive studies. Thus Parsons’ taxonomy guided Stouffer and Toby to a descriptive study which presented the distribution of some college students on the variable “particularism-universalism” defined by Parsons. This variable is one among others in a set called “the pattern variable schema” which has proven useful in characterizing any social relation.¹² Since sociology, like geology, botany, and geography, is largely a descriptive science, the importance of sociological taxonomy must be taken for granted. However, it should be emphasized again that a concern with taxonomy and descriptive studies does not furnish any explanations.

To know the labels of phenomena and to know their distribution is not to explain them. In the best case, these sets of definitions and maps of distributions leave you where Linnaeus left biology in the eighteenth century — that is with denotations of species and studies of their distribution. When Darwin formulated the principles of the origin of any one species from others, he pushed biological thinking toward something more wor-[p.27]thy of being called a theory. He not only formulated definitions of categories to investigate what cases fall into these categories; he formulated propositions and started to verify them. Sociological thinking, if it is to progress scientifically, is also bound to add some propositions to the already long array of definitions, and to let some of the effort now going into the making of descriptive studies be allocated to the verification of these propositions. In so doing, we should, of course, use as many of the previously formulated definitions as we can. Darwin was greatly aided by Linnaeus’ definitions, and some — but not all — of Linnaeus’ definitions became definitions in Darwin’s theory.

A second and related resolution to the dilemma between diversity of subject matter and the need for specialization enters here. It is represented by the program for sociology set forth by Georg Simmel over half a century ago:

. . . we shall discover the laws of social forms only by collecting such societary phenomena of the most diverse contents, and by ascertaining what is common to them in spite of their diversity.¹³

The assumption here is that sociology will eventually discover a small number of propositions that are valid in several diverse contexts. This idea, that there are [p.28] sociological propositions that hold in diverse contexts, is gradually becoming more of an established fact and less of a wishful hope. In George Homans’ The Human Group we find a few hypotheses confirmed by such diverse subject matter as an industrial work group, a Polynesian kinship structure, a city street gang, and a small New England community. This approach represents what we see as the main task of the sociological theorist — that is, the discovery of general propositions.

The systematically interrelated propositions that result from this effort are theories. Only at this stage does it make sense to speak of “testing a theory,” “derivation,” and — most important of all — “explanation.” To “test” a theory, we check how well each of its propositions conforms to data and how well several propositions in conjunction with each other account for the outcome of a given situation. If such a “derivation” (or prediction) is successful, we call the outcome “explained”; that is, we claim that observed events conform to known propositions. Thus, Homans is able to explain the friendly feelings between brothers on the island of Tikopia by a reference to his already established proposition that a higher frequency of interaction results in a greater mutual liking.¹⁴

Theories summarize and inspire, not descriptive studies, but verificational studies — studies construed to test [p.29] specific hypotheses. The number of such studies has grown to a gratifying extent in recent years, and every volume of the sociological journals seems to have at least a few articles in which the author formulates specific hypotheses and then reports data that bear on them. If the 1gso’s were particularly hospitable to taxonomies and descriptive studies, the 1960’s seems more hospitable to theories and verificational studies. The growing use of theory in applied sociology is helpful to this development.¹⁵

The following listing of some key words may serve as a summary of the kinds of activities we have discussed:

	I	II
Unit	Definition	Proposition
Interrelated units	Taxonomy	Theory
Application of unit to new subject-matter	Diagnosis	Explanation
Research summarized by or inspired by unit	Descriptive study	Verificational study

We can round out this listing by noting that some contemporary sociologists prefer the term “frame of reference” to our “taxonomy,” and some, perhaps distressed at the corruption of the concept of social theory, prefer the term “model” to our “theory.” The words used make little difference so long as we remember to keep separate the two enterprises depicted in our discussion.

Notes to Chapter 2

1. One physicist has gone so far as to characterize physics as a humanistic discipline. See the postscript to T R Gerholm, Physics and Man, Totowa, N J, The Bedminster Press, 1965

2. Bernard Berelson and Gary Steiner, Human Behavior: An Inventory of Scientific Findings, New York: Harcourt Brace and World, 1964.

3. Terence K. Hopkins, The Exercise of Influence in Small Groups, Totowa, N.J. The Bedminster Press, 1964.

4. Pitirim A. Sorokin, Social and Cultural Dynamics, 4 vols., Totowa, N.J.: The Bedminster Press, 1962. (Originally published 1937-1941.)

5. Robert K. Merton, Social Theory and Social Structure, revised and enlarged edition, New York: The Free Press of Glencoe, 1957, pp. 5-10.

6. A work by Pareto from 1901 contains a rather full blown propositional theory, and my introduction to the English translation of the latter claims that it is the first propositional theory in sociology. See Vilfredo Pareto, The Rise and Fall of Elites, Totowa, N.J.: The Bedminster Press, 1965.

7. See the autobiographical introduction to his collection of essays entitled Sentiments and Activities, New York: The Free Press of Glencoe, 1962, pp. 1-49.

8.Talcott Parsons and Edward A. Shils (eds.), Toward A General Theory of Action, Cambridge: Harvard University Press, 1959.

9. Cf. Robert S. Lynd, Knowledge for What, Princeton: Princeton University Press, 1939.

10. Leonard Broom and Philip Selznick, Sociology, second ed., Evanston: Row, Peterson and Company, 1958, p. 3.

11. Talcott Parsons and Neil J. Smelser, Economy and Society, New York: The Free Press of Glencoe, 1956, p. 9.

12.Talcott Parsons, et al., ‘Values, Motives, and Systems of Action,” in Parsons and Shils, op. cit., pp. 76 ff and Samuel A. Stouffer and Jackson Toby, “Role Conflict and Personality,” ibid., pp. 481-496.

13. Georg Simmel, “The Persistence of Social Groups,” American Journal of Sociology, vol. 3 ( 1898 ), pp. 829-836.

14. George C. Homans, The Human Group, New York: Harcourt Brace & World, 1950, p. 242 ff.

15. Hans L. Zetterberg, Social Theory and Social Practice, Totowa, N.J.: The Bedminster Press, 1962, p. 189.

[p.30]

3. On Definitions in Sociology

Definitions should be used to facilitate communication and argument, and used only to the extent that they make it possible to say something more easily and clearly than would otherwise be possible. Sociologists have spent much energy in developing technical definitions, but to date they have not achieved a consensus about them that is commensurate with their effort. At present there are so many different competing definitions for key sociological notions such as ‘status’ and ‘social role’ that these terms are no more valuable than their counterparts, “position” and “social relation,” in everyday speech.

Yet definitions are indispensable to the sociological enterprise. To the annoyance of many critics, they set sociological writings apart from historical and biographical writing. History describes and often explains the development of a society. In their accounts historians focus particularly on factors determining change. [p.31] In all, the goal of the historian is very similar to that of the sociologist. However, the historian in his accounts uses, by and large, the language of the sources. He would say that a certain political document stresses the individual pursuit of happiness over the obligation to serve the king, and that a certain theological document indicates a shift in interpretation of the Bible’s rules about taking usury so that it is taken to refer to the Jews of ancient time and not to contemporary Christians. The sociologist would use his own terms to describe this; he might say, for example, that the political writer and the theologians were both expressing ‘achievement norms.’ The sociologist thus replaces the language of the sources with his technical terms. Since historians have sifted enormous amounts of source material to describe and explain change in a society, they must implicitly have made many of the distinctions necessary and crucial for the study of society. Macro-sociology translates these distinctions from the language of the sources into a general terminology applicable to any society. In the same way, micro-sociology translates accounts of small groups and the encounters reported in biographical documents into a general vocabulary. Of course, such translations are likely to be abstract approximations of the rich local color of the language of the sources; this is one reason why so many artists and humanists feel irritated at sociology.

[p.32] Discussions of definitions often separate three different but related topics. They may be represented by a triangle:

A term (word or sign) is used to designate certain objects or events; the objects or events are included in a conception; the term means this conception.

Terms

All sciences have their technical terms (and it follows from a sociological law that this technical vocabulary will be called “jargon” by outsiders). Much of the terminology of sociology is also found in the language of educated speech, but some of it is more specialized. Even highly educated speech is sometimes too imprecise or cumbersome for sociological discourse. As everyone knows, the word ‘behave’ might mean any activity, [p.33] but sometimes it means ‘to act with good manners’; the word ‘society’ might mean the largest social system, but it might also mean the ‘upper crust.’ The sociologist needs some words of everyday speech which have such an emotive or affect value in ordinary speech that they must be re-introduced as formal definitions in sociological discourse; ‘culture’ and ‘bureaucracy’ are examples.

Our personal conviction is very much in favor of having a sociological vocabulary that, in the main, is understood by most every educated person, but in which each term has a more precise meaning to the sociologist than to the layman. Unfortunately, the opposite tendency prevails at present. Instead of speaking of ‘equal rights,’ some sociologists have learned to say ‘universalistic standards,’ or, instead of speaking of the ‘familiarity’ that prevails in some social relations, sociologists have learned to say that ‘diffuseness’ characterize the relations, and so forth, almost ad nauseum. This makes sociology unnecessarily incomprehensible to outsiders. We do not deny that ‘diffuseness’ in some contexts is a more precise term than ‘familiarity,’ and that the sociologist may sometimes need this added precision. We rather suggest that the sociologist might still use the word ‘familiarity’ but give this word a restricted and precise meaning. Among sociologists there would then be no loss in clarity and outsiders would [p.34] still have some understanding of sociological discourse. Sociological terminology, in other words, should have incipient similarities with good literary prose.¹

Varieties of Conceptions

In sociology, as in other sciences, conceptions may vary in content and also in formal structure. Let us review some varieties of forms of definitions.

An ‘ostensive’ definition is “any process by which a person is taught to understand a word otherwise than by use of other words.” ² ‘Morale,’ for example, is ostensively defined by pointing to a situation and saying, [p.35] “this is good morale.” It is entirely conceivable that the average spectator at a ball game would be unable to give a verbal definition of the morale of the playing teams. Yet he would not hesitate long to point out instances of good or bad morale. In the same way, an officer might be unable to give a satisfactory definition of morale except by pointing to those of his units that have high morale.

In psychology and sociology, several studies have started with ostensive definitions. In a study of marital adjustment, a sociologist asked a representative sample of an Indiana county to point out to him couples among their acquaintances who were unusually happy and couples who were unusually unhappy. In this way he obtained two ostensively defined categories, one with high marital adjustment and one with low. These two groups he later compared on a variety of criteria to discover factors associated with happiness in marriage. ³

When a term is defined by means of other words we are, of course, dealing with ‘verbal’ definitions. It is of some importance to distinguish between conceptions that assert something that can be more or less truthful, and conceptions that merely express linguistic conventions without assuming anything that can be proved true or false. The rather misleading term ‘real definition’ usually stands for [p.36] a truth-asserting definition: an agreement to use in a given way certain notions which have empirical relationships with each other. The other “which-clause” is the heart of the matter, indicating that these definitions are in the last analysis genuine hypotheses which require testing before they can be accepted. An example is furnished by a current (rather monstrous) definition of ‘social system’. It is suggested that “reduced to the simplest possible terms, then, a social system consist of a plurality of individual actors interacting with each other in a situation which has at least a physical or environmental aspect, actors who are motivated in terms of a tendency to the optimization of gratification and whose relation to their situation to their situation, including each other, is defined and mediated in terms of a system culturally structured and shared symbols”. ⁴ This definition has a built-in proposition about optimization of gratification. However true a proposition like this may sound, we might hesitate as a matter of principle to have it as an integral part of a definition. Actually a good argument can be made that only under special circumstances — albeit common in our society — do people “optimize” their gratification; the universal tendency seems to be that people are motivated to “maintain” their accustomed gratifications.

There are several varieties of these truth-asserting [p.37] definitions. As in the above example, one criterion to be met by any case belonging to the defined category might be that it verifies a given proposition. When electricity is defined as anything that satisfies Maxwell’s equations we have a pure definition of this type. Likewise, we have such a definition when a ‘crowd’ is defined as anything that verifies LeBon’s law of mental unity. It is immediately realized that the value of this type of definition rests upon the validity of of the propositions involved. If the proposition turns out to be false, then the usefulness of the definitions is nil. Since our knowledge of sociological events rarely deserves unquestioned acceptance, we might suggest that definitions in this form should be employed with caution.

A second variety of truth-asserted definitions is found in many ideal types developed by sociologists. Suppose we define a “primary group” as one of small size, high morale, and of early position in the individual’s life cycle. Then we can define a secondary group as one of large size, low morale, and late position in the individual’s life cycle. So far, all is well. But sooner or later the writer who has defined his terms in this way is likely to be found saying that “this group is more primary than that group”. Such an innocent remark implies that he assumes a whole series of hypotheses to be true. He assumes that the three variables defining a primary group are interrelated: in other words, that a) the smaller the group is, the higher its morale, b) the earlier it oc-[p.38]curs in an individual’s life cycle, the smaller it is, and, c) the earlier it occurs in an individual’s life cycle, the higher its morale. None of these hypotheses is particularly well tested or tenable. It should be clear that such assumptions revealed in the usage of many an “ideal type” cannot be accepted in advance of empirical testing and proof.

Often we are drawn into truth-asserting by the use of analogous terms. In social science it has been common to draw analogies from physical science. An example is found in the definition of group ‘cohesiveness’. Cohesiveness has been defined as the sum total or resultant of all forces that keep a member of a group. ⁵ Borrowing from the field of physics of the term ‘force’ might seem innocent enough were it not for the fact that usage of the term implies at least two propositions. In Newton’s days these propositions were grand discoveries, but since then they have become so self-evident that we take them for granted. One of these hypotheses is that whatever the origins of the forces — whether from the moon or from an apple — they have the same consequences. Now, the forces keeping a member in a group may vary greatly. He may stay in the group because of the the prestige the group offers him, because of the friends he has there, because of his need to be punished by an authoritarian leader, and so on. To assume without test-[p.39]ing, that all these forces have the same consequences would indeed be presumptuous. ⁶ The second assumption involved in the use of the term ‘force’ in the definition of cohesiveness, is that whenever several sources of cohesiveness are present their effects are cumulative. This principle has proved to be immensely useful in physics: when several forces act simultaneously, the effect is the same as if they had acted in turn. This hypothesis is much less likely to be successfully maintained in social science than in physical science. The consequences of family cohesiveness deriving from both adequate communication and adequate sexual adjustment during one year of marriage are likely to be very different from the consequences of a family cohesiveness based on one year of adequate sexual adjustment and poor communication, followed by one year of adequate communication but poor sexual adjustment. ⁷ Thus, we see how the person who borrows a term from another science runs the risk of borrowing more than a word: inadvertently he may borrow also some propositions of this science. Clearly, definitions in the form of analogous [p.40] terms deserve an extra careful examination prior to their use in social theory.

In contrast to all the above varieties of truth-asserting definitions, a ‘nominal’ definition is a suggestion to name a phenomenon in a given way without implying anything about the scientific propositions relating to this phenomenon. Thus, nominal definitions are devoid of hypotheses. They cannot be true or false. They can be clumsy or elegant, appropriate or inappropriate, effective or worthless — but not true or false.

A common and loose form of nominal definition is the “enumeration” of ostensively or otherwise defined terms. Military morale is difficult to define for research purposes. It is most readily defined by enumeration of several factors: confidence in officers, confidence in training, confidence in equipment, confidence in rear echelons, identification with the war effort, hatred of the enemy, satisfaction with the task assigned, friendship with fellow soldiers, satisfaction with the military system or rewards, and so forth. ⁸ Definitions by enumeration give the scientist easy directions for concrete empirical references to a concept. There seem to be, however, at least two problems involved in the usage of definitions by enumeration. For one, we have the risk that the factors enumerated are empirically unrelated. As a [p.41] matter of fact and research, the various dimensions of military morale enumerated above show rather low intercorrelations, occasionally negative ones. To pool them and call them ‘morale’ may be convenient for some purposes but misleading for others. Secondly, the factors enumerated may have no conceptual attribute in common. In what way is, for example, ‘identification with the war effort’ conceptually similar to ‘confidence in rear echelons’? Enumerations, thus, easily become conceptual patchworks, more confusing than illuminating.

This last danger is avoided in the conventional ‘Aristotelian’ definition. The phenomena defined by such a definition always have two attributes in common. One attribute — genus proximum — they share with a larger class; another attribute — differentia specifica — is peculiar to the category defined. Many definitions of morale follow this pattern. For example, morale has been defined as “a disposition to act together (genus proximum) toward a goal (differentia specifica).” ⁹ This type of definition is so well known and its virtues are so obvious that many a scholar thinks of this variety as the definition.

Lately, however, our attention has been called to another form of definition which is rapidly gaining ground [p.42] in many scientific fields. Carnap calls it a ‘dispositional concept.’ ¹⁰ It is employed, for example, to define the electrical resistance of a wire; the resistance is a given number of ohms when a given number of volts produce a current of a given number of amperes. The relation ohms = f (volts, amperes) might be said to define resistance dispositionally. In physiology or animal psychology, we could likewise define hunger in terms of an equation involving hours of starvation and number of calories in recent meals. Somewhat simplified, the reasoning follows this pattern: if an object (wire, animal) is subject to A (volts, hours of starvation) and to B (ampere, calories), then we define its X (resistance, hunger) as X = f (A,B).

In discussions of morale — to stay with the previous example — one finds statements to the effect that the test of morale lies in the way a person or a group meets adverse circumstances. For example, “we ascribe morale to a group to the extent that it maintains (its) steadfastness of purpose, maintains its solidarity, its integrity and its will to victory even in the face of adversity.” ¹¹ This idea, that morale is a measure of the extent a group sticks together under adverse circumstances, is a rudimentary dispositional definition of morale. A more precise formulation would be: if a group is placed into a [p.43] situation of A degrees of adversity and loses (or gains) B degrees of steadfastness of purpose, then its morale, X, is X = f (A,B).

One is impressed with the elegance of the dispositional concepts. However, their use in sociology might at times be somewhat restricted. The very formulations of dispositional concepts indicate that they require the scientist to manipulate his object of study. For example, previously we used the phrase “If an object (wire, animal, group) is subject to etc.” Now, our mores are such that we can easily manipulate metal wires and laboratory animals but not so readily human beings, human groups, and social institutions. We are not, and do not want to be, in a position to assign disasters and adversities to individuals and groups and societies in order to measure their morale. Thus the practicality of the above dispositional definition of morale is questionable. This, however, does not mean that other dispositional concepts might not be useful in other contexts of sociology, particularly in micro-sociology. A person’s ‘attitude,’ ‘self-image,’ ‘action repertoire,’ ‘commitment’ might best be conceptualized as dispositional concepts.

The Ordering of Definitions

Max Weber has made the most successful attempt so far to provide a taxonomy for sociology. It appears in [p.44] the opening sections of his posthumously published Wirtschaft und Gesellschaft.¹² There are many reasons for the success of this particular taxonomy: it takes into account numerous crucial distinctions made by historians; it allows a rather easy translation of the language of the sources into a technical vocabulary; it points out those factors we cannot afford to overlook in a routine sociological description; its terms have been used in the formation of some interesting propositions.

Moreover, Weber’s taxonomy is readily seen as an orderly progression from the simplest terms to the most complex ones. At any point we know what any two conceptions have in common and where they differ.¹³ [p.45] Consider, for example, this string of definitions: A ‘social relation’ is the presence of a probability that a social action will occur; thus, teachers and students have a social relation as long as it is likely that they will meet for class or conference. A ‘social order’ consists of social relations guided by a set of prescriptions; thus, education is a social order that includes social relations between teachers, students, administrators, and outsiders. The ‘closure’ of a social relation indicates the extent to which persons are restricted from entering it; thus we normally can say that an advanced seminar is [p.46] more closed than a lecture course for beginners. An ‘organization’ is defined as a rather closed social relation whose order is enforced by a common leader or staff; thus a particular college is an organization separated from other colleges because it has its own president, deans, departmental chairmen, and so forth. Organizations may be ‘compulsive,’ that is, impose their order on anyone with given characteristics, such as an elementary school which all children of a certain age have to attend. Or, an organization may be ‘voluntary,’ that is, impose its order only on those who have given it its personal adherance, such as a college which we are free to leave at any time. A ‘state’ is a compulsive organization that imposes its order on anyone living in a given territory and that may legitimately use violence in doing so. Thus Weber builds from the simplest (action) to the most complex (state). Herein lies much of his appeal as a taxonomist.

Let us explore some formal aspects of such a process of taxonomy construction. In all schemes of definitions we shall find words belonging to the vocabulary of logic and mathematics. Words such as “and,” “or,” “not,” “imply,” “equal to” belong here. They are called ‘logical terms.’ ¹⁴ Unlike logical words, they are not shared by all sciences but are specific to one or a few. Samples are ‘entrophy’ in thermodynamics, ‘reinforcement’ in [p.47] learning theory, ‘homeostasis’ in physiology, ‘social role’ in sociology. For example, in a list of kinship terms, ‘father,’ ‘mother,’ ‘son,’ ‘daughter,’ ‘brother,’ ‘sister,’ ‘uncle,’ ‘aunt’ would be extralogical terms. On the other hand, the words ‘any,’ ‘and,’ ‘or,’ ‘of,’ ‘is called’ are the logical words in the definition: “Any brother of father and/or mother is called uncle.”

In an ideal theory it should, furthermore, be possible to find a small group of extralogical words, the ‘primitive terms,’ which in different combinations with each other and with logical terms can define all other extralogical terms of the theory, the ‘derived terms.’ Any derived term, in short, is obtained by combinations of the primitive terms and the logical words. There have been very few efforts to systematically separate primitive and derived terms in sociology. Talcott Parsons, in his analysis of the works by Marshall, Pareto, Durkheim, and Weber, suggested that their theories can be, viewed in terms of a handful of primitives, the so-called means-end schema. Although no formal demonstration of this has been made and the author himself eventually abandoned the schema,¹⁵ there can be no doubt about its logic. A small set of building blocks, the primitive [p.48] terms, can be combined into new units, the derived terms, to create a sociological taxonomy.

Typologies are constructed along similar lines but without the requirement that the building blocks be primitive terms. The complex terms may be obtained also by making use of derived terms previously defined. Consider, for example the typology used in Merton’s discussion of anomie.¹⁶ Merton sees the normative order of society under two headings. First, there are norms stating goals (e.g., to be successful, to rise from rags to riches); second, there are norms setting forth the approved means to reach these goals (e.g., to attend college, to take financial risks). Deviance from these norms may now occur in several ways and we obtain the following typology:

	Acceptance
Goals	Yes	Yes	No	No
Means	Yes	No	Yes	No
	Conformity (e.g., Babbit)	Innovation (e.g., Robber Baron)	Ritualism (e.g., Bureau- crat)	Retreatism (e.g., Hobo)

Here ‘goals,’ ‘means,’ and ‘acceptance’ correspond to lower-order terms such as the primitive ones, and ‘conformity,’ ‘innovation,’ ‘ritualism,’ and ‘retreatism’ to complex derived terms. As we see from this illustration, there is no logical difference between a system of defini-[p.49]tions and a system of classification. That is why we can call both by the same word: taxonomy.

Minimum Terms: The Uniqueness of Our Subject Matter

Some primitive terms of one science may also occur in other sciences. Geography provides a simple example. Among its primitive terms are “Greenwich” and “the North Pole.” Bertrand Russell has indicated that remaining primitive terms of geography can be drawn from chemistry, physics, or geometry:

The relation of “west of” is not really necessary, for a parallel of latitude is a circle on the earth’s surface in a plane perpendicular to a diameter passing through the North Pole. The remainder of the words used in, physical geography, such as “land,” and “water,” “mountain,” and “plain,” can now be defined in terms of chemistry, physics, or geometry. Thus it would seem that it is the two words “Greenwich” and “north pole” that are needed in order to make geography a science concerning the surface of the earth, and not some other spheroid. It is owing to the presence of these two words (or two others serving the same purpose) that geography is able to relate the other discoveries of travelers. ^l7

[p.50] Thus it appears that certain primitive terms are unique to a given academic discipline while others are shared with other academic disciplines. Let us call those primitive terms that are unique to a given theory its ‘minimum terms’ and those that are shared with other theories its ‘borrowed terms.’ ¹⁸

The distinction between minimum and borrowed terms helps us clarify an old issue. A number of sociologists from the time of Comte have tried to answer questions such as “What is the proper subject matter of sociology?” “What subject matter, if any, should be treated in sociology but by no other science?” Much of this discussion has been futile. However, our classification of primitive terms provides a new and clearer criterion for questions relating to the limits and uniqueness of the subject matter of any theory.

It should be plain that a theory does not have any subject matter that can be called exclusively its own if all its primitive terms are borrowed terms from other sciences. If so, anything that it talks about can be exhaustively presented within the frameworks of theories from other sciences. On the other hand, if we have one [p.51] or more terms that can properly be called minimum terms, we also have a unique subject matter. Any phenomenon, then, that has to be defined in these minimum terms is our exclusive subject matter. Therefore, instead of asking the old question, “What, if any, is the proper subject matter of sociology?” we instead ask, “What, if any, are the minimum terms of a sociological theory?”

Most sociologists have held that sociology has a minimum vocabulary of its own and that the terminology of physics and biology is not relevant in sociological discourse. In this vein, MacIver writes:

There is an 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 fear to its bodily concomitants we merely substitute the concomitants for the reality experienced as fear. ¹⁹

A few sociologists have objected to this. For example, Lundberg:

The principle of parsimony requires that we seek to bring into the same framework the explanation of all flying objects…. From the latter point of view a paper flying before the wind is interpreted as the be-[p.52]havior of an object of specified characteristics reacting to a stimulus of specified characteristics within a specified field of force. Within this framework we describe the man and the crowd, the paper and the wind. ²⁰

The mainstream of sociological thinking on this issue sides with MacIver. The primitive terms we need in social theory do not consist exclusively of terms found in physical and biological sciences. We find it plainly impossible to say much of sociological significance in a vocabulary based on the terms of physics and biology.

A Consideration in Selecting Primitive Terms for Sociology

Pareto and Weber, as well as most contemporary social theorists, have assumed that the building blocks of sociological definitions are terms that denote human beings and their actions. The rationale for this choice is found in a suggestive analogy between the position of ‘actions’ in all modern sociological theorizing, and the position of ‘primitive terms’ in any taxonomy. The sociologists say that all social events consist of combinations of human beings and their actions. The logicians say that all terms of a theory can ultimately be defined by combinations [p.53] of primitive terms. It therefore seems useful — at least as a first approximation — to assume that the primitive terms of sociology should be words that denote human agents and their actions.

The dangers involved in departing from the rule that, in the last analysis, we can only use terms representing combinations of human actions are well illustrated in the history of social thought. As a dramatic case in point consider the term volonté générale in Rousseau’s Contrat Social. Rousseau uses two terms to denote popular will: volonté de tous, which is the expressed majority opinion of the citizens, and volonté générale, which is the society’s will apart from what any citizen might say or do. The latter term, then, is not readily seen as a combination of actions. The questionable innocence of this distinction could be seen already in 1797 in the first popular election in the Cisalpine Republic, a small country on the Italian peninsula, newly constituted as a democracy by the victorious armies of the French Revolution who had deprived the local clergy and nobility of their traditional power. However, in the election, contrary to the expectations of the revolutionaries, the popular majority supported the old reactionaries, the nobility, and the clergy. Confused and bewildered at this turn of events, the revolutionaries wrote to Paris for directives. In response, a letter of April 7, 1797, ordered General Bonaparte to take over the functions of the legislature and void the election. The motivation was that the peo-[p.54]ple had been misled: their “real” desires were with the revolutionaries.^2l Volonté générale was to prevail over volonté de tous. Thus democracy was suppressed in democracy’s name, a procedure commonly experienced in modern dictatorships, and a notion of the “real” desires of a people apart from what they express as their desires was used to legitimatize the suppression. However, such a notion of “real” desires is mysticism. Volonté générale slides into metaphysics; we do not know any constellation of observables that define it.

The same holds for any sociological conception which does not represent a combination of observable human beings and their actions. Thus, as social theorists, we are well advised to select primitive terms that stand for actors and types of actions. Since these primitives then are used as building blocks, which in various combinations furnish more complex terms, we are assured that even very complex ideas — e.g., property, institution, feudalism, or class — will remain on this side of metaphysics.

The Generation of Derived Terms

Psychological definitions are constructed by operations combining primitives referring to actions of one actor. Sociological definitions are constructed by operations [p.55] combining primitives of several actors. Let us assume that our primitives are verbal actions such as ‘descriptions,’ ‘evaluations,’ and ‘prescriptions.’ As a first operation consider any procedure used to find a ‘central tendency.’ Central tendencies of descriptions, evaluations, and prescriptions within one individual thus became defined as his ‘cognitions,’ ‘attitudes,’ and ‘expectations.’ Central tendencies of the same action types among an aggregate of individuals become their ‘social beliefs,’ ‘social valuations,’ and ‘social norms.’ Any other operation can be used to manipulate the primitives; the outcome is other derived terms. For example, if we select the operation of ‘dispersion’ of the action types within one individual we get a definition of his ‘rigidity’; if ‘dispersion’ is applied to actions in the aggregate of individuals we obtain a definition of their ‘consensus.’ We might also apply an operation finding ‘proportions’ to the primitives. An individual with a high proportion of prescriptions among his actions might be defined as ‘dominant.’ As the economic geographer divides the earth into production areas, so the sociologist can divide society into realms according to the proportion of actions of a certain type. The realm of society with a high proportion of prescriptions (laws, ordinances, executive orders, platforms, decisions, programs, commands, etc.) might then be defined as its ‘body politic.’

Of particular importance to sociologists are the operations we use to characterize social units or collectivities.[p.56] Lazarsfeld and Menzel ²² have delineated some types, the most important of which are:

Analytical properties, which are obtained by performing some operation on each subunit of the unit to be characterized. The average income in a city would be an example of an analytical property of a city.
Structural properties, which are obtained by performing some operation on relations between subunits of the unit to be characterized. The ‘star-dominance’ of a social group as indicated by the concentration of sociometric choices upon one or a few individuals would be an example of a structural property of a group.
Relational properties, which are obtained by performing some operation on relations between the unit to be characterized and its neighboring units. Any definition of social isolation of an individual or a group would be a relational property.
Contextural properties, which are obtained by performing some operation on the relation between the unit to be characterized and its superunit. Classification of election districts into rural or urban would probably be done by a contextural property of the district.

[p.57]The development of conceptions of this kind make it possible to have a ‘holistic’ approach to sociology without loss of precision in reasoning and research.

Descriptive Schemas

The procedure of breaking up a set of definitions into its primitive components and to identify the operations or formation rules used in arriving at any derived term helps us in getting a precise understanding of what is covered by each term.

However, for research purposes one often has to organize definitions differently from the logical schema of primitive and derived terms. The researcher needs terms so arranged as to guide him to the phenomena to which he shall pay attention and presented in the order in which he shall pay attention to them. He needs a set of terms in the form of a check list for the observation he is supposed to record. Terms organized for this purpose constitute a schema for routine description. Such “shopping-lists” tell him what he must know to have a standard sociological account of a person, a social role, a group, and institutional realm, a society. Some contemporary theorists have taken great pains in providing such descriptive schemas. Thus, Parsons has formulated a much used schema for the description of social roles in [p.58] terms of five ‘pattern variables,’ Merton has compiled a list of 26 attributes of social groups, and Gross has provided a descriptive schema for a contextual account of roles. ²³ Other descriptive schemas have their origin in research practice. In no case can we claim complete agreement among sociologists as to what constitutes the best descriptive schema for any given topic; however, there is usually enough consensus to warrant criticism when a sociologist has omitted a common item.

As an example of a fairly well-standardized descriptive schema let us cite the schema sociological pollsters use in describing a person. They usually call it “face sheet variables” and the social theorist identifies most of them as status variables. The descriptive schema used is generally a selection from the following lists:

I Past Contextual Variables

1. Place of birth (native or foreign); sometimes also parents’ place of birth

2. Type and size of community in which most of childhood was spent (rural, small town, city, metropolis)

II Present Contextual Variables

1. Type and size of community in which the respondent lives

2. Geographical region of country

III Contemporary Statuses: Ascribed

1. Sex [p.59]

2. Age

3. Ethnic background

4. Religious affiliation

IV Contemporary Statuses: Achieved

1. Occupation; sometimes also husband’s or wife’s occupation

A. classified according to occupational rank (upper, middle, lower )

B. classified according to work situation (salaried, self-employed)

C. classified according to institutional realm (business or industry, civil service or politics, education or science, religion, art, welfare institutions, private household)

2. Family statuses

A. Marital (single, married, widowed, divorced)

B. Parental (no children, children living at home, children living away from home)

3. Memberships in voluntary associations (including business associations and unions); political party affiliation or preference

V Past Statuses

1. Father’s occupation

2. Type of schools attended

3. Military service

4. Past full-time occupations

VI Stratification

1. Riches

A. Family income

B. Family property

a. Residence (owns, rents, boards) [p.60]

b. Consumer goods (e.g., auto, TV)

2. Knowledge or competence

A. Years of schooling; sometimes also husband’s or wife s years of schooling and children’s education

3. Power

A. Executive position

B. Political office

C. Office involuntary associations

This list is notable not only for what it includes — primarily the dominant statuses occupied by a person in the various institutional realms of society — but also for what it leaves out, that is, reference to any specific action (e.g., opinion and attitude), physical attributes (e.g., height and weight), biological attributes (e.g., diseases and deformities), and psychological attributes (e.g., authoritarianism and intelligence). The list rather serves as a framework for the account of the distribution of such attributes throughout society. It should be noted, however, that no study we know has utilized all the categories of the above list to describe the persons studied; selections (and additions) are made depending on the special interest of the researcher. But all categories listed have appeared in several of the best sociological survey studies published in the past two decades.

The methodology and technique used in developing successful descriptive schemas has not yet been made explicit. A statistical technique known as factor analysis [p.61] can be used to develop descriptive schemas for quantifiable variables; so far it has proved more successful in psychology than in sociology. Some sociological schemas have grown out of the analysis of classical works. Thus Parson’s pattern variable schema grew out of a reanalysis of Töennie’s discussion of Gemenschaft und Gesellschaft. Merton’s schema of group properties was inspired by a less systematic discourse by Simmel. Gross’ language of role analysis was apparently developed to fit a specific research topic and grew out of a specific research situation. The pollsters’ face sheet schema emerged through trial and error and tended to include all factors that consistently seemed to differentiate opinions and attitudes.

In a seminar jointly conducted by Professor Sigmund Diamond and the author, a descriptive schema for routine accounts of total societes was developed through the following steps:

A number of books by historians, sociologists, and anthropologists that contain attempts to characterize total societies (Western and non-Western) are read. Lists are made comprising the main factors to which each one of these authors paid attention.
The lists are compared for similarities and differences and all reasonably common items are recorded.
The common items are reworded (often rephrased in more abstract sociological terminology) and organized as a descriptive schema. [p.62]
The resulting master schema is used to describe several societies and modified in the process.

There is a considerable gap at present between systems of definitions developed by social theorists and descriptive schemas used by researchers. This is unfortunate. The theorists should be encouraged to show how their terms can be used in descriptive schemas, and the researchers should be encouraged to contemplate how the descriptive schemas they employ are related to the terms developed by the theorists.

Notes to Chapter 3

1. Sometimes one feels tempted to argue that there is an inherent clumsiness in any general language dealing with sociological phenomena. Listen to the awkwardness of a great poet and stylist in his attempt to make an elementary sociological distinction: “Each individual, who composes the vast multitude which we have been contemplating, has a peculiar frame of mind, which, whilst the features of the great mass of his actions remain uniform, impresses the minuter alineaments with its peculiar hues. Thus, whilst his life, as a whole, is like the lives of other men, in detail it is most unlike; and the more subdivided his actions become, that is, the more they enter into that class which have a vital influence on the happiness of others and his own, so much more are they distinct from those of other men. … This is the difference between social and individual man. Not that this distinction is to be considered definite, or characteristic of one human being as compared with another, it denotes rather two classes of agency, common in a degree of every human being.” (Percy Bysshe Shelley, “Speculations on Morals” in Works in Verse and Prose, Reeves & Turner, London, vol. 6, pp. 317-318.)

2. Bertrand Russell, Human Knowledge — Its Scope and Limits, New York: Simon and Schuster, 1948, p. 63. (Italics added here.)

3. Harvey J. Locke, Predicting Adjustment in Marriage, New York: Holt and Company, 1951.

4. Talcott Parsons, The Social System, New York: The Free Press of Glencoe, 1951, pp. 5-6.

5. Leon Festinger, et al, Social Pressures in Informal Groups, New York, Harper and Bros, 1950, p 164.

6. Kurt Back has shown that cohesiveness based on friendship, prestige, and monetary gratifications resulted in roughly the same changes on certain selected variables such as communicaton and conformity. See his “Influence Through Social Communications”, Journal of Abnormal and Social Psychology, vol. 46, 1950, pp. 9-23.

7. In an unpublished papper, “Cohesiveness as a Unitary Concept — Some Further Evidence”, the author refuted the hypotheses about such additivity of consequences by using the same procedure Back (op. cit.) used to prove the similarity of consequences of different sources of morale.

8. These dimensions are taken from the studies of military morale in Samuel Stouffer et al., The American Soldier, Vols. I and II, Princeton: Princeton University Press, 1949.

9. Herbert Blumer, “Morale” in W F Ogburn, ed., American Society in Wartime, Chicago: University of Chicago Press, 1943, p. 211. (The quotation occurs in italics in the original.)

10. Rudolp Carnap, “Testability and Meaning. Part IV”, Philosophy of Science, vol. 4, 1937, pp. 1-40.

11. Louis Wirth, “Morale and Minority Groups”, American Journal of Sociology, vol. 45, 1941, p. 425.

12. 4th edition, edited by J Winkelmann, Tübingen, Germany: J. C. B. Mohr, 1956. English translation, edited by Guenther Roth, Economy and Society, Totowa, N J: The Bedminster Press, 1966.

13. This is not always the case in sociology. Consider, for example, some of Cuber’s definitions in his Sociology: A Synopsis of Principles. The following list samples some major definitions as they appear in the third edition (1955):

Culture:	“the continually changing patterns of learned behavior and the products of learned behavior (including attitudes, values, knowledge, and material objects) which are shared by and transmitted among the members of society” (p. 56).
Mores:	“the must behaviors, the basic and important patterns of ideas and acts of a people” (p. 116).
Folkways:	“somewhat less compulsive than the mores of the same society” (p. 116).
Ideal patterns:	“models of exemplary conduct which are held up as standards of perfection” (p. 122).
Real patterns:	“what the people actually do, irrespective of what they are ideally supposed to do, or what they themselves believe they should do” (p. 122).
Social norm:	“the accepted or required behavior for a person in a particular situation” (p. 208).
Social role:	“the culturally defined pattern of behavior expected or required of a person in specific social positions” (p. 295).

Here we have illustrated many ways of making loose definitions, and one can only pity the college students who had to memorize them. Two sources of lack of precision are particularly obvious. First, it appears that many terms are defined by words that in turn might require further definitions. For example, what is meant by “position” in the definition of social role? Second, there is no way of knowing how different terms relate to one another. No one can read the definition of mores, folkways, ideal patterns, social norm, and social role without realizing that they have a component in common. Yet this common component (of prescription) is never identified and the reader is quite uneasy about just where these terms overlap and where they differ. Both these remarks boil down to the observation that most of the time that Cuber defines a new term he ignores all his previous definitions and starts again from scratch.

14. Carl G Hempel, Fundamentals of Concept Formation in Empirical Science, University of Chicago Press: Chicago, 1952, p. 15.

15. Talcott Parsons, The Structure of Social Action, New York: McGraw-Hill, 1937, p. 44. For his later theorizing Parsons has expanded his list of primitives to comprise a score of terms, the so-called ‘components of the frame of reference of the theory of action’. See Talcott Parsons and Edward Shils, Toward A General Theory of Action, Cambridge: Harvard University Press, 1952, pp. 56-64.

16. Robert K. Merton, Social Theory and Social Structure, 2nd ed, New York: The Free Press of Glencoe, 1957, ch. 4.

17. Bertrand Russell, op. cit., pp. 243-244.

18. In summary, then, we can organize the terms of any scientific theory into the following categories:

I Logical terms

II Extralogical terms

1. Primitive terms

a. minimum terms

b. borrowed terms

2. Derived terms

19. Robert M. MacIver, Social Causation, Boston: Ginn & Co, 1942, pp. 476-477.

20. George A. Lundberg, Foundations of Sociology, New York: Macmillan, 1939, pp. 13, 14.

21. Guglielmo Ferrero, The Gamble: Bonaparte in Italy, 1796-1797, London: Bell, 1939, ch. 13.

22. Paul F Lazarsfeld and Herbert Menzel, “On the Relation between Individual and Collective Properties” in A Etzioni (ed), Complex Organizations: A Sociological Reader, New York: Holt, 1961.

23. Parsons and Shils, op. cit., pp. 76-88; Merton, op. cit., 310-326; Neal Gross et al, Explorations in Role Analysis, New York: Wiley, 1958, ch. 4.

[p.63]

4. On Propositions in Sociology

Confronted with a proposition, we tend to ask, “What does it mean?” and “Is it true?” This chapter will deal with the former question, which we will subdivide into three separate questions: “What are the determinants and results entering the proposition?” “What linkage is presumed between them?” “What is the informative value of the proposition?”

The propositions we will use as illustrations in this book are all phrased in ordinary language. In recent years it has been increasingly common to state propositions in some artificial language, such as mathematics or symbolic logic. The use of mathematics is not only an escape from the well-attested inability of many sociologists to write an attractive, literary prose. Mathematics adds precision to theory construction. We have much more explicit rules for manipulating mathematical expressions than we have for manipulating ordinary sentences.

The appropriate degree of precision for a theory must [p.64] be chosen with an eye to the quality of the data submitted in its support and with another eye to the possible use of the theory in research and practice. Precision per se is a dubious and boring virtue. It will be some time before we have accumulated enough experience to assess the advantages and disadvantages of phrasing our theories in artificial languages. The present generation of theorists seems to be able to proceed without mathematics; the next generation will, in all likelihood, rely much more on it. The most fruitful compromise at present seems to be a very disciplined ordinary language in theoretical sociology, occasionally supported by mathematical expressions and graphs.

Variates: Determinants and Results

Propositions relate variates to each other. We say, “The more knowledge a man has, the higher his prestige,” and have thus uttered a proposition that relates the two variates ‘knowledge’ and ‘prestige’ to each other. When we know or assume the direction in which the variates influence each other, we can designate one as a determinant (cause or independent variable) and the other as a result (effect or dependent variable). In our example, ‘knowledge’ is the determinant and ‘prestige’ the result; we get prestige from knowledge but no knowledge from prestige.

Note that we need, at the very minimum, two variates [p.65] to have one proposition. Many propositions contain more than two; in sociology we have to consider it as normal that events have multiple determinants and/or multiple consequences. It is the predominance of such multivariate propositions that justify us in saying that ours is a complex field. Given this complexity, one gets understandably impatient with the number of statements in sociology including only one variate — e.g., “x varies” — and also slightly suspicious of the unqualified propositions with two variates — e.g., “if x varies, then y varies.” I will not dignify one-variate statements by calling them propositions, and we will not deal with them here; they belong to descriptive sociology. (If the editors of the sociology journals made it a rule of thumb never to print any article presuming to give a theoretical discussion in which the conclusion is a one-variate statement, they would add scientific maturity to their product. )

Propositions with two variates are acceptable as intermediary steps in theory construction even if they do not tell the whole story. Once formulated they lend themselves to amendments. For example, Homans’ two-variate proposition, “if the frequency of interaction between two or more persons increases, the degree of their liking of one another will increase and vice versa” ¹ was a good start, even though later theorizing makes it plain [p.66] that two additional variates have to be introduced — viz. cost of avoiding interaction, and availability of alternative rewards. A woman suffering punishment in her interaction with her husband may not break off the relation because of the high cost of divorce; but her liking for her husband does not increase as their interaction goes on. And if she finds her husband’s behavior at least in part punishing but does not have an alternative man around the corner who would be more rewarding, she may continue to interact with her husband without liking him more.

Malewski adds these factors into a new formulation of the proposition: “If the costs of avoiding interaction are low, and if there are available alternative sources of reward, the more frequent the interaction, the greater the mutual liking.” ² This is a multivariate proposition, and more adequate than the original one with only two variates. However, it is not likely that one would have arrived at the multivariate one without having the two-variate one as a convenient intermediary step. Two-variate propositions can thus be justified on tactical grounds. From a pedagogical point of view, two-variate propositions are also useful, since they are so easy to grasp. Since our purpose in this book is pedagogical, it [p.67] is perhaps understandable that many of our examples will contain only two variates.

The first requirement of a proposition is that the determinants and the results be precisely defined. A celebrated monograph that has been subject to much misunderstanding because it fails to state precisely its key proposition is Max Weber’s The Protestant Ethic and the Spirit of Capitalism.³ Its proposition is hinted in its very title: the Protestant ethic is the determinant and the spirit of capitalism is the result. There are, however, at least four different ways of specifying the determinant and the result in this proposition. If the terms in italics stand for the variates that may be related, we have these possibilities:

The Protestant Ethic and the Spirit of Capitalism
The Protestant Ethic and the Spirit of Capitalism
The Protestant Ethic and the Spirit of Capitalism
The Protestant Ethic and the Spirit of Capitalism

If we emphasize the first mode, we study the frequency with which persons who are Protestants become capitalists and compare it with the frequency with which persons who are Catholics become capitalists. If the second [p.68] interpretation is made, we look for ethical precepts in Protestantism which are more conducive to the emergence of capitalism than the corresponding precepts of Catholic ethics — for example, the more lenient attitude of the Protestants toward usury. If the third interpretation is made, we look for a different spirit of entrepreneurship and hard work among Protestants compared with Catholics. If the fourth interpretation is made, we presume that some ethical precepts in Protestantism, such as its invisible stratification in religion and ethics (i.e., the concept of predestination and the denial of good works as a measure of one’s ethical worth ), lead to a particular spirit which is manifested as concern with one’s rank in the more visible aspects of stratification — that is, with achievements in the worlds of money, power, and science. All four ways of interpreting the thesis are in varying degrees present in Weber. Also, Weber’s critics often touch upon some of these ways of interpreting the thesis in a haphazard way. Thus, claims by his critics that Weber has been proved right or proved wrong are usually restricted to one or two of these possibilities. Much confusion could have been avoided if the determinant and the result of the proposition had been more clearly specified. [p.69]

The Varieties of Linkage Between Determinants and Results

In Weber’s proposition, as in most sociology, the relation between determinant and result is left vague: it is said that the Protestant ethic “led to” or “contributed to” or was “functional” for the spirit of capitalism. In stricter theorizing we must spell out in greater detail what kind of relation is assumed in a given proposition. The topic of causal linkages is complicated, and it would carry us too far from the everyday problems of the working theorist in sociology to give it systematic presentation. Instead we will present a mere listing of varieties of causal linkage encountered in sociology and illustrate them with examples. Several of these varieties would have to be treated on different levels in a more systematic presentation.

A relation may be reversible (if X, then Y; and if Y, then X) or irreversible (if X, then Y; but if Y, then no conclusion about X). Reversible propositions are not unusual in sociology. A well-known one is Homans’ previously mentioned law about frequency of interaction and liking: the higher the frequency of interaction between two or more persons, the greater their liking for one another, and conversely, the greater the liking for one another among two or more persons, the higher the [p.70] frequency of their interaction. ⁴ When we say that a proposition is reversible, we assume, in fact, that it contains two separate ideas requiring two separate tests.

Second, a relation may be deterministic (if X, then always Y) or stochastic (if X, then probably Y). Deterministic relations seem very rare in sociology. A possible example might be Simmel’s proposition: if there is an increase in the number of members in a completely unstructured group, then there is always an increase in the anonymity of the actions of the group.⁵ Stochastic relations are more common, and they range from the quite strong to the highly attenuated. Consider, for instance, this hypothetical statement: if a person must choose between conforming to a norm and abandoning a high rank, he is likely to keep the high rank. There are, of course, many men who would choose to obey the norm in such a dilemma, but in a large aggregate of men a majority is predicted to deviate from the norm and keep the rank.

Third, the relation may be a sequential one (if X, then later Y) or a coextensive one (if X, then also Y). An illustration of the former might be Lazarsfeld’s cross-pressure proposition: if voters are subject to contradictory influences in their primary groups during an election campaign, then they are likely to delay their voting [p.71] decision. ⁶ A coextensive relation is illustrated by the statement: the higher the rate of social mobility, the less the extent to which the lower classes accept militant class ideology. ⁷ No assumption is made here that mobility occurred before or after the spread of a working-class ideology. (It might be noted in passing that reversible coextensive relations are often called “functional” ones. In sociology, however, “functionalism” has so many special meanings that I will not use the word “functional” in this context.)

Fourth, a relation may be sufficient ( if X, then Y, regardless of anything else) or contingent (if X, then Y, but only if Z). Sufficient propositions are rare in sociology, and contingent propositions are the rule. This is one way in which multivariate propositions dominate over two variate propositions. For instance, all propositions about interpersonal influence assume some kind of interpersonal contact (e.g., social visibility) to be valid.

Fifth, a relation may be necessary (if X, and only if X, then, Y) or substitutable (if X, then Y; but if Z, then also Y). Halévy appears to assume that the presence of groupings located between the state and the family, such as non-conformist sects, was a necessary factor in the process that spared England from the bourgeois revolutions that occurred in France and other parts of [p.72] Europe. ⁸ Propositions with substitutable determinants are otherwise very common in sociology, as seen in the wide usage of the phrase “functional equivalence.” For example, in work groups paid on a piece-rate basis, we find that norms prescribing restriction of output and norms prescribing secrecy about earnings and production records may be functional equivalents in one specific sense: both are likely to reduce interpersonal tensions resulting from invidious comparisons of work achievements.

Any proposition may now be characterized according to the above check list of attributes. For example, Max Weber’s thesis about a relation between the Protestant ethic and the spirit of capitalism is best interpreted as an irreversible, stochastic, sequential, contingent, and substitutable proposition.

The above five attributes of a causal relation are well known in any science. Sociologists, however, might take special pains in identifying an additional type of relation. It is actually a combination of a reversible, sequential, and contingent attributes, but it is so uniquely applicable to sociological subject matter that it deserves a special name and a separate discussion. It is interdependent relation.

Let x and y be small increments in variables x [p.73] and y, respectively. An interdependent relation is present when the following conditions are met:

If x changes from x₁ to x₂, and x₂ = x_l + x, then and only then, y changes from y₁ to y₁ + y; further, when y changes from y₁ to y₂ and y₂= y₁ + y, then and then only, x changes from x₂ to x₂ + x, etc.

Thus, in an interdependent relation, a small increment in one variable results in a small increment in a second variable; then, the increment in the second variable makes possible a further increment in the first variable, which in turn affects the second one, and so this process goes on until no more increments are possible. Note, however, that an immediate large change in one variable will not bring about a large change in the other variable. The only way a large change is brought about in an interdependent relation is through a series of interacting small changes. It is as if the two variables were flirting with each other; an almost imperceptible hint gives the first and necessary encouragement for a braver hint, and so on. A big initial hint, however, would have no effect.

The operations of an interdependent relation are found in many social processes. It is said, for example, that voluntary associations develop with the urbanization and industrialization of a society. Migration to cities creates tensions for the former rural resident, who re-[p.74]solves them by joining a voluntary association; the latter makes him a more involved urbanite and industrial employee, which in turn generates further tensions and further involvement in voluntary associations. Thus, membership in voluntary associations and participation in urban and industrial life stand in a kind of piecemeal give-and-take relation that we call an interdependent relation.

The types of causal linkage should be kept in mind in all manipulations of propositions. So long as all propositions used in our theorizing are of the same type, there are few dangers involved. However, when they are of different varieties, pitfalls appear, and one must proceed with caution.

Functional Propositions

No review of varieties of sociological propositions would be complete without a mention of the common practice of making a functional assessment of propositions. Many times sociologists use the word “function” to mean only “result,” and the term “functional prerequisite” to mean only “determinant.” However, in stricter thinking, “function” has a more special meaning:

Functions are those observed consequences which make for the adaptation or adjustment of a given sys-[p.75]tem; and dysfunctions, those observed consequences which lessen the adaptation of the system. ⁹

In an analysis of the functionalist position, Galtung points out that the crucial words in this statement by Merton are “adaptation” and “adjustment”:

But “adaptation” and “adjustment” to what? The answer “to S [the system] as it is, to status quo” can be discarded at once — this would make all consequences implying social change dysfunctional by definition. The answer “to a social change in or of S” can likewise be discarded, as we do not believe today that consequences implying social change are necessarily beneficial for the system. ¹⁰

Galtung proceeds to give the answer in terms of some shared value within a subsystem. By voluntarily keeping their output roughly at the same modest level, workers paid according to piece rates do not envy each other’s take-home pay, and they help preserve well-paying piece-rate contracts. Expressed as a functional proposition this would read: The functions of informal norms prescribing restriction of output are to reduce invidious comparisons of wages, and to keep stable, high wages. Both these functions can be justified by values held by the majority of the workers, but not by the values held by the majority of employers; the latter want maximum [p.76] output for the lowest possible wage costs. What is functional and dysfunctional thus depends on the values in the social system or subsystem taken as a point of departure.

Before accepting this radical solution, it may be appropriate to consider the use of functionalism in other fields. Professor Nagel sums up a review of functionalist formulations in various sciences in the following way:

Accordingly, functional statements are regarded as appropriate in connection with systems possessing self-maintaining mechanisms for certain of their traits, but seem pointless and even misleading when used with reference to systems lacking such self-regulatory devices.¹¹

This makes sociological functionalism dependent on the discovery of self-regulatory mechanisms that keep a given variable at a certain level or “goal state,” in Nagel’s terminology. Those factors are “functional” which keep a variable at the goal state, while those factors that tend to move it from the goal are “dysfunctional.” Biology can demonstrate many such goal states, which are maintained within very narrow limits, for example, the sugar level in the blood. Whether sociology will discover some is an open question.

Two less rigid aspects of functionalism are illustrated [p.77] in the works of Talcott Parsons. He hypothesizes that all social systems (and all subsystems, etc.) have to solve four problems: (1) “adaptation,” (2) “goal-attainment,” (3) “integration,” and (4) “latent-pattern maintenance and tension management.” These are the “functional imperatives” of any system of action.¹² Imagine that we can measure how well each is resolved in a social system by reading four master gauges, and that all subsystems have similar gauges; and that we also can rate the subsystems according to the input and output they make of these four quantities in the larger system. Each gauge will now have at its lower end a red danger zone; if the dial on any one of the four master gauges falls in this danger zone, Parsons would predict that the whole system would perish. The same would be the case with a subsystem if any of its four gauges fell below their critical points; the larger system would still remain, but only if the disappearance of the input contributed by the subsystem does not reduce the quantities necessary to maintaining the larger system beyond the critical point. Thus we see that the functional formulation here is applied, not as in Nagel’s case involving the maintenance of a narrow range or goal state of a variable, but only in the maintenance of minimum values on variables. [p.78]

The second and more important use of functional formulations by Parsons concerns the operation and change of on-going systems. He assumes that the reading of the dials above the danger zone would show certain interdependencies. For example, if an advance is made in adaptation — caused by added resources channeled to this area by internal or external events — modifications would have to be made also in the other problem areas. Parsons, being a taxonomist, has not written a set of specific propositions about these interdependencies, but it is clear that he expects the quantities represented by the dials to form a moving equilibrium. Thus propositions in functional language are used here in a way that has long standing in the social sciences: that is, to specify an equilibrium theory. Whether they have an advantage over conventional propositions or equations about determinants and results, which normally are used to write equilibrium theories, remains to be seen. It should also be noted that Parsons’ type of functionalism, while clearly aimed at inclusive theory, uses an approach that at best will produce a partial theory. Assuming that sociologists would agree on what are the imperative problems of any social system — itself a rather remote possibility — no assurance can be given that all sociological propositions are relevant to these problems. Thus there would be some sociological knowledge that is not included in the functionalist formulations.

At present, functionalist formulations enjoy wide cur-[p.79]rency in sociology. I have some misgivings as to their usefulness to the sociological theorist, since they assume either self maintaining “goal states” that have not yet been discovered, or universal “imperatives” that are subject to disagreement and probably are unrelated to at least some sociological knowledge. Less doubt can exist, however, of the usefulness of functionalist formulations to the social practitioner. The logic of functionalism — to judge consequences in terms of “adjustment” and “adaptation,” or “goal states,” or “imperative” problems specified in advance of analysis — is the typical logic of applied theory.

Ordinary and Theoretical Propositions

To present precisely defined determinants and results and specified relations between them are two ways of answering the question: What does this proposition mean? A third way of answering the same query phrases the answer in terms of the informative value of the proposition.

In general, the larger the number of different ways in which a proposition can conceivably be proved incorrect, the higher its informative value. Put differently, the higher the informative value of a proposition, the greater is the variety of events for which it can account. A critical task for the theorist in any science is to sub-[p.80]sume a large number of propositions of low informative value under a few propositions of higher informative value. When the theorist asks about a proposition, “What does it mean?” he wants to know also (1) what are the less informative propositions that are implied in the one under consideration, and (2) what are the more informative propositions that imply the one under consideration.

Propositions of low informative value are legion, and I shall simply call them ordinary propositions. Propositions of high informative value deserve to be called theoretical propositions.

Since theoretical sociology is already very abstract, it is essential for both researchers and practitioners to learn to extract the ordinary propositions from theoretical ones. Researchers need them for their research designs, and practitioners need them as bases for concrete advice to clients.

Suppose we ask for the ordinary propositions implied in this theoretical one:

Persons tend to engage in actions that maintain the evaluations they receive from their associates.

A key term here is ‘evaluations.’ Like so many other terms in sociology, it is a broad tent covering a multitude of phenomena that look different to common sense. We find the special cases of this proposition by searching in our taxonomy for all terms that have ‘evaluations’ [p.81] as a component. We find, among many others, that ‘approval’ is defined as an evaluation of an action; ‘esteem’ is defined as an evaluation of a person; and ‘rank’ is defined as an evaluation of a position in a social structure. We can now specify our proposition so that it deals separately with these three instances. For example, the last of the three propositions so obtained would read:

Persons tend to engage in actions that maintain the rank they enjoy in their social structure.

This is the well-known story that men tend to do everything to avoid demotion.

We may proceed further by decomposing terms other than evaluation in our original proposition. Suppose we take the term ‘action.’ One taxonomy divides it into ‘physical actions’ and ‘communicative actions,’ and the latter in turn into ‘descriptions’ (“Mr. X is a Senator”), ‘evaluations’ (“Mr. X is a great man”), and ‘prescriptions’ (“Re-elect Senator X!”). This sensitizes us to the variety of actions that may be involved in maintaining approval, esteem, and rank. To single out just one of the many propositions specifying each variety of evaluation, we take the category of evaluation we last discussed and the last category of action, and obtain the following:

Persons tend to issue prescriptions that maintain the rank they enjoy in their social structure. [p.82]

In other words, a person would try to issue rules that help him retain tenure in his rank.

We may further decompose the term ‘social structure,’ for example, into ‘organization,’ structures with common leadership, and ‘markets,’ structures without common leadership. The latter could be further broken down into markets in various institutional realms, e.g., economic markets such as commodity exchanges, scientific markets such as fields of social science, political markets such as electorates, and so forth. Taking only the last mentioned as an illustration, we have:

Persons tend to issue prescriptions that maintain the rank they enjoy in their electorate.

Thus, if a person has any elected rank at all, he will work for those rules, suggestions, and laws that maintain him in office. This proposition of political sociology is thus a special case of our original theoretical proposition. We may proceed to apply it to a specific electorate, e.g., the American, to specific persons, e.g., the House of Representatives, at a specific time, e.g., 1963. This gives us one of the necessary propositions for relating the work of the 88th Congress to the sentiments of the American people. We have gone from the theoretical to the ordinary.

The type of causal linkage in the special case is the same as in the original proposition. If we assumed that the original proposition was an irreversible, stochastic, [p.83] coextensive, sufficient one, then its ordinary implications would have the same causal linkage.

If we want to investigate whether two or more ordinary propositions can be assumed under the same theoretical proposition, we first must establish whether they have the same type of causal linkage. If such a similarity in type can reasonably be assumed, we may proceed by analyzing the terms in the ordinary propositions which indicate their determinants and results. If these terms have common elements, we can then formulate a theoretical proposition by using these common elements. An illustration may clarify this procedure.

Suppose we have the following findings:

Students at Bennington College in the middle 1930s, who were elected worthy by popular vote of representing the school in a meeting with other colleges, were more affected by the liberal values predominant among their teachers and fellow students than were others.¹³

Subjects in a social psychological experiment in Ann Arbor in the late 1940s, who were told that their instructor-experimenter selected them as a model group, agreed with each other in the writing of a story connecting three pictures more than did others who were not told that they were a model group.¹⁴ [p.84]

From reading the studies by Newcomb and by Back which report these findings, it seems reasonable to assume that we have in both instances a stochastic substitutable proposition, contingent on social visibility of attitudes and cognitions. It is less easy to say from the studies whether the findings are sequential or coextensive, reversible or irreversible. My guess would be that they are reversible and coextensive. I shall, however, assume that whatever one is, the other is the same.

The findings state that whatever happened took place in “a college” and in a “social psychological experiment,” among “students” and “teachers” in one instance and “subjects” and “instructor-experimenter” in the other instance. We subsume college and social psychological experiment under the term “group” and all the persons involved under the term “group member.” If a single theoretical proposition can be found that implies the two findings, it will be one that deals with groups and group members. However, all measures were taken among the rank-and-file members of the groups (the students), so the effects should be stated as valid only for the rank-and-file.

We now analyze the determinants in the findings. They are:

“elected worthy by popular vote of representing the school in a meeting with other colleges” [p.85]
“told that their instructor experimenter had selected them as a model group”

Both of these may be subsumed under:

“received more favorable evaluations”

Proceeding to the results we have in the findings:

”more affected by the liberal values predominant”
“agreed with each other more in writing a story”

Both of these may be subsumed under:

“their ideas converged more with those of other group members”

We can now formulate a theoretical proposition which contains our two findings:

The more favorable evaluations rank-and-file members receive in a group, the more their ideas converge with those of other group members.

We have gone from the ordinary to the theoretical. The resulting proposition, like its component findings, is assumed to be a stochastic, contingent one with a substitutable determinant.

Note that in the process of formulating the theoretical proposition we dropped references to Bennington and [p.86] Ann Arbor and to the fact that the studies were made in the 1930s and 40s. In highly theoretical propositions we do not make references to time and space; these propositions are presumed valid in all places at all times. Nor do they contain proper names (e.g., of specific individuals); they are presumed valid for all.

In reviewing our illustration, it is clear that we arrived at our theoretical proposition by a process of analyzing terms used in the findings. If we had had no general terminological schema into which the words of our findings could have been fitted, it would have been difflcult to arrive at the theoretical proposition. Here, then, is a type of problem whose solution requires a good taxonomy.

Notes to Chapter 4

1. George C. Homans, The Human Group, New York: Harcourt Brace & World, 1950, p. 112.

2. Andrzej Malewski, “Levels of Generality in Sociological Theory” in Hans L. Zetterberg and Gerda Lorenz (eds.), A Symposium on Theory and Theory Construction in Sociology, Totowa, N.J.: The Bedminster Press, 1965.

3. Max Weber, Die Protestantische Ethik und der Geist des Kapitalismus in Gesarnmelte Aufsatze zur Religionssoziologie, Vol. I, Tübingen: J. C. B. Mohr, 1922.

4. George C. Homans, op. cit.

5. Georg Simmel, Soziologie, 3rd ed., Berlin: Duncker~Humblot ,1958, ch. 2

6. Paul F. Lazarsfeld, et. al., The People’s Choice, 2nd ed., New York: Columbia University Press, 1948, pp. 59-61.

7. Werner Sombart, Warum gibt es in den Vereinigten Staaten keinen Sozialismus?, Tübingen: J. C. B. Mohr, 1906.

8. Eli Halévy, Histoire du peuple anglais au XIX siècle: L’Angleterre en 1815, 2nd ed., Paris: Hachette, 1913.

9. Robert K. Merton, Social Theory and Social Structure, revised and enlarged ed., New York: The Free Press of Glencoe, 1957, p. 51.

10. Johan Galtung, “An Outline of Structural Functional Theory Applied to Social Change,” unpublished manuscript, Ch. I, p. 6.

11. Ernest Nagel, Logic Without Metaphysics, New York: The Free Press of Glencoe, 1961, pp. 251-252.

12. Talcott Parsons, “An Outline of the Social System,” Talcott Parsons, et al (eds.) in Theories of Society, Vol. I, New York: The Free Press of Glencoe, 1961, pp. 38-41.

13. Theodore M. Newcomb, Personality and Social Change, New York: Dryden Press, 1943.

14. Kurt Back, “Influence through Social Communication,” Journal of Abnormal and Social Psychology, vol. 46,1951, pp. 9-23.

[p.87]

5. The Ordering of Sociological Propositions

Any sociological topic is likely to bring to mind many propositions. These propositions are identified and presented in a variety of ways in theoretical publications. Some writers present them in the normal course of a paragraph, and the reader is warned merely by the surrounding text that a proposition has been advanced. Some authors help their readers by presenting their propositions in italics. Still others set them apart by indentations or other typographical devices. Some present them as listings under special subheadings such as “Hypotheses.” Many give them numbers or proper names to facilitate their identification. No uniform rules prevail here, nor are they needed so long as the reader is made aware that certain sentences are propositions.

Regardless of how propositions are identified, the problem of ordering them becomes important as soon as they reach a number beyond two or three. While it is possible to simply list the propositions — as do the authors of Voting in a useful appendix — more efficient [p.88] modes of ordering propositions are often possible. Let us examine some currently used formats.

Inventory of Determinants

All factors that affect certain phenomenon are systematically listed in an inventory of determinants. A good illustration is provided in a paper by Kingsley Davis and Judith Blake containing propositions of factors determining a society’s fertility rate. ¹ The factors are sorted into three main categories: (1) those affecting the likelihood of sexual intercourse, (2) those affecting the likelihood of conception, and (3) those affecting the likelihood of fetus survival. Each one is subdivided to allow formulation of specific propositions. The first category contains a variety of propositions:

The higher the customary age for entry into marriage (or other sexual union), the lower the fertility rate.
For example, property rules in Ireland, where a father relinquishes control over farm property at his son’s marriage, is conducive to a high age at marriage (29 years for women) and this factor depresses the fertility rate.
The greater proportion of women in permanent celibacy, the lower the fertility rate.[p.89] For example, the low proportion of never-married women in Ceylon (.8%) and India (3.4%) make for high fertility, while the high proportion in Ireland (26.3%) or Sweden (20.9%) leads to a low fertility rate.
The longer the time of celibacy, after or between unions, the lower the fertility rate.
For example, prohibition of remarriage of widows and divorcées in some societies or religions depresses the fertility rate. The time-lapses between common-law marriages for women in Jamaica reduces the fertility rate by 37%.

And so the authors continue to develop over a dozen propositions in which the determinants (independent variables) vary but the result (dependent variable) is always fertility.

Inventory of Results

A list of propositions in which the determinant is one and the same but the dependent variables are different is an inventory of results. An illustration is furnished by Janowitz in a paper on the consequences of mobility. ² The author organizes his propositions into two categories: (1) those dealing with consequences for pri-[p.90]mary groups, and (2) those dealing with consequences for secondary groups. To sample:

The greater the social mobility of a family, the greater the instability of a family.
The greater the social mobility of a person, the stronger his ethnic and racial prejudices.
Upward mobility produces the political behavior typical of the new (higher) stratum.
Downward mobility produces the political behavior typical of the old (higher) stratum.

Janowitz is able to show that all propositions about mobility and primary groups can be subsumed under one more informative proposition first suggested by Durkheim: “Increased social mobility leads to increased disruption of primary relations.” No such theoretical proposition can subsume his statements about the consequences of mobility on secondary groups.

Chain Patterns of Propositions

When we deal with two or more sequential propositions in which a result in one reappears as a determinant in another, we can order them as a chain. An illustration is furnished by Terence Hopkins,³ who has reviewed stud-[p.91]ies of small groups, focusing, among other things, on four aspects:

The knowledge possessed by a person of the needs and attitudes of other group members;
The prestige of a person, that is, the extent to which others give him a favorable evaluation;
The authority of a person, that is, the extent to which he issues directions to the group that are acceptable to the group members;
The centrality of a person in the group, that is, the extent to which he maintains interaction with many other group members.

Studies can be cited showing that all these variables are positively correlated. If a person possesses one of these attributes he is likely to possess the others as well. Twelve separate propositions — or, better, six reversible propositions — can be written to show these relations.

However, the studies make it reasonable to assume that we deal here with sequential propositions. One possible flow of determinants and results is the following:

Persons who occupy central positions, that is, who interact with many other group members, tend to obtain better knowledge of their needs and attitudes;
Persons who have better knowledge of the needs and attitudes of others can more easily issue directives acceptable to others and thus tend to obtain higher authority; [p.92]
Persons of higher authority tend to receive more prestige;
Persons with prestige become sought-after interaction partners, and thus tend to obtain central positions in the group.

Chains like these can be illustrated by a schema of arrows:

One should not expect that every chain in this way becomes a circle; all kinds of geometric patterns are possible. Even for the problem at hand Hopkins has suggested several alternatives.

Chain patterns of great complexity can be simulated by electronic calculators. I will not discuss this here, except by noting that electronic simulations are most useful when we deal with complex patterns of causal linkages of the sequential type. Their usefulness when the causal linkages are of other types is less certain.[p.93]

Matrixes of Propositions

Another form of presenting propositions is the matrix. Here a certain number of factors are given and all their interrelations are specified. An example is furnished by the early part of Homans’’book, The Human Group. Three variables are given: “activity,” “interaction,” and “sentiment.” They are all considered both as determinants and results. Thus we get the matrix:

		Results
		Activity	Interaction	Sentiment
Determinants	Activity	—	H_ai	H_as
	Interaction	H_ia	—	H_is
	Sentiment	H_sa	H_si	—

In various parts of his text Homans can then spell out the six possible interrelations between these variables. They happen to be three reversible propositions, presumably of a coextensive type:

	H_is and H_si:	“If the frequency of interaction between two or more persons increases, the degree of their liking [sentiment] for one another will increase, and vice versa” ⁴
	H_ai and H_ia:	“If the scheme of activities is changed, the [p.94] scheme of interaction will, in general, also change, and vice versa.”⁵
	H_as and H_sa:	“A motive sentiment gives rise to an activity . . . but if either side of the relationship is changed, the other will be affected.” ⁶

It is plain that if we read across the rows of the matrix of propositions we obtain inventories of results, for example, “if high interaction, then much activity”; “if high interaction, then much sentiment.” If we read down a column we have an inventory of determinants, for example, “if much activity, then high interaction”; “if much sentiment, then high interaction.” Unlike an arrow schema of a chain pattern, a matrix like this is not restricted in its usefulness to sequential propositions.

Axiomatic Format with Definitional Reduction

Inventories and matrixes list every relevant proposition. A sophisticated theorist, however, might want to reduce the size of the matrix. This leads to an axiomatic theory. Axiomatic theories can have many different patterns. We shall illustrate only two possibilities, one obtained by the reduction of a matrix through the manipulation [p.95]of the definitions, and one obtained through the manipulation of the propositions. Normally, both manipulations are done at the same time; however, it may be more instructive for us to discuss each separately.

A brief example of a definitional reduction in a list of propositions can be constructed on the basis of a discussion of social aggregates by Arnold Rose.⁷ Let us assume as given this inventory of propositions about the emotional excitement and membership turnover in social aggregates.

Groups have less turnover than publics
Publics show less emotion than crowds
Groups show less emotion than masses

We begin the reduction of these propositions by an analysis of the key terms:

Groups are social aggregates interacting in terms of specified roles and with a common leader (e.g., a voluntary association).
Masses are social aggregates interacting (if at all) in terms of unspecified roles but with a common leader (e.g., a radio audience).
Publics are social aggregates interacting in terms of specified roles but without a common leader (e.g., a market).
Crowds are social aggregates interacting in terms of [p.96]unspecified roles and without a common leader (e.g., milling in Times Square).

Comparing these with our original propositions, we find that the aggregate with common leader is assumed to have less turnover, and the aggregates with interaction in terms of specified roles show less emotion. Thus our original propositions are reduced to two theoretical propositions:

If a social aggregate has a common leader, then its turnover is low.
If a social aggregate interacts in terms of specified roles, then its level of emotion is low.

The most interesting part of this procedure is that these two propositions do not merely imply the three that we had as our starting point but also a fourth. Proposition (I) and Definition (B) imply that “masses have less turnover than crowds.” This is a novel hypothesis which, to the best of my knowledge, is presented here for the first time. Thus we see how an axiomatic format not merely organizes existing propositions but generates new ones implicit in the existing ones.

Axiomatic Formats with Propositional Reduction

In the previous example we obtained a reduction in a list of propositions by combining propositions with defi-[p.97]nitions. It is also possible to obtain a reduction by combining propositions with other propositions. From the list of original propositions (inventories or matrices) a certain number are selected as postulates. The postulates are chosen so that all other propositions, the theorems, are capable of derivation from the postulates and no postulate is capable of derivation from other postulates. One generally strives to use as few postulates as possible. Assume, for example, that the following propositions are given:

If national prosperity increases, then the middle classes expand.
Economists are fairly well in agreement that the ranks of service occupations, dealers, and brokers expand during periods of prosperity and in countries with a growing GNP.
If the middle classes expand, the consensus of values in the society increases.
While disproportionate expansion of lower or upper classes leads to a polarization of values (as Marx argued), a similar expansion of the middle classes promotes the convergence of values in the society.
If the middle classes expand, the social mobility increases.
The expanding ranks of the middle classes must be filled by persons from other classes, thus promoting mobility.
If social mobility increases, the consensus of values in the society increases and vice versa.
Social mobility creates families in which fathers, sons [p.98]and brothers belong to different classes and family loyalties modify class ideologies. This is a reversible proposition: if there is much consensus of values between social strata, then social mobility between them becomes easier.

From this list we may select propositions (1), (2), and (4) as postulates. Let us restate them with roman numbers:

If national prosperity increases, the middle classes expand.
If the middle classes expand, the consensus of values increases.
If social mobility increases, the consensus of values increases, and vice versa.

The implications of these propositions can now be spelled out in the form of theorems. Postulates II and III combine into the familiar:

If the middle classes expand, the social mobility increases;

thus completing the set of propositions we had at the beginning. In addition, Postulates I and II render this theorem:

If national prosperity increases, the consensus of values increases.

Furthermore, if Theorem 3 is combined with Postulate I, we obtain: [p.99]

If national prosperity increases, the social mobility increases.

The last two theorems are novel in the sense that they were not included in our original set. Theorem 5 is not trivial; it suggests, for example, that if we want to promote social stability in the form of less political and ideological cleavages in a society, we should maximize its national income. (This is one argument for foreign aid to less prosperous societies.) Theorem 6 has been mentioned — in fact, among others by Lipset and myself⁸ — in the literature; I was, however, unaware of its logical ties with our other propositions.

Our experience in axiomatizing sociological propositions is limited. However, I believe the above instances are fairly typical: attempts toward axiomatization often generate some propositions that were not explicitly mentioned in the original set. Some of these added propositions may be novel; others may be well known by themselves but not in their connections with other propositions. An axiomatic schema renders this service because it makes visible all ideas implicit in some given ideas.

I do not hesitate, therefore, to recommend to a theorist that he arrange his propositions in the axiomatic [p.100]way: it forces him to spell out his assumptions, to make explicit his deductions; and it will remind him of any bypassed implications. This does not necessarily mean that his final publication should have an axiomatic organization. The way propositions are presented to the public is an editorial question. There may be instances in which axiomatic thinking is most efficient but an axiomatic editorial format becomes so cumbersome that it gets in the way of efficient communication. Theoretical sociology can never surrender logic to taste or style; however, as soon as we know from an axiomatic exposition that our logic is good, there is every reason to proceed in the best of taste and style.

Notes to Chapter 5

1. Kingsley Davis and Judith Blake, “Social Structure and Fertility: An Analytic Framework”, Economic Development and Cultural Change, vol. 5, 1956, pp. 211-235.

2. Morris Janowitz, “Some Consequences of Social Mobility in the United States”, Transactions of the Third World Congress of Sociology, vol. 3, 1956, pp. 191-201.

3. Terence K Hopkins, The Exercise of Influence in Small Groups, Totowa, N J: The Bedminster Press, 1964.

4. George C Homans, The Human Group, New York: Harcourt Brace & World, 1950, p. 112.

5. Ibid. p. 102.

6. Ibid. p. 99.

7. Arnold M. Rose, Sociology, New York: Alfred A. Knopf, 1954, ch. 9.

8. Seymour Martin Lipset and Hans L. Zetterberg, “Social Mobility in Industrial Societies” in Seymour Martin Lipset and Reinhard Bendix, Social Mobility in Industrial Society, Berkeley and Los Angeles: University of California Press, 1959, p. 27.

[p.101]

6. The Confirmation of a Proposition

So far, we have dealt with the question of what propositions mean and how they may be ordered. Now let us turn to the equally crucial issue of the truth of the propositions, the evidence that supports them.

Let us begin by noting that propositions supported by evidence are called invariances, and propositions for which more evidence is needed are called hypotheses. This distinction cross-cuts our previous one between ordinary and theoretical propositions, and we get this important fourfold division:

	Low informative value	High informative value
Empirical support wanting	Ordinary hypothesis	Theoretical hypothesis	HYPOTHESES
Empirical support sufficient	Ordinary invariance: Finding	Theoretical invariance: Law	INVARIANCES
	ORDINARY PROPOSITIONS	THEORETICAL PROPOSITIONS

[p.102]An ordinary invariance is what we know as a finding; a theoretical invariance is what we know as a law. (Our current speech habits do not give separate words for ordinary and theoretical hypotheses.)

There is an embarrassment of riches of ordinary hypotheses about social life. Most sociologists at present take this as a great challenge to test the hypotheses and turn them into findings of high probabilities. However, as Popper points out:

Science does not aim, primarily, at high probabilities. It aims at high informative content, well backed by experience. But a hypothesis may be very probable simply because it tells us nothing, or very little. ¹

I think we should allow ourselves a little more courage in taking the abundance of available propositions about social life as a challenge to turn ordinary hypotheses into theoretical ones without first maximizing the evidence that supports them. This is one way in which sociology can avoid its current painstaking triviality. In particular, I think sociology should make a more serious effort to incorporate in its theories the best thoughts (theoretical hypotheses) of the human condition found in Homer, Dante, Shakespeare, Cervantes, Twain and other great writers, who now provide the lion’s share of [p.103] any educated layman’s conception of the human drama. ² In the end, however, the outcome of the theoretical enterprise should be ”high informative content, well backed by experience,” that is, laws. Only experience through trial and error can teach us whether we arrive from ordinary hypotheses to laws more easily via findings or via theoretical hypotheses. For the present, we shall only discuss how hypotheses are turned into invariances, and leave this question open.

Our focus will be the confirmation of a single hypothesis. However, it should be said at the very outset that it is usually very hard to confirm a single hypothesis but quite easy to confirm a theory, a system of hypotheses. For pedagogical purposes, however, we may focus on a single hypothesis in this chapter, and postpone the discussion of systems of hypotheses to Chapter 8.

It sounds so simple when we say that in order to confirm a hypothesis we check it against observations. However, the actual procedure is amazingly complicated, and errors of many different kinds can easily creep in and disqualify the results. Some major steps in verification will be set forth in this chapter, and some of them will receive special attention in the next chapter. We will, however, avoid treating the very intricate problem of the possibilities of induction in general. We [p.104] will instead tie our treatment to some current practices among sociologists which appear sound and reasonable.

An Overview of Steps in Confirming a Proposition

As an illustration of the confirmation of a single proposition, let us discuss a test of the hypothesis: ”The more a member conforms to the norms of a formal organization, the greater the likelihood that he will be promoted.” This will be called the Conformity-Promotion Proposition. Relevant data for testing are available in The American Soldier. ³

The indicators of ”conformity to Army norms” consist of six questions, for example: ”In general, how serious an offense do you think it is for a soldier to go ‘AWOL’ (Absent without official leave)?” The conformist answer to this question is ”very serious”; other answers were rated as non-conformist. The conformist answer to all six questions were fitted into a Guttman-type quasi-scale with a reproducibility coefficient of .82. In all, this index for measuring conformity to Army norms appears valid and fairly reliable. It was part of a questionnaire given to privates in November 1943. According to [p.105] their scores on the scale, they were classified as Strict Conformist (score 5-6), Medium Conformists (score 3-4), and Poor Conformists (score 0-2). To record ”promotion,” a search was made in the records to find out which of the same privates had made the rank of non-commissioned officer (mostly corporals) by March1944. There is no need to question the validity or reliability of this simple indicator. The sample consisted of 374 men from an Infantry division who had entered the Army during the summer of 1943.

We now proceed to check whether the data trend fits the trend predicted by the hypothesis. The following summary gives the necessary information:

Conformity to Army Rules	Prediction from hypothesis: Likelihood of promotion	Data: Per cent promoted
Strict conformists (N=68) Medium conformists (N=138) Poor conformists (N=112)	High Medium Low	31% 28% 17%

Thus, we find that the trend in data parallels the trend predicted from the theoretical proposition.

The fact that data and proposition point in the same direction is comforting. To be more certain, however, we might also want to appraise to what extent this parallelism exists. In our case the differences appear small, particularly between the strict and medium conformists. Our design is not precise enough to allow a strict [p.106] test of the ”goodness of fit”; sociological models rarely make detailed predictions about the behavior of the indicators, only over-all predictions. In our case, we can at best check how often differences of the magnitude we found occur as chance fluctuations in sampling. A chi-square test renders X² = 5.916, which corresponds to the probability .02<p.<.03 with one-tail test of significance. This gives us only a modest assurance that that trend in our data would be replicated in new samples. The representativeness of the sample could not be checked for this particular instance, but information about similar samples in War Department studies allow us to assume that this one is fair. The scope of the population — the US Army in World War II — is a more serious limitation; ideally we would like to see the proposition tested with data from other institutional realms, e.g., a religious hierarchy, a business enterprise, a civilian government bureaucracy, a university.

The most intriguing problem of appraising our test remains: to control for alternative explanations. We know from other parts of the Army study that likelihood of promotion increases with length of service, and that it increases with education. How can we make sure that seniority or educational qualifications (or any other known or unknown factor) cannot account for our finding? As for seniority, we know that it could not play any part, since all our subjects were in the Army equally long. The part played by education was [p.107] checked in this and two other samples through the technique of multivariate analysis. ⁴ The tables are not published, but the authors report the result in the text:

When the data . . . are broken down into two educational classes, the same consistency appears in all three samples for high school graduates and college men and in two of the three samples for other men, in spite of the small number of cases.⁵

Thus we have some assurance that differences in educational qualifications do not account for the findings about likelihood of promotion. Whether other alternative factors, unknown at present, can account for the trend in our data remains a question. Only experimental designs can control for unknown alternative hypotheses.

Most research workers would probably stop their test at this point. However, one more appraisal ought to be done: how well integrated is the tested proposition in available social theory? We know from a large number of studies that the proposition ”the more a member conforms to the norms of his group, the more favorable valuations does he receive from his group” ⁶ is valid. Let us call this the Sanction Proposition. There is also a well-[p.108]known sociological proposition about rank equilibrium: a person with high rank in one sphere of life tends to move toward high ranks also in other spheres of life.⁷ This proposition can be generalized into one which has higher informative value: ”the more a person receives favored valuation on one dimension, the greater the likelihood that he receives favored valuations also on other dimensions.” This we might call the Halo Proposition. However, ”rank” is a social valuation of a position. Thus we can specify our derivation to read: ”the more a member conforms to the norms of his group, the greater likelihood that he is given higher rank.” Now we have only to note that a ”formal organization” is a kind of group and that to be ”given higher rank” is to be ”promoted” and we have the proposition of our test: ”the more a member conforms to the norms of a formal or organization, the greater the likelihood that he will be promoted.” Thus we see that the hypothesis we test is consistent with other confirmed propositions. This greatly adds to our confidence in accepting it as plausible.

The steps in our appraisal of this study are illustrated in the adjoining flow chart. [p.109]

In summary, we base our decision to call the Conformity-Promotion Proposition confirmed on the following criteria: [p.110]

the validity of the indicators;
the reliability of the indicators;
the fit between the data trend from the indicators and the trend predicted by the tested proposition:
(a) the extent to which the direction of the trends coincide;
(b) the likelihood that the data trend is a chance fluctuation;
the control of alternative propositions;
the representativeness of the sample and the scope of the population;
the extent to which the tested proposition is an integral part of established theory.

All these criteria have to be weighted into a composite judgment of acceptance or rejection. The fact that we can get quantitative estimates of criteria (2) and (3b) should not tempt us to give undue emphasis to them. The beginner would probably reject the tested hypothesis because the reproducibility .83 is not quite the desired .90. In our opinion, a rejection would be a mistake. The validity seems good, the fit (3a) is fair, one important alternative hypothesis is ruled out, and the proposition is integrated in established theory. The reliability and statistical significance are not so far off that they subtract much from the good impression the test gives on these more important criteria. Thus we accept for the time being the Conformity-Promotion Proposition as tentatively confirmed. [p.111]

At this point a lingering doubt might occur: should we not, after all, play it safe and reject the proposition? Even if we are 85 per cent sure, would it not be correct, in the name of science, to reject it? The answer is no. Scientific advance is as hampered by the error of rejecting something true as by accepting something false.

The Separation of Definitions and Indicators

To some writers, the confirmation procedure concerns only the indicators. It is, therefore, tempting to dispense entirely with everything else. Why not simply call the indicators ”operational definitions” and not use any other definitions of determinants and results? This is the position taken by orthodox operationalists. This movement has been strong in contemporary sociology and deserves some attention. Lundberg expresses its spirit when he writes about Thurstone’s operational definition of attitudes:

Thus, Thurstone records his observation of certain behavior. This behavior explicitly defined operationally, he calls an attitude. Whereupon his critics vigorously proclaim that this is not an attitude at all. Attitude is something else — and proceed to define it not by other operations than Thurstone’s but by another series of noises, which have an expressive function [p.112] comparable to exclamations of joy or sadness, laughter, or lyric poetry, but which have no objective representative function at all. ⁸

To the extent that this statement means that we can never avoid being explicit about our research procedures and measurement descriptions, we do not object to it. But when operationalism is construed to mean that anything social that can be recorded and measured should command our attention as sociologists, then we object. In terms of confirmation of theories, it is plain that only those indicators that have counterparts in definitions of determinants and results are worthwhile. When verifying a theory, other measurement devices may very well be worthless and irrelevant.

The vindications for the use of conventional nominal definitions are many. One is that they can enter into the logical relationships that make possible the advantages of theory — particularly of axiomatic theory — more readily than operational definitions can. If we accept orthodox operationalism, we make it unduly difficult to obtain the advantages of theorizing. The insistence that all definitions should be operational also leads to other rather undesirable consequences. If fully accepted, it means that a change of operational definition implies a [p.113] change in the proposition being tested. Then it would be impossible to disprove an earlier accepted proposition with new and better indicators.

One may here question the place of operationalism in sociology. A very legitimate aspect of operationalism concerns the definitions of score values on variables. When we are asked, not what variable a certain scale measures, but what value a certain score on this scale signifies, we give our answer in terms of a description of the scoring technique, the standardization group, and so forth — in short, an operational definition. ⁹

Nothing said above should be construed as an appeal to keep definitions and indicators at arm’s length. They should instead embrace each other in the most intimate way. When we ask how ”valid” the indicators are, we are asking about the intimacy of this embrace.

Notes to Chapter 6

1. Karl Popper, ”Degree of Confirmation”, British Journal for the Philosophy of Science, vol. 6, 1955, p. 146.

2. A successful illustration of this approach is found in Hugh Dalziel Duncan, Communication and Social Order, Totowa, N.J.: The Bedminster Press, 1962.

3. Samuel A. Stouffer, et al., The American Soldier: Adjustment during Army Life, Princeton: Princeton University Press, 1949, vol. I, pp. 258-265.

4. Infra. pp. 144-146.

5. Samuel A Stouffer, op.cit., p. 263.

6. Henry W. Riecken and George C. Homans ”Psychological Aspects of Social Structure” in G. Lindzey, Handbook of Social Psychology, Cambridge, Mass.: Addison-Wesley, 1954, pp. 786-832.

7. Emile Benoit-Smullyan, ”Status, Status Types, and Status Interrelations”, American Sociological Review, vol. 9, 1944, pp. 151-161.

8. George A. Lundberg, Foundations of Sociology, New York The Macmillan Company, 1939, p. 59. In his Social Research revised edition, New York: Longmans, Green and Co., 1941, p. 7, Lundberg, however, advocates the same approach to the relation between theorizing and research operations as we have given here.

9. Cf. Gösta Carlsson, Dimensions of Behaviour, Lund: C. W. K. Gleerup, 1949, ch. 2.

[p.114]

7. On the Decisions in Verificational Studies

In this chapter we propose to comment on the different decisions required in the enterprise of confirming a sociological proposition by research. We want to cover a large number of decisions in the research process, using only the briefest illustrations. Those who have not participated in social research or read textbooks about techniques of social research may want to read only the conclusions of this chapter.

Internal Validity

Validity, loosely speaking, is the extent to which an indicator corresponds to a definition. The question of validity thus goes to the core of the relation between theory and data. Part of Newton’s genius is that he could see that the indicators used in astronomy could be validly coordinated to definitions used to interpret [p.115] data from small scale experiments. The progress toward validity lies in a continuous adjustment of theorizing to the techniques of research, and in a continuous adjustment of techniques of research to theorizing. Unfortunately, contemporary sociologists sometimes seem to lack good understanding of this principle. New methods are often developed in a theoretical vacuum, sometimes in response to practical needs. And whole conceptual systems are published without the slightest hint as to how their concepts should be translated into research operations.

Guttman has divided the issue of validity into “internal” validity and “external validity.” ¹ The major difference is that the former expresses a “logical” relationship, while the latter expresses an “empirical” relationship. Internal validity, in other words, can be appreciated without empirical studies, while the determination of external validity is a test of a hypothesis. Let us begin by discussing the former.

Perfect validity means that the indicator has the same scope of content as the definition. Some typical problems of internal validity might now be illustrated. Let us assume that our nominal definition is one of “work satisfaction.” Let us represent it with a circle — [p.116]

Let the indicators, e.g., a questionnaire to record work satisfaction, be represented by a broken circle —

Or, more generally, the solid circle represents sentences that enter into the nominal definition, and the broken circle represents the sentences describing the indicators.

We may distinguish the following typical problems of validity.

The definition implies the indicator and, in addition, something other than the indicator.

[p.117] This would, for example, be the case if we had the response, “I am satisfied with the ventilation where I work” as the operational definition of work satisfaction. Obviously the term “work satisfaction” implies more than satisfaction with the ventilation.
The indicator implies the definition and, in addition, something other than the definition:

This would, for example, be the case if we had the response, “I like it here in X town” as the indicator of “work satisfaction.”
The indicator implies the definition and vice versa:

The response, “I like my friends and acquaintances here in X town” would represent this situation if it were used as an operational definition of “work satisfaction.” The satisfaction with fellow workers would belong to “work [p.118] satisfaction,” while satisfaction with leisure time friends would not.

These three kinds of errors of internal validity may account for some of the contradictory results we have sometimes found when similar studies have been done around the same topic. One usually tries to minimize the effects of these errors by combining many indicators into one index. Indexes are formed according to rules for combining indicators. For verification purposes, one should use rules of index-formation which maximize the likelihood that the index will record the determinant (or result, as the case may be) of a proposition and distribute the errors, that is, all other factors it records at random. An index thus tends to let errors among its component indicators cancel out. However, it is well worth remembering that one single valid indicator is worth more than an index made up of numerous indicators of low validity. And too often one reads research reports in which the author has allowed several less valid indicators to contaminate one really valid indicator by pooling them all into one and the same index.

In the last analysis, the validity of our indicators can be judged only in the context of success of a theory. The acceptance by the scientific community of a theory gives content to its definitions and meaning to its indicators, and makes it possible to speak of them as having a close correspondence.

Validity can be achieved not only by changing one’s [p.119] indicators but by changing one’s definitions. This interplay between definitions and indicators can be illustrated by two typical “aha-experiences”:

“This dimension which my theory said was one and the same is actually several distinct dimensions in my data.”
One might assume that Neal Gross and his coworkers had this experience when they studied the role of high school superintendent.² Sociological theory has long assumed the concept “consensus of role prescriptions,” the idea that many persons agree on what is expected of the occupant of a given position. The indicators used by Gross, however, showed that there is considerable difference between what a group of school superintendents agree is their role and what members of school boards, for example, agree is their role. This difference in the data leads Professor Gross and his coworkers to reformulate the concept of consensus of role prescriptions into two concepts — “interposition consensus” and “intraposition consensus” — and to develop separate indexes for them.

“These dimensions which my theories said were several distinct ones are actually one and the same in my data.”
In a review of small groups research we were struck by a peculiar circumstance: the indicators that members of a group evaluate each other favorably is in one research tradition linked to the definition of “soci-[p.120]metric popularity,” in another to the definition of “cohesiveness,” in a third to a definition of “sentiment,” and in a fourth school of thought it is linked to “morale.” It is, therefore, possible to reduce these to one — namely, the “favored valuations from associates.” Here, then, several indicators in the research literature could be validly coordinated to one definition rather than to several definitions.

External Validity

Let us now proceed to external validity. External validity becomes important whenever we want to use one indicator as an index or prognosis of another indicator. The most common case of this in sociology is when we use a verbal expression as an index to other behavior. When we ask people about their social participation, we would like to know how they actually participate, not what they tell about their participation. The validity of a social participation item in a questionnaire is limited by the extent to which our respondent tells the truth. This is the case with many other indicators too. The number of sociometric choices received is a valid index of popularity, provided the respondents have told the truth about who their friends are. Monthly earnings, the number of school years completed, are likewise valid indexes of income and education only to the extent that they are accurately reported.³ [p.121]

Now we know that some data used in sociological studies probably are inaccurate. There are reasons to believe that births from certain areas are underreported; there are suspicions that crime statistics from some countries may be falsified for political reasons, and so forth. We know that people often lie in response to our interviewers. In Elmira, New York, 9 per cent of the individuals who told the interviewers that they had voted had actually not done so, according to the records of the Election Board.⁴ The same is the case in a survey I did in Uppsala, Sweden. Out of 19 persons who did not vote we found that 11 told the interviewers that they had voted. The point can be duplicated in other surveys.

In the example of the voting question, we can check the validity of responses because we know of an external, quite valid and reliable criterion. In most cases, however, we lack such a criterion. Exploring the latter, the Uppsala survey received these responses to the following questions: [p.122]

Do you have as good table manners when you are at home as when you are at other people’s homes?	Yes	98%
Do you ever think badly of your closest friends?	No	91%
Are you always sincerely happy over your friends’ successes?	Yes	82%
Do you sometimes have sexual thoughts which you think are improper or immoral?	No	79%
Do you always take time to listen to other people’s problems?	Yes	79%

A psychologically sophisticated person is likely to believe that most of these responses are thoughtless untruths It appears that most of the persons interviewed in this survey responded in accordance with what they believed to be the standards of decent people. The majority refused to admit that they ever violated these standards. In short, they showed a tendency toward conventionalized answers.

An assumed conventionality is, of course, only one of the many personality traits that render interpretations of statements about facts hazardous in surveys and historical documents. Other personality traits may upset them in similar ways. The invalidity caused by conventionalized answers, however, seems to have traceable effects. Studies employing the interview method in the area of marital satisfaction are in amazing agreement with the conclusion that the person likely to love his [p.123] wife also loved his parents, had a happy childhood, goes to church regularly, and, in a number of ways, acts and feels according to what is considered to be virtuous. The hypothesis about a generalized tendency to conventionalize answers to interview questions makes these results less remarkable.

Lack of validity because of false information is, in principle, possible to detect. In practice, however, it is very troublesome, and we generally prefer to discover other facts than the extent to which informants lie. But, like the historians, we are obliged to check the truth of a piece of information before we use it. When the interview method is employed to obtain factual information, cross-checks can often be made. We can ask two respondents about the same facts. We can come back to the same respondent and ask him again. Such tests of agreement and consistency can never fully prove validity, but they can fully disprove validity.

Reliability

Reliability is the extent to which an indicator renders unambiguous readings. Reliability is a necessary prerequisite for validity. Unreliable instruments always lack validity.

The current discussion of reliability lacks precision, a fact pointed out by Ekman.⁵ The term reliability as it is [p.124] used in psychology and sociology involves at least four different measures:

The congruency of several indicators, that is, the extent to which several indicators measure the same thing
The precision of an instrument (intra-individual reliability), that is, the extent to which the indicator registers in a consistent way for one observer.
The objectivity of an instrument (inter-individual reliability), that is, the agreement of one scientist’s reading of the indicator with the readings made by other scientists.
The constancy of an object measured, that is, the extent to which the object measured does not fluctuate.

One may conceive of the general method of separating these components as a complex analysis of variance according to a factorial design:

Time	T₁				T₂				T₃
Observer	A	B	C		A	B	C
Indicator	1 2 3	1 2 3	1 2 3	. . .	1 2 3	1 2 3	1 2 3	. . .	. . .
Reading	x x x x x x x x x	x x x x x x x x x	x x x x x x x x x	. . . . . . . . .	x x x x x x x x x	x x x x x x x x x	x x x x x x x x x	. . . . . . . . .	. . . . . . . . .

[p.125] The variance between indicators reveals congruency; the variance between readings indicates precision; the variance between observers indicates objectivity; the variance between different times indicates constancy. There will remain the possibility of a residual error, represented by the different interaction effects.

No one has ever — to my knowledge — carried out this complex design and received estimates for all these types of errors of measurement. (The calculations, however, are not overly difficult, and this could be done.) Yet, we have many studies that include estimates of some of these sources of error and estimates of the total magnitude of measurement error. The cumulative experience of these studies has taught us to have confidence in certain measurement techniques, particularly those used in surveys and quantitative content analysis. Persons using these techniques, therefore, no longer feel obliged to go through the tedious process of checking their reliability. But whenever a novel research technique emerges, the question of reliability becomes pertinent. When sociological research uses sources and methods of historical research, possibilities for quantitative estimates of reliability become rare, perhaps non-existent. Historians have developed a special sense for these problems, but it seems largely uncodified and has to be learned in apprenticeships with established historians. Since sociological problems often call for historical data, efforts to make explicit the rules [p.126] for evaluating the reliability of historical indicators should be encouraged.

The role of measurements in the development of theoretical sociology should not be exaggerated. Comparatively few theoretical advances in other sciences seem to have been inspired by refined measurement techniques, and I cannot think of any existing sociological proposition that owes its existence and plausibility primarily to a careful control of the errors of measurement.

Scope

By ‘scope’ I mean the proportion of all possible sources of data that is represented in a given research.

A favorite example of the importance of scope is found in the generalization, “All swans are white.” This was held true until Australia was discovered and her black swans became known. Social science knows of similar occurrences. Freud, developing his theory of the Oedipus complex, worked with cases from a population of rather authoritarian Viennese families. But since cultural anthropologists discovered a society in which a man lives in the home of his wife’s parents, supports not his own but his sister’s children, behaves like a Western uncle toward his own children while the children’s uncle assumes the role of the father, it is not surprising [p.127] that Freud’s theory has not been confirmed.⁶ In both cases the theory could claim plausibility only in a limited population. When the scope was enlarged, the theory had to be amended.

However, most theorizing claims universality. The universality of a sociological proposition is an assumption which we have to confirm by extensive replications of our studies. When applicable, we have to confirm our propositions on different subject matters (political, religious, etc.) in different categories in the same society (income brackets, educational levels, etc.) and in different societies (civilized and primitive, ancient and contemporary). This is a demanding process to undertake before we can claim a hypothesis verified.

Current practice nevertheless allows a great deal of extrapolation beyond the scope of our original data. As indicated in the previous chapter, these lax standards are reasonable. However, on two scores I have learned to be suspicious and suggest that an extrapolation should not be accepted in advance of proof. These are (1) generalizations from micro-sociology, i.e., encounters, organizations, and markets, to macro-sociology, i.e., institutional realms, systems of stratification, and culture; (2) generalizations from executive actions and realms, i.e., economy, polity, and science, to emotive actions and realms, i.e., art, religion, and morals. [p.128]

We may ask here what point there is in a single test of a hypothesis performed on a population with limited scope. If any bet has to be placed on the outcome of a test of the hypothesis outside this population, the following principle applies: “It is more probable that a hypothesis holds true outside the population on which it has been confirmed than that the contrary of the hypothesis holds true in the new population.” It is hard to prove this point strictly — except, perhaps, through studies in the history of science — but this is actually a principle that we employ daily in common sense judgments. For example, when we stand in front of new doors, we treat the door handles according to this principle and turn them in the direction indicated by our experience with other door handles. In the long run, the principle seems to do more good than harm.

Representatives

Sociologists in general seem to have a very advanced conception of the role of samples in research. This sophistication has developed in descriptive sociology — not the sociology concerned with testing theories. Descriptive sociology has found it very convenient to describe a strictly defined universe through the use of various sampling techniques. The procedures for statistical evaluations of the representative samples are well [p.129] known, and there is no reason for entering into this topic. The question we have to answer is rather what importance we shall allocate to the representativeness of samples in the total verification process.

The relationships expressed in theoretical propositions are presumed to be universally present. They are accordingly, present both in representative and in nonrepresentative samples. To disprove or demonstrate their existence is, hence, possible in any kind of sample — biased or unbiased. This important, and perhaps surprising, consideration should immediately be qualified. When using a biased sample for a verification, we must have assurance that the relationship we want to prove is not introduced into our data by selective sampling. This possibility, however, is in most cases rather unlikely. Also, when using a biased sample for verification, we should realize that we have no knowledge of the population to which the result can be safely generalized. Furthermore, if the test of a given hypothesis involves, for example, mathematical constants computed on the basis of a representative sample, a biased sample might be avoided for further tests.

Hence we conclude that, on balance, nonrepresentative samples apparently are not significantly inferior to representative samples when we want to disprove a theoretical hypothesis. This relatively minor importance of representativeness in verification studies is in sharp contrast to the overwhelming importance of representa-[p.130]tiveness of samples in descriptive studies. When the law of falling bodies was demonstrated by our physics teacher, he used various materials — stones, metals, wood, cloth and cotton — to show that they all fell equally fast in a near-vacuum. He did not take representative samples of all these materials, but chose a wide scope for the population of material — the scope ranging from metal to cotton. Galileo, who first proved the hypothesis, proceeded in the same way, disregarding representativeness in favor of the scope of the population. To our knowledge, he has never been blamed for this.

Representativeness should not be confused with randomization. Randomization can be used to obtain representativeness. However, it is also used as a method of controlling irrelevant factors when testing a hypothesis. Randomization in the latter function is of the utmost importance, as will be shown in the next section.

Designs

We use the word “design” to indicate the way we arrange to produce the readings of our indicators. Different designs give different plausibilities to a test of a proposition, and we must learn to evaluate designs. As mentioned, we can never prove a proposition in any strict sense. The best proof we have is that our proposi-[p.131]tion can predict observations. This is an incomplete proof, since we always run the risk that new observations will disqualify our prediction or that alternative propositions will predict observations equally well. A conventional test of a proposition is to disprove its opposite, the null-hypothesis. The outcome should be based on: (a) fit, or the extent to which our data fall in the direction predicted by the hypothesis; (b) significance, or the certainty with which we have disproved the null-hypothesis; and (c) control, or the certainty and extent to which we have disproved alternative hypotheses. Designs might be evaluated in these three respects.

Fit. When the motions of the planet Uranus were studied, a discrepancy was found between the actual orbit and the orbit predicted by Newton’s law of gravitation. The discrepancy could be explained by hypothesizing an additional planet at a certain position. Eventually a planet, Neptune, was discovered in the vicinity of this predicted position. We realize that this was possible through the existence of very precise propositions phrased in quantitative variables. A sociological illustration of the same process of verification has been given by Coleman and his co-workers.⁷

We may hypothesize that the flow of scientific information among physicians is due to the fact that they [p.132] talk with each other and pass on new information in personal communication. Assuming that each physician has an equal probability of communicating an item to the others per time unit, we can then express the rate of diffusion of information about, say, a new drug by a “logistic” formula:

X_{t+ l} = X_t + k_lX ( l – X )

(1)

An alternative proposition is that information about the new drug, like a pure advertising campaign, reaches a fixed number of persons per time unit, independent of the number who have already learned about it. We then obtain a “constant source” formula:

Y_{t+ l} = Y_t + k₂ ( l – Y)

(2)

In this case, our propositions make rather specific predictions, and a good quantitative measure can be obtained for the dependent variable. The theory says that the logistic pattern will prevail when doctors associate with each other and that the more linear pattern will predominate when they are isolated from each other. Sociometric questions about friends can be used to separate the integrated from the isolated doctors. Their respective rates of adoption of a drug is shown in the chart, together with the rates predicted from the theory. We see that the predictions are reasonably close to the data. When complex predictions of this kind turn [p.134] out to be correct, one is tempted to accept the theory without further controls.

[p.133]

Significance. It is common to distinguish between ‘cross-sectional’ and ‘longitudinal’ designs for tests of hypotheses. In sociology they are represented by, for example, the survey method and the panel technique, respectively. Suppose that x is the determinant and y the result. In a cross-sectional design we measure a sample of n units, at the time t₁ with regard to x and y. The data we obtain consists of two series such as the following:

^t¹X₁^t¹X₂^t¹X₃_|^t¹X_n

^t¹Y₁^t¹Y₂^t¹Y₃_|^t¹Y_n

In the longitudinal design, we measure a sample of n individuals at the times t₁, t₂, t₃, etc., with regard to x and y. The data we have are series like the following:

^t¹X₁^t¹X₂^t¹X₃^t¹X₄_|
|^t¹X_n

^t¹Y₁^t¹Y₂^t¹Y₃^t¹Y₄_|
|^t¹Y_n

^t²X₁^t²X₂^t²X₃^t²X₄_|
|^t²X_n

^t²Y₁^t²Y₂^t²Y₃^t²Y₄_|
|^t²Y_n

[p.135] The longitudinal design is more effective than the cross-sectional. If we are concerned with disproving the hypothesis that x is the cause of y on a common-sense level, the demonstration of the superiority of the longitudinal design could be carried out like this. In the cross-sectional design we would order the individuals according to their values of x. If the hypothesis is correct, the values of y should follow the same order after this procedure. In the longitudinal design we may first carry out the very same test as in the cross-sectional design. In addition, we can perform a second test. We can compare the individuals with regard to x at the two different times, t₁ and t₂. If we arrange them in proportion to the extent they gained or lost in x between t₁ and t₂, the hypothesis states that their gains or losses in y should follow the same order. It should be realized that the latter test is different from the former. It is conceivable that the first test in the longitudinal design may make us inclined to accept the hypothesis while the latter test rejects the hypothesis. Since the first test is the one employed in the straight cross-sectional design, we conclude that longitudinal designs are more sensitive than cros-sectional designs. The former stand a better chance of disproving the null hypothesis. The latter, however, are far more common in sociology.

Longitudinal designs can be further subdivided into ‘retrospective’ and ‘prospective’ designs. These designs [p.136] represent two somewhat different ways of performing a test about determinant and result. We may observe what our proposition has called the result and then inquire as to whether it was preceded by the hypothesized determinant. Let us call this procedure, which advances from the establishment of effects to the establishment of causes, the retrospective design. On the other hand, we can observe what our proposition terms the determinant and then investigate whether it is followed by the hypothesized result. This procedure, which goes from the establishment of causes to the establishment of effects, we may term the prospective design.

X₁ ¾ Y₁X₁ ¾ Y₁X₁ ¾Y₁X₁ ¾ Y₁_{|          |
|          |
|          |}X₁ ¾ Y₁

In order to evaluate the retrospective and the prospective design, let us again assume that we have the hypothesis, “x is a determinant of y.” If all cases in our population have both the properties X and Y, it is plain that both the retrospective and prospective design [p.137] would confirm the hypothesis. Suppose, however, that some of the cases fail to show the property Y, and let us indicate such a case by writing X¾ O:

Pattern I:

X₁ ¾ Y₁X₂ ¾ Y₂X₃ ¾ Y₃_{|              |
|              |
|              |}X_i  ¾  Y_iX_i+1 ¾  O
X_i+2 ¾  O_{|              |
|              |
|              |}X_n ¾  O

If our data show this pattern, the retrospective design would make us accept the hypothesis, since Y is always preceded by X. However, using the prospective design on this pattern of data, we reject the hypothesis, since all X are not always followed by Y.

The case is reversed when some of our cases fail to show the property X: [p.138]

Pattern II:

X₁ ¾ Y₁X₂ ¾ Y₂X₃ ¾ Y₃_{|           |
|           |
|           |}X₁ ¾ Y_i        O  ¾  Y_i+1
        O  ¾  Y_i+2_{|          |
|          |
|          |}   O  ¾  Y_n

Using the retrospective design on this pattern of data, we would reject the hypothesis, since all Y are not preceded by X. The prospective design, however, would make us accept the hypothesis, since all X are followed by Y.

These peculiarities become important for research evaluations once we realize what kind of causal relationships these patterns allow us to assume. In Pattem I we are free to assume that x is a necessary and contingent cause of y. In other words, x is essential, but so are other factors in order to produce y. The retrospective design is, then, adequate for the acceptance of the hypothesis about a necessary but not sufficient cause. [p.139]

In Pattern II we are free to assume that x is a sufficient and substitutable cause of y. In other words, it is true that y is a result of x and that y may also be produced by other factors than x. The prospective design, then, is adequate for the acceptance of the hypothesis about a sufficient but not necessary cause. When a hypothesis is accepted by the retrospective design and accepted by the prospective design, we may deal with a sufficient but not necessary cause. When both designs make us accept the hypothesis, we may have a necessary and sufficient cause.

Quantification and statistical analysis are helpful in making the decisions discussed above. However, we should make clear that the use of statistics is no substitute for theorizing. Churchman concludes as follows concerning the proper place of statistics:

One cannot simply take a set of data, make certain distribution hypotheses about their populations, and proceed to a statistical test; one cannot do so and expect a meaningful answer will be the result. To paraphrase Kant, statistical tests without theory are blind: no general results can be asserted, no predictions made unless one assumes that the statistical hypotheses are consequences of a general theory within which prediction can be made independent of specialized restrictions…. We may therefore take the following to be the criterion for the meaningfulness of statis-[p.140]tical tests: every statistical hypothesis should be a consequence of a formal theory of nature. ⁸

Nor do we believe that statistics is the only acceptable method of evaluating a test of a proposition. Non-quantitative methods, under certain circumstances, give equally plausible or more plausible results than some quantitative methods. The main reason for the use of quantitative variables, treated statistically, is that we obtain through them a quantitative expression of the plausibility of the null-hypothesis. As is well known, we must be aware of two kinds of error in testing a null-hypothesis:

Errors of Type I:	The null-hypothesis is actually true, but we reject it on the basis of our test. In other words, a false hypothesis is accepted
Errors of Type II:	The null-hypothesis is actually false, but we accept it on the basis of our test. In other words, a true hypothesis is rejected.

The use of statistical significance tests renders a measure, expressed as a level of probability, which can be used in evaluating the risk of making errors of Type I. Not so that this significance level is the probability of making errors of Type I, but by consistently applying a given significance level, such as the .05 level, we know [p.141] that in the long run we have rejected only 5 per cent of the true null-hypotheses.

Control. The problem of the ruling-out of alternative hypotheses is known as the problem of ‘control’ in a verification enterprise. We should distinguish between the control of known alternative hypotheses and the control of unknown alternative hypotheses. The best method of verification controls both known and unknown alternatives. A method of this extraordinary kind does exist and is known as the ‘experimental’ design.

The experimental design controls alternatives by producing the hypothesized determinant and by randomization of known and unknown factors. The experimenter does not merely observe what his hypothesis assumes as the determinant, but, in addition, himself produces it. In an effort to obtain a base line for measuring the possible effect, he would use a minimum of two groups in his experiment, one in which the determinant is produced — the experimental group — and one in which nothing is done — the control group. In order further to control the influence of alternative determinants, he assigns subjects at random to the experimental group and the control group. By so doing, he obtains maximum likelihood that the groups are similar in all respects except one: the introduced determinant in the experimental group. Accordingly, he is reasonably confident that any difference between the control group [p.142] and the experimental group is due to the introduced determinant.

Experimental designs may be longitudinal, involving repeated measurement of the same subjects. These and other more effective designs have developed in close connection with the statistical techniques accompanying analyses of the results — notably, analysis of variance. The latter has also provided the possibility of testing several hypotheses in the same experimental design.

The advantages of the experimental design, however, rest with the possibility of a random assignment of cases to the experimental and control groups and on the possibility of producing what the working hypothesis terms the cause. Unfortunately, in sociology we rarely have these possibilities. Certainly many factors are intentionally introduced into a society by politicians, educators, welfare agencies, and so forth. But these phenomena are seldom or never introduced, because they are termed causes in a scientific social theory. Furthermore, when compulsory education, socialized medicine, public housing projects, and like measures, are introduced into a society, the very complexity of the new phenomena does not make them suitable as indicators of concepts of a theory.

In the second place, we can rarely introduce randomization of the persons supposed to enjoy these intentionally produced phenomena without violating strong [p.143] moral sentiments. As to the social programs of the welfare state, Chapin makes the comment:

The conventional method of equalizing factors that are known and also unknown (by R. A. Fisher’s design of experiment) is to select at random both the experimental group that receives treatment and the control group that serves as a reference group for comparison. In social research the program of social treatment cannot be directed toward a randomly selected group because the prevailing mores require that this treatment be directed to a group of individuals who are eligible because of greater need. Thus precise control of the unknown is impossible and the only factors that can be controlled are factors that are known to be in the particular social situation because of previous studies. ⁹

It seems that this inability to satisfy the conditions for profitable use of the experimental design would definitely curtail the sociologist’s prospect to verify his theories. However, the situation is by no means disastrous: sciences like meterology and astronomy have verified theories without employment of the experimental method.

For control of alternative hypotheses, the sociologist is to a large extent dependent on what might be called [p.144] pseudo-experimental designs. These designs control determinants known from alternative propositions, but, unlike the experimental designs, these designs cannot control unknown alternatives.

The most commonly used method in sociology for control of known alternative propositions is multivariate analysis, which has been formalized by Paul Lazarsfeld. ¹⁰ Skill in its use has become essential for most sociological research. The technique controls alternative propositions by testing the hypothesis in subsamples that are homogeneous with respect to the determinants specified by the alternative propositions. It can be used to control all known alternative determinants, provided the sample used is large enough.

The simplest relation between two variates x and y is a fourfold table:

[p.145] To discover whether a third variable, z, accounts for any of the relations found in such a table, we break it into two parts:

If the relation between x and y still holds in all subclasses of z, we may retain, for the time being, our trust in the proposition that x affects y. To this kind of design many new alternative determinants can be added, and it works equally well for qualitative and quantitative varieties.

However, the advantages do not end here. We can tabulate:

and also: [p.146]

The purpose of these tabulations is to discover the actual link between the three variables. It would carry us far to review all the rules of interpretation involved here. However, if certain assumptions about the time lag between the variates can be made, it is possible to use such tabulations to disentangle a wide variety of causal chains, as shown in the adjoining diagram adapted from Dahlström.¹¹

Another method of pseudo-experimental control is that of matching.¹² An experimental group and a control group are made equal on some criteria by discarding cases in one group for which no “twin” can be found in the other group. One disadvantage of this procedure is that the matched groups so obtained are not representative of the original groups. We do not quite know to what population the results can be generalized. Control in pseudo-experimental design can be obtained through [p.147] the use of other statistical adjustments. Various applications of the multiple regression approach can be made, provided variables fitting the rather rigid assumptions are used. The most common methods are those of par-[p.148]tial correlation and analysis of covariance. These methods become rather laborious if the number of factors to be controlled is more than three or four.

Experimental designs and pseudo-experimental designs may be cross-sectional or longitudinal. We have already pointed out that longitudinal designs are more effective than cross-sectional designs and that experimental designs are more effective than pseudo-experimental designs. We can now reach a typology of designs:

		The test of the null-hypothesis
		Cross-sectional	Longitudinal
The control of alternative hypotheses	No control
	Pseudo- experimental
	Experimental

The closer a design comes to the longitudinal experimental the better it is. However, we know little or nothing about how to evaluate crosswise combinations of the two criteria. We have no way in which to tell whether a pseudo-experimental longitudinal design (such as a panel with multi-variable analysis) is as effective as the cross-sectional experimental design (the conventional laboratory experiment).

Most quantitative sociological research — particularly [p.149] public opinion polling — is cross-sectional without controls. Most non-quantitative sociological research — the case studies, for example — is longitudinal without controls. A superior user of the survey method introduces pseudo-experimental controls in his designs; perhaps he makes repeated interviews with the same group, thus rendering his design longitudinal in addition. This is about as far as field studies can ever reach in precision. A person using non-quantitative methods can also use pseudo-experimental controls. Max Weber attempted to find situations in the history of Asia that resembled the situation in urban Europe after the Reformation, with the exception of the religious factor present in Europe. ¹³ His findings became additional evidence for his hypothesis that Protestantism played a role in the creation of capitalism. Such a design should count upon a fair plausibility.

A limited range of problems of sociology is open to the experimental design. The results from methodologically adequate experiments in group dynamics should be viewed as highly plausible, and ingenuity should be encouraged when bringing problems into the laboratory. Since symbolic interaction is a major realm of sociological study and language variables are easily taken into the laboratory, it should be possible to use experi-[p.150]mental design to a greater extent than is now the case. The very fact that laboratory experiments have the higher verificational power is, however, no sufficient reason immediately to bring all research problems to the laboratory. Lippitt thinks that it is often most strategic to start with a survey about a problem, go on to a field experiment, and then subject the problem to a laboratory experiment. ¹⁴

The Composite Judgment of Acceptance or Rejection

We have now completed our discussion of some of the various components of a decision to accept or reject a proposition in sociological research. In looking back at the complexities of evaluating internal and external validity of the indicators, their precision and objectivity, the representativeness of the sample, the scope of the data, the control of alternative hypotheses, the fit between predictions and observations, etc., one conclusion stands out: No presently known mechanical or mathematical device can help the sociologist in his decision to accept or reject a proposition; only good training and much experience can guide him. [p.151]

How stiff should our criteria be? The ideal of science prescribes standards that few, if any, concrete research projects ever meet. The surest way of damning any research report is to compare it with the ideals of science. The best way of evaluating a research report is to compare it with other research reports, the most reputable ones in our field. Looking at the most reputable specimens of sociological research, we find, not unexpectedly, that standards vary from place to place, from time to time, from topic to topic. What is acceptable at Columbia University may be unacceptable at Stockholm University; what was acceptable in the 1930’s is unacceptable in the 1960’s; what is acceptable in macro-sociology is unacceptable in micro-sociology. To test satisfactorily a single given proposition by means of a sociological research project is clearly extraordinarily difficult. To a considerable extent, sociological research is the art of the possible, and it must be judged accordingly.

In the last analysis, the verification enterprise is a comparison of two broad classes of sentences, those in a theory and those about indicators and data. They should not contradict each other, nor vary independently of each other, but be in consonance. It does not make any difference whether the theory has preceded the research or vice versa; which class of sentences was written first is irrelevant. But they must agree according to the set of rules that we have tried to make explicit. In-[p.152]spite of these rules, however, there is no specific amount of supporting or contradictory data that automatically makes us accept or reject a theory.

All theoretical propositions make claims that go far beyond any data that scientists cite in their support. Everyone knows that this is true for the speculative propositions for which we at best can cite informal illustrations. But in large measure this is also true for propositions that are well established laws. Even the latter are actually tested only with an in infinitesimal fraction of all cases to which the law applies. “Physicists seem to be satisfied with far fewer observations than logicians would expect them to make; one finds in practice none of that restless accumulation of confirming instances which one would expect from reading books on logic.” ¹⁵ As soon as one takes the step from ordinary propositions to theoretical ones, there is a change of scale: from the summit of theory, a one-story building with data does not look very different from a fifteen-story building with data. And neither can reach more than the smallest fraction of the way to the summit.

One often asks if under such circumstances one should bother with theoretical propositions at all. Why not stay with ordinary low-level hypotheses and find-[p.153]ings, in other words, stay where the propositions do not go much beyond the research data? The answer to this question depends on what we see as the goal for the scientific enterprise and how we want our own contribution to be judged. No one has stated the alternatives better than Malewski:

If the main thing in science were to avoid formulations, which could turn out to be untrue even in part, we should always have to make our propositions, as little general as possible. However, such knowledge limited or almost limited to summaries of what was observed would have very low informative value and would permit a very limited range of prediction. Consequently there are numerous investigators who believe that our knowledge of the regularities of human behavior can be far more effectively promoted by formulating hypotheses, as general as possible, if only these hypotheses meet the requirements of testability and agree with earlier findings. Theoretical knowledge is considered here not as a set of infallible and absolute truths, but rather as a system of hypotheses which present a challenge to other investigators who through gradually modifying them make them somewhat more adequate hypotheses. Accordingly, the process of building up a science is one where the various hypotheses and their systems, while offering an approximate picture of regularities of human behavior, inspire further investigations which lead to the formulation of more and more adequate systems of hypotheses. In assessing from this angle a scientist’s [p.154] contribution to the body of theoretical knowledge the decisive factor is not so much whether his hypotheses have been maintained entirely unchanged, but the role his hypotheses have played in the development of theoretical knowledge. Personally I fully accept such a view of science.¹⁶

A consequence of this view is that, to the theorist, a huge accumulation of supporting evidence is hardly more impressive than a few strategically selected cases. This may, of course, lead a theorist to such a cavalier attitude toward data that he looses touch with the work of his discipline and, indeed, with science. But short of such failures, it gives him the freedom he needs to move on with his work.

In due course, theorists tend to develop confidence in certain theoretical propositions in spite of the fact that they go beyond available data. Their internal consistency and their ability to account for a selection of very different events, some of which may be paradoxical or puzzling to common sense, justify this confidence. Once this confidence has emerged in a set of propositions, the theorist takes in stride a fair amount of observations and findings which seemingly contradict his propositions. Theories of genetics, theories of intelligence, theories of sub-culture, and the methodology of IQ-testing, all combine into the proposition that “all races possess the abil-[p.155]ities needed to participate fully in the democratic way of life and in modern technological civilization.” ¹⁷ The social scientists have found this conclusion compelling in spite of the fact that virtually all comparisons of IQ-scores of White and Negro Americans show that the latter have lower average scores. They have enough confidence in their theories to know that the lower score is due to extraneous factors that have nothing to do with race.

Once confidence in a proposition has developed, contradictory findings are not a signal to abandon or recast the proposition, but to search the extraneous, undetected factors that operate in a particular situation to obscure the impact of the proposition. The mature scientist acts so as to save the theoretical propositions in which he has developed confidence; this is the normal attitude in all sciences. One elaborates and amends established theoretical propositions that are contradicted by data; one does not abandon them. The student who, when taught a new sociological proposition, triumphantly says, “But I know a case in which it does not fit,” thinking he has killed a giant, is not really as smart and scientific as he thinks he is.

Of course, theories do change in confrontation with research, also in other ways than in form of amend-[p.156]ments that save a cherished proposition. When an accumulation of negative findings have occurred, and someone finds a new set of postulates that provides for a more regular agreement between predictions from theory and observations of events, the time is ripe for a change of theory. Then we have a genuine scientific revolution. Characteristically such a revolution tends to involve at least one very fundamental postulate and a whole set of propositions related to it. The new postulate that thus gains the seat of honor is then accorded much benefit of doubt in the face of adverse evidence, and the process starts over again.

One must be knowledgeable, impartial, and reasonable in attempting to change a theory, whether by amendment or revolution. One needs also a very delicate balance between tolerance of dissonant findings and a sense of time for change. For there is no standard answer to the question how much evidence is needed to make or break a theory.

Notes to Chapter 7

1. Louis Guttman, “The Problem of Attitude and Opinion Measurement,” in Samuel Stouffer (ed.), Measurement and Prediction, Princeton: Princeton University Press, 1950, pp. 57-59.

2. Neal Gross, et. al., Explorations in Role Analysis, New York: John Wiley and Sons, 1958, ch. 7.

3. This analysis is not possible when the expressions to be analyzed are statements that are neither true nor false. Indicators of attitudes, values, and role-prescriptions belong, for example, here. Expressions like:

“I like X”
“X is good”
“Buy me X”

are, according to the philosophy we assume, neither empirically true nor empirically false. Accordingly, statements indicating attitudes, values, and norms are neither true nor false.

4. Alice S. Kitt and David B. Gleichner, “Determinants of Voting Behavior,” Public Opinion Quarterly, vol. XVI, 1950, p. 407.

5. Gösta Ekman, Reliabilitet och konstans, Uppsala: Almqvist & Wiksell, 1947. Our definitions of the components of reliability differ somewhat from those given by Ekman. The differences are due to our desire to deal with the reliability of any indicator, not only of psychological tests.

6. Bronislaw Malinowski, Sex and Repression in Savage Society, New York: The Humanities Press, 1927.

7. James C. Coleman, Elihu Katz and Herbert Menzel, “The Diffusion of Innovation Among Physicians,” Sociometry, vol. 20, (1957), pp. 253-270. A fuller report is contained in a forthcoming book by the same authors.

8. C. West Churchman, Theory of Experimental Inference, New York: The Macmillan Company, 1948, p. 218. (Italics in the original.)

9. F. Stuart Chapin, “Experimental Designs in Social Research,” American Journal of Sociology, vol. 55, 1950, p. 402. (Italics in the original.)

10. Paul F. Lazarsfeld, “Interpretation of Statistical Relations as a Research Operation” in Paul F. Lazarsfeld and Morris Rosenberg (eds.), The Language of Social Research, New York: The Free Press of Glencoe, 1955, pp. 115-125.

11. Edmund Dahlström, “Analys av surveymaterial” in Georg Karlsson, et al., (eds.), Sociologiska metoder, Stockholm: Svenska Bokforlaget, 1961, p. 193.

12. F. Stuart Chapin, Experimental Designs in Sociological Research, New York: Harper and Brothers, 1947.

13. For a brief summary of the variables that enter into Max Weber’s comparative studies of world religions, see Reinhard Bendix, Max Weber: An Intellectual Portrait, Garden City, N.Y.: Doubleday, 1960, ch.8.

14. Ronald Lippitt, The Strategy of Socio-Psychological Research” in James G. Miller (ed.), Experiments in Social Process, New York: McGraw Hill Book Company, 1950, pp. 17-30. Cf. Leon Festinger, “Laboratory Experiments: The Role of Group Belongingness,” ibid., p. 33.

15. Stephen Toulmin, The Philosophy of Science, London; 1953, p. 110.

16. Andrzej Malewski, “Two Models of Sociology”, The Polish Sociological Bulletin, vol. 1, 1962, p. 21.

17. Resolution passed by the American Anthropological Association in November 1961, reported in the Fellow Newsletter, December 1961, p. 1.

[p.157]

8. The Confirmation of Complex Theories

As we have seen, the confirmation of a proposition is a complex and rather tedious enterprise. Since the proofs are so long and life is so short, it is essential to devote our research efforts to hypotheses that are strategic. Some of the more sophisticated ways of ordering our propositions can aid us in selecting the most strategic ones and guide us in the expenditure of research efforts. It sounds paradoxical, but it is actually easier to test systems of propositions, that is, theories, than single propositions.

This holds primarily for theories organized as matrixes, chains, or axiomatic systems. But such simple modes of ordering propositions as inventories of determinants and results also aid the strategy of research. Such inventories give us listings of factors to include in an empirical study. It is important to make sure that the recurrent variable — the result in the case of inventories of determinants, and the determinant in the case of inventories of results — is given the best possible measure-[p.158]ment and that special research efforts are made in constructing its indicator. The reason for this is simple: this variable occurs in all tests, and if it is faulty the entire research effort is wasted.

Using the more sophisticated ways of ordering propositions, we employ the strategic principle of testing those propositions that have the greatest pay-off value in the form of deduced additional propositions. However, in any concrete instance, we will probably work with theories in which some of the propositions are already adequately supported by data. Here we may proceed differently. First, we assess the amount of support that past research gives to each proposition. Second, we make selections of unsupported propositions that, in combination with each other and with supported propositions, can derive the largest number of unsupported propositions. The shortest selection then becomes the one with the highest research priority. However, the shortest selection may not contain propositions that are easily subject to test, and we may for practical reasons consider some of the longer selections. In short, we make compromises between the difficulty of testing a proposition and its deductive power. In this way we get the most out of our research effort, the most in the form of direct or deduced support to previously unsupported propositions in our theory. In no instance should we have to test every proposition in a well-organized theory. [p.159]

Let us review some of the interplay between the ordering of propositions and the efforts of research. We may use an axiomatic theory as our example, since an axiomatic theory best illustrates the advantages we can obtain from theorizing when doing research.

Axiomatic Theories and Research

We have earlier illustrated some modes of axiomatic theories. We already know that this type of ordering of propositions serves the theorist in one essential way: it spells out all propositions implicit in some propositions. Now let us see how it serves the researcher.

Let us assume that we have reviewed or conducted research on a number of social groups with respect to (a) the number of associates per member in the group; (b) the solidarity of the group; (c) the consensus of the beliefs, values, and norms in the group; (d) the division of labor in the group; and (e) the extent to which persons are rejected (excluded) from the group when they violate group norms. Assume that these variables were found to be related in the following way:

The greater the division of labor, the greater the consensus.
The greater the solidarity, the greater the number of associates per member.
The greater the number of associates per member, the greater the consensus. [p.160]
The greater the consensus, the smaller the number of rejections of deviants.
The greater the division of labor, the smaller the number of rejections of deviants.
The greater the number of associates per member, the smaller the number of rejections of deviants.
The greater the division of labor, the greater the solidarity.
The greater the solidarity, the greater the consensus.
The greater the number of associates per member, the greater the division of labor.
The greater the solidarity, the smaller the number of rejections of deviants.

Let us assume that our research suggests that the link between determinant and result in these propositions is necessary and reversible.

These propositions can be ordered axiomatically in a variety of ways. If we select as postulates the last four findings, we obtain a somewhat distorted version of Durkheim’s theory of division of labor.¹ Let us restate them with roman numbers:

The greater the division of labor, the greater the solidarity.
The greater the solidarity, the greater the consensus. [p.161]
The greater the number of associates per member, the greater the division of labor.
The greater the solidarity, the smaller the number of rejections of deviants.

These four propositions can be used to derive the other findings which thus become theorems. I and II render (1); I and III render (2). II and (2) render (3); II and IV render (4) . I and IV render (5), and III and (5) render (6) . The ten findings were reduced to four. ²

This illustrates the first virtue of theorizing for the researcher: a theory can be used to provide the most parsimonious summary of actual or anticipated research findings.

Suppose now that we did not ourselves conduct all the ten studies that resulted in our findings. We have instead ten different researchers who do not know of one another and who, independently of one another, each confirm one of the ten propositions. On the basis of their investigation, they have some, but not much confidence in their findings. They know how hard it is to confirm a single proposition. Let us, for the sake of argument, say that this confidence can be represented by a probability of .85; they and their colleagues judge that there are about 85 chances out of 100 that they have really hit upon something true.

Suppose further that we have a more theoretically [p.162] oriented sociologist, who formulates a theory like the one above, and that he has performed exactly the same tests as our ten researchers, and that he, too, assigns a plausibility of .85 that each single test supports its hypothesis. However, when this theorist now talks of any of the ten propositions he can claim that their plausibility goes way beyond .85.

The reason for this claimed gain in probability is that the scientist working with theory — although he performs the same test as the scientist without a theory — also verifies several implications of his hypotheses. His procedure is practically identical with replications of a statistical test. The evidence from the tests of the implications reflects on the hypotheses as additional support according to a well-known law of probability calculus.

This virtue of the axiomatic theory can be used also in another way. Suppose our theoretically oriented researcher is satisfied with a probability of .95. To obtain this he may verify only a selection of hypotheses — for example, the first five or six in our list. If he establishes them with a probability of .95, he can claim that his whole theory has about the same probability. Through the use of his theory he has saved a great deal of experimental work. The amount of probability transferred is a matter of some debate. Most writers hold that a deduction carries the same probability as the proposition from which it is deduced. However, in sociology we should not claim too much from the transfer of proba-[p.163]bility, since our deductions are not too precise, so long as our concepts are defined in normal prose and the deduction rules of ordinary language are used.

Thus we claim as the second and cardinal virtue of theorizing for the researcher that a theory can be used to coordinate research so that many separate findings support each other, giving the highest plausibility to the theory per finding.

Here, then, is the reason why it is comparatively hard to confirm a proposition but comparatively easy to confirm a theory. We can give token empirical support to any of our propositions. A theory then can coordinate these modest supports into high support for its postulates.

A further advantage of a theory is that we can, at any stage of the verification enterprise, figure out what parts of the theory are confirmed and what parts remain as uncertain hypotheses. This is particularly useful when we want to economize our research efforts by locating research topics that will contribute most to the confirmation of a theory.

Using the same example as before, assume that we want to test whether division of labor leads to greater solidarity. Assume further that we do not have indicators of division of labor and solidarity for the same collectivities, nor can we conceive of any within the limits of our research budget. We have, however, informal or formal observations indicating that greater solidarity [p.164] results in few outright rejections of deviants. Our reasoning is, then, the following. We want to test by implication:

The greater the division of labor, the greater the solidarity.

We may assume:

The greater the solidarity, the smaller the number of rejections of deviates.

Hence we need to test:

The greater the division of labor, the smaller the number of rejections of deviates.

We find, thus, that we can test the latter hypothesis in order to give support to our postulate that division of labor leads to solidarity. The hypothesis thus selected is easier to test. We can easily coordinate its determinant and result to available indicators: for example, the division of labor may be indicated by the number of occupations in a society, and the number of rejections of deviants is indicated by the statistics on the persons executed, or exiled, or confined to correctional institutions for a long time.

Thus we find a third virtue in theorizing for the researcher: a theory can be used to locate the most strategic or manageable propositions for testing. [p.165]

Finally, let us consider an instance when our research fails to support a proposition which is part of a theory. Let us assume, for example, that research proves theorems (1), (2), and (3) but conclusively fails to support our proposition (5), that: The greater the division of labor, the smaller the number of rejections of deviants. Apparently, something must now be wrong with our theory, and the question arises which parts of the theory would now have to be rejected. We find this by deriving our postulates from our theorems, including the theorem that was proved wrong. We note that the false theorem was derived from Postulates I and IV. Hence either, or both, of these are false. However, as mentioned, we have good evidence for theorems (1), (2), and (3). Consider first (1) and (3) .

The greater the division of labor, the greater the consensus.

The greater the number of associates per member, the greater the consensus.

Hence we obtain:

The greater the division of labor, the greater the number of associates per member;

which, combined with (2) renders —

The greater the solidarity, the greater the number of associates per member; [p.166]

The greater the division of labor, the greater the solidarity;

which is Postulate I. Thus our findings support Postulate I, and the falsehood is thus localized to Postulate IV. We must thus drop Postulate IV from our theory and also all theorems derived by means of Postulate IV — i.e., theorems (4), (5), and (6) . The rest of the theory is still tenable.

Here, then, is a fourth virtue of theory for the researcher: a theory provides a limited area in which to locate false propositions when a hypothesis fails to meet an empirical test.

To these four advantages could be added others. The ones we have reviewed seem to me to be the most relevant for the present state of sociology. They are so important that no sociological researcher can afford to be ignorant of theory construction.

Testing Total Theories Through Their Gross Predictions

An impressive way of testing a theory is to use its component propositions to make one joint prediction and to demonstrate that this is an accurate prediction. Scientists in a hurry and with a flair for the spectacular have done this in several instances, and the theories so tested [p.167] have become accepted by their colleagues. However convincing this method may appear, it always contains elements of risk: several wrong premises may, of course, render the correct prediction. A critical colleague is never quite sure of the solidity of theories confirmed in this fashion. However, since we sociologists are in a hurry to deliver something else than promises and hopes to the society that supports us, a moderate encouragement of this procedure may be in order.

The joint predictions from several propositions can be arrived at through a careful use of theories phrased in ordinary language. Since deductions are sometimes complex, it may be most efficient to restate the propositions as mathematical equations and let the gross prediction be a solution to a series of equations. This is not because mathematical language adds anything of substance to theoretical propositions; it does not. However, mathematical language has stricter rules for making derivations than does usual scholarly prose, and the derivations needed for gross predictions may be complex and in need of this extra precision.

Another way of making gross predictions is the use of allegories. These allegories, or simulations, as they are often called, are either verbal, mechanical, or electronic. A “utopia” is a verbal allegory. The “census clock” in the lobby of the Department of Commerce, Washington, D.C., is a well-known mechanical allegory which predicts the size of the population of the United [p.168] States at any given time. It is a machine analogous to the simple proposition that any change in the size of population depends on a change in the number of births and deaths and/or the number of immigrants and emigrants. Each decennial census provides a check on the adequacy of this simulation. It has recently been found that electronic calculators can be wired to function as flexible allegories to social processes. So far, only a few electronic simulations have been tested as to their accuracy in gross predictions; however, developments in this field move very fast and carry great promise.

We may illustrate simulation procedures by using a form of verbal allegory. Consider the following stochastic propositions relevant to voting behavior:

Members of a primary group (family, friendship clique, informal work group) tend to vote for the same party.
The higher the occupational stratum of the members of a primary group, the greater the likelihood that the group will vote for a rightist (conservative) party.
The more a primary group takes on the style of life of an occupational stratum, the more likely it is to vote like other primary groups in this occupational stratum.
The higher the proportion of salaried or wage-earning (as opposed to self-employed) persons in an occupational stratum, the more likely are members in its primary groups to vote for a social welfare party. [p.169]

Add to these the following findings:

At present, families in the higher occupational strata in Western countries have fewer children than families in the lower occupational strata.
At present, the proportion of people in all occupational strata in Western countries who work for a salary or wage is increasing, and the number of self-employed is correspondingly decreasing.

Can we test all these propositions in one master stroke by checking how well they can account for the outcome of elections? To do this, we need facts about the location of primary groups in occupational classes, the style of life in these groups, the mobility between occupational classes, the trends away from self-employment, and information about differential fertility in these classes, and attitudes toward welfare policies among salaried and self-employed. All this makes a number of tables, which, happen to be available in a large Swedish study from 1954-55 (N = 2554). Furthermore, we need to make some assumption regarding the time element in several of the above propositions. This is not readily available and must be subject to some guessing. Finally, we need to know which parties are Leftists and Rightists and which parties are in favor of social welfare measures. An allegorical statement of the behavior of the electorate in Sweden may now read as follows: [p.170]

The age group that during 1950-55 has seen their children move into voting age consists of 51% Leftists (Social Democrats and Communists) and 49% Rightists (Conservatives, Liberals, and Agrarians). To avoid speaking in percentages, let us put it this way: We have 51 Leftist homes and 49 Rightist homes with children who become voters. Let us see what happens to them over the next four national elections.

During a few years, 54 children growing up in Leftist homes come of voting age. Fourteen of these children happen to acquire Rightist friends or workmates at an early age. Three of them cannot resist the attraction of these friends or workmates and convert to the bourgeois view before they cast their first ballot. Remaining are thus 51 who at their first election opportunity vote with the Leftists, their parents’ party.

Fifteen of these advance to become white-collar workers or entrepreneurs or marry into this group. Nine of them keep their past style of life; for example, they do not acquire an automobile. Three of them nevertheless adopt Rightist party preferences. Remaining are 48.

The remaining six who become white-collar workers or entrepreneurs buy a car, drink wine instead of beer with their food now and then, and four of them become bourgeois also in political aspects. Remaining are 44. One dies. Remaining are 43. However, four were gained from other parties, so their final count becomes 47.

During the same time, the following happens in the [p.171] 49 Rightist homes. One way of paying for their higher standard is to have fewer children than the Leftists. Only 48 grow to voting age in the bourgeois homes. Four of these children acquire Leftist friends or workmates, and one converts to the Left before his first election; 47 are left.

Ten fall below their parents’ station. Of them, four fail in their studies or marry below their station, but they keep their style of life and enjoy a car. A woman converts, however, to the Leftists. Remaining are 46.

Six of the others move from their parents’ middle or upper class into the working class; some abandon their family farm in the country and appear in the cities as workers or workers’ wives. Their incomes do not allow for a car. Two become Socialists. One dies, which leaves 4 for the Rightists. However, as mentioned, they gained ten from the Leftists’ parties, which gives them a final count of 53.

Not much seems to have changed, but the apparently calm electoral surface conceals a great deal. In the beginning the forecast for the Socialists seemed good. The parental vote had been 51 for the Leftists against 49 for the Rightists. The Leftists had more children than the Rightists, and the children voted at their first election, 52 against 49 for the Rightists. The difference could have been even greater. In the course of time, almost three times as many deserted the Leftists as the Rightists. Rightist politicians are thus more successful than Leftist politicians as political evangelists and get the upwardly mobile as converts in their nets. However, the Leftists keep afloat as more diligent midwives. The final score after 20 years be-[p.172]comes 47 against 53 in favor of the Rightists. During the years that this process has taken place, the socialist parties in this age cohort have declined from 52 to 47 and the bourgeois parties increased from 49 to 53.

So far, we have told this story assuming that the Rightist and the Leftist parties have identical policies about social welfare. This was true in 1952 and virtually true in 1956. Let us amend this now by admitting that the Leftists in 1958 and 1960 proposed more generous social benefits in the form of pensions than did the Rightists. This has consequences. In the course of every four-year period, one of our 20 families with a self-employed member of the middle class faces a major problem: the man gives up being on his own, sells his shop or farm, and starts working for someone else. The result is that more people during the period here considered become concerned over their pensions and are attracted to the welfare program of the Leftists. Half of these, or 2 persons, get to the point of voting for the Leftists. Furthermore, this greater generosity of the Leftists makes it a little harder to get converts from them among the upwardly mobile. Only 5 are gained for the Rightists instead of 10, as before. And it becomes easier for the Leftists to gain converts; 5 are gained instead of the 4 before. All this changes the balance of our age cohort:

	In 1968 if parties have identical welfare program	In 1968 if Leftists have more generous pension program
Rightists	53	48
Leftists	47	52

[p.173] In 1968, the all-important majority is solidly among the Leftists. If the parties had maintained virtually identical welfare programs, there would have been a Rightist majority by 1962; the more generous Leftist program of the late ’50s served to solidify and increase the Leftist majority.

This is, admittedly, a rather freely constructed story, which is anchored at some points but not at others in statistical facts from 1954-55. Its gross prediction is fairly accurate. The actual division of the popular vote for the entire Swedish electorate, not just our age cohort, is shown in the following table:

	Per Cent
	Bourgeois Parties	Socialist Parties
1952	49.6	50.4
1956	50.4	49.6
1958	50.4	49.6
1960	47.4	52.7
Source: Statistisk Årsbok 1962

Our allegory can be further used to reveal that even if the parties of the Right in 1962 change their mind — as has been done by some — and adopt the Socialist welfare plan, they will not gain a majority until the ’70s, if present rates of mobility, differential fertility, and spread of the middle-class way of life prevail. This type of conclusion indicates a key feature of sociological simulations. They can estimate outcomes of alternative [p.174] possibilities and give guidance in policy choice. And, once a simulation is established, one can keep it realistic by adjusting the rates of the variables involved and by adding new variables as they become relevant.

It would take us too far afield to discuss how simulations are done on an electronic computer. Suffice it to say that this way of formulating and testing theories can be done with ease and great speed on the standard models of electronic calculators that are now available.

Many persons can correctly predict an election, and the fact that a simulation gives a correct master prediction is no important confirmation of its truth. To test a simulation one makes many partial predictions which can be checked. In an election simulation these may be predictions of votes by sexes, by areas of the country, by different age groups. Only when a very high proportion — say, 90 to 95 per cent — of a large number of such partial predictions also prove accurate does one begin to trust the simulation. In reality, one achieves this degree of accuracy only by systematic tinkering with the variables.

Notes to Chapter 8

1. Émile Durkheim, De la division du travail social, Paris, Felix Alcan, 1893, (English edition available from The Free Press, New York). It should perhaps be stressed that nothing in this chapter is intended as a review or criticism of this classical work.

2. We ignore the complications posed by the task to reduce also the number of reversible propositions.

[p.175]

Concluding Remarks

Nothing in our discussion has indicated that it is impossible to obtain a sociological theory which is as well verified as theories in other sciences. We can only express the hope that theorizing and verificational studies contributing to this end will attract the attention of the sociologists of the future to the same extent that taxonomy and descriptive studies do today.

If anything stands out from our review, it is the realization that there are great difficulties in testing a detached hypothesis compared with testing a hypothesis integrated into a theory. This has some important implications for the strategy of advancing sociological knowledge: verification is easier in studies suggested by sociological theory than in other studies.

Yet most problems for social research at present are not suggested by theoretical sociology. They are suggested by the whim or wisdom of foundation officials; by clients who want sociological help to acquire a lar-[p.176]ger share of markets, commodities or votes; by journalists, clergymen, and others who choose to debate certain issues of the day as social problems. The typical sociological research project is an ad hoc study of topics suggested by non-sociologists. It is sheer accident when these topics can be integrated in sociological theory. The studies of these topics, accordingly, face difficult problems of verification. Moreover, they are not cumulative; the best that can be said for them is that they foster methodological advances.

By contrast, the studies suggested by existing theory are cumulative, and their problems of verification are modest. This state of affairs, when a theory guides the choice of research topics, is “normal science,” and it prevails until the research findings no longer seem in regular agreement with the postulates of the theory and alternatives are formulated. At such a point, it no longer suffices merely to amend and elaborate the theory; the scientific community goes through a “revolution” and the Young Turks emerge with a novel theory, which serves as a guide to novel research topics. ¹ Such has been the historical pattern of most sciences, and we would be bad sociologists if we assumed that our science would progress differently. [p.177]

Thus, the task for sociology is to continue with great dedication to sum up its knowledge in the form of theory and to use this theory to gain control over its research efforts.

Note

1. Thomas Kuhn, The Structure of Scientific Revolutions, Chicago: University of Chicago Press, 1962. It might be noted that ”normal science” serves the needs of practitioners more through applied theory than applied research.