Method, Generalization, and Prediction in Social Psychology
The data of social psychology lie in the field of social events, which include both the subjective attitudes and meanings as well as the overt behavior of the individual in interaction with his fellows. The major methods of the field fall really into two large classes: the quantitative or statistical, and the historical or case study. The former deal only with mass data considered as independent units in a continuum or as a series of attributes. The latter rest upon what Boas call's the "specific form" or the individual. Within the limitations of the method, predictions of a probability character have been worked out for certain types of social psychological material. The statistical approach, however, does not afford much predictive value for the individual case. Prediction in the field of case studies must deal with the individual. This may be undertaken, within limits, provided the psychological conditioning of the earliest years is relatively permanent in effects, and that the culture remains likewise relatively constant. It is evident that social psychology must not neglect, as its central focus, the meanings and attitudes of the individual, and if their study is unprofitable for generalization and prediction from the statistical method, then we may have to have recourse to the case method even though the latter may approach more to an art than it does to a science strictly defined.
The principal discussion in this paper concerns generalization and prediction, but since these matters cannot be treated without reference to method, some attention to basic methodological problems will be presented.
Social psychology may be defined as the study of the individual as he is affected by his fellows and as he, in turn, affects his fellows. It is clear that our subject matter is the individual. But since the precipitates of social interaction which we call culture play a part in the individual's behavior, we must deal not only with person-to-person interaction outside of any particular cultural framework but also with interaction as it is affected by the impress of culture upon the persons concerned.
If social psychology is to be a science it must conform to the criteria of other sciences. Science is used, of course, in two senses, as content or facts, hypotheses and generalizations, and as a logical method of exploring events. As Blumer says we cannot have a
( 21) "science without concepts." Nor can we have one without facts. Logical concepts clearly depend upon mental organization, but some naive persons forget that facts do not speak for themselves. These individuals suffer from what Walter Wheeler Cook so aptly calls the "pebble theory of facts," the notion being that the mind just picks up facts as one may collect pebbles in a basket. Facts depend, as any amateur philosopher knows, upon certain sensa or givens plus a meaning added from past experience.
Our concern in social psychology is in a train or manifold of events taken always from the standpoint of the observer of these events. Our particular set of events has to do with the individual in an interactional relation with other individuals. Thus we select only those events which throw light on our particular purposes, ignoring, as science always does, other factors of the totality of events. Out of the analysis of these events we hope to develop hypotheses and generalizations. Some workers wish to go farther and develop such generalizations which will help us control the world of social interaction. Others doubt that we can follow the footsteps of the natural sciences, which have so assisted us in controlling the material world, and do not believe prediction and consequent control of social events is possible.
The more optimistic view is held by no less a person than R. A. Fisher who discounts the old separation of natural from social science now that the former has moved away from earlier concepts of absolute laws of nature to conceptions based on statistical probabilities. We have come to realize that the operations of nature offer us regularity and hence predictability because, after all, the properties of nature "are in some sort the average or total of a large number of independent items of behavior." In other words we deal best with the events of nature as a set of independent variables whose average performance moves in one direction or another in terms of probable frequencies. From this Fisher is led to say: "On this so to speak, statistical view of the world, the meaning of law, and the possibilities s of the scientific application of ideas of natural causation, are the same in the physical and the social sciences.” 
Such a statistical view of social events, however, is only possible, according to Fisher, when the units under investigation are considered as independent. Moreover, these units must be, from the standpoint of the observer, reliable "in respect to the constancy of (their) reactions to outside circumstances." A great many studies have been made in the field of social behavior in which assumption of independence of units has proved effective in dealing with certain sorts of data. Reference to these studies will be made below.
On the other hand, even Fisher issues an important caution when he points out that such factors as public opinion and culture make it difficult to handle behavior variables as absolutely independent of each other. Other writers like Cooley hold that such factors as social organization and culture make a purely statistical view of the world absurd, because they enter into the individual's subjective trains of events and cannot be handled quantitatively.
Some social scientists, realizing the difficulties involved in dealing with culture, social organization, and subjective events, deny the possibility of predictive generalizations entirely. Among them are certain philosophers and many anthropologists and historians who hold that social and personal events are so unique and indeterminate that we must dismiss any hope of generalizations leading to prediction in the sense of recurrent regularities. The social event or the personal event therefore differs from the natural event which is recurrently regular and hence predictable. This extreme view in social science arose in part as a revolt against the so-called "scientific history" and against the alleged objectivity of the Spencer-M organ School with its overemphasis upon evolutionary generalizations. As Boas puts it, "It is our general experience that attempts to develop general laws of integration of culture do not lead to significant results." And again, "The attempts to reduce all social phenomena to
( 23) a closed system of laws applicable to each society and explaining its structure and history do not seem a promising undertaking."
Since social psychology, like ethnology and sociology, must deal in large measure with historical series of events, with events in retrospect, such a generalization about generalizations may come as a shock to some. Furthermore, in social psychology we deal with the individual as a "specific form," as Boas puts it. Ile says that no matter how much we may generalize the "generalizations will cling to certain specific forms." This implies for social psychology that since the individual is our primary object, prediction for the individual is well-nigh impossible.
The doubt about generalization and prediction for social psychology rests really upon the difficulties of getting at both overt and subjective events in retrospect. We certainly should not neglect the opportunity to use statistical devices in the study of those variables which lend themselves to such treatment, and from this we may hope for some practical use in prediction. The approach to our data from this standpoint we may refer to as the statistical or quantitative. On the other hand, we dare not dismiss as useless the problems of attitude and meaning, that is, the subjective events in the interactional configuration.
Before taking up the matter of generalization and prediction within each of these two approaches to social psychology, let us note briefly certain advantages and disadvantages of the respective methods. In such a brief statement one is likely to seem both dogmatic and too clear-cut in distinctions, but in order to offer adequate caution about prediction and control some such summary is necessary.
The advantages of the statistical method may be summarized as follows: (1) The units of measurement or counting must be carefully defined and fixed, and, hence, interchangeable. A review of recent research in our field shows that these seem best confined to the study of overt responses and to measurable aspects of the stimuli. The work of Willard C. Olson, Florence Goodenough, J. E. Anderson, Mildred Parten, and Dorothy Swaine Thomas, who have used the time-sampling technique, illustrate this method. (2) Some ap-
( 24) -proximation to an experimental program may be set up to examine various aspects of change in verbal and overt response. Some of the studies using the time-sampling techniques have carried on such work in child psychology. Other investigations approaching an experimental technique are those of C. W. Bird and A. D. Annis on planted newspaper content. So, too, L. L. Thurstone's research of the effect of certain motion pictures on children is particularly apt. Other such studies have been made or are projected by Lowell Carr, Clifford Kirkpatrick, and others. These new departures are extremely promising even though they are confined to very limited problems since it is increasingly evident that it is useless to measure any situation or behavior unless one can reasonably control both the situation and the individuals as subjects. (3) With proper sampling, which is so essential, a more adequate picture of the distribution of units is obtainable. (4) The statistical reliability of these measures may be fairly easily determined by well-known statistical devices. And in more careful work validity may be fairly well established. (5) Generalizations may be determined in terms of probabilities and correlations, and from this we may hope for certain predictive uses within the limited range of social organization and culture.
Certain limitations may be offered: (1) In the determination of units of measurement or counting there is a tendency to select only the simpler features of behavior, which behavior may not be as "significant" as other behavior which does not lend itself to being broken down into smaller units. There is a common assumption that all science proceeds by breaking down more complex wholes into simpler parts, but this process applied to complex social psychological data may neglect the very configurations of behavior and of situations which we wish to investigate. This is especially true in regard to so-called subjective or inner events. This criticism has been made of the time-sampling techniques by such men as James W. Woodard and E. B. Wilson. (2) Sometimes statistical studies have masked their real difficulties by combining many variables into single units or categories. This misleads the unwary reader and tends to false
( 25) conclusions. (3) Certain data are not obtainable in sufficient homogenous samples, and, moreover, the reliability and standardization of tests are difficult to establish. (4) Often statisticians pooh-pooh the case study method for being subjective, and yet the validation of many social psychological measures is quite dependent on subjective judgments, no more sound than judgments of validity regarding case materials. This is illustrate(] by the validation of moral and character tests by means of a few so-called "experts" on ethics or character training. The overused method of validating one set of measures by another set is open to errors hidden in the original validation of the first test. In the better studies, notably Thurstone's, many of these difficulties have been removed.
We have, then, in the statistical approach an extremely usable method for certain types of data. We may deal with overt action and the stimulating situation, leaving inferences to be stated in quantitative terms or developed from the use of non-quantitative data. From these methods may we anticipate any use of these types of studies for making forecasts of behavior? That is, may we hope at all for prediction and hence control? Let us look at some examples.
Given an adequate distribution of unit variables, say of intelligence scores, at a given time, may we predict that this same population will fall in the same arrangement of frequencies on the same scales a month, a year, or a decade hence? A number of extensive studies of intelligence scores, for example, show pretty conclusively that the relative change in position of the individual on the scale is slight. That is, true constancy of the intelligence quotient for the large bulk of school children seems pretty definitely established. There will be exceptions, some of them notable, definitely the many variables which determine performance on an intelligence test sometimes combine to make for radical changes, although for most children they do not seem to disturb the relative position on the scale very much. Whether we can get any such constancy in tests of emotions and social attitudes, we are not able to say. Their development is too recent. So, too, whether we find any such constancy in units of response in time-sampling research we do not know, but we expect it.
More important than these simple distributional constancies is the matter of the correlational techniques. If we find that intelligence scores and high-school grades give a certain correlation, may we predict what the grades of the student in college will be assuming intelligence to remain constant? Here again our results do not yield perfect correlations in time-sequences, but such investigations prove much better for prediction than off-hand guesses. Of course, these results hold only for large numbers in terms of probabilities. We cannot, for example, say that John Smith who may stand in the first 5 per cent of the distribution of high-school grades and in the first 5 per cent of test scores will retain the identical percentile rank in grades as a college Senior. The probabilities are high, other things being equal, that is, other things remaining constant or canceling each other out, that he may. But for the individual, other things do not remain constant. John may get the measles, or a girl, or lose interest and drop down in his work. We can only say that, these other variables remaining stable, he probably will stand in the first 5 per cent. Yet for the total student population we seem able to say that certain percentages of the students will remain constant on the scale. Thus for placing students in classes, for predicting withdrawals for low grades from college, this probability prediction has enormous practical significance. Again, whether the correlation of tests of nonintellectual factors, emotions, behavior traits, and unit habits will prove so valid for prediction remains to be seen. And yet, as J. E. Anderson remarks, if in an experimental set-up using the time-sampling technique where we place various pairings of two children and one toy in a room and find that one child in ninety-five times out of one hundred always makes off with the toy, we have a strong case for predicting that this child will tend to be dominant in other social situations whereas another child with only a record of 5 per cent dominance will not.
In the same manner we may make some fairly reliable probability predictions in relating social-economic status to all sorts of social behavior: delinquency, school success, and the like. Shaw's ecological studies of delinquency areas seem to confirm this. In this way we are in a position to make adequate changes in social-economic conditions in the hope that conduct may change to that which we label socially desirable.
One caution, however, must be repeated. These statistical predictions are built around mass data, not around single individuals, and we must not be misled into thinking that we can ever predict fully the individual case from this sort of treatment. It is very easy in this connection to draw an analogy from quantum physics about the "fine scale phenomena" to the complex social world in which again the individual as a unit looms so large and yet in which, due to the variables that play upon him, it seems so difficult to predict his behavior. We must not forget the caution of Fisher that only when we deal with our statistical data in the mass, as a group of independent variables unaffected by social organization, are we on safe ground. Hence we must not forget that when we use categories such as delinquency, crime, broken homes, and social disorganization in the stimulus series (the situations) we are masking perhaps dozens of variables in one category. So, too, when we speak of success in college work, of success of a parolee, of reaction to prejudiced stimuli, etc., we are likewise masking many variables in one category of behavior. If we assume, as we doubtless may in many instances, that sets of uncontrolled variables cancel each other out, or if we equate our materials in terms of carefully defined socio-economic or cultural status, age, sex, and marital condition by means of partial correlational techniques, we are laying the foundation for making certain probable predictions which may prove of practical value. Certainly for mass data these quantitative formulas are preferable to mere verbal guesses.
Turning now to the case-study or historical-genetic method, let us state briefly the advantages and limitations of, this approach before we take up the matter of generalization and prediction.
Among the advantages of this standpoint we may mention the following: (1) It gives a more or less continuous picture through time of the individual's interpretation of his own experience and often of that of others. As W. I. Thomas puts it, "If men define situations as real, they are real in their consequences." In other words, the nature of social reality is revealed only when we know the meaning which people put upon their experience. The researches of Thomas and Znaniecki on the Polish peasant, of Healy, Shaw, and
(28) Burgess on the delinquent and criminal, of Park and his students on urban personalities, of Grace Marcus on the relations of social worker to client, of Lasswell on political leaders, and of Lowry, Levy, and Kenworthy on child conduct cases, are sufficient to indicate the wide range of material used. (2) It furnishes a picture of past situations which gave rise to new meanings and new responses. This is particularly valuable in giving information as to crises which are significant in the development of new attitudes, meanings, and habits. Thus, the immigrant diary or letter may show the changes in situation, in attitude or meaning, and in response, which take place in America. Or the delinquent reveals to us his changed attitudes and behavior in the face of punishment or a foster home. (3) The psychoanalytic method, in spite of harsh criticism, is the most highly developed method of eliciting this material and in the bands of the enterprising Harold D. Lasswell has been applied to normal persons. This method furnishes materials on the earliest conditioning of the person to others which is not available by any other known method. (4) Repetitions of situations, meanings, and responses may be noted and used for comparative purposes in forming generalizations. Certainly such logical methods as those of agreement and difference, and even what L. L. Bernard, first, and, later, Clifford Kirkpatrick called an "informal statistical method," may be employed in working up generalizations. Thus Thomas and Znaniecki in their well-known "Methodological Note" have evolved certain formulas for describing in social psychological terms the social process. They say:
The fundamental methodological principle of both social psychology and sociology-the principle without which they can never reach scientific explanation-is therefore the following one:
The cause of a social or individual phenomenon is never another social or individual phenomenon alone, but always a combination of a social and an individual phenomenon.
Or, in more exact terms:
The cause of a value or of an attitude is never an attitude Dr a value alone, but always a combination of an attitude and a value. 
(5) Inferences and generalizations in this field rest upon an intimate knowledge of the situation and of the habits and attitudes of the persons interacting. There is furnished us a kind of moving-picture of events in which interpretation grows out of judgments made in a comparative manner. There is hardly any attempt to get at or to formulate standard units of measurement. Rather the thinking relates to the totality or configuration of situation and persons. Inferences are drawn from a kind of intuitive judgment, insight, or feeling for the whole social Gestalt. This is exemplified in the work of C. H. Cooley and of J. M. Williams whose generalizations rest rather in a verbalization of their impressions and comparisons, but which are certainly not expressed in quantitative form.
The limitations of the case study method are numerous and often seem to far outweigh the advantages. They may be summarized as follows: (i) The records are open to errors of perception, memory, judgment, and unconscious bias with a special tendency to overemphasize unusual events. Nevertheless the discovery of internal consistency in the record is believed to offset this difficulty in part. Sometimes, too, a check may be made by the use of records from social agencies, the school, or correctional institutions, or from the case records of other individuals in the same groups. (2) No interchangeable units of behavior or of the stimuli, to say nothing of units of subjective events, are used in these reports; hence it is impossible to make any quantitative check upon them without additional data, such as Stouffer used in his comparison of autobiographical statements on prohibition and responses to a standard test. (3) Sampling is usually neglected and generalizations may thus be false, but this may be due to lack of numerous records rather than to the nature of the data itself. So much data now used represents atypical cases that an adequate check-up with typical, normal cases in sufficient number is greatly needed. Moreover, the sheer difficulty of managing a large number of case records puts a further limit on the hope of very adequate sampling.
Before taking up the possibility of prediction in the case study field, something must be said of the relation of this approach to the problem of the subjective factors already noted. In social psychology we cannot escape the importance of the problem of meaning or sub-
( 30) -jective event. Any fundamental generalizations about social behavior must take into account the meaning which the individual gives the external situation and his own overt responses unless we wish to ignore those more complex levels of interaction which are the heart of social psychological analysis. To ignore the matter of meaning would be to reduce social psychology to the mere mechanics of the interaction of one bodily mass upon another which it would be the province of physics or physiology to undertake. If the statistician has no means of handling meaning quantitatively, especially in the time sequence, then we shall have to take refuge in some other approach. This the case study attempts to do. The problem is frankly one of the interrelation of subjective and objective factors in the total behavior situation. We wish, in other words, to correlate subjective meaning with overt response and with the stimulating situation.
It is interesting that many years ago Kulpe, Marbe, Ach, and others associated with the so-called Wurzburg School of Psychology tackled this problem in the experimental laboratory. I think social psychologists will find their work enlightening and perhaps even encouraging. For example, Ach in experiments on volition and various higher thought processes used precise reaction times and introspection with excellent results. Out of his work and that of others came a new view of the place of the Aufgabe, the determining tendencies or attitudes, of purpose or motives in the development of meaning, judgment, volition, and other higher thought processes, the very materials with which we are here concerned in a somewhat different logical setting. 
It seems to me that the analogy of our work in social psychology and the approach of Ach and others is very close. By statistical measurements of stimuli and of units of time applied to responses
( 31) we may secure something comparable to Ach's use of controlled stimulus and reaction time. The use of case study material affords something analogous to the introspective description of the process so essential to formulating the conceptual system which will satisfy our data and purpose.
When we come to deal with predictive categories in regard to the individual from the case study method, we are on somewhat different ground than when we deal with a mass of individuals measured on some unit scale. Not only have we but one case, with possibly other cases for comparative purposes, but the material is in non-quantitative form. Even the data on the situations to which the individual is exposed are in non-measurable form, although some may hope for improvement in this connection, with the development of more adequate methods of record-taking in schools, institutions, business, and industry. As we have just noted, moreover, so many subjective factors seem impossible of quantification. In fact, to most workers in the field, statistical quantification would indicate nothing because they are treating the situation, meaning, and the overt reactions in a time-series with the emphasis upon the individual as a unit.
With these cautions in mind, may we hope for any predictions founded on case studies? From any complete case story, such, say, as some of those collected by Shaw, Burgess, and Landesco, is it possible for us to predict how the individual will behave in the future? It seems to me that two things are involved: the constancy of the situation, and the constancy of the organism. The Freudians, in particular, are convinced that there are certain persistencies built up within the individual from inherited background and early learning and that from these we may state certain constants for the adult, provided that the psychoanalysis has been adequate. Lasswell has this in mind in his studies of political leaders. Burgess has taken this over in his concept of the "personality type" which, established from inheritance and early conditioning, tends to remain the same throughout life. Thus Burgess and Shaw see in Stanley, the Jack-Roller, what they call an "egocentric" personality with certain rather distinctive character traits. Even though the Jack-Roller is reformed, and is now a successful salesman, he remains this sort of fellow. They are able to state with considerable assurance that he will
( 32) remain egocentric and that he will do better in certain businesses than in others. Were they more courageous they might even venture to forecast his behavior in any given social situation provided it had to do with his interaction with others in a situation of dominance and submission. Can they go beyond these rather general guesses? Can they foresee details of Stanley's behavior twenty-five years hence? This is asking a good deal. The unpredictability of situations makes it impossible. Moreover, even granting that his egocentricity is deeply rooted, we are still not altogether certain of its permanence, especially in the light of our ignorance regarding the possible effects of somatic maturation and possible endocrine and other changes which may appear in middle life and later. Yet, in the light of long observations on psychopathic, neurotic, and even certain normal persons, the weight of evidence seems to favor the view that a probable prediction, not stated quantitatively of course, may be made of Stanley's future.
A nice problem is thus introduced, namely, the relation of personality type to social role. The latter seems to change in the light of social situations, while the former remains more constant. One may ask, is the social role of any person predictable, accepting the concept of social role as logically valid? This strikes me as being much more difficult to answer, partly because we do not yet have a very clearly defined set of categories for the social roles we play. To get at prediction from this angle we should know something of the likelihood of our Stanley remaining in one culture or another. We may say with some assurance-how much I do not know-that if Stanley remains in his present culture, he will-the personality type remaining constant-continue to have an accepted status. So, too, of another person we might feel certain that, if be remains in the culture of gangland, his social role will continue to fit this culture.
It is here that the anthropologist and historian enter again with their recurrent caution about the interplay of the culture and the individual. While the impress of culture upon the personality type has not been clearly worked out, there seems good evidence that for the most part the personality type is determined by what I have called the personal-social rather than the cultural factors. If this prove a correct assumption, then we may say that underneath the
( 33) variations of cultural conditioning there do lie basic social psychological patterns, the core of which seems to be established early in life and to fluctuate but slightly. On the other hand, the social role is so distinctly determined by the play of culture that we shall be forced to proceed with caution regarding constancies here. Even so, given a relatively stable and unchanging culture, the play of these forces upon the individual seems to be also somewhat predictable. We know that the boy and girl reared in the Chinese home of fifty years ago were going to take on certain attitudes toward marriage and their places in the family. In contrast, when at present external cultures have intruded themselves into China, such a prediction would be difficult to make. True, no problem in social psychology and sociology is more challenging than that of social change, still our ignorance of the interplay of divergent cultures upon the individual is so great that prediction in a period of transition is indeed a risk few would dare to take.
In summary then, let me say that, while I am not so pessimistic as some of our historians and anthropologists, I am not any too optimistic in regard to prediction. We have been overimpressed with the role of prediction and control in the natural sciences and suffer from a certain feeling of inferiority because we cannot produce the miracles of control in human behavior which we see around us in the material universe. If our data are different, our conceptual systems will be different. We may well find that we can get on toward certain generalizations, without recourse to the notions of prediction, even if one of our predictions is that there can be none.
Stated tentatively, however, I hope that prediction is possible under the following conditions: (1) within small segments of behavior where the variables are determinable; (a) within the limits of early psychological conditioning; (3) within restricted cultural constancies. Larger social events determined by the interaction of cultures and races, for example, seem to me impossible to predict and control, and, since the anthropologist and the historian deal with these larger events, there is no wonder they issue us a caution at all times to beware our predictive inferences lest they lead us into a cul-de-sac of nonsense.
( 34) For myself I cannot get as anxious as some folks who fear that, if our data cannot be quantified, we cannot have a science, at any rate above the level of verbal description. Let us reserve judgment on this point for the present. Suppose we do grant the contention of ardent statistical fact-finding researchers that without quantification there can be no science. If understanding or interpretation of motivation and meaning be the core of social psychology as some believe and if these things cannot be approached statistically or experimentally, then we may well have recourse to some other means of arriving at our goal. Perhaps some social psychologists lack the temerity to admit that art as well as science may throw light on our subject matter. I for one do not. I do not care whether the physicist, chemist, or biologist believes there can be a social science or not. Most of them are either indifferent or else convinced that there cannot be; so in any case why should we worry about the matter? Rather let us continue our work with the best tools at hand and leave to the future the question about a science or an art in the fields of social psychology.