Frequently we describe behavior not with verbs which specify action but with adjectives describing characteristics or aspects of action. Instead of saying, “He shook hands and said, ‘Hello, hello.’” we may say, “He was most cordial.” The adjective “cordial” is one of a list of about 4,500 English words compiled by Allport and Odbert which refer to more or less enduring traits of human behavior. If we add terms which refer to temporary conditions, such as “embarrassed” or “hazy,” the number is about doubled. Most of these trait-names are nontechnical words which we use in our daily affairs. They are an essential tool of the novelist; literature is, in fact, responsible for adding many of them to the language. By describing human behavior in characteristic situations, the storyteller creates convenient expressions for later use—such as “dog in the manger” or “a Daniel come to judgment.” The list has also been extended by the invention of such technical terms as “phlegmatic” and “melancholic” or, more recently, “inhibited,” “introverted,” and “cerebrotonic.”
Staying at a single letter in the Allport and Odbert list, a biographer might describe the behavior of a subject as follows: “There was a remarkable change in his behavior. Where he had been happy-go-lucky, he grew hesitant and heavy-handed. His natural humility gave way to a sustained haughtiness. Once the most helpful of men, he became heedless and hard-hearted. A sort of histrionic horseplay was all that remained of his fine sense of humor.” A passage of this sort tells us something important. If it were a description of an old friend, for example, it would prepare us to deal with him more effectively when we saw him again. But it may come as something of a surprise to discover that no behavior has actually been described. Not a single action has been mentioned. The passage might be describing a series of letters—of a colleague or business acquaintance, perhaps. On the other hand, it might be describing a wholly nonverbal scene from a ballet. It might concern a shopkeeper, a plant foreman, a salesman, a diplomat, a schoolboy—in short, any one of dozens of different kinds of people whose behavior would have nothing in common except those aspects to which the passage refers.
There are practical circumstances under which it is useful to know that a man will behave in a given manner even though we may not know precisely what he will do. To be able to predict, for example, that a proposal will probably be “received favorably” is valuable even though the specific form of the reception remains to be seen. Under certain circumstances everything else about the behavior may be irrelevant, and a description in terms of traits is then highly economical. But are terms of this sort valuable in a functional analysis? And if they are, how are they related to the variables which we have so far considered?
A common and unchanging property of the behavior of all members of a species would not usefully be referred to as a trait at all. It is only because people differ from moment to moment or from person to person that trait-names arise. We may look for the equivalents of traits in a functional analysis, therefore, by asking in how many ways we should expect a person to differ from other persons or from himself from time to time.
Differences in variables. Some differences are due to the differences in the independent variables to which people are exposed. Although we may be struck by the effect upon behavior, the original individuality lies outside the organism. Differences in experience between the “ignorant” and the “learned,” the “naive” and the “sophisticated,” or the “innocent” and the “worldly” refer mainly to differences in histories of reinforcement. Such terms as “enthusiastic,” “interested,” and “discouraged” describe the effects of different schedules of reinforcement. People are “inhibited,” “timid,” or “cowed” because of special contingencies involving punishment. The “discriminating” individual has made distinctions among stimuli which are not made by the “undiscriminating.” Differences in deprivation lead us to distinguish between the “voracious” and the “finicky,” the “libidinous” and the “sexless.” Differences in hereditary endowment, which are too conspicuous to be overlooked when we compare different species but presumably are also present to a lesser extent between members of a single species, account for other differences in repertoire, as do differences in age (“youthful,” “senile”) or in development (“infantile,” “adolescent”). The field of emotion has scarcely advanced beyond an aspect-description, where more or less transitory differences in behavior are attributed to various exciting circumstances (“frightened,” “angry,” and “embarrassed”).
Traits of this sort are simply a way of representing the repertoire of an organism, with some indication of the relative strength of its parts and with certain inferences regarding relevant variables. The “tests” which measure such traits are inventories, which list responses falling within certain classes and estimate their relative frequencies of occurrence. Surveys of attitudes and opinions are usually of this sort, as are tests of achievement. The Kinsey reports on sexual behavior are surveys of frequencies of certain types of responses from which we may infer certain conditions of deprivation, a history of sexual reinforcement, and the health and hereditary endowment of the organism.
Differences in processes. A second kind of difference in behavior arises from a difference in the rate at which changes in behavior take place. The “intelligent” individual, for example, is commonly supposed to show more rapid conditioning and extinction, to form discriminations more rapidly, and so on. The resulting effect upon behavior is not always distinguishable from that of “experience.” When an individual scores high on an achievement test, the result may be traced either to an exposure to certain variables or to the rate at which these variables have taken effect or to both. Vocabulary tests, for example, presumably reflect differences both in exposure and in rate of conditioning. When we distinguish between the “phlegmatic” and the “sanguine” or “the slow-to-anger” and the “truculent,” the differences are not in degree of deprivation or emotional circumstances but in the speed with which behavior changes as a function of such circumstances.
Traits of this second sort cannot be measured by an inventory. If we want to know simply whether a given set of conditions will make a man angry or lead him to take any other sort of action, a survey of his behavior under those conditions will suffice. If, however, we want to know how quickly he becomes angry or how alertly he takes action, we need a measure appropriate to a functional process. Differences of this second sort can eventually be expressed in quantitative form as differences in the values of certain constants in the equations describing the appropriate processes. Once these values are available, they will characterize an individual just as the physical constants of thermal conductivity, electrical conductivity, specific gravity, and so on characterize materials. (It is significant that these “individual differences” among physical objects were once attributed to essences or principles which strongly resemble traits as the term is now commonly used.)
Traits which can be reduced to inventories of behavior, to the relative strengths of parts of a repertoire, or to the speed with which behavioral processes take place have acceptable scientific dimensions, and their relation to a functional analysis is clear. Those who are currently engaged in the study of traits, however, almost always quantify their data in a quite different way. The intelligence test is a classic example. When a man takes such a test, he makes a score. This is numerical, but it is not an acceptable measure of a trait because it is arbitrary: it depends upon the length of the test, its nature, the time allowed in taking it, and so on. To obtain a less arbitrary measure, the same test is given to a number of people under comparable conditions, and each raw score is converted into a standard score which gives the position of the individual with respect to the group. Even this standard score is not a quantitative measure of a trait; it simply shows that the performance of an individual exceeds that of a certain percentage of the group. But the group is, like the original score, arbitrary. Trouble will arise when we try to use such a measure in a different group,
The difference between a measure based upon a population and a measure based upon frequency of response is clear when we consider a population of only one man. Robinson Crusoe, before the advent of his man Friday, must have shown a certain repertoire of behavior, certain frequencies of response, and certain rates of change in frequency. Occasionally he must have been hungry in the sense of being inclined to eat at a given rate, angry in the sense of being disposed to injure animals or objects, and intelligent in the sense of being quick to solve the problems of his daily life. His behavior must have been modified at given rates as the result of certain contingencies. He must have been able to discriminate stimuli of given complexities or subtleties. All this he himself might have observed and measured in a quantitative way. He could not, however, have measured his own I.Q., since he could not have devised a test on which his score would be divested of the arbitrary features of length, level of difficulty, or allotted time.
The use of a population to measure a trait is illustrated by a scale commonly used to designate the hardness of minerals. The scale ranges from 1 (talc) to 10 (diamond). When we say that quartz has a hardness of 7, we mean that it will scratch or cut all minerals having a hardness of 6 or less, and that it in turn can be cut by minerals of hardness 8, 9, or 10, If the world were constructed of quartz alone, the number 7 would have no significance. Such a scale is unquestionably useful for technological purposes, but it does not greatly advance the study of the hardness of minerals. The physicist accounts for different positions on the scale as differences in molecular structure. A measure of the hardness of quartz expressed in terms of structure is meaningful without reference to minerals of other hardnesses. Insofar as we can express differences in intelligence as differences in repertoire, in exposure to variables, or in rates of change, our measure is similarly independent of a population.
A test is simply a convenient opportunity to observe behavior—to survey or sample our dependent variable. The score may be used to predict some aspect of the larger universe of behavior from which the test is drawn. Thus a test of mechanical ability, or intelligence, or extroversion may enable us to predict success or failure in a job in which these traits are important. But the causal relation invoked in this kind of prediction is not the same as that which appears in a functional analysis. Certain variables in the history of the individual and in the current environment are responsible for the behavior in the test situation, and they also determine the behavior in the larger situation. The prediction is not from cause to effect, but from one effect to another. This is shown by the fact that we use tests as a basis for prediction without knowing what variables are responsible for the score obtained or for the behavior predicted. It is shown also by the fact that if we extend a test without limit, adding more and more items, it eventually coincides with the behavior to be predicted. No true prediction then survives. There is no way in which we could extend a true independent variable so that it would become identical with the dependent variable in a functional analysis.
A prediction from effect to effect is, of course, sometimes useful. It may enable us to dispense with the direct observation of variables. This is particularly important when the variables are clearly out of reach. For example, whether an individual shows certain patterns of behavior characteristic of his species or where he stands with respect to other members of his species in relative frequency of response, as in the Kinsey reports, can at the moment be determined only through a survey, since we have no direct control over the independent variables. A complete survey of such behavior would be easily understood; an incomplete survey constitutes a “test” from which the result of a complete survey may be inferred.
We may also find it convenient to survey the current effects of variables which, though manipulable, lie in the remote history of the individual. We use body weight as a current indication of a history of food deprivation in predicting the probability that an experimental animal will eat, and we might use some collateral test of the “trait of voracity” for the same purpose. Rate of eating in a test situation would enable us to predict rate of eating in a larger experimental situation. Similarly, by making an inventory of current aggressive tendencies, we may dispense with the possibly difficult study of early environmental factors responsible for aggressive behavior.
The principal advantages of a functional analysis are lost, however, when we resort to these alternative practices. Perhaps the most conspicuous feature of an aspect-description is its failure to advance the control of behavior. By measuring a set of traits, we judge the suitability of an individual to a given task, but the only practical step is to accept or reject him. The measurement of the trait does not suggest a way of altering his suitability to the task, for it does not bring us into contact with variables which may be manipulated in generating or eliminating the behavior which it describes. The only practical advantage we gain is that we may make better use of relevant variables already in our possession.
Instead of predicting performance from the test of a trait, we may be interested in predicting one trait from another or from some other sort of variable. Thus a personality in all its manifestations is often attributed to the physique of the organism, a relation which is presumably capable of being expressed trait by trait. Often personality is attributed to variables which are immediately controllable. For example, the “oral,” “anal,” and “erotic” personalities of Freud refer to groups of traits which are assumed to have been generated by the early history of the individual—a history which is presumably modifiable, if taken in time, or at least capable of being masked by a later history superimposed upon it. A similar controlling relation is suggested for a single trait when it is asserted that aggressiveness is a function of frustration. There are, however, certain inherent limitations in a functional analysis in which the dependent variable is a trait.
The usefulness of any lawful relation depends upon the sharpness of reference of the terms in which it is stated. We may predict and control only as much as we specify in our laws. We have seen that there are practical circumstances under which it may be useful to predict traits, but in general the trait-name tells us little about behavior. It is not only lack of specificity, however, which makes the trait-name unsuitable for a functional analysis. In the chapters which follow we shall turn to certain complex processes. Interlocking systems of responses will be traced to complex arrangements of variables, and a workable conception of the individual as a responding system will be set up. The trait-name does not refer to a unit of behavior which makes such an analysis possible.
The fact that a conception of the individual as a behaving system seems to lie beyond the reach of an aspect-description is exemplified by a practical problem in current clinical psychology. Through an extensive use of tests and other measurements of aspects of behavior the individual is characterized for diagnostic purposes. But the resulting information is of little or no help in therapy—in dealing with the individual as a dynamic system. The clinician must turn from a “psychograph” of the personality to “common sense” or to an entirely different conceptual system—such as that of psychoanalysis, which, as we shall see later, is similar to a functional analysis. Currently, little or no effort is being made to reconcile these two ways of dealing with human behavior, perhaps because a reconciliation seems hopeless. The measurement of aspects of behavior is likely to be associated with the belief that the business of science is primarily to supply information which is then used to further the art of dealing with people, not only in the clinic, but in salesmanship, education, family counseling, labor problems, diplomacy, and so on. But the special wisdom which this art presupposes, the special insight into human behavior which is needed to make effective use of such information, is precisely what a functional analysis supplies.
We are all thoroughly familiar with descriptions of behavior in terms of traits, and trait-names are an extensive part of our daily vocabulary. As a result, we feel at home in describing behavior in this way. But the familiarity is misleading. The fact is that we can predict and control a response much more readily than a trait. A response is easier to define and identify, and its probability varies more sensitively. Even when we define a trait as a group of responses, the unity or coherence of the group needs to be proved. Do all the responses which are taken to be evidences of aggressiveness, for example, vary together with a given condition of frustration? And are all conditions of frustration equally effective? In order to be sure of the unity of the trait, we have to show that each of the acts which “expresses” it is controlled by each of the conditions specified as its cause—that each aggressive act, for example, is controlled to the same degree by every condition which can be described as frustrating. But this is the program of a functional analysis. We have not reduced the labor of such an analysis by resorting to summary statements in terms of traits.
Almost any characteristic may be set up as a dimension of the personality, but this extended coverage is of little value until something is achieved beyond mere naming. The additional work required to establish traits as scientific categories is just as laborious and just as detailed as the analysis of discrete responses. The effort required to make any account comprehensive is determined by the subject matter itself. Unfortunately, behavior is complex.
Trait-names usually begin as adjectives—“intelligent,” “aggressive,” “disorganized,” “angry,” “introverted,” “ravenous,” and so on—but the almost inevitable linguistic result is that adjectives give birth to nouns. The things to which these nouns refer are then taken to be the active causes of the aspects. We begin with “intelligent behavior,” pass first to “behavior which shows intelligence,” and then to “behavior which is the effect of intelligence.” Similarly, we begin by observing a preoccupation with a mirror which recalls the legend of Narcissus; we invent the adjective “narcissistic,” and then the noun “narcissism”; and finally we assert that the thing presumably referred to by the noun is the cause of the behavior with which we began. But at no point in such a series do we make contact with any event outside the behavior itself which justifies the claim of a causal connection.
Efforts have been made to put the matter in better scientific order by establishing the validity of the trait as a conceptual cause. A search for the smallest number of traits which will “explain” behavior has worked in this direction. Since trait-names come from many sources and may be multiplied at will, the kinds of behavior to which they refer often overlap. The overlap may be discovered by analyzing the forms of behavior specified in tests of two traits or by showing that the result on one test enables us to predict the result on another. When two traits are found to be almost identical, one is simply dropped. When the overlap is not complete, we appear to be measuring a trait which is common to both tests, yet not measured exclusively by either one. The trait therefore appears to have different dimensions from the behavior from which it is inferred, and this fact has encouraged those who are concerned with finding a minimal set of these causes.
The smallest number of traits needed to account for the performances of a group of people on a number of tests may be determined through certain mathematical procedures. From such a result we may say that a given individual does well in one group of tests because he possesses a certain amount of a certain trait, and on another group of tests because he possesses a certain amount of a different trait. Since these procedures take us some distance away from the observed data, it is tempting to identify the resulting traits or factors with physiological states or psychological faculties and to give them additional dimensions not found in the measures of behavior from which they were inferred. Regardless of the length of the mathematical procedure, however, a trait or factor is derived from the observation of the dependent variable only. This limitation is not changed by any mathematical operation. A fairly exhaustive set of tests may enable us to evaluate traits and to predict performances in a wide range of situations, but the prediction is still from effect to effect. The mathematical refinement has not brought the trait under control. We do not change behavior by manipulating a trait.