A Case History of a Variable

In this article the construct of generalized expectancies for internal versus external control of reinforcement is used as a model to present a brief on the importance of broad theory and training in theory construction and evaluation in psychology. Specifically, the extraordinary heuristic value of this construct is attributed to four characteristics: the nature of the definition, the careful imbedding of the construct in a theoretical context, the use of a broad behavior theory in construction of a measure of individual differences, and the programmatic nature of the original research as well as the format of the early publications. Reevaluation of the characteristic criteria for publication of research studies and for the evaluation of promotion or advancement for psychologists involved in research activities is also discussed. Internal versus external control of reinforcement, often referred to as locus of control, is currently one of t h e most studied variables in psychology and the other social sciences. Frequent studies have been done in fields as diverse as political science and public health, and a test of individual differences in locus of control has been translated into many different languages. Current research continues at approximately the same high rate as it did 20 years ago. Perhaps even more surprising has been the relative success of a test of individual differences in predicting behavior for many different populations in a way that made good sense, rather than providing after-thefact explanations that strained credulity. Briefly, internal versus external control refers to the degree to which persons expect that a reinforcement or an outcome of their behavior is contingent on their own behavior or personal characteristics versus the degree to which persons expect that the reinforcement or outcome is a function of chance, luck, or fate, is under the control of powerful others, or is simply unpredictable. Such expectancies may generalize along a gradient based on the degree of semantic similarity of the situational cues (Rotter, 1966). Numerous reviews of internal-external control research and applications (Lefcourt, 1976, 1981; Phares, 1976) have been published, and last year Bonnie Strickland's (1989) APA presidential address brought much of this literature up to date. I do not intend another review. But the enormous and somewhat surprising popularity of this variable is, in itself, interesting, and I will try in this article to speculate on the reasons for the widespread interest in locus of control. Naturally, I rejected luck as an explanation, but at first, I thought (Rotter, 1982b) that the social upheaval of the times might be responsible. The ~g~etnam War, Watergate, the inner-city riots, and political assassinations were disturbing in themselves, but were also concerning many people--including social scientists---because of the perceived lack of control over their own lives. Such factors probably played some role in the interest in locus of control, as did the interest in the F Scale, a measure of implicit antidemocratic or potentially fascist attitude (Adorno, Frenkel-Brunswick, Levinson, & Sanford, 1950), following World War II. However, on further reflection I felt that there have frequently been such periods of social upheaval in the past and that the explanation for the many research studies published was probably due more to scientifically technical reasons. In the remainder of this article, I will discuss four propositions that may account for the heuristic value of internal-external control, propositions that I believe are particularly relevant to the field of personality theory and personality measurement, but also to the study of psychology as a whole. The Importance of Precise Definition The first proposition I would suggest is that the heuristic value of a construct is partially dependent on the precision of its definition. One might wonder why so many social scientists in the last 30 years have been involved in studying a concept that could not be regarded as anything new. Surely in prerecorded history humans were concerned with ideas of causality, and early recorded history is replete with stories and myths describing events controlled by capricious gods and fate, as well as the results of one's own behavior. When social scientists emerged, they also dealt with interpretations of causality, beliefs in luck, fate, notions of alienation (Seeman, 1959), and celestial intervention. An interest in the perception of causality has been present for a long time. What was needed was a April 1990 • American Psychologist Copyright 1990 by the American Psychological Association, Inc. 0003-066X/90/$00.75 Vol. 45, No. 4, 489--493 489 concept that was clear enough to lead to acceptable measurement. A good definition, especially of a cognitive or subjective variable, must be stated in language that is careful and precise and leads to a common understanding. It needs to be illustrated with many behavioral examples of its consequences if its presence or absence is not directly observable. It should be stated in such a way that the operations for its measurement are not only clear but are widely accepted as logical and reasonable. Terms such as self-actualization, coping, intrinsic versus extrinsic tootivation, unconditional love, and cognitive psychology create interest and enthusiasm, but each scientist reads his or her own meaning into the concept, and operations for measurement either do not exist or vary so much that the resulting research produces a series of contradictory or nonreplicable studies. In 1945, Herbert Feigl provided a set of criteria (p. 258) for a good operational definition that would still provide dividends to psychology, if every graduate student in psychology would learn and understand them. Even with precise definitions, it is usually necessary to distinguish between the construct being defined and other constructs used in the past or present with which it can be confused, as well as making clear the connections to other constructs so that previously collected data can be interpreted and built on. T h e I m p o r t a n c e o f I m b e d d i n g a C o n s t r u c t in a B r o a d e r T h e o r y The second proposition is that the heuristic value of a construct is considerably enhanced if it is imbedded in a broader theory of behavior. Although the majority of studies of locus of control have dealt with applied problems, it is important to recognize that the concept originated both from theoretical and clinical concerns, with social learning theory organizing our thinking in both cases. More specifically, in several of our studies involving increments and decrements of expectancies following both negative and positive outcomes, a large number of our subjects were not raising their expectancies after success or lowering them after failure, and we began to see a pattern of difference in situations in which the subject believed that success was dependent on one's own skill versus those situations in which it was clear that the experimenter was manipulating success and failure indeEditor's note. This article was originally presented as a Distinguished Scientific Contributions award address at the meeting of the American Psychological Association in Atlanta in August 1988. Award-based manuscripts appearing in the American Psychologist are scholarly articles based in part on earlier award addresses presented at the APA convention. In keeping with the policy of recognizing these distinguished contributors to the field, these submissions are given special consideration in the editorial selection process. Author's note. Correspondence concerning this article should be addressed to Julian B. Rotter, Department of Psychology, U-20, University of Connecticut, Storrs, CT 06269-1020. pendently of the subject's behavior. My ideas about these findings crystalized in discussing a psychotherapy client with Jerry Phares who I was supervising. As supervisor, I was trying to understand and interpret the client's behavior from a social learning point of view. This client, whom we had persuaded to try out some new behaviors that met with success, persistently explained away the successes as a matter of luck and not likely to happen again. He appeared, in most situations, to feel that what happened to him was entirely beyond his control. This led to our hypothesis that not only did learning take place differently in chance versus skill situations, but it took place differently among individuals in situations that might be considered ambiguous or novel or that had elements of both chance and skill. Although the implications for human learning theory were of prime importance in the development of this construct, it was a measure of relatively stable, cross-situational, individual differences that captured the interest of many and accounted for at least 90% of the publications that followed. Two things about this puzzled me. One of these was that although many social scientists were eager to use the test of individual differences, they clearly did not understand the theory in which it was imbedded. Specifically, they regarded these individual differences as fixed traits, or types. The notion of a generalized expectancy, of course, involves the learning theory principle of generalization and a gradient of generalization. Such a gradient implies both generality and specificity, but nevertheless, numerous articles were written and published challenging the notion of generality because some specificity could be demonstrated (Rotter, 1975). The theory does not specify independent traits, faculties, or types, but numerous psychologists have taken a 23-item test, subjected it to an orthogonal factor analysis, and concluded mistakenly that the concept had no generality because some specificity could be demonstrated. Generality-specificity is a matter of degree, not kind. The second thing that puzzled me was the reaction to the studies of performance. Many of the original studies of internal-external control were done comparing groups on their performance on skill and chance tasks. Sometimes the tasks were different and sometimes the same ambiguous task was described differently to groups of subjects. The substance of these studies was that changes in expectancies and, consequently, both acquisition and performance were very different for chance versus skill tasks, not merely in degree but in kind. For example, in a number of replicated studies, we discovered that the well-known finding that 50% reinforcement is more resistant to extinction than 100% reinforcement was true only in chance or experimenter-controlled tasks. In skill tasks in which the subject believes that reinforcement is contingent on his or her own skill, the opposite is true (Blackman, 1962; Holden & Rotter, 1962; James & Rotter, 1958; Rotter, Liverant, & Crowne, 1961). One hundred percent reinforcement is more resistant to extinction than