In the mid-1950s, the results of my doctoral dissertation were nearly ruined; it appeared that I might have treated my experimental subjects in such a way as to lead them to respond in accordance with my experimental hypothesis, or expectancy (the details are given in Rosenthal, 1985). All of this was quite unwitting, of course, but it did raise a sobering question about the possibility of interpersonal expectancy effects in the psychological laboratory. If it was my unintentional interpersonal expectancy effect or my "unconscious experimenter bias" that had led to the puzzling and disconcerting results of my dissertation, then presumably my students and I could produce the phenomenon in our own laboratory, and with several experimenters rather than just one.
Early Laboratory Experiments
The earliest studies were conducted with human participants. Experimenters instructed the participants to rate photographs of people, with half the experimenters led to expect high ratings and half led to expect low ratings. In the first several such studies, experimenters expecting high ratings obtained substantially higher ratings than did experimenters expecting low ratings (Rosenthal & Fode, 1961).1
To investigate the generality of these interpersonal expectancy effects in the laboratory, my colleagues and I conducted two studies employing animal subjects. Half the experimenters were told their rats had been specially bred for good maze (or Skinner box) performance, and half were told their rats had been specially bred for poor maze (or Skinner box) performance. In both experiments, when experimenters had been led to expect better learning from their rat subjects, they obtained better learning from their rat subjects (Rosenthal & Lawson, 1964).
Pygmalion Effects in the Classroom
If rats became brighter when expected to by their experimenters, then it seemed not far-fetched to think that children could become brighter when expected to by their teachers. Jacobson and I conducted a study (1968) to test this hypothesis. All of the children in the study were administered a nonverbal test of intelligence, which was disguised as a test that would predict intellectual "blooming." The test was labeled "The Harvard Test of Inflected Acquisition." There were 18 classrooms in the school, three at each of the six grade levels. Within each grade level, the three classrooms were composed of children with above-average ability, average ability, and below-average ability, respectively. Within each of the 18 classrooms, approximately 20% of the children were chosen at random to form the experimental group. The teachers of these children were told that their scores on the "Test of Inflected Acquisition" indicated they would show surprising gains in intellectual competence during the next 8 months of school. The only difference between the experimental group and the control group children, then, was in the minds of the teachers.
At the end of the school year, 8 months later, all the children were retested with the same test of intelligence. Overall, the children from whom the teachers had been led to expect greater intellectual gain showed a significantly greater gain than did the children of the control group, thereby supporting the "Pygmalion" hypothesis2 (Rosenthal & Jacobson, 1968).
Generality and the Magnitude of the Effect
A dozen years after the Pygmalion-in-the-classroom study was completed, the research literature on interpersonal expectancy effects had broadened to include 345 experiments in eight domains of research. For each of these domains, Table 1 gives an example of the types of studies conducted and the average magnitude of the effect obtained.3 After another dozen years had gone by, there were 464 studies, with an overall d of .63, r of .30, and r2 of .09.4, 5
Unfortunately, evaluating the social importance of these results is problematic because all three of these effect-size estimates suffer from the same problem: the underestimation of the practical importance of the effect of a behavioral or biomedical intervention. Something more is needed to address the question:
What is the effect on the success rate (e.g., survival rate, cure rate, improvement rate) of any intervention? One approach is to use the binomial-effect-size display (BESD), which shows the change in outcome attributable to the independent variable (Rosenthal & Rubin, 1982). To employ the BESD, one calculates the success rates for the treatment and control groups by adding 1/2r to .50 (treatment group) and subtracting 1/2r from .50 (control group). Table 2 illustrates the BESD for the overall mean effect size (r = .30) of the 464 studies of interpersonal expectancy effects and, for comparison, for the results of a recent study of the effects of aspirin in the prevention of heart attacks (r = .04; Steering Committee of the Physicians Health Study Research Group, 1988).6
The 10-Arrow Model
For many years, the central question in the study of interpersonal expectancy effects was whether they existed. The meta-analytic evidence has answered that question sufficiently so that simple replications will add little new knowledge. Today, the central focus in the study of interpersonal expectancy effects has changed to include the isolation of the variables that (a) moderate expectancy effects and (b) mediate expectancy effects. Moderator variables are preexisting variables, such as sex, age, and personality, that influence the magnitude of interpersonal expectancy effects; mediating variables are the behaviors by which expectations are communicated. The 10-arrow model, designed to clarify the study of interpersonal expectancy effects, posits 10 links between five groups of variables: (a) distal independent (moderator) variables (e.g., stable attributes of the expecter and expectee), (b) proximal independent variables (the expectancies), (c) mediating variables, (d) proximal dependent variables (e.g., short-term outcome measures such as achievement on tests), and (e) distal dependent variables (long-term outcome variables). A useful feature of this model is that its 10 arrows represent the types of relationships that can be examined in research on expectancy effects (see Figure 1).
The arrows in the model are described in detail elsewhere (Rosenthal, 1981, 1985), so I discuss here only the two links relevant to the topic of mediation: The B - C and C - D arrows. B - C relationships describe the effect of the expectancy on the expecter's behavior. These are the relationships most often investigated in research on mediation. Equally important to undfirstanding mediation, however, are the C - D relationships between the expecter's behavior and outcome variables. Research bearing on the B - C link tells which behaviors are induced by a given expectancy, but research bearing on the C - D link shows that these behaviors affect the expectee so as to create a self-fulfilling prophecy. As is evident, the two types of relationships address different questions, making the distinction between B - C and C - D critical (for meta-analyses making this distinction, see Harris & Rosenthal, 1985, 1986).7
The Four-Factor Theory
On the basis of the first 30 or so published studies relevant to mediation, a four-factor "theory" of the mediation of teacher expectancy effects was proposed (Rosenthal, 1973, 1974). Table 3 summarizes these four factors, and Table 4 gives the average magnitude of the role of each factor separately for the B - C and C - D links. Although all four factors had significant effects, the magnitudes of the effects for the climate and input factors were especially impressive. Teachers appear to teach more and to teach it more warmly to students for whom they have more favorable expectations.
From these results, one cannot infer that selecting warmer teachers who present more material will result in children learning more. One also cannot infer from these results that training teachers to be warmer and to present more material will lead to improved learning. The results do suggest, however, that conducting the research required to determine the benefits of selection and training for climate (or affect) and input (or effort) may well yield substantial benefits both for science and for society.
The most recent work has examined, and work currently in progress continues to examine, interpersonal expectancy effects in an ever-widening circle of contexts. Pygmalion effects in management (Eden, 1990), in courtrooms (Blanck, Rosenthal, Hart, & Bernieri, 1990; Halverson, Hallahan, Hart, & Rosenthal, 1997), in nursing homes (Learman, Avorn, Everitt, & Rosenthal, 1990), and in a variety of classrooms (Babad, 1992) are under investigation. Studies have demonstrated that organizational effectiveness can be increased by raising leaders' expectations, that verdicts of guilty can be increased by having juries instructed by judges who believe the defendants to be guilty, that depression among nursing home residents can be reduced by raising the expectations of caretakers, and that teachers' expectations can serve as self-fulfilling prophecies in other countries and for more than simply intellectual tasks. In all these cases, the mediating variables are receiving special attention, with the growing evidence indicating that much of the mediation is occurring by means of unintended nonverbal behavior (Ambady & Rosenthal, 1992).
Ambady, N., & Rosenthal, R. (1992). Thin slices of expressive behavior as predictors of interpersonal consequences: A meta-analysis. Psychological Bulletin, 111, 256-274.
Babad, E. (1992). Teacher expectancies and nonverbal behavior. In R. S. Feldman (Ed.), Applications of Nonverbal Behavioral Theories and Research (pp. 167-190). Hillsdale, NJ: Erlbaum.
Blanck, P. D., Rosenthal, R., Hart, A. J., & Bernieri, F. (1990). The measure of the judge: An empirically-based framework for exploring trial judges' behavior. Iowa Law Review, 75, 653-684.
Cooper, H., & Hedges, L. V. (Eds.). (1994). The Handbook of Research Synthesis. New York: Russell Sage Foundation.
Eden, D. (1990). Pygmalion in Management: Productivity as a Self-Fulfilling Prophecy. Lexington, MA: Lexington Books.
Halverson, A. M., Hallahan, M., Hart, A. J., & Rosenthal, R. (1997). Reducing the biasing effects of judges' nonverbal behavior with simplified jury instruction. Journal of Applied Psychology, 82, 590-598.
Harris, M. J., & Rosenthal, R. (1985). Mediation of interpersonal expectancy effects: 31 meta-analyses. Psychological Bulletin, 97, 363-386.
Harris, M. J., & Rosenthal, R. (1986). Four factors in the mediation of teacher expectancy effects. In R. S. Feldman (Ed.), The Social Psychology of Education: Current Research and Theory (pp. 91-114). New York: Cambridge University Press.
Learman, L. A., Avorn, J., Everitt, D. E., & Rosenthal, R. (1990). Pygmalion in the nursing home: The effects of caregiver expectations on patient outcomes. Journal of the American Geriatrics Society, 38, 797-803.
Raudenbush, S. W. (1984). Magnitude of teacher expectancy effects on pupil IQ as a function of the credibility of expectancy induction: A synthesis of findings from 18 experiments. Journal of Educational Psychology, 76, 85-97.
Rosenthal, R. (1973). The mediation of Pygmalion effects: A four factor "theory." Papau New Guinea Journal of Education, 2, 1-12.
Rosenthal, R. (1974). On the Social Psychology of the Self-Fulfilling Prophecy: Further Evidence for Pygmalion Effects and their Mediating Mechanisms. New York: MSS Modular.
Rosenthal, R. (1981). Pavlov's mice, Pfungst's horse, and Pygmalion's PONS: Some models for the study of interpersonal expectancy effects. In T. A. Sebeok & R. Rosenthal (Eds.), The Clever Hans Phenomenon, Annals of the New York Academy of Sciences (Vol. 364). New York: New York Academy of Sciences.
Rosenthal, R. (1985). From unconscious experimenter bias to teacher expectancy effects. In J. B. Dusek, V. C. Hall, & W. J. Meyer (Eds.), Teacher Expectancies. Hillsdale, NJ: Erlbaum.
Rosenthal, R. (1987). Pygmalion effects: Existence, magnitude, and social importance. Educational Researcher, 16, 37-41.
Rosenthal, R. (1991a). Meta-Analytic Procedures for Social Research (Rev. ed.). Newbury Park, CA: Sage.
Rosenthal, R. (1991b). Teacher expectancy effects: A brief update 25 years after the Pygmalion experiment. Journal of Research in Education, 1, 3-12.
Rosenthal, R., & Fode, K. L. (1961). The problem of experimenter outcome-bias. In D. P. Ray (Ed.), Series Research in Social Psychology. Washington, DC: National Institute of Social and Behavioral Science.
Rosenthal, R., & Jacobson, L. (1968). Pygmalion in the Classroom. New York: Holt, Rinehart and Winston.
Rosenthal, R., & Lawson, R. (1964). A longitudinal study of the effects of experimenter bias on the operant learning of laboratory rats. Journal of Psychiatric Research, 2, 61-72.
Rosenthal, R., & Rubin, D. B. (1978). Interpersonal expectancy effects: The first 345 studies. The Behavioral and Brain Sciences, 1, 377-415.
Rosenthal, R., & Rubin, D. B. (1982). A simple, general purpose display of magnitude of experimental effect. Journal of Educational Psychology, 74, 166-169.
Smith, M. L. (1980). Teacher expectations. Evaluation in Education, 4, 53-55.
Steering Committee of the Physicians Health Study Research Group. (1988). Preliminary report: Findings from the aspirin component of the ongoing physicians' health study. The New England Journal of Medicine, 318, 262-264.
Rosenthal, R. (1976). Experimenter Effects in Behavioral Research (Enlarged ed.). New York: Irvington.
Rosenthal R., & Jacobson, L. (1992). Pygmalion in the Classroom: Teacher Expectation and Pupils' Intellectual Development. New York: Irvington.
1 That this research was received with ambivalence is illustrated by the receipt of two letters on the same day: The first letter rejected the paper for publication in a prestigious journal, and the second letter announced that the paper had received the Socio-Psychological Prize for 1960 from the American Association for the Advancement of Science.
2 A surprising finding was that the more children in the control group gained in IQ, the more unfavorably they were judged by their teachers. Apparently there were hazards to unpredicted intellectual growth. Also surprising was the strength of both the favorable and the unfavorable reactions to our research. For a summary of the criticisms and replies to them, see Rosenthal (1987).
3 Effect sizes are expressed in terms of both d and r. The former is the difference between the experimental and control groups divided by the standard deviation of both groups combined. The latter is the point-biserial correlation between experimental versus control group status (e.g., coded 1 for experimental and 0 for control) and the outcome score (e.g., gain in performance). The effect sizes of Table 1 are based on Rosenthal and Rubin (1978).
4 These meta-analyses are also described in Rosenthal (1991b). For technical details on meta-analytic procedures see Rosenthal (1991a) and Cooper and Hedges (1994).
5 More specialized meta-analyses for the effects on pupils' IQ test performance are also available (Raudenbush, 1984; Smith, 1980). Both Raudenbush and Smith found significant overall effects of interpersonal expectations on students' IQ. Smith reported a d of 0.16 and r of .08. Raudenbush found very strong evidence (r = .67) that substantial effects of teachers' expectancies could be demonstrated only when the induction of the expectancy was credible (i.e., when teachers had known pupils only 2 weeks or less at the time they were given the expectation). For the seven studies in which teachers had known pupils 1 week or less, as was the case for the Pygmalion study, the mean effect-size d was 0.29, r was .14.
6 This study was discontinued on the grounds that it would be unethical to continue depriving placebo-control participants of the beneficial effects of aspirin, given the effect size (r = .04, r2 = .00).
7 Although all the arrows in the model point to the right (or to the future), they may usefully be viewed as often going in both directions. Thus, improved student performance (D) can affect 24 subsequent teacher expectations (B) and behavior toward the student (C).
Author note: This paper is based on an invited Psi Chi Distinguished Lecture presented at the annual meeting of the Eastern Psychological Association in Boston, February 28, 1998. It provides an overview of research cited in the footnotes and references and is an updated version of a paper published in Current Directions in Psychological Science under the title of "Interpersonal Expectancy Effects: A 30-Year Perspective" (1994; reprinted with the permission of Cambridge University Press; all rights reserved).
I am grateful to Kay Wilson, Executive Officer of Psi Chi, for the opportunity to prepare this updated article for Eye on Psi Chi. I thank the many students, colleagues, collaborators, and tutors who have been educating me for over 40 years. Much of the research reviewed here was supported in part by the National Science Foundation, and preparation of this review was supported in part by the Spencer Foundation, though the contents of this review are solely the responsibility of the author.
Address correspondence to Robert Rosenthal, Department of Psychology, Harvard University, 33 Kirkland Street, Cambridge, MA 02138.