Interpersonal Expectancy Effects:A 30-Year PerspectiveRobert Rosenthal

In the mid 1950s, the results ofmy doctoral dissertation were nearlyruined; it appeared that I might havetreated my experimental subjects insuch a way as to lead them to re-spond in accordance with my exper-imental hypothesis, or expectancy.'Al! of this was quite unwitting, ofcourse, but it did raise a soberingquestion about the possibility of in-terpersonal expectancy effects in thepsychological laboratory. If it wasmy unintentional interpersonal ex-pectancy effect or my "unconsciousexperimenter bias" that had led tothe puzzling and disconcerting re-sults of my dissertation, then pre-sumably my students and I couldproduce the phenomenon in ourown laboratory, and with several ex-perimenters rather than just one.

EARLY LABORATORYEXPERIMENTS S

The earliest studies were con-ducted with human subjects. Exper-imenters instructed the subjects torate photographs of people, withhalf the experimenters led to expecthigh ratings and half led to expectlow ratings. In the first several suchstudies, experimenters expectinghigh ratings obtained substantially

Robert Rosenthal is Professor ofSocial Psychology at HarvardUniversity, where he studies inter-personal expectancy effects, non-verbal communication, and data-analytic procedures. Addresscorrespondence to RobertRosenthal, Department of Psychol-ogy, Harvard University, 33 Kirk-land St., Cambridge, MA 02138.

higher ratings than did experiment-ers expecting low ratings.'

To investigate the generality ofthese interpersonal expectancy ef-fects in the laboratory, my col-leagues and 1 conducted two studiesemploying animal subjects. Half theexperimenters were told their ratshad been specially bred for goodmaze {or Skinner box) performance,and half were told their rats hadbeen specially bred for poor maze(or Skinner box) performance. Inboth experiments, when experi-menters had been led to expect bet-ter learning from their rat subjects,they obtained better learning fromtheir rat subjects."*

PYGMALION EFFECTS INV THE CLASSROOM

If rats became brighter when ex-pected to by their experimenters,then it seemed not farfetched tothink that children could becomebrighter when expected to by theirteachers. Jacobson and 1 conducteda study to test this hypothesis. All ofthe children in the study were ad-ministered a nonverbal test of intel-ligence, which was disguised as atest that would predict intellectual"blooming." The test was labeled"The Harvard Test of Inflected Ac-quisition." There were 18 class-rooms In the school, three at each ofthe six grade levels. Within eachgrade level, the three classroomswere composed of children withabove-average ability, average abil-ity, and below-average ability, re-spectively. Within each of the 18classrooms, approximately 20% ofthe children were chosen at randomto form the experimental group. The

teachers of these children were toldthat their scores on the "Test of In-flected Acquisition" indicated theywould show surprising gains in intel-lectual competence during the next8 months of school. The only differ-ence between the experimentalgroup and the control group chil-dren, then, was in the minds of theteachers.

At the end of the school year, 8months later, all the children wereretested with the same test of intelli-gence. Overall, the children fromwhom the teachers had been led toexpect greater intellectual gainshowed a significantly greater gainthan did the children of the controlgroup, thereby supporting the "Pyg-malion" hypothesis/

GENERALITY AND .MAGNITUDE OF THE EFFECT |

A dozen years after the Pygma-lion-in-the-classroom study wascompleted, the research literatureon interpersonal expectancy effectshad broadened to include 345 ex-periments in eight domains of re-search. For each of these domains.Table 1 gives an example of thetypes of studies conducted and theaverage magnitude of the effect ob-tained.^ After another dozen yearshad gone by, there were 464 stud-ies, with an overall d of 0.63, r of.30, and r^ of .09.^'^

Unfortunately, evaluating the so-cial importance of these results isproblematic because all three ofthese effect-size estimates sufferfrom the same problem: the under-estimation of the practical impor-tance of the effect of a behavioral orbiomedical intervention. Somethingmore is needed to address the ques-tion: What is the effect on the suc-cess rate (e.g., survival rate, curerate, improvement rate) of any inter-vention? One approach is to use thebinomial-effect-size display (BESD),which shows the change in outcome

Published by Cambridge Universily Press

CURRENT bimCTICmm F^VCHOLDClGAi SGtENCE

Table 1. Interpersonal expectancy

Domain

Laboratory interviews

Reaction time

Learning and ability

Person perception

Inkblot tests

Everyday situations

Psychophysical judgments

Animal learning

Weighted mean*'Unweighted meanMedian

effects in eight research domains

Mean effect

d

0.14

0.17

0.54

0.55

0.84

0.88

1.05

1.73

0.700.740.70

Note. The effect sizes shown are based

size

r

.07

.08

.26

.27

.39

.40

.46

.65

,33,35.33

an Rosentha•"Weighting is by number of studies in each domain.

Example of typeof study

Effects of sensory restrictionon reports of hallucinatoryexperiences

Latency of word associationsto certain stimulus words

IQ test scores, verbalconditioning (learning)

Perception of other people'ssuccess

Ratio of animal to humanRorschach responses

Symbol learning, athleticperformance

Ability to discriminate tones

Learning in mazes andSkinner boxes

and Rubin,^

attributable to the independent vari-able.^ To employ the BESD, one cal-culates the success rates for the treat-ment and control groups by addingy2rto .50 (treatment group) and sub-tracting V2rfrom .50 (control group).Table 2 illustrates the BESD for theoverall mean effect size (r - .30) ofthe 464 studies of interpersonal ex-pectancy effects and, for compari-son, for the results of a recent study

of the effects of aspirin in the pre-vention of heart attacks (r = .04).^

THE 10-ARROW MODEL

For many years, the central ques-tion in the study of interpersonal ex-pectancy effects was whether theyexisted. The meta-analytic evidence

Table 2. Two binomial-effect-size displays

Condition

TreatmentControl

Total

TreatmentControl

Total

Treatment r

Favorable

Pygmalion effects {r = .30, d6535

100

Aspirin effects (r ^ .04, d ^5248

100

esult

Unfavorable

- 0.63, r^ = ,09)3565

TOO

0.08, r - .00)4852

100

Note. The Pygmalion effects are based on 464 studies. The aspirin effects22,000 participants.

Total

100100200

100100200

are based on

has answered that question suffi-ciently so that simple replicationswill add little new knowledge. To-day, the central focus in the study ofinterpersonal expectancy effects haschanged to include the isolation ofthe variables that (a) moderate ex-pectancy effects and (b) mediate ex-pectancy effects. Moderator vari-ables are preexisting variables, suchas sex, age, and personality, that in-fluence the magnitude of interper-sonal expectancy effects; mediatingvariables are the behaviors by whichexpectations are communicated.

The 10-arrow model, designed toclarify the study of interpersonal ex-pectancy effects, posits 10 links be-tween five groups of variables: (a)distal independent (moderator) vari-ables (e.g., stable attributes of theexpecter and expectee), (b) proximalindependent variables (the expec-tancies), (c) mediating variables, (d)proximal dependent variables (e.g.,short-term outcome measures suchas achievement on tests), and (e) dis-tal dependent variables (long-termoutcome variables). A useful featureof this mode! is that its 10 arrowsrepresent the types of relationshipsthat can be examined in research onexpectancy effects (see Fig. 1).

The arrows in the model are de-scribed in detail elsewhere,^^ so 1discuss here only the two links rele-vant to the topic of mediation: theB-C and C-D arrows. B-C relation-ships describe the effect of the ex-pectancy on the expecter's behav-ior. These are the relationships mostoften investigated in research on me-diation. Equally important to under-standing mediation, however, arethe C-D relationships between theexpecter's behavior and outcomevariables. Research bearing on theB-C link tells which behaviors areinduced by a given expectancy, butresearch bearing on the C-D linkshows that these behaviors affect theexpectee so as to create a self-fulfilling prophecy. As is evident, thetwo types of relationships addressdifferent questions, making the dis-

Copyright © 1994 American Psychological Society

IHDBPEKDEHT(Pradictor)

DISTAI.(Modarator)

PROXIHAL(Expectancy)

OP VARIABLES

H8DIATINO(A,B) (D,E)

PROCESSc t a r Baha

DEPEKDEHT(Ou-tCOB*)

PROXIMAL(Ianaediate) DISTAL

(POllOM-Up)

Fig. 1. Ten-arrow model for the study of interpersonal expectancy effects.

Table 3. Four factors in the mediation of teacher expectancy effects

Factor Summary of the evidence

Central factors:1. Climate (affect)

2. Input (effort)

Additional factors:3. Output

4. Feedback

Teachers appear to create a warmer socio-emotionalclimate for their "special" students. This warmthappears to be at least partially communicated bynonverbal cues.

Teachers appear to teach more material and moredifficult material to their "special" students.

Teachers appear to give their "special" studentsgreater opportunities for responding. Theseopportunities are offered both verbally andnonverbally (e.g., giving a student more time toanswer a question).

Teachers appear to give their "special" students moreinformative feedback, both verbal and nonverbal, asto how these students have been performing.

Table 4. Meta-analytically derived average correlations representing theimportance (effect sizes) of the four factors

Factor

Correlation betweenexpectation and

expecter's behavior(B-C link)

Correlation betweenbehavior of expecter

and response ofexpectee {C-D link)

Geometricmean

correlation

1. Climate (affect)2. Input (effort)3. Output4. Feedback

.23

.26

.18

.13

.36

.28

.16

.08

.29

.27

.17

.10

Note. These data are from Harris and Rcsentbal (1986)." All correlations aresignificantly greater than zero at p < .002. The correlation between the magnitudes ofthe average B-C and C-D links is .88.

tinction between B-C and C—D crit-ical.^'

THE FOUR-FACTORTHEORY"

On the basis of the first 30 or sopublished studies relevant to media-tion, a four-factor "theory" of themediation of teacher expectancy ef-fects was proposed.^^ Table 3 sum-marizes these four factors, and Table4 gives the average magnitude of therole of each factor separately for theB-C and C-D links. Although allfour factors had significant effects,the magnitudes of the effects for theclimate and input factors were espe-cially impressive. Teachers appearto teach more and to teach it morewarmly to students for whom theyhave more favorable expectations.

From these results, one cannot in-fer that selecting warmer teacherswho present more material will re-sult in children learning more. Onealso cannot infer from these resultsthat training teachers to be warmerand to present more material wiiilead to improved learning. The re-sults do suggest, however, that con-ducting the research required to de-termine the benefits of selection andtraining for climate (or affect) and in-put (or effort) may well yield sub-stantial benefits both for science andfor society.

LOOKING FORWARD g |

The most recent work has exam-ined interpersonal expectancy ef-fects in an ever-widening circle ofcontexts. Pygmalion effects in man-agement,'^ in courtrooms/"^ in nurs-ing homes,'^ and in a variety ofclassrooms'^ are under investiga-tion. Studies have demonstrated thatorganizational effectiveness can beincreased by raising leaders' expec-tations, that verdicts of guilty can beincreased by having juries instructed

Published by Cambridge University Press

by judges who believe the defen-dants to be guilty, that depressionamong nursing home residents canbe reduced by raising the expecta-tions of caretakers, and that teach-ers' expectations can serve as self-fu l f i l l ing prophecies in othercountries and for more than simplyintellectual tasks. In all these cases,the mediating variables are receivingspecial attention, with the growingevidence indicating that much of themediation is occurring by means ofunintended nonverbal behavior. ^

Acknowledgments—I thank the many stu-dents, colleagues, collaborators, and tu-tors who have been educating me formore than 40 years. Much of the researchreviewed here was supported in part bythe National Science Foundation, andpreparation of this review was supportedin part by the Spencer Foundation, thoughthe contents are solely the responsibilityof the author.

Notes

1. The details are given in R. Rosenthai, Fromunconscious experimenter bias to teacher expectan-cy effects, in Teacher Expectancies, |.G. Dusek,V.C- Hall, and W.). Meyer, Eds. (Eribaum, Hills-dale, NI, 1985).

2. R. Rosenthai and K.L. Fode, The problem ofexperimenter outcome-bias, in Series Research inSocial Psychology, D,P. Ray, Ed. (National Instituteof Social and Behavioral Science, Washington, DC,1961). That this research was received with ambiv-alence is illustrated by the receipt of two letters onthe same day: The first letter rejected the paper forpublication in a prestigious journal, and the secondletter announced that the paper had received theSocio-Psychological Prize for 1960 from the Amer-ican Association for the Advancement of Science,

3. R. Rosenthai and R. Lawson, A longitudinalstudy of the effects of experimenter bias on the op-erant learning of laboratory rats, journal of Psychi-atric Research, 2, 61-72 (1964).

4. R. Rosenthai and L. Jacobson, Pygmalion inthe Classroom (Holt, Rinehart and Winston, NewYork, 1968). A surprising finding was that the morechildren in the control group gained in IQ, the moreunfavorably they were judged by their teachers. Ap-parently there were hazards to unpredicted intellec-tual growth. Also surprising was the strength of boththe favorable and the unfavorable reactions to ourresearch. For a summary of the criticisms and repliesto them, see R- Rosenthai, Pygmalion effects; Exis-tence, magnitude, and social importance, fduca-tional Researcher, Ib, 37-41 (1987),

5. Effect sizes are expressed in terms of both dand r. The former is the difference between the ex-perimental and control groups divided by the stan-dard deviation of both groups combined. The latteris the point-biserial correlation between experimen-tal versus control group status (e.g., coding 1 forexperimental and 0 for control) and the outcomescore (e.g,, gain in performance). The effect sizes inTable 1 are based on R. Rosenthai and D.B. Rubin,Interpersonal expectancy effects: The first 345 stud-ies. The Behavioral and Brain Sciences, 3, 377-386(1978).

6. Tbese meta-anaiyses are also described in R.Rosenthai, Teacher expectancy effects: A brief up-date 25 years after the Pygmalion experiment, jour-nal of Research in Education, I, 3-12 (1991). Fortechnical details on these meta-analytic procedures,see R. Rosenthai, Me[a-,4na/yl/c Procedures for So-cial Research, rev. ed. (Sage Publications, NewburyPark, CA, 1991),

7- More specialized meta-analyses for the ef-fects on pupils' IQ test performance are also avail-able: S.W. Raudenbush, Magnitude of teacher ex-pectancy effects on pupil IQ as a function of thecredibility of expectancy induction: A synthesis offindings from 18 experiments, journal of Education-al Psychology, 76, 85-97 (1984); M.L. Smith,Teacher expectations. Evaluation in Education, 4,53-55 (1980), Both Raudenbush and Smith foundsignificant overall effects of interpersonal expecta-tions on students' IQ. Smith reported a d of 0.16 andf of .08. Raudenbush found very strong evidence(r = .67) that substantial effects of teachers' expec-tancies could be demonstrated only when the induc-tion of the expectancy was credible (i.e., whenteachers had known pupils only 2 weeks or less atthe time they were given the expectation). For theseven studies in which teachers had known pupils 1week or less, as was the case for the Pygmalionstudy, the mean effect-size d was 0.29, and r was.14.

8. R. Rosenthai and D.B. Rubin, A simple gen-eral purpose display of magnitude of experimentaleffect, journal of Educational Psychology, 74, 166-169 (1982).

9, Steering Committee of the Physicians HealthStudy Research Group, Preliminary report: Findingsfrom the aspirin component of the ongoing physi-

cians' health study. The New England journal ofMedicine, 318, 262-264 (1988). Tbis study was dis-continued on the grounds that it would be unethicalto continue depriving placebo-control participantsof the beneficial effects of aspirin, given the effectsize (r - .04, r^ - .00).

10. R. Rosenthai, Pavlov's mice, Pfungst'shorse, and Pygmalion's PONS: Some models for thestudy of interpersonal expectancy effects, in TheClever Hans Phenomenon, Annals of the New YorkAcademy of Sciences, Vol, 364, T,A, Sebeokand R,Rosenthai, Eds. (New York Academy of Sciences,New York, 1981); also see Rosenthai, note 1.

11. For meta-analyses making this distinction,see M.J. Harris and R. Rosenthai, Mediation of in-terpersonal expectancy effects; 31 meta-analyses.Psychological Bulletin, 97, 363-386 (1985); M.J.Harris and R. Rosenthai, Four factors in the media-tion of teacher expectancy effects, in The Social Psy-chology of Education, R,S. Feldman, Ed, (Cam-bridge University Press, New York, 1986), Althoughall the arrows in the model point to the right (or tothe future), tbey may usefully be viewed as oftengoing in both directions- Thus, improved studentperformance (D) can affect subsequent teacher ex-pectations (B) and behavior toward the student (C)-

12. R. Rosenthai, The mediation of Pygmalioneffects: A four factor "theory," Papua New Guineajournai of Education, 9, 1-12 (1973); R. Rosenthai,On the Social Psychology of the Self-FulfillingProphecy: Further Evidence for Pygmalion Effectsand Their Mediating Mechanisms (MSS ModularPublications, NewYork, 1974)-

13. D. Eden, Pygmalion in Management (Lex-ington Books, Lexington, MA, 1990).

14. P.D. Blanck, R. Rosenthai, A,J- Hart, and F.Bernieri, The measure of the judge: An empirically-based framework for exploring trial judges' behav-ior, Iowa Law Review, 75, 653-684 (1990),

15. L.A, Learman, J, Avorn, D,E, Everitt, and R-Rosenthal, Pygmalion in the nursing home: The ef-fects of caregiver expectations on patient outcomes,journal of the American Geriatrics Society, 38, 797-803 (1990).

16. E. Babad, Teacher expectancies and non-verbal behavior, in /\pp/jcat(ons of Nonverbal Be-havioral Theories and Research, R.S. Feldman, Ed.(Eribaum, Hillsdale, NJ, 1992),

17. N, Ambady and R. Rosenthai, Thin slices ofexpressive behavior as predictors of interpersonalconsequences: A meta-analysis. Psychological Bul-letin, 111, 256-274 (1992).

Recommended Reading

Rosenthai, R. (1976). Experimenter Effects in Be-havioral Research (Irvington, New York).

Rosenthai, R., and Jacobson, L, (1992), Pygma-lion in the Classroom (Irvington, New York).

Copyright © 1994 American Psychological Society