An Experimental Study of Stereotypes[1]

Muzafer Sherif
Harvard University

EVERYBODY knows that in perceiving or responding to any stimulus, besides the external factors, internal factors play an important part. Psychologically these internal factors are often referred to as needs, drives, attitudes, affects, or emotional states. Their effect is especially striking in cases where the external field of stimulation is not so sharply and rigidly structured as to play the predominating part in the perception or judgment. Although psychologists are familiar with this interplay of external and internal factors as exemplified in the perception of ambiguous figures, the principles underlying such perceptual phenomena have not yet been extended to the explanation of the common social phenomena known as prestige-suggestion and stereotypes.

The words of a person with whom we are acquainted or the pronouncements of a person who has prestige in our eyes, are experienced against the whole background of our relationship to that person. The same statement made by two different people with whom an individual is in different affective relationships arouses quite different experiences and responses in that individual. The influence of these differential affective relationships constitutes a very important problem in social psychology. The following experiments are an attempt towards a quantitative study of this problem.[2]


A mimeographed sheet containing the names of sixteen authors arranged in alphabetical order (Barrie, Joseph Conrad, James Fenimore Cooper, Dickens, Thomas Hardy, Hawthorne, Kipling, Poe, Ruskin, Scott, Stevenson, Thackeray, Tolstoy, Mark Twain, Walt Whitman, Thornton Wilder) was presented to the subjects with the following instructions at the top of the sheet. 

Place beside each name a figure indicating the order of preference you have for the following authors. Make your judgment solely on the grounds

(372) of subjective liking for the works of the writer. If you have no feeling of like or dislike for a certain author, you may omit his name. Place the figure (1) beside the name of the writer whose work you like best, (2) beside the next, and so until you have arranged all with whom you are acquainted.

These ratings were designated as Series A.

One month later the same subjects were presented sixteen mimeographed slips each containing a short passage of three or four lines and, so far as three judges could determine, of about the same literary value. Under each passage was placed the name of one of the sixteen authors used in the first part of the experiment. Each passage was ascribed to a different author, but in reality all the passages were taken from one author, namely, Robert Louis Stevenson. No subject suspected the deception. The instructions for the second part of the experiment were as follows:

Attached are sixteen descriptive passages with the authors' names. Kindly arrange these in order of your preference by inserting a figure (1) at the left of the passage which appeals to you most. Make your judgments simply on the grounds of liking or disliking of the passages. Place (2) against the next most pleasing passage. If it is difficult to distinguish closely between the different passages, you are asked simply to “guess" if no more accurate judgment is possible.

These ratings were designated as Series B.

After the ratings in Series B were completed a written introspective report was secured from each subject as to whether he or she had made a special effort to eliminate the influence of the authors' names in evaluating the passages, either by covering them while the judgments were being made or by intentionally disregarding them.

The correlation between the order of preference for authors (Series A) and the judgment of the literary merit of the passages ascribed to the same authors (Series B) was calculated for each subject.

The results are classified into two groups on the basis of the introspective reports of the subjects: (1) those who did not make a special effort to overcome the influence of the authors' names, (2) those who did make a special effort to overcome this influence or to ignore the authors' names.

In the Harvard group there were thirty-three subjects. Twenty-five come under Group 1, that is, those who did not, make a special effort to overcome the influence of authors' names or to ignore them altogether. The average correlation for this group is +.45. Eight subjects come under Group II, those who made a special

(373) effort to overcome the influence of authors' names or to ignore them altogether. The average correlation for this group is -.03. From Radcliffe there were nineteen subjects. Eleven subjects come under Group I. The average correlation for this group is +.53. There were eight subjects in Group II. The average correlation is +.04.

The fact that the results obtained from those who made a special effort to overcome the influence of the authors' names gave practically zero-correlation shows that. the intrinsic value of the passages was not a factor working in any direction. In contrast with this, the positive correlations (+.45 in the case of Harvard, +.53 in the case of Radcliffe) obtained from those who took a natural attitude indicate the influence upon the evaluations of the names attached to the passages.


The second experiment[3] was carried on in the psychological laboratory of Gazi Terbiye Institute, Ankara, Turkey, the following year. In general the procedure followed the Harvard experiment with certain slight modifications and controls.

Here again all the passages were taken from one single author. Twelve Turkish authors' names and consequently twelve passages were used. As a control the passages were prepared in two forms. In both forms the passages were the same, but the names of the twelve authors were distributed in a different order, so that a different author's name appeared under the corresponding passages of Form I and II. The subjects were students of four different colleges. In each case the first form was given to one-half of the group and the second form to the other half. This was done as a control to check upon the influence of the intrinsic literary value that any passage might have.

This experiment was carried out on 106 subjects; sixty-seven subjects were male and thirty-nine, female. The records of a few subjects who reported that they deliberately ignored the names of the authors were discarded.

The correlations may be summarized as follows :


Average correlation for 67 male subjects +.45
Average correlation for 39 female subjects +.50
Average correlation for 106 subjects +.47

Between the correlations obtained from Form I and II there were no significant differences. Therefore one may conclude as in

(374) the first experiment, that the intrinsic value of the passages did not play a part in giving these correlations.


The third experiment was carried on at Harvard two years after the first experiment. Sixty-six subjects were used, twenty-five students from Harvard, twenty from Radcliffe, nine from the School of Education, twelve adults attending an extension course.

To make the results comparable with the first experiment the same material was used. The procedure was the same, but the same control as in the second experiment was used. That is, the material was prepared in two forms as a check upon the intrinsic value of the passages. The results here also indicate that the correlations were not influenced by the intrinsic value of the passages. The results obtained from the subjects who did not make a consistent special effort to overcome the influence of the authors' names or to ignore them altogether are as follows

Average correlation for 22 Harvard students +.33
Average correlation for 17 Radcliffe students +.45
Average correlation for 10 extension course adults +.30
Average correlation for 8 School of Ed students +.31
Average correlation for 57 total +.36

There were nine subjects who simply ignored the names of the authors. The average correlation for these nine subjects is +.03. The results of the third experiment correspond therefore with the results of the first and second experiments.

In the first experiment the average correlation for the Harvard group was +.45, for Radcliffe, +.53; in the second experiment the average correlation for men was +.45, for women, +.50; in the third it is +.33 in the case of Harvard and +.45 in the case of Radcliffe. These results seem to indicate that the judgments of women may be more influenced than those of men by their preexisting attitudes toward the authors.

On the basis of these experiments it may be concluded that prestige-suggestion or stereotype plays a considerable part in people's judgments. In other words the attitudes towards authors serve as reference points. Authors rated high tend to pull up, and conversely authors rated low tend to pull down, the rating of the passages attributed to them. This is but a specific case of a general psychological principle. It appears that our judgments like our perceptions, are organized in relation to definite reference points or in relation to a general level of reference.

( 375)

In the written introspective records obtained right after the experiments several subjects spontaneously reported "they wished that the names of the authors were not there", or that "they were irritated by the presence of the author's name". Such emotional remarks indicate that the point of reference which causes the bias is sometimes a matter of considerable concern to the subject himself.

The method described in this paper may be extended to measure the influence of prestige or stereotype in economic, political, religious and theoretical fields as well. It has in fact points of similarity to the previous work of Rice [4] and Zillig.[5] But it is suggested that the interpretations offered in the present paper in terms of "reference points", and the hypothesis that judgment is psychologically in close kinship to perception, constitute an advance in the explanation of the nature of stereotype.


  1. Received and acknowledged January 15, 1934.
  2. For assistance in the execution of this project the writer wishes to thank Prof. G. W. Allport and Dr. H. Cantril.
  3. M. Hurrem of Gazi T. Institute, Ankara, cooperated in this experiment.
  4. Rice, S. A., " Stereotypes": A Source of Error in Judging Human Character. Journal of Personnel Research, Vol. V, No. 7, November, 1926.
  5. Zillig, M., Einstellung and Aussage.. Zeitschrift für Psychologie, Vol. 706 (1928), pp. 58-106.

Valid HTML 4.01 Strict Valid CSS2