Professional Documents
Culture Documents
Department Xocial and Administrative Studies, University Oxford
Department Xocial and Administrative Studies, University Oxford
101-114 101
Printed in Great Britain
1. INTRODUCTION
The purpose of the present paper is to report experiments designed to test some of
the hypothesis put forward in a previously published paper (Tajfel, 1 9 5 9 ~ ) The .
principal trend of the argument in that paper was that shifts and biases in stereotyped
judgements can be subsumed under similar shifts existing in absolute judgements of
series of physical quantities. This is best shown when one takes into account the effects
on judgement of those aspects of judging situations that cannot be included in the
description of the physical properties of the stimuli. Examples of such ‘non-physical ’
determinants of judgement are : the relevance to the subject of differences between
the stimuli in a series (e.g. Bruner & Tajfel, 1961; Tajfel, 1957; Tajfel, 1959b;
Tajfel & Cawasjee, 1959); the subject’s perception of some stimuli as relevant or not
relevant to his judgement of other stimuli (e.g. Davidon, 1960); and the manner
in which a series is classified by the subject.
Most work on psychophysical phenomena has been concerned with the formulation
of functions relating orders of physical magnitudes to series of corresponding dis-
criminative responses. The existence of predictable relations between these two
orders of events does not preclude the possibility of existence of second-order laws
which would state coherent and predictable relations between judgement and some
subjective aspects of the judging situation. These second order laws would be most
directly relevant to prediction of shifts and biases in social situations. However,
they should first be established through investigations of scales of response pertaining
to series of physical magnitudes in simple and well-controlled judgement tasks.
The experiments reported here are concerned with classification as a variable. A
series of stimuli may be classified in a number of ways, and there can be a number
of possible relations between a classification and the physical magnitudes which
are subjected to judgement. A series of stimuli may be classified in terms of a dis-
continuous attribute (e.g. a group of people classified into, say, Swedes and Italians) ;
the dimension along which the judgements are made (e.g. height) may be a
continuous one. If in our group of people all the Swedes were taller than all the
Italians, there would be a perfect correlation between the classification and the
7 Gen. Psych. 64. 2
102 HENRITAJFELAND A. L. WILKES
physical attribute under judgement, though the series was not classified in terms of
that attribute. If our group of people consisted of, for example, French and Italians,
and there was no difference between the means and the spreads of the heights of
individuals in both groups, we would have a classification standing in no relation to
the physical attribute under judgement.
These two imaginary cases may be considered as two extremes of a continuum,
ranging from an entirely predictable relation between a classification and a physical
dimension to a complete lack of relation. Most real life cases would fall, of course,
between the two extremes; though some (e.g. skin colour and some racial classifica-
tions) approach very nearly to a perfect correlation, and many could be found at the
uncorrelated end of the continuum.
The problem with which the present experiments are concerned can be stated as
follows : how does the subject’s knowledge that stimuli fall into various classes affect
the judgements of a physical dimension of the stimuli when this dimension stands i n
a certain correlation to the classification Z
The predictions stated in the paper previously referred to (Tajfel, 1 9 5 9 ~ were
) as
follows :
‘ (i) When a classification in terms of an attribute other than the physical dimen-
sion which is being judged is superimposed on a series of stimuli in such a way that
one part of the physical series tends to fall consistently into one class, and the
other into the other class, judgements of physical magnitudes of the stimuli falling
into the distinct classes will show a shift in the directions determined by the class
membership of the stimuli, when compared with judgements of a series identical with
respect to this physical dimension, on which such a classification is not superimposed.
‘ (ii) When a classification in terms of an attribute other than the physical dimen-
sion which is being judged is superimposed on a series of stimuli, and the changes
in the physical magnitudes of the stimuli bear no consistent relationship to the
assignment of the stimuli to the distinct classes, this classification will have no effect
on the judged relationships in the physical dimension between the stimuli of the
series’ (pp. 20-21).
The effects predicted in paragraph (i) above depend on the information which the
classification offers regarding the differences between the stimuli. Each stimulus is
judged individually, and each has a label identifying it as belonging to one or another
of two classes. I n the subject’s past experience, the stimuli belonging to the distinct
classes differed consistently from each other along the quantitative dimension which
he is judging. The effect of this situation on judgement can be stated in terms bor-
rowed from the experimental literature on anchoring.
An anchoring stimulus placed outside a series of stimuli may determine either
a contrast or an assimilation effect in the judgements of the stimuli of the series. I n
the case of contrast, judgements of the stimuli shift away from the value of the anchor
stimulus; in the case of assimilation they shift towards that value. There exists a
good deal of evidence that when the anchor stimuli are placed relatively far outside
the original series, they determine a contrast effect; when they are placed very near
or a t the extremes of the series, they often lead to an assimilation effect.
It is a fair assumption thai; the judgemental process underlying the contrast and
assimilation effects reduces itself to the perceived extent of similarity and of difference
ClassiJication and qwantitative judgement 103
between the anchoring stimuli and the stimuli of the series. When the anchoring
stimulus is perceived as definitely different from those nearest to it in the original
series, the judgements of these stimuli rdflect this differentiation by shifting away
from the judged value of the anchor; when the anchoring stimulus is perceived as
very similar to the stimuli of the series, the opposite happens: the judgements of the
stimuli nearest to the anchor reflect the subjective equivalence by shifting towards
the judged value of the anchor stimulus.* This would fit in with the general results
of experiments on anchoring which show that when the values of anchoring stimuli
progressively approach the values of the end stimuli of the original series, the extent
of contrast is reduced (e.g. Heintz, 1050) until finally assimilation takes over (Sherif,
Taub & Hovland, 1958).
A classification superimposed on a series of stimuli may be regarded as determining
the same type of shifts in the extent of judged similarities and differences between
the various stimuli. The nature of these shifts will depend on the relationship between
the division of the stimuli into classes and the pattern of variation in the magnitudes
of the stimuli which the subjects are requested to judge.
I n a series in which a classification is directly and consistently related to the
physical dimension under judgement, the class identification of a stimulus provides
a supplementary source of information about the relationship of its magnitude to
the magnitudes of other stimuli, whether identified as belonging to the same class or
to a different class. In absolute judgements upon a series of stimuli which are
presented in successive random orders, ail identity in the lahelliiig should therefore
produce assimilation or convergence in the assessments, whereas a difference should
produce contrast or divergence. Therefore, the differences between the judgements
of magnitude of stimuli belonging to the distinct classes should in such a series be
greater than in an identical series on which a classification has not been super-
imposed. At the same time, differences between the judged magnitudes of stimuli
belonging to the same class should show a tendency to be smaller than the corre-
sponding differences in an unclassified series.
There is no reason to assume that any consistent shift in the differences between
the judged magnitudes of stimuli belonging to distinct classes should be observed
when a classification superimposed on a series stands in no coherent relation to the
magnitudes of the stimuli. I n this situation, ‘the physical magnitude of a stimulus
does not help in its assignment to one of the various classes, and the identification
of a stimulus as belonging to one of the classes does not provide any cue as to its
physical magnitude’ (Tajfel, 1959a, p. 23).
Two subsidiary hypotheses concerning the effects on judgement of a consistent
relationship between a classification and physical magnitudes can be briefly stated
as follows:
(i) The predicted effects of an increase in the differences between the judged
magnitudes of stimuli belonging to the different classes and of a decrease in such
differences between stimuli belonging to the same class should become accentuated
as the classification is made more salient to the subjects. The term ‘salience’ is used
* This assumption was independently validated in experiments (Tajfel & Wilkes, in preparation) in
which the relative extent of contrast was seen to depend on the manner in which the anchoring stimuli
were classified in relation to the classifications imposed on the stimuli of the original series.
7-2
104 HENRITAJFELA N D A. L. WILKES
here in a restricted sense. A classification may become more salient to a subject
because of an increase in the amount of past experience with i t ; or because of its
unvarying consistency. I n the present experiments salience was manipulated by
presenting the subjects with varying conditions of visual impact of the classification.
(ii) These effects should also become accentuated in direct relation to the amount
of past experience in judging the classified series.
11. METHOD
The purpose of the experiments was to assess the effects of classifications on simple
quantitative judgements. Judgements of length of a series of lines were used. This
choice was dictated by the relative simplicity of judgements of length; if it is found
that classification exercises its predicted effects in this context, that ‘stereotypes ’
can be formed about the length of lines, then such effects are likely to be even more
marked when the complexity of the task of judging increases, with the corresponding
tendency of the subject to rely increasingly on all the available sources of informa-
tion about the stimuli.
Two pilot experiments showed that the series of stimuli described below was
clearly discriminable in an absolute judgement situation, and that the effects of
classification clearly tended in the predicted directions.
StiWlUli
A series of eight lines differing from each other by a constant ratio of approximately 5 yo of
length were used as stimuli in all the experiments. The shortest line was 16.2cm. long; the
longest 22.9 em. Each line was drawn diagonally on a sheet of white cardboard whose dimensions
were 63.5 cm. x 50.8 cm. The relatively large size of the sheets of cardboard was chosen in
order to minimize the cues about the differences in the length of lines that could be provided
by comparisons with the frame. Each length of line was presented several times (see below),
but a separate sheet was prepared for each presentation of each stimulus, in order to avoid
the possibility of an identification of a stimulus on the basis of irrelevant cues.
Subjects
Sixty-one subjects took part in the first two experiments (Expts. I a and I I a ) . They were
volunteers, men and women, drawn from students a t Oxford University, Westminster Training
College, and Manchester University. Of those sixty-one subjects, fifty-four were available for
the second sessions (Expts. I 6 and I1 b).
Procedure
(i) Procedure common to all experimental situations
The eight lines were presented one by one six times in successive random orders. The subjects
were requested to judge tho length of each line in centimetres. Judgements in centimetres
rather than inches or fractions of inches were used because in this way the differences in length
between the stimuli could be made fairly small while avoiding at the same time the tendency
for rough judgements in terms of the nearest inch. Before the beginning of judgements, a ruler
was shown and several examples of relationship between centimetres and inches were given.
The subjects were tested individually. They were seated facing the stimuli which were presented
by the experimenter a t a distance of about 8 ft. They called out their judgements which were
recorded by the experimenter. There was no time limit for the presentation of a stimulus which
was held in front of the subject until the judgement was made. The interval between the presen-
tation of successive stimuli was 4 see. The subject was not told how many different lengths of
lines were presented t o him. Each subject was requested to come for a second session a week
a.fter thc first.
Table 1 presents the summary of the procedures used in the four experiments.
ClassiJication and quantitative judgement 105
(ii) Classification
Three groups of subjects served in Expts. I a and I b , and two groups in Expts. I I a and I I b .
Experiment l a . Group C (classified). This group was presented with a classification super-
imposed on a series in such a way that there was a stable and predictable relationship between the
lengths of lines and their labels. Each of the four shorter lines had a large letter A drawn above
its middle a t each presentation; each of the four longer lines a letter B drawn in the same way.
The group consisted of twelve subjects.
Group R (randomly classified). This group was presented with a classification which bore no
relationship t o the lengths of lines. The stimuli were labelled with letters A and B as for group C;
but each of the eight stimuli was labelled A a t half of its presentations and B a t the other half.
The order of appearance of labels A and B was randomized for each stimulus. The group con-
sisted of thirteen subjects.
Group U (unclassified). Conditions of presentation of stimuli for this group were identical
to thosc for groups C and R, the only difference being that the lines were presented without any
labels. The group consisted of twelve subjects.
Experiment I b . A week later, the subjects in groups C, R and U were tested again. The con-
ditions of presentation of stimuli were the same for each group as in Expt. I a . In group C, ten
out of the original twelve subjects were still available; all the thirteen subjects from group R
reappeared; and eleven out of twelve from group U.
Experiment I I a . Group C, (classified). Conditions of presentation of stimuli were tho sa.mo
aa for group C in Expt. I a . The group consisted of twelve subjects.
Group U , (unclassified). Conditions of presentation of stimuli were the same as for group U
in Expt. I a . The group consisted of twelve subjects.
Experiment 11b. The same subjects were tested again a week later. Tho conditions of presenta-
tion of stimuli were the same for each group as in Expt. I I a . Eleven out of twelve subjects
were still available for group C,, and nine out of twelve for group U,.
111. RESULTS
(i) Description of the general pattern of results
Table 2 sets out the general results: means of all the control groups (randomly
classified and unclassified) were combined, as their patterns are highly similar. The
table shows the mea,ns of the judgements of each stimulus. Stimuli have been
numbered from 1 to 8 in order of increasing length. Stimuli 1 to 4 belong to class A,
stimuli 5 to 8 to class B in the classified series (groups C and Cl). Therefore, the
inter-class difference is between stimuli 4 and 5. All the other differences between
adjacent stimuli are intra-class differences. The data from Expt. I b include only
the first six presentations of the series.
Groups C and C , 15.6 16.5 17.2 18.3 20.3 21.6 22.4 24.4
Groups R, U 16.6 17.4 17.9 19.0 20.3 21.3 22.8 24.6
and U,
ClassiJcation and quantitative judgement 107
I n Figs. 1 and 2 the actual difference between each two adjacent stimuli was
taken as a basis, and the deviations of the judged differences from the actual ones
are expressed as percentages of the actual differences. These deviations are plotted
as positive when the judged differences are greater than the actual ones ; as negative
when they are smaller.
D
C
2
& +loo -
-3-
z
+75-
:
g +50- Classified series
22
-$
.-
.*
+25-
$ 0
-u
0
2 -2j-
B
2z -50 I I I l l l l
1-2 2-3 3-4 4 5 5-6 6-7 7-8
Adjacent stimuli
Fig. 1. Comparison of actual and apparent differences between adjacent stimuli in
Expts. Ia and IIa.
Adjacent stimuli
An inspection of Figs. 1 and 2 will show that, as predicted, the inter-class differences
(between stimuli 4 and 5) are by far larger for the classified series (groups C and C,)
than for the randomly classified and unclassified ones, both in the first and in the
second sessions. I n the first sessions the relevant percentage figures are as follows:
apparent difference 100 % greater than the actual difference for the classified series ;
11 yogreater for the randomly classified and unclassified ones. In the second sessions :
122 % greater than the actual for the classified series, 44 % for the unclassified ones.
This represents, of course, a considerable exaggeration of the actual inter-class
difference for the classified series. There are two other points a t which a difference
between two adjacent stimuli is markedly exaggerated for these groups. The first
108 AND A. L. WILKES
HENRITAJFEL
is between the stimuli 7 and 8 ; but this seems t o be a consistent phenomenon for all
the groups of subjects and is therefore probably inherent in the nature of the range
of stimuli and in the pattern of presentation. The differences between the experi-
mental and control groups are very small as compared with corresponding differences
at the point of the inter-class break.
The second point at which the intra-class differences are considerably larger than
the actual differences for the classified series is between the stimuli 1 and 2 in the
first sessions for groups C and C,. At this point, the difference is not only larger than
the actual one, but also larger than the differences in the control groups (the percen-
tages are : apparent differences 62 yogreater than the actual differences for the classi-
fied series ; 13 yo greater for all the control series). The direction of this relationship
between the classified and control series is maintained in the second sessions (see
Fig. 2). I n the classified series the apparent difference between the stimuli 1 and 2
is 13 % greater than the actual difference ; in the control series the apparent difference
is 11 % smaller than the actual one. This problem will be dealt with in the section
devoted to the discussion of results.
Table 3 provides the group means of the scores relevant to the analysis, namely:
mean intra-class slopes +(ma+ m,) and the inter-class differences (yS - Y4).
The results of the statistical analysis were as follows :
( a ) The analysis of variance for the inter-class differences (y5-Y4) in Expts. I a
and I I a shows that the classification, as predicted, determines an increase in the
apparent inter-class difference (P ratio for the classification is 6.12: P < 0.025).
Differences in the extent of salience (manner of prior presentation) do not lead to any
changes in the apparent inter-class differences.
( b ) The analysis of variance for mean slopes &(m,+m,) shows that neither the
classification nor its salience determined any consistent differences in the apparent
intra-class differences This does not confirm the prediction that an introduction of
the classificationwould lead to an increase in the judged similarity of stimuli belonging
to the same class.
110 AND A. L. WILKES
HENRITAJPEL
( c ) Analysis of variance for Expts. I b and I I b shows results parallel to the previous
analysis. The predicted effects of classification for inter-class differences are significant
(F ratio = 4.40;P < 0.05). No other differences between the groups are significant.
( d ) A comparison of apparent inter-class differences between groups C (classified)
and R (randomly classified) in the first session (Expt. I a ) yields a strong tendency for
the differencesin group C to be larger, as predicted; but this tendency does not reach
statistical significance. In the second session (Expt. I b), this difference between the
groups is significant in the predicted direction at P < 0.01. A comparison of the two
groups taking as a score for each subject the means of his performance in the first
and second sessions is significant at P < 0.025 in the direction of larger inter-class
differences for group C. There are no differences between the intra-class slopes of the
two groups, either in the first or in the second session.
Stimuli ... 1 2 3 4 6 6 7 8
Actual values 16.2 17-0 17.9 18.8 19.7 20.7 21.7 22.8
Group C 17.1 17.9 18.6 19.1 21.4 22.0 22.8 24.2
Groups R and U 16.9 18.1 18.6 19.8 20.4 21.2 22.2 23.8
Adjacent stimuli
Fig. 3. Comparison of actual and apparent differences between adjacent stimuli in the
second part of Expt. Ib.