Professional Documents
Culture Documents
Ekman and Kuennapas, Scales of Aesthetic Value
Ekman and Kuennapas, Scales of Aesthetic Value
Ekman and Kuennapas, Scales of Aesthetic Value
'This investigation was supported by the Swedish Council for Social Science Research.
The computational work was done by Miss Inga Sommarstroem. A transcript of data
has been filed as Document N o . 6931 with the American Documentation Institute,
Auxiliary Publications Project, Photoduplicarion Service, Library of Congress, Washing-
ton 25, D. C. Remit $1.25 for 35-mm. microfilm or photocopies.
G. EKMAN & T. KUENNAPAS
0 1 2 3 1
arranged in three groups on the basis of the preliminary sorting. Each group
contained 7 stimuli, 2 of which were common to another group (Group I: A, B,
C , D , E , F , G ; GroupII: F , G , H , I , J,K,L; GroupIII: K , L , M , N , O , P , Q ) .
The number of combinations within each group is 21. Since two of these
combinations belong to more than one group, the total number of pairs to be
compared is 61.
The drawings were photographically reduced, and the 61 different pairs
were printed in random order on test sheets, two or three pairs on each sheet.
One special set of sheets was printed for category estimation. Each sheet
showed drawings A and Q as anchoring points, and one of the other drawings
to be estimated, so that there were 15 such sheets. The order of sheets was
randomized among Ss.
Procedure.-For the pairs of drawings Ss were instructed ( a ) to mark that draw-
ing in each pair which they liked best, and ( b ) to estimate in per cent how much they
liked the other drawing i n relation to the one they preferred. T h e instruction for
category estimation was ( c ) to regard d r a w ~ n g sA and Q as points 1 and 9 o n a scale
extending both below and above these polnts, and to assign a proper scale value to
each of the other drawings; it was emphasized char the estimate should reflect subjective
difference with regard to aesthetic value.
Subjects.-The number of Ss was 100 in pair comparisons and ratio estimation; 85
of these Ss also participated in category estimation. Most Ss were psychology students,
and about a dozen were staff members. The majority of the group had some previous
experience of psychophysical judgment.
Scale const~action.-The ratio scale was constructed from geometric means
of ratio estimates by a computational procedure suggested by one of the present
authors (Ekman, 1958). This was first done separately for each of the sub-
matrices. Then a scale with a common unit of measurement was constructed
in the following way. The scale from the second sub-matrix was multiplied
by a constant factor chosen so that the average value of Stimuli F and G was
SCALES OF AESTHETIC VALUE 21
made equal to the average value of the same stimuli in the scale from the first
sub-matrix. The third scale was transformed in an analogous way. By this
procedure, two slightly different scale values were obtained for each of Stimuli
F, G , K, and L; they were averaged. Finally, the lowest of all scale values
(Stimulus C) was chosen as the arbitrary unit. The final scale is shown
in Table 1.
The interval scale was constructed according to Thurscone's Case V (Thur-
stone, 1927; cf. Torgerson, 1958). First, one separate scale was obtained from
each sub-matrix. Then a constant was added to the scale of the second sub-
matrix so as to make the average scale value of Stimuli F and G equal to the
average value of the same stimuli in the scale from the first sub-matrix. The
scale from the third sub-matrix was transformed analogously. Means were
TABLE 1
THE FINAL SCALES
Drawing Ratio scale Interval scale Category scale
A 1.27 4.73 (1.00)
B 1.34 3.97 0.13
C 1.00 1.28 0.78
D 1.01 0.00 0.97
E 2.10 10.58 4.15
F 1.58 6.94 3.09
G 1.58 6.99 2.63
H 2.32 11.03 5.56
I 1.58 5.97 2.68
J 2.40 12.37 4.78
K 2.22 10.89 4.41
L 2.40 11.53 4.77
M 3.85 18.01 7.79
N 3.08 15.25 6.86
0 3.57 17.90 7.17
P 3.09 17.46 6.44
Q 4.60 22.58 (9.00)
computed for Stimuli F, G, K, and L from the two slightly different scale
values obtained by this procedure. The final scale was obtained by shifting
the arbitrary origin to the lowest of all scale values (Stimulus D ) ; it is shown
in Table 1.
The category scale was obtained simply by computing the average scale
(category) value for each of the 15 stimuli estimated in the experiment. The
scale is also shown in Table 1.
DISCUSSION
OF RESULTS
The ratio scale.-The total range of the ratio scale in the present experi-
ment is 1:4.60 in terms of the ratio between the lowest and the highest scale
value. A sample of six stimulus drawings, so selected that they represent
nearly equal steps on the ratio scale, are shown in their proper scale positions
in Fig. 1.
22 G. EKMAN & T. KUENNAPAS
00 10 2n 30 LO 50 00 01 01 03 04 05 06 0.7
FIG. 2 A. The "interval" scale plotted against rhe ratio scale. 2 B. The "inrerval"
scale plotted against the logarithm of the ratio scale.
that the other scale is a true ratio scale, which, however, we will assume is the
case for the time being.
The "interval" scale was constructed according to Case V, i.e., on the
assumption of constant "discriminal dispersion." As is well known from the
work of Stevens and others, there is evidence that che subjective uncertainty
(intra-individual variability) increases with the magnitude of the subjective
variable, at least in "prothetic" continua (Stevens & Volkmann, 1940; Harper &
Stevens, 1948; Stevens, 1957, 1960; Ekman, 1956). Some experiments in-
dicate that this kind of dispersion is directly proportional co the scale position
(Ekman, 1956,1959).
It is conceivable or even plausible that a similar relation holds for the
dispersion between individuals, with which we are concerned here as one
SCALES OF AESTHETIC VALUE 23
FIG. 3 A. The category scale plotted against the ratio scale. Filled circles represent
anchoring stimuli. 3 B. The category scale plotted against the logarithm of the ratio
scale.
The category scale.-The category scale is plotted against the racio scale
in Fig. 3 A. The trend of the data is clearly curvilinear, in about the same way
as for the "interval" scale in Fig. 2A. The result is in general agreement wich
the results obtained by Stevens and Galanter ( 1957). As compared wich the
racio scale, the trend of the category data is "concave downwards."
Stevens and Galanter offer an explanation of this trend in terms of the in-
24 G . EKMAN & T. KUENNAPAS
creasing uncertainty upward on the scale; this is the same type of interpretation
as has been discussed above in connection with the "interval" scale. An alterna-
tive explanation for category scales is, however, possible. It is based on the
interpretation of a perceptual "distance" in terms of perceptual "similarity."
This interpretation and its possible consequences for scaling theory will be
briefly outlined below.
The problem of how a perception of similarity arises, has been investigated
in several experiments in this laboratory. In the first experiment the similarity
of pure tones of varying pitch and equal loudness was studied (Eisler & Ekman,
1959). A pitch scale was constructed, and a scale of subjective similarity was
obtained. Then, it was possible to investigate similarity as a function of pitch.
It was found that
This means that eqaal steps or distances on the category scale will correspond
to eqzial successive ~ a t i o son the ratio scale.
As has been pointed out in previous papers, this hypothetical interpretation
of category scaling implies a logarithmic relation between the category scale
and the ratio scale (Ekman, Goude, & Waern, 1961; cf. Junge, 1960). The
hypothesis has been tested in the present experiment by plotting the category
scale (R,) against the logarithm of the ratio scale (R,) in Fig. 3B.
The relation is clearly linear, with the equation R, = - 0.02 +13.33 R,.,
and the hypothesis may be considered verified. This is, however, so far an
isolated finding. If the category scales and ratio scales obtained by Stevens
and Galanter (1957) are plotted as in Fig. 3B, it will be seen that the relation
is logarithmic for the greater part of the range, but that this relation breaks
down completely in the low range of the scales; the same is true for several
sets of data obtained in this laboratory. On the other hand, power functions
SCALES OF AESTHETIC VALUE 25
may always be fitted to such data; the exponents are always below 1 (cf. Ekrnan
& Kuennapas, 1960b). The category scale of the present experiment was ob-
tained in a way that differs somewhat from the usual procedure. Ss were in-
suucted to regard the stimuli denoted 1 and 9 as anchoring points which, how-
ever, are not necessarily the end points of the scale. The usual break-down of
the logarithmic relation in the low range of the scales may, thus, be a con-
sequence of an end effect, which was eliminated in this particular experiment.
W e are not accepting this interpretation; it is mentioned merely as one point
of consideration for further research on the relation between ratio scales and
category scales.
"Interval scale"
FIG. 4. The category scale plotted against the "interval" scale. Filled circles rep-
resent anchoring stimuli.
In Fig. 4, finally, the category scale is plotted against the "interval" scale.
This relation is linear, as can be expected from the linear relations in Figs. 2B
and 3B; the straight line in the graph represents the equation R , = - 0.04 +
0.41 Ri,as determined from the equations relating the "interval" scale and the
category scale to the ratio scale. The graph illustrates that in this experiment
similar scales were obtained by the direct method of category estimation and the
indirect method of pair comparisons on the assumption of Case V. In a sense,
it also illusuates that the assumption of Case V is not necessarily unrealistic:
it appears applicable when the observer is instructed in terms of subjective
di~tancebut not when he is instructed to give ratio estimates. This apparent
paradox constitutes an interesting problem.
G. EKMAN & T. KUENNAPAS