LOOKING AT UPSIDE-DOWN FACES1
ROBERT K. YIN 2
Massachusetts Institute of Technology
Memory for faces was compared with memory for other classes of familiar and complex objects which, like faces, are also customarily seen only in one orientation (mono-oriented). Performance was tested when the inspection and test series were presented in the same orientation, either both upright or both inverted, or when the two series were presented in opposite orientations. The results show that while all mono-oriented objects tend to be more diffi- cult to remember when upside-down, faces are disproportionately affected. These findings suggest that the difficulty in looking at upside-down faces involves two factors: (a) a general factor of familiarity with mono-oriented objects; and (6) a special factor related only to faces.
It is a well-known fact that pictures of hu- man faces, when viewed upside-down, are extremely difficult to recognize (Arnheim, 1954, p. 86; Attneave, 1967, p. 26; Kohler, 1940, p. 60). Kohler not only noted this, but also speculated that the difficulty was attributable to the loss of “facial expression” in the inverted picture. More recently, in- vestigators have examined this phenomenon in several ways. Brooks and Goldstein (1963) showed that recognition of inverted faces is worse than that of upright faces when children are asked to identify snapshots of their classmates. That memory for in- verted faces is poorer than memory for up- right faces among adults has been shown in a paired-associate task (Goldstein, 1965) and a recognition task (Hochberg & Galper, 1967).
These studies have not indicated the ex- tent to which the difficulty in viewing an up- side-down face is related specifically to the face. An alternative hypothesis would be that any set of objects customarily seen in one orientation, i.e., mono-oriented, might be more difficult to recognize when inverted. Some evidence for this hypothesis was re-
1 This study was supported by a grant from the John A. Hartford Foundation, Inc. (New York, N. Y.) to H.-L. Teuber and a predoctoral award to the author from the National Science Foundation. The author gratefully acknowledges the advice and encouragement of H.-L. Teuber throughout all phases of this work.
2 Requests for reprints should be sent to Robert K. Yin, Department of Psychology, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139.
ported by Henle (1942), who showed that alphabetic letters were correctly perceived more frequently than their mirror reversals by 5″s familiar with the letters, and by Ghent (1960), who found that young children are markedly dependent on familiar orientation for recognizing realistic figures. In addition, Dallett, Wilcox, and D’Andrea (1968) re- ported that memory for upright magazine pictures was better than that for the same pictures when presented upside-down. The investigators did not indicate, however, the extent of homogeneity among the pictures or the degree to which the pictures were of ob- jects that are customarily mono-oriented.
The present experiments were designed to test whether a general impairment on mono- oriented objects when inverted could account for the difficulty with viewing upside-down faces. More specifically, performance on up- right and inverted tasks for faces was com- pared with that for other classes of every- day objects having a priori properties similar to faces in being mono-oriented, familiar, complex, and not easily verbalized, i.e., ob- jects that are not distinguished from each other by the use of simple labels.
To test performance, a forced-choice recog- nition memory task was used. In this task, 5s were shown individual pictures (an in- spection series) and then presented with pairs of pictures (a test series). In the test series they indicated the one of the pair they thought they had seen in the in- spection series. Three experiments were conducted. In the first and third, the
142 ROBERT K. YIN
orientation of the materials in both the in- spection and test series was the same (both upright or both inverted). In the second, the orientations were opposite (inspected upright and tested upside-down or inspected upside-down and tested upright).
EXPERIMENT I Method
Subjects.—There were 26 paid volunteers, 13 men and 13 women, ranging from 18 to 31 yr, of age (mean age = 21.7 yr.). These were under- graduate and graduate students attending summer schools in the Boston area and represented a wide variety of geographical origins and academic interests.
Materials.—There were 64 different pictures, all black and white, within each of four types of materials: faces, houses, airplanes, and men in motion. All pictures were pasted on a 3 X 5 in. card for presentation.
The faces were studio pictures of adult males, chosen to be similar with respect to general age, expression, and lack of outstanding distinguishing features, such as glasses, beards, or unique marks. All poses were full face, and the pictures were trimmed just under the chin to eliminate as much clothing as possible. The houses were generally of the same architecture, but were not as uniform as the faces in orientation of view or size of pic- ture. In addition, since all were actual photo- graphs, the pictures included trees and other nat- ural features, although they were trimmed to minimize the presence of distinguishing features, such as fences, front stoops, and roof markings.
Neither the airplanes nor the men in motion were real photographs, but were caricatures. The planes were sideview silhouettes of all types and models (military, commercial, and private) of planes found in the world today. The men-in-mo- tion pictures consisted of the same cartoon stick figure engaged in various everyday movements and postures, with no other objects present in any of the pictures.
Procedure.—Each S1 looked at an inspection series of 40 pictures, presented singly and turned
TABLE 1 MEAN ERRORS, Exp. I
Faces Houses Airplanes Men in motion
Test and Inspection series upright
.89 2.23 3.65 2.35
1.09 1.60 1.69 1.27
Test and inspection series inverted
4.35 3.42 3.85 3.27
1.41 1.36 2.03 1.58
by E at a rate of 3 sec. per picture. Then a test series, consisting of 24 pairs of pictures, was pre- sented. Each pair contained 1 old picture (an exact duplicate of a picture in the inspection series) and a new picture (one not previously shown), and S had merely to indicate which picture in each pair was the old one. The S proceeded at his own rate in the test series. Since only 24 pairs were in the test series, there were 16 pictures in the inspection series which did not recur in the test series.
Each inspection and test series constituted a block and was a mixed list, containing two different types of materials, 20 of each in the inspection series and 12 pairs of each in the test series. The order of presentation of the 40 inspection series pictures was randomized, with the two exceptions that neither of the two materials was shown for more than four consecutive cards and that there were always at least 2 of the nonrecurring pictures, one of each type of material, at either end of the series. The order of the 24 test series pairs was dictated by the order of pictures in the inspection series, so that there was a constant lag between each inspec- tion picture and its occurrence in the test series.
All 5s went through four such blocks of inspec- tion and test series, viewing two blocks rightside- up (both series upright) and two upside-down (both series inverted). Thus each S performed in all experimental conditions, viewing the four materials in two presentations. The order of presentation among the blocks was balanced in the following manner: (a) Each 5″ was shown all four materials (two blocks) first; half of the 5″s saw these two blocks upside-down first, the other half rightside-up first; (6) the mixing of the materials was such that roughly one-third of the ,9s had blocks consisting of faces-houses or airplanes- men-in-motion, one-third had blocks of houses-air- planes or faces-men-in-motion, and the remaining third had blocks of airplanes-faces or houses-men- in-motion; (c) the blocks were alternated so that each picture was shown rightside-up as often as it was upside-down; and (d) the sexes were balanced with regard to all of these conditions.
The mean errors, with their standard de- viations, appear in Table 1. An analysis of variance of the error scores showed that there were significant differences as a func- tion of presentation, F (1, 25) = 90.90, p < .0005, materials, F (3, 75) = 6.63, p < .001, and their interaction, F (3, 75) = 9.18, p < .0005.
Although all materials were more difficult in the inverted presentation, the extent to which each type of material contributed to this effect varied. Using t tests for matched pairs, two-tailed, the effect of inversion was
RECOGNITION OF INVERTED FACES 143
greatest for faces, t (25) = 8.48, p < .001, significant but not as great for the houses, t (25) = 3.01, p < .01, and the men in mo- tion, t (25) = 2.15, p < .05, and not signif- icant for the airplanes, t (25) < 1.
The materials also differed in their overall difficulty. Although this finding is not of primary interest here, the major reason for it was that the airplanes tended to be the most difficult material in either presentation.
Of greater interest is the fact that the Presentations X Materials interaction was significant. Further analysis showed that this was due mainly to the faces, which were easier than all the other materials when viewed upright, * (25) = 7.31, p < .001, but more difficult than the rest when viewed upside-down, t (25) = 2.53, p < .02. Ex- amination of the individual scores produced added evidence of the existence of a dif- ference between faces and the other ma- terials. In general, those who did better in the inverted orientation also tended to be the ones who did better in the upright orientation. However, for faces, the reverse was true. Taking the average inverted score for houses, airplanes, and men in motion, and arbitrarily assigning all 5″s to a better group (n = 14, average error = 2.88) and a worse group (» = 12, average error = 4.25), the better group is also better on the upright task (average error = 2.36), while the worse group is still the worse one (aver- age error = 3.19). Using a t test for inde- pendent samples, two-tailed, the difference between the two groups in their upright scores is significant at the p < .05 level, * (24) = 2.46.
On the other hand, arbitrarily assigning all 5*8 by their score on the inverted-face task to a better group (« = 14, average error = 3.29) and a worse group (» = 12, average error = 5.58), we find that the better group is now the worse one on the upright-face task (average error =1.29), while the worse group is the better one (average error = .42). This difference on the upright-face task is significant at the p < .05 level, t (24) = 2.14.
Sex differences.—Men and women did not differ in their total upright or inverted scores. There were differences between ma-
TABLE 2 MEAN ERRORS, EXP. II
Faces Houses Airplanes Men in motion
3.81 2.86 3.19 4.05
1.71 1.83 1.94 1.79
5.14 3.43 4.14 4.24
1.39 1.47 1.58 1.72
terials, however, in that the men’s average upright and inverted score for airplanes was better than that of the women, t (24) = 2.26, p < .05, while the women’s average upright and inverted score for houses was better, but not significantly, than that of the men, t (24) = 1.91, p < .10. In both cases the t test was for independent samples and two-tailed. There were no sex differences for either the faces or the men in motion.
Order of presentation.—There were no differences when the groups were character- ized by viewing order, upright first or in- verted first, or by the mixture of the ma- terials in the different blocks.
Experiment II required 5″s to make a mental inversion of the materials, presenting the inspection and test series in opposing orientations.
Method Subjects.—There were 21 paid volunteers, 13
men and 8 women, ranging from 18 to 26 yr. of age (mean age = 21.1 yr.). The general nature of the sample was the same as that of Exp. I.
Materials and procedure.—The materials were the same as those used in Exp. I, and the overall procedure was exactly the same with one excep- tion: For each 5 the two presentations were up- down (inspection series presented upright and test series inverted) or down-up (inspection series pre- sented inverted and test series upright). As in Exp. I, each 5 performed in all experimental con- ditions, viewing the four materials in both pres- entations.