our experimental conditions than did equally strong stimuli. For this reason we made the first and second stimulus in successive presentations, respectively 40 and 20 msec. When we used pictures of 20 msec (see fig. 1), the picture consisted of either the odd or the even lines. In this way of course the second stimuli also have a lower resolution than the first stimuli and consequently less contour in formation. But though a favourable strong-weak relation of first and second portraits might function as a sort of catalyser, it cannot explain differentiation as such between our experimental conditions.
In case of simultaneous (SOA = 0) presentations of the same stimulus pairs, both photographs were presented completely (40 msec, odd and even lines). However, to control brightness differences between simultaneous (for example the background would be much brighter now) and successive presentations, we only used half intensities for each of the photographs in case of simultaneous presentation. This electronic integration of the two portraits in case of simultaneous presentations might be different from the one accomplished by the eye in case of successive presentations. One can imagine a relative
emphasis on local similarities between both photographs because of a diffusion of local differences. However, again this situation is equal for all our experimental conditions. Eventually we may find a general difference between simultaneous and successive presentations (see also our discussion of the main effect of SOA). But neither these differences in stimulus quality, nor those between the first and the second stimulus, can explain the eventual differential effect of our experimental conditions with increasing SOA.
As pre- and post-adaptation fields, and during the empty periods of the SOA, we used a square field of the same size as the photographs, which was illuminated at the same intensity as the backgrounds of the portraits.
A dim light was arranged in such a position that the S could not see any reflections in the glass-scope. When the S pushed a starting-key the first trial of the series was presented.
In the experiment the S was sitting in front of a display-scope. The visual angle of the faces on the screen was approximately 5 degrees. We did not use a fixation mark. Nevertheless most Ss will mostly more or less fixate the centre of the screen. But sometimes an S might prefer to look at the right or the left side. In the same way the S could try to attend only to the first or the second portrait. It should be noticed, however, that a crucial point of our experimentation is a complete balancing of stimulus-pairs
over all conditions. Together with a specific correction for guessing this has to prevent that any unintended strategy of our Ss can produce a Type I error. For this purpose we are prepared to pay the price of a larger risk of a Type II error, i.e. unjustly accepting the null hypothesis.
When both faces are in similar positions the S seems to see only one face. When both faces are in different positions the S sees something like a Janus-face. If the S has identified one face, correct identification of the other face is, however, seldom much higher than about chance-level. Therefore we
instructed the S to identify only one model at a trial. This instruction does not offer many problems: the S simply identifies the face he or she has seen most clearly. After the S had signalled a choice to the computer by means of one out of six response-keys, which corresponded to the alphabetically arranged names of the six models, the starting-key could be used again for a new trial. Presentation of a trial started within 20 msec after the starting-key was pushed. Response-errors could be corrected by means of a correction-key, which was on the same response-panel as the starting-key and the response-keys.
Scoring and correction for guessing
To make sensible comparisons between different conditions we should normalise for unequal numbers of trials and correct for blind guessing as well as for guessing on the basis of only seeing the spectacles on the first and/or the second photographs. For this purpose we developed a specific "two-stepped" correction for guessing. From here on we shall simply call the corrected proportion of correct identifications of the first and/or the second photographs in a condition, respectively Pc1, Pc2 and Pc1 + Pc2.
Because there will always be some masking as a result of superposition of the two faces, a crucial test will have to compare different conditions with respect to the relative shifts from the identification of the first photograph to the identification of the second photograph. If correct detection of position is indeed primary and essential for identification, there will be stronger initial shifts towards identifying the second photograph as a function of increasing SOA when positions of the two coincide than will be the case
when the two positions are different. Therefore our first prediction is that with increasing SOA there will be a specific interaction: Position X Recency X SOA.
However, in Calis' (1974) earlier experimentation with a dichoptic stimulus presentation this interaction was actually destroyed because already with short SOAs the preference for the second photograph in case of similarity of position and spectacles increased about as much as it decreased in case of position similarity and spectacle dissimilarity. Thus if the first portrait has the same position as the second one but then belongs to a different set, it may very effectively put the perceiver off the scent. Of course this does not falsify the theory, because it also predicts that an identification process that has already established position as well as the presence or absence of spectacles on the basis of the first photograph, is now "looking" for specific individual features. But precisely these features cannot be found in the second photograph because this belongs to a different set. In this case the identification process would be derailed in such a late phase, that the critical iconic data are already too deficient to make a new round successful. Perhaps this effect was due to the dichoptic stimulus presentation, which might produce less
interference, or speed up the identification process in comparison with the binocular presentation of portraits we used in this experiment. For us it means that finally our second prediction might be more crucial. Once position is correctly established, the shift to the second photograph can be enhanced even more if there is also continuity with respect to the presence or absence of spectacles. This effect would be maximal of course if both photographs are completely identical. However, because we only have Pc1 + Pc2 scores in this case, we cannot directly see such effects in our data. In our analysis of variance,
which for this last reason only concerns the identifications with portraits of two different persons, this crucial test means that we predict a significant higher order interaction: Spectacles X Position X Recency X SOA.
Fig 1. Stimulus presentations.
t0 = start
t1-t0 <= 20 msec.
t2-t1 = 40 msec.
t3-t2 = 20 msec.
t4-t3 = 20 msec.
SOA = Stimulus Onset Asynchrony
S1 = first portrait
S2 = second portrait
S1/S2 = simultaneous superimposed
(each half intensity)
B = blank field (same intensity and quality
as the background of the portraits)
Results and conclusion
The results are presented in fig. 2 and table 1. As can be seen from a comparison of predictions and results, there is significant confirmative evidence for our hypothesis. Hence we believe we have demonstrated some facts concerning the hierarchical-sequential nature of immediate perceptual recognition of familiar faces.