Assume that guessing became more important as the task became more difficult.
First, consider Experiment 3, in which the probe was always presented at the immediate
offset of the stimulus array. It seems reasonable to assume that guessing would not affect
p(L) because there was virtually no location uncertainty. The difficulty of the task was
determined solely by how closely the mask followed the stimulus array. The "genuine"
probability of recalling item identity may be called p'(I) in order to distinguish it from
p(I) (the observed probability of recall of item identity). The reasonable assumption here
is that p'(I) was smaller than p(I) at short mask-ISIs. That is, the "genuine" p(L|I) at short
mask-ISIs might have been underestimated as a result of guessing. Consequently, the
left-hand end of the dotted-line function in the right bottom panel of Figure 5 should have
been higher than it currently is. However, such an elevation of the dotted-line function
might not be sufficient to turn the function into one with a significant negative slope. In
any event, the data would still be inconsistent with the expectation of the dual-buffer
model.
The mask (when applicable) was presented at the immediate offset of the stimulus
array in Experiment 2. Only the delay of the probe was varied. It is reasonable to assume
that guessing became more important at longer probe-ISIs. That is, p'(1) is smaller than
p(I), and p'(L) is smaller than p(L) at longer probe-ISIs. Consequently, the right-hand
ends of all the functions in the middle and bottom panels on the right of Figure 4 might
have to be elevated. However, the two functions in the right bottom panel of Figure 4
would still be inconsistent with the expectation of the dual-buffer model.
The situation is more complicated in the case of Experiment 1, in which the delay
of the probe and of the mask were manipulated simultaneously. Delaying the mask
should reduce the contribution of guessing because the task became easier at longer
mask-delays. At the same time, delaying the probe would render the task more difficult.
The dual-buffer model is not explicit regarding the relative importance of mask delay and
probe delay. The reasonable assumption seems to be that the two may cancel each other
out. Given this assumption, the data as depicted in Figure 3 would remain unchanged.
Some probable effects that guessing might have on the outcomes of the three
experiments have been considered. In all cases, it seems reasonable to suggest that
guessing could not have biased the outcomes in ways which are unfavorable to the
dual-buffer model.
In terms of the new measures, it was demonstrated that the systematic decline in
partial report was matched by a systematic decrease in item information when the
partial-report probe was delayed. When performance improved with increases in the
delay of the masking array, the improvement was matched by an increase in the
availability of item information. Apart from not vindicating the dual-buffer model, this
study also serves to exclude Mewhort et al.'s (1981) attempt to question the empirical
foundation of iconic memory in the sense of informational persistence.