Memory in visual search: Do the eyes have it?

Todd S. Horowitz, Jeremy M. Wolfe, and Megan Hyle

Theories of visual search have generally assumed that rejected distractors are marked so as to avoid further processing of these items (memory-driven search). To test this assumption, Horowitz and Wolfe (1998) developed the randomized search paradigm, in which standard static search is compared to dynamic search in which items are randomly replotted at new locations throughout a trial. Memory-driven search predicts that search slopes computed from means should double in the dynamic condition (with medians, the dynamic:static ratio should be 1.38:1). We have repeatedly found that slopes are similar in the two conditions. Recently, a number of studies have utilized the random search paradigm to explore this issue. Particular interest has been devoted to the question of whether there is memory for overt shifts of attention as measured by eye movements. Some studies have replicated our results, some have not. We present new data addressing this question. We had subjects perform a 2-AFC search  for items which required fixation. Stimuli were lower-case trigrams presented in 9 point Palatino font. Targets were "bab" and "hoh". Distractors were drawn from the set {"beb", "bib", "bob", "bub", "hah", "heh", "hih", "huh", "kak", "kek", "kik", "kok", "kuk", "dad", "ded", "did", "dod", "dud"}. Averaged over 16 subjects, dynamic and static median slopes were 211.06 ms/item and 217.88 ms/item, respectively. These were not significantly different (F(1, 15) < 1). While error rates were greater for dynamic (0.073) than static (0.023) searches (F(1, 15) = 12.75, p < .005),  slope differences are not significant even when corrected for errors (F(1, 15) < 1). We propose a framework to explain when we should and should not expect to observe evidence for memory in visual search.