Abstract
Past theories of visual search for a feature singleton in a display dictate that visual search is highly efficient and RTs are basically unrelated to the number of elements in the display (Treisman & Gelade, 1980). Here, we show evidence that this view is fundamentally wrong. Lleras, Cronin and Buetti (submitted) proposed a new theory of visual search (Information Theory of Vision, ITV) whereby reaction times in a search task are proportional to the amount of information in the display. The uncertainty in a display is given by the number of possible target locations and the uncertainty about the identity of an item. The theory predicts that, initially, all items in a display will be processed and then, items most unlikely to be target (i.e., the ones carrying the most information) will be first discarded from consideration, reducing the location uncertainty of the target. Processing will continue on the remaining items, and so forth, until a final set of candidate targets is found and scrutinized in finer detail. Here, we tested the sequential discounting argument using a singleton search task (only one candidate in the display). Importantly, we included traditional visual pop-out conditions: the singleton (a red triangle) embedded in a uniform field of visual lures (elements that produce flat search slopes). The lures could be blue squares, yellow triangles or orange diamonds. In support of ITV, we found that RTs for uniform displays for all three different types of lures increased in logarithmic fashion. Further, we found clear evidence of sequential discounting by dissimilarity: using the slope estimates of the uniform displays, we were able to predict 94% of the variance of RT in heterogeneous trials, assuming discounting of most dissimilar elements first. These results clearly challenge current theories of so called "feature" search.
Meeting abstract presented at VSS 2014