Models of large-scale visual search
Investigators have for some time been interested in identifying factors which influence visual search performance and the effect these factors have on the scanning process which is a major component of visual search. The factors receiving the most attention include the stimulus to response mapping (varied or consistent), display set size, and level of target-distractor discriminability. Information on how the clustering of stimuli and varying the probability of targets appearing in particular locations affects the search process has been reported. These factors have been shown to affect the scanning process, determining both search strategies and whether the process occurs in serial or parallel. Previous experiments have typically used visual fields smaller than that subtended by a single computer monitor and few have systematically manipulated all the factors mentioned. Actual search tasks frequently involve complex targets in much larger visual fields whose stimuli frequently vary in terms of: stimulus to response mappings; discriminability; display set sizes; and physical clustering. How these factors affect large field visual search is not understood as well as their role in smaller visual fields. Two experiments were run to determine how these factors effect large field visual search. In contrast to previous research, display set size effects were found in consistent mapping conditions in both experiments. Under certain discriminability conditions, consistent and varied mapping response times mimicked each other. Three different stimuli configurations were researched: uniform distribution, equal sized clusters, and clusters of greatly different size. The uniform distribution showed the longest response times in both experiments. When letter stimuli were used, there was no significant difference between response times in the other clustering conditions but when word stimuli were used, participants tended to search the smallest clusters first.
A model of the process was developed. A very good fit was found between the model's predictions and observed results. Results from the model suggest that participants did not match their search strategies to the probabilities of the target appearing in a given location and that they failed to optimize their performance.