Purchase this article with an account.
Evan Palmer, David Fencsik, Todd Horowitz, Jeremy Wolfe; Signal Detection Evidence for an Attentional Bottleneck in Spatial Configuration Visual Search. Journal of Vision 2011;11(11):1311. doi: https://doi.org/10.1167/11.11.1311.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Do we need to assume capacity limitations to explain visual search performance on simple tasks? Attention-limited models propose two qualitatively different stages of perceptual processing: an unlimited capacity preattentive stage and a limited-capacity selective attention stage. Noise-limited decision models propose a single unlimited capacity perceptual processing stage, with decision processes influenced only by stochastic noise. For briefly presented displays, these two approaches can be modeled in a signal detection framework as different decision rules evaluating signal and noise. Consider two decision rules: Under a MAX rule, all items are processed with unlimited capacity and the maximum-value item is selected, while under a LIMITED CAPACITY rule, only a subset of items can be processed in a brief exposure and the MAX rule is applied to the subset.
In two-stage models, a MAX rule would govern stimuli that can be processed preattentively (e.g., feature searches) with a LIMITED CAPACITY rule required for other stimuli (e.g., spatial configuration searches). One stage, noise-limited models would either propose one decision rule for all stimuli (e.g. MAX) or must explain why different rules are required for different stimuli.
Five observers searched for either a tilted line among vertical lines or a 2 among 5s. Performance over all set sizes was equated for the two displays by varying superimposed noise. Task and relevant set size (1, 2, 4, or 8) were randomly intermixed to prevent observers from adopting different strategies for each stimulus. With stimuli thus equated, any single rule would predict essentially the same accuracy by set size function for both tasks. Two rules predict a crossover interaction with the 2v5 task easier than feature search at set size 1 and harder at set size 8. This crossover pattern was seen in the results, consistent with two-stage models and requiring a modification of one stage models.
This PDF is available to Subscribers Only