Evidence against late selection: stimulus quality effects in previewed displays

Strong late-selection theories of visual attention assert that when multiple stimuli belonging to familiar categories are presented, their identities are computed automatically and tagged for their locations. When selection by location is required, the identities are said to be retrieved without any need to repeat the perceptual processing. Five experiments designed to test this account are reported. All included a condition in which a display of eight characters was previewed for several hundred ms; a bar probe then designated one character the target for speeded classification. Stimulus factors that slow the character encoding process were manipulated. If selection is late, then such factors should have no effect in this condition because the probe occurs after automatic encoding is complete. There was no evidence of any such reduction in these factors' effects on reaction times or errors. The results were unchanged when catch trials with postdisplay masks were included, to discourage any optional delay of encoding. Several possible accounts are considered of how the strong late-selection model may be wrong, even if parallel encoding occurs in various situations.