The preregistration of this experiment can be found here.

In a prior study, when subjects had to attend to or ignore a set of objects consisting of four different colored shapes (black, red, yellow, and purple), they treated an unexpected object in a unique color (green) no differently than one bearing a color in the display. This suggests that attention sets are based on categories, such as "white" and "nonwhite," rather than on the features of the objects they contain.

However, we may have obtained these results not because attention sets in inattentional blindness are always categorical, but because we forced them to be by overloading subjects' working memory. It could be that, in the absence of load, subjects use feature-based setting instead of category-based setting.

To test this, we will vary the color of the nonwhite shapes across trials, rather than within. On each trial, white shapes will be pitted against red, yellow, and purple shapes in a counterbalanced sequence. The unexpected object will be either the current color of the nonwhite shapes, a color that appeared on a previous trial, or a unique color (green). If attention sets are broadly and persistently category-based, we expect no differences in noticing rates between the three colors; they should be noticed frequently when attending nonwhite, and noticed infrequently when ignoring nonwhite. Conversely, a feature-based account would predict high noticing for any non-display color when ignoring colors. See the uploaded Predictions folder for sample graphs.