Detecting conjunctions of color and form: Reassessing the serial search hypothesis

Treisman and Gelade (1980), who had subjects search for conjunctions of color and form, observed that the slope relating reaction times to number of items in the display was twice as steep for negative (target absent) responses as for positive responses. This result is widely accepted as evidence that a serial self-terminating search is necessary to detect conjunctions. In Experiment 1 of the present study, the subjects searched for a conjunction target (a green T) in displays of 2, 4, or 8 items. Positive and negative display-size slopes were essentially the same in this range. In Experiment 2, the subjects searched for conjunction targets in displays of 2 to 24 items. In displays of 2 to 8 items, the positive and negative slopes were again comparable and the 2:1 ratio was again rejected. In displays of 8 to 24 items, on the other hand, the negative slopes were approximately twice the positive slopes. The same result was obtained in Experiment 3, which was modeled after the conjunction search experiment of Egeth, Virzi, and Garbart (1984). In summary, a 2\3n1 slope ratio appears to describe the function quite well in the range above 8 items (or if display size is coarsely sampled, as in previous research), but not in the lower range, in which the true ratio is closer to 1\3n1. These nearly parallel functions in the lower range probably do not, however, reflect a serial exhaustive search: in Experiment 4, positive reaction times were substantially reduced when redundant targets were added to the display. An alternative model of conjunction search is proposed on the basis of these data. This model postulates a molar serial self-terminating search over relatively large clumps of items, and a parallel search within these clumps. It appears, then, that the nature of the attentional limits arising in the process of feature conjunction must be very different from that suggested by Treisman and Gelade.