Figure 1.
A schematic of the FOA and divided resource models indicating how they explain set-specific capture.
In this example, participants are maintaining two attentional sets: “search for green letters in the central display” and “search for orange letters in the central display.” Peripheral displays contain distractors that are sometimes target-colored. Items in each of the three RSVP displays change identity every 166 ms. In this trial, a green distractor (X) matches one set and appears two frames prior to an orange target (Y). In the FOA model, depicted on the left, attentional sets reside in memory during active search. The focus of attention is empty until a target-colored item is detected, at which time the item’s corresponding set enters the focus. Only one set can be inside the focus at any given time. Thus, in this example, the orange target is not able to be identified because the green set still occupies the focus of attention when the orange target appears. In the divided resources model, depicted on the right, resources are divided unequally among concurrently-maintained attentional sets. Once a target-colored distractor is detected, its corresponding set is enhanced relative to other concurrently-maintained sets. When a target appears shortly afterward, its corresponding set is also enhanced. However, this enhancement is reduced as it stems from resources that are already partially allocated to enhancing the distractor’s set. In this example, target detection fails because too many resources were allocated to enhancing the distractor’s set when the target appeared.
Figure 2.
Task, trial types, and stimuli.
In each trial, participants viewed 15 colored letters in rapid succession. One to three of these letters were targets defined by color (orange, green, or lavender) that participants were supposed to identify at the end of the trial. Non-target colors were tan, magenta, and turquoise. Each letter appeared for 150 ms followed by a blank space lasting 16 ms. In the figure, the T1 Only trial type is illustrated using a full sequence of 15 letters. The other trial types are illustrated more briefly by showing only targets and intervening non-targets. The first target (T1) appeared randomly between the fourth and eighth positions in the RSVP stream (inclusive). AA and AB trials contained two targets (T1 & T2) that were the same (AA) or different (AB) colors. T1 and T2 were separated by one (lag 2) or three (lag 4) intervening items. BAA, BAB, and BAC trials contained three targets (T1, T2 & T3), each of which was separated by one intervening item. In BAA trials, T2 and T3 were the same color while T1 was a different color. In BAB trials, T1 and T3 were the same color while T2 was a different color. In BAC trials, T1, T2, and T3 were all different colors.
Table 1.
Trial counts for each condition.
Figure 3.
T2 accuracy in trials with two targets.
The left panel plots T2 accuracy given that T1 was identified. Consistent with our prior findings of set-specific capture, T2 accuracy was lower in AB than in AA trials at both Lag 2 and Lag 4. A double asterisk indicates a p-value less than 0.0005. The right panel plots T2 accuracy given that T1 was missed or reported incorrectly. In these trials, T2 accuracy did not vary with T1’s color.
Figure 4.
T3 accuracy in trials with three targets.
The left panel shows T3 accuracy in BAA, BAB, and BAC trials given that both T1 and T2 were identified. Consistent with set-specific capture, T3 accuracy was lower when T2 and T3 matched different attentional sets (BAB and BAC trials) than when they matched the same attentional set (BAA trials). Critically, in line with the FOA model, T3 accuracy did not vary with whether T1 and T3 matched the same attentional set: T3 accuracy in BAB and BAC trials did not significantly differ. The right panel shows T3 accuracy given that T1 was identified and T2 was missed. In line with the FOA model, in these trials T1’s attentional set did influence T3 accuracy, while T2’s set did not. That is, T3 accuracy was higher in BAB trials than in either BAA or BAC trials, but did not differ between BAA and BAC trials. A double asterisk indicates p<0.0005, and a single asterisk indicates p<0.05.
Figure 5.
A comparison of the predictions made by the FOA and divided resource models for T3 accuracy in BAB vs. BAC trials when T1 and T2 are both identified.
In the FOA model (left), identifying a target letter requires the corresponding attentional set to be enhanced inside the FOA. In each trial, the focus is empty prior to the appearance of the first target, as indicated by the circle with the dotted line. If a target is identified, the corresponding set enters and occupies the FOA, as indicated by the change in color of the FOA. No other set can be enhanced at the same time, as indicated by a solid line around the focus. Given these assumptions, the FOA model predicts that T3 is equally unlikely to be identified in BAB and BAC trials because T2’s set already occupies the FOA when T3 appears. In the divided resources model (right), an attentional set can be activated in a graded manner. In order for a target to be identified, its corresponding set must be activated above a certain threshold level, as depicted by the dotted line. Critically, the divided resources model predicts that T3 is more likely to be identified in BAB than in BAC trials because, in BAB trials, there is lingering enhancement of T1’s set even after T2’s set has been activated and T2 has been identified. This lingering enhancement should make it relatively easy for T3’s set (which matches T1’s set) to reach the activation threshold for identification. In BAC trials, there is no lingering enhancement of T1’s set, which should make it relatively hard for T3’s set to reach the activation threshold.