Set-specific capture can be reduced by preemptively occupying a limited-capacity focus of attention

Abstract

Recent work has shown that contingent attentional capture effects can be especially large when multiple attentional sets for color guide visual search (Moore & Weissman, 2010). In particular, this research suggests that detecting a target-colored (e.g., orange) distractor leads the corresponding attentional set (e.g., identify orange letters) to enter a limited-capacity focus of attention in working memory, where it remains briefly while the distractor is being attended. Consequently, the ability to identify a differently-colored (e.g., green) target 100–300 ms later is impaired because the appropriate set (e.g., identify green letters) cannot also enter the focus of attention. In two experiments, we investigated whether such set-specific capture can be reduced by preemptively occupying the focus of attention. As predicted, a target-colored central distractor presented 233 ms before a target-colored peripheral distractor eliminated set-specific capture arising from the peripheral distractor. Moreover, this effect was observed only when the central distractor’s color (e.g., orange) (a) matched a different set than the upcoming peripheral distractor’s color (e.g., green) and (b) matched the same set as the upcoming central target’s color (e.g., orange). We conclude that the same working memory limitations that give rise to set-specific capture can be preemptively exploited to reduce it.

INTRODUCTION

Selective attention directs limited resources to stimuli that are important for achieving behavioral goals (Yantis & Egeth, 1997). Moreover, selective attention appears to be enabled by creating and maintaining one or more attentional sets, which specify the perceptual and/or conceptual attributes (e.g., color, location, size, shape, or semantic category) that define relevant stimuli (Adamo, Pun, Pratt, & Ferber, 2008; Ansorge & Heumann, 2003; Ariga & Yokosawa, 2008; Folk, Remington, & Johnston, 1992; Pashler & Huang, 2007; Pashler & Shiu, 1999). Guided by such attentional sets, top-down signals enhance the processing of relevant stimuli while limiting the processing of irrelevant stimuli (Corbetta & Shulman, 2002).

Sometimes, however, top-down signals have the opposite effect. For example, contingent attentional capture refers to a phenomenon in which an irrelevant stimulus that possesses a target-defining attribute (e.g., a particular color) attracts attention as though it were a target (Bacon & Egeth, 1994; Folk, Leber, & Egeth, 2002; Folk, Leber, & Egeth, 2008; Folk et al., 1992; Serences et al., 2005). Because limited resources have been allocated to the irrelevant stimulus, the identification of a target that appears within a few hundred milliseconds afterwards is impaired. Such capture has been observed in a variety of paradigms including those involving spatial cuing (Bacon & Egeth, 1994; Folk et al., 1992; Gibson & Kelsey, 1998), visual search (Olivers, 2008), and rapid serial visual presentation (RSVP) (Folk, Ester, & Troemel, 2009; Folk et al., 2002; Folk et al., 2008; Leblanc & Jolicoeur, 2007; Moore & Weissman, 2010; Serences et al., 2005).

Given that minimizing distraction is crucial for enabling purposeful behavior, it is important to investigate whether contingent attentional capture can be reduced. Support for this possibility comes from a recent study by Folk and colleagues (Folk, Ester & Troemel, 2009). In this study, participants searched for letters of a particular color (e.g., red) within a centrally-presented RSVP stream. Similar to prior studies of contingent attentional capture, a peripheral distractor impaired the identification of a subsequent target more when it was target-colored (e.g., red) than when it was not target-colored (i.e., grey). However, this effect vanished when a target-colored distractor was presented in the central RSVP stream a few hundred milliseconds before the peripheral distractor was presented. Folk et al. (2009) argued that detecting the central distractor led to the opening “of a gate between perceptual processes and higher level cognitive processes,” which allowed target-colored items in the central RSVP stream, but not at different locations, to access higher level cognitive processes. Thus, contingent attentional capture arising from the subsequent peripheral distractor was reduced.

Reducing contingent attentional capture may be even more important when multiple attentional sets guide target selection because contingent capture effects can be especially large under such conditions. For instance, in one of our recent studies (Moore & Weissman, 2010), participants searched a heterogeneously-colored central RSVP stream for occasional target letters that appeared in either of two possible colors (e.g., orange and green). As in other studies of contingent attentional capture (Folk et al., 2009; Folk et al., 2002; Folk et al., 2008; Serences et al., 2005), a peripheral distractor impaired subsequent target identification more when it was target-colored (e.g., orange) than when it was not target-colored (e.g., lavender). Critically, at short stimulus onset asynchronies (SOAs) of approximately 100–300 ms, this effect was two to three times larger when the peripheral distractor’s color (e.g., orange) matched a different attentional set than the upcoming target’s color (e.g., green) as compared to the same attentional set (e.g., orange). We call this phenomenon set-specific capture.

Referring to a current model of working memory, we argued that set-specific capture reflects a redistribution of processing resources among the attentional sets that guide visual search (Moore & Weissman, 2010). Our view states that detecting a target-colored item (e.g. a green letter) leads the corresponding attentional set (e.g. identify green letters) to enter a “focus of attention” in working memory that is limited to a single item (Jonides et al., 2008; McElree, 2001; Klaus Oberauer, 2002)¹. Evidence to support the existence of such a focus in working memory comes from a variety of sources including studies of working memory (Berti, 2008; Garavan, 1998; McElree, 2001), task switching (Hsieh & Allport, 1994; Monsell, 2003), and the attentional blink (Juola, Botella, & Palacios, 2004; Vachon, Tremblay, & Jones, 2007). For example, when participants maintain separate counters in working memory corresponding to the number of times that circles and triangles have been presented, they are faster to update a particular counter (e.g., the number of circles) just after the same (versus a different) counter has been updated (Berti, 2008; Garavan, 1998). This effect is not due to perceptual priming as it disappears when the updating task permits the counters to be bound into a single representation (Bao, Li, & Zhang, 2007). Therefore, it has been argued that only a single item or representation can occupy the focus of attention at time (Berti, 2008; Garavan, 1998). Notably, this view is consistent with various theories of selective attention. For example, coherence theory states that focused attention can be allocated to only one object at a time (Rensink, 2000).

Our theory of set-specific capture (see Figure 1) posits that once a target-colored item is detected its corresponding attentional set remains in the focus of attention for up to a few hundred milliseconds while the detected item is being identified (Moore & Weissman, 2010). During this time, it is possible to identify a second target-colored item if that item’s color (e.g. green) matches the same attentional set (e.g., identify green letters) as the first item’s color². However, if the second item’s color (e.g., orange) matches a different attentional set (e.g., identify orange letters), then it will not be identified.

Figure 1.

Schematic of the set enhancement / focus of attention model proposed in Moore & Weissman (2010). (A) The components of the model. Attentional sets (e.g. “identify green letters” and “identify orange letters”) reside in memory during active search—the model does not specify whether they reside in working or long-term memory. An attentional set will enter the focus of attention (limited to a single item) if a target-colored item is attended, but the focus of attention remains empty during active search. (B) The model performing during a trial in which the peripheral distractor matches a different attentional set than the target. During active search (e.g., stimulus frame “H J T”), the focus of attention is empty, as indicated by the dotted line around its border. As soon as the target-colored green “X” appears in the periphery, the corresponding attentional set is enhanced as it enters the focus of attention. As long as “X” is attended, its corresponding attentional set remains in the focus. When the different-target-colored orange “Y” appears, it cannot be attended because the focus of attention is still occupied with the “identify green letters” attentional set, as indicated by the solid line around the focus of attention.

Further support for our theory comes from two additional findings in our prior study (Moore & Weissman, 2010). First, when all possible target colors could be maintained in the same attentional set (i.e., identify any colored letter in a central RSVP stream that contains mostly grey letters), set-specific capture vanished, even though peripheral target-colored distractors still captured attention. This finding fits with our view because directing processing resources more strongly to a global attentional set for color should facilitate the identification of a subsequent target, regardless of its exact color. It also rules out bottom-up perceptual priming of the target’s color as an alternate account of set-specific capture. Second, set-specific capture was reversed when the peripheral distractor appeared 116 ms after the central target. That is, the peripheral distractor disrupted performance when its color (e.g., green) matched the preceding target’s color but not when its color matched the other target color (e.g., orange). This finding fits with our view because detecting a target (e.g., a green letter) should lead to a relative increase of processing resources toward the corresponding attentional set (e.g., identify green letters). Thus, a distractor that appears soon afterward is more likely to be attended and, consequently, to impair performance if its color matches the same (versus a different) attentional set as the preceding target’s color. In sum, multiple findings suggest that set-specific capture reflects a redistribution of processing resources among the attentional sets that guide visual search.

Critically, our view leads to a hypothesis about how to reduce set-specific capture arising from a target-colored peripheral distractor. Namely, such capture should be reduced if a target-colored distractor is detected within a few hundred milliseconds before the peripheral distractor is presented. Moreover, this reduction should occur only when the first distractor’s color (e.g., orange) and the peripheral distractor’s color (e.g., green) match different attentional sets. In such trials, detecting the first distractor’s color (e.g., orange) should lead the corresponding attentional set (e.g., identify orange letters) to enter the limited-capacity focus of attention, where it should remain for a few hundred milliseconds while the first distractor is attended. If, during this time, a peripheral distractor that possesses a different target color (e.g., green) is detected, then the attentional set corresponding to its color (e.g., identify green letters) should be unable to enter the already-occupied focus of attention. Thus, it should not be possible to attend to the peripheral distractor and, consequently, set-specific capture arising from the peripheral distractor should be reduced.

Given the sizable literature investigating interactions between attention and working memory, it is important to clarify two aspects of our view. First, we posit that it is only after a target-colored (e.g., orange) item is detected during a visual search task that the corresponding attentional set (e.g., identify orange letters) enters the focus of attention. In other words, active visual search before a target-colored item is detected does not require an attentional set to occupy the focus of attention. Consistent with this view, some researchers have argued that attentional sets are maintained in working memory during active visual search, but not necessarily in the focus of attention (e.g., (Olivers & Meeter, 2008). Also consistent, attentional sets are able to guide visual search even when they are not actively maintained in working memory (Thompson, Underwood, & Crundall, 2007). Finally, the magnitude of contingent attentional capture that is observed when a distractor’s color (e.g., orange) matches an upcoming target’s color (e.g., orange) does not vary with the number of attentional sets for color that guide visual search (Moore & Weissman, 2010). This result suggests that searching for multiple target colors does not involve rapidly switching which attentional set occupies the focus of attention. In sum, multiple findings suggest that it is only after a target-colored item is detected that the corresponding attentional set enters the focus of attention.

The second aspect of our view worth clarifying is that it does not relate to a controversy regarding the interaction between attention and working memory. Researchers disagree about whether stimuli capture attention when they are not the targets of an ongoing visual search task, but nonetheless have features that match the contents of working memory (e.g., Houtkamp & Roelfsema, 2006; Olivers, 2009; Soto, Heinke, Humphreys, & Blanco, 2005; Woodman & Luck, 2007). In contrast, our view is that such stimuli capture attention when they are the targets of an ongoing visual search task. This view is not connected to the current debate (Olivers, 2009). Moreover, it has received support from numerous prior studies of contingent attentional capture (e.g., Folk et al., 2002; Folk et al., 1992; Leblanc, Prime, & Jolicoeur, 2008; Serences et al., 2005).

EXPERIMENT 1

The goal of Experiment 1 was to investigate whether it is possible to reduce set-specific capture. As in our prior study, participants searched for target-colored letters that appeared unpredictably in either of two possible colors (e.g., orange and green) within a central RSVP stream. In some trials, a target letter was preceded by a target-colored distractor that appeared in one of two peripheral RSVP streams. Consistent with our prior findings of set-specific capture (Moore & Weissman, 2010), we expected target identification accuracy to be lower when the peripheral distractor’s color (e.g., green) differed from the target’s color (e.g., orange) than when it matched the target’s color. To investigate whether set-specific capture can be reduced, in some trials we presented a target-colored distractor in the central RSVP stream 233 ms before presenting the peripheral distractor. We predicted that this central distractor would reduce set-specific capture arising from the peripheral distractor, but only if its color matched a different attentional set than the peripheral distractor’s color.

Method

Participants

Thirty University of Michigan students (15 female) participated in exchange for course credit. All participants (age range: 18–25) reported normal or corrected vision and no history of neurological injury or disease. Participants gave informed written consent before the experiment in accordance with the University of Michigan Behavioral Sciences Institutional Review Board.

Procedure

The task was similar to those in prior contingent attentional capture experiments that have used rapid serial visual presentation (RSVP) displays (Folk et al., 2009; Folk et al., 2002; Folk et al., 2008; Moore & Weissman, 2010; Serences et al., 2005). Participants viewed three simultaneously presented RSVP streams, each of which was composed of letters and digits (Figure 2A). In the two peripheral RSVP streams, most of the characters were grey, aside from an occasional target-colored (e.g., orange or green) letter distractor. In the central RSVP stream, each character appeared in one of six possible colors. Participants were instructed to identify occasional target letters in the central RSVP stream that appeared unpredictably in either of two target colors (e.g., orange and green) while ignoring characters that appeared in any of the four other colors. They were also told to ignore target-colored letters in the two peripheral RSVP streams. As in a similar prior study (Serences et al., 2005), when a target-colored letter appeared in the central RSVP stream, participants were to indicate via a key press whether the target was from the first half of the alphabet (J key, right index finger) or the second half of the alphabet (K key, right middle finger). Participants were not pressured to respond quickly.

Figure 2.

(A) Examples of the stimulus displays used in Experiment 1. Participants searched for target letters that could appear in either of two possible colors (e.g., orange and green) within a heterogeneously-colored, central rapid serial visual presentation (RSVP) stream while ignoring occasional target-colored distractors that appeared in either of two peripheral RSVP streams. Each frame was presented for 116 ms. D1 was a digit in the central RSVP stream, which was either non-target-colored (i.e., tan, magenta, or turquoise in the _AA and _AB trial types) or presented in one of the two possible target colors. D2 was a letter in one of the two peripheral RSVP streams. D2 appeared two frames after D1 and was always presented in one of the two possible target colors. Finally, the central target appeared two frames after D2. (B) A list of the trial types in Experiment 1, which includes an example stimulus sequence for each trial type. (C) The 6-color wheel from which stimulus colors were selected in Experiments 1 and 2. Color names and RGB values are indicated beneath the color wheel.

As in our previous study (Moore & Weissman, 2010), the three RSVP streams were presented continuously and the inter-target interval varied randomly across trials. Thus, there were no perceptible breaks between trials because non-target stimuli appeared in the RSVP display between successive targets. An advantage of this procedure is that it does not allow participants to form expectations about when a target letter will appear. We therefore reasoned that it would result in relatively low false alarm rates. To construct an appropriate set of inter-target intervals, we took into account an analysis of response times from our previous study of set-specific capture in which there was also no pressure to respond quickly (Moore & Weissman, 2010). This analysis revealed that more than 95% of responses occurred within 2000 ms of target onset with an average response time of about 800 ms. The same was true in the present study. Thus, we felt confident in making our shortest inter-target interval 2333 ms. Other inter-target intervals were 2916, 3500, and 4083 ms. To maintain consistent criteria for defining a target response across the four inter-target intervals, we defined a target response as the first key press logged within 2200 ms following target onset. Thus, participants responded while non-target stimuli continued to appear in the RSVP display.

The experiment began with instructions explaining the task followed by 48 practice trials. During the practice trials, the rate of presentation of items in the three RSVP streams was slow at first (250 ms per item), but accelerated to full speed (116 ms per item) by the 24^th trial. During the test trials, participants were given a self-paced rest break every 32 trials (about every 2 minutes).

Design

Targets in the central RSVP stream were preceded by zero, one, or two critical distractors. The first possible distractor (D1) was a digit that appeared in the central RSVP stream four frames (466 ms) prior to a target. D1 was either target-colored or non-target-colored. If it was target-colored, then it was either the same color as the upcoming target or the other target color. The second possible distractor (D2) was a target-colored letter that appeared equally often in each of the peripheral RSVP streams. D2 always appeared two frames (233 ms) prior to a target and was either the same color as the upcoming target or the other target color. As we describe next, various combinations of D1, D2, and the target led to the creation of nine trial types. In each of these trial types, a single trial consisted of a target and all stimuli appearing four or fewer frames prior to the target (i.e., D1, D2, and several non-target stimuli).

Three trial types did not include D1 and D2. In target alone trials, the display ran at normal speed (116 ms per item) while a target appeared in the central RSVP stream and non-target-colored characters appeared in the peripheral RSVP streams. In target catch trials, the display paused for 1000 ms while a target appeared in the central RSVP stream and non-target-colored characters appeared in the peripheral RSVP streams. In non-target catch trials, the display paused for 1000 ms while a non-target-colored letter appeared in the central RSVP stream that participants were not supposed to identify and non-target-colored characters appeared in the peripheral RSVP streams. The purpose of including catch trials was to provide a relatively pure test of participants’ ability to remember the target colors by eliminating the severe encoding limitations that were imposed by the RSVP display. In other words, these trials were included to ensure that participants were keeping track of the target colors. They were not, however, important for testing our main hypotheses.

Six additional trial types did include D1 and D2. To simplify our discussion of these trial types, we name each one using a three-letter sequence. In this sequence, the first letter represents D1, the second letter represents D2, and the third letter represents the target. “A” and “B” refer to different target colors. For example, “AAB” refers to a trial in which D1 and D2 appeared in the same target color (e.g., orange), while the target appeared in a different target color (e.g., green). An underscore in the first position indicates that D1 appeared in a non-target color (e.g., lavender). For instance, “_AA” refers to a trial in which D1 was not target-colored (e.g., lavender) and D2 was the same color as the upcoming target (e.g., orange). Figure 2B indicates the three-letter label that corresponds to each of the six trial types involving distraction. It also provides an example stimulus sequence for each trial type, which is further illustrated in Figure 2A. Table 1 indicates the number of trials that were included for each of these trial types in Experiments 1 and 2.

Table 1.

The number of trials in each trial type of Experiments 1 and 2.

Trial type	N, Exp1	N, Exp2
Target Alone	64	64
Target Catch	32	18
Non-target Catch	32	18
_AA	64	60
_AB	64	60
AAA	64	60
AAB	64	60
BAA	64	60
BAB	64	60
BAC	0	60

Finally, we controlled for effects of response congruency in our design. Whether or not D2 and the target were from the same part of the alphabet was fully crossed with whether or not D2’s color and the target’s color matched the same attentional set (this control was unnecessary for D1, which was a digit whose identity was not mapped to a task-relevant response). Set-specific capture effects arising from D2 were therefore not confounded with potential effects of response congruency.

Apparatus and Stimuli

Stimuli were displayed on a 19” Viewsonic CRT monitor with a 60 Hz refresh rate, controlled by a Dell PC running Windows XP. Presentation® software (Neurobehavioral Systems, Inc.) was used to control stimulus presentation and to record participants' responses. A viewing distance of 80 cm was enforced by a chin rest.

Three RSVP streams containing letters and digits (character size, 2.07° × 1.88°) were presented simultaneously on a black background: one stream was centered at fixation, while two others were centered 4.22° to the left and 4.22° to the right of fixation. A new character appeared in each RSVP stream every 100 ms, followed by a blank gap that lasted 16 ms. Target and D2 letters were drawn from the beginning (A, B, C, D and G) and the end (T, V, X, Y, Z) of the alphabet, so that target identification would not rely on an overly demanding decision (i.e., “is ‘M’ from the first or second half of the alphabet?”). D1 digits included 2, 3, 4, 7 and 9. Other characters in the RSVP streams included the entire alphabet except for I, O, and W, and all of the digits excluding 0 and 1 (see Figure 2A).

The stimulus colors were identical to those in the “light” color scheme of our prior study (Moore & Weissman, 2010). These colors were drawn from a 6-color wheel (Figure 2C), in which each color had approximately the same CIELAB lightness value (L* = approx. 70). The two target colors and single non-target color (i.e., D1’s color in _AA and _AB trials) were drawn from three non-adjacent colors in the wheel (orange, green, and lavender: colors 1, 3, and 5). A control experiment in our prior study confirmed that each of these three colors was equally discriminable from the other two colors in the triplet (Moore & Weissman, 2010). Moreover, the target and non-target colors were counterbalanced across participants.

We took several steps to ensure that participants could only perform the task if they maintained distinct attentional sets for each of the target colors. First, as mentioned earlier, the target and non-target colors in the central RSVP stream were drawn from three non-adjacent colors in the wheel. The remaining three non-adjacent colors in the wheel (tan, turquoise and magenta: colors 2, 4, and 6) were randomly assigned to other characters in the central RSVP stream. Second, the non-target color appeared in the central RSVP stream once between successive targets, such that the non-target color appeared in this stream as frequently as a target color. Third, as mentioned earlier, all six colors in the central RSVP stream were nearly equated for luminance and saturation. For these reasons, participants could perform the task correctly only if they maintained distinct attentional sets for each of the target colors.

Results

Mean accuracy was the dependent measure in the main analyses. At the outset, we excluded eight participants (three female) whose performance indicated that they were unable to keep track of the target colors. These participants failed to correctly discriminate the target letter in more than 20% of target catch trials and/or produced false alarms in more than 20% of non-target catch trials.

Among the remaining 22 participants, we observed evidence of attentional capture (Figure 3). Specifically, target identification accuracy was lower when a target was preceded by a target-colored peripheral distractor (79%, _AA trials) than when it was not preceded by such a distractor (84.0%, target alone trials), [t(21) = 3.80, p < 0.001]. Although isolating contingent attentional capture requires a comparison of target identification accuracy following target-colored versus non-target-colored distractors³, the contrast above typically leads to the same conclusion about the presence or absence of contingent attentional capture (Folk et al., 2009; Folk et al., 2002; Folk et al., 2008; Moore & Weissman, 2010; Serences et al., 2005). As this conclusion is not critical for testing our main hypotheses, we now turn to discuss set-specific capture effects.

Figure 3.

Target identification accuracy plotted separately for each of the main trial types in Experiment 1. Consistent with prior studies and demonstrating an attentional capture effect, performance in all conditions involving distraction was significantly worse than performance in target alone trials. In line with a set-specific capture effect, performance in _AB trials was worse than performance in _AA trials. Critically, when D1’s color matched a different attentional set as D2’s color, set-specific capture vanished as indicated by (a) no difference in performance between BAB and BAA trials and (b) better performance in BAB than in _AB trials. In contrast, when D1’s color matched the same attentional set as D2’s color, set-specific capture was still present as indicated by worse performance in AAB than in AAA trials. Error bars illustrate the standard error of the mean.

An analysis of such effects supported our hypothesis about reducing set-specific capture. Replicating our previous finding of such capture, performance was better in _AA (79.8%) than in _AB trials (71.2%), [t(21) = 3.80, p < 0.001]. Consistent with predictions, however, this effect was absent when D1’s color matched a different attentional set than D2’s color: performance did not differ in BAA (74.7%) and BAB (77.8%) trials [t(21) = 1.71, p = 0.11]. A repeated-measures ANOVA verified that set-specific capture was significantly reduced when D1’s color matched a different attentional set than D2’s color (BAA – BAB), relative to when D1 was absent (_AA – _AB ) [F(1, 21) = 13.02, p < 0.002]. Moreover, this reduction occurred because the addition of D1 led to improved performance in _AB trials containing set-specific capture: performance was better in BAB than in _AB trials [t(21) = 2.62, p < 0.016]. Thus, as predicted, D1 reduced set-specific capture arising from D2 when D1’s color matched a different attentional set than D2’s color.

Also as predicted, D1 did not reduce set-specific capture arising from D2 when D1’s color matched the same attentional set as D2’s color. Under such conditions there was still a significant set-specific capture effect, as reflected by better performance in AAA (80.8%) than in AAB (70.7%) trials [t(21) = 4.08, p < 0.001]. A repeated-measures ANOVA confirmed that set-specific capture was not reduced when D1’s color matched the same attentional set as D2’s color (AAA – AAB) relative to when D1 was absent (_AA – _AB) [F(1,21) = 0.172, p = 0.683]. Finally, the addition of D1 did not lead to improved performance in _AB trials containing set-specific capture: performance in AAB trials did not differ from that in _AB trials [t(21) = 0.17, p = 0.87].

We also investigated whether D1 reduced contingent attentional capture effects arising from D2 when D1, D2, and the target were all the same color, as in Folk et al.’s (2009) study. Surprisingly, we did not observe significantly higher target identification accuracy in AAA than in _AA trials [t(21) = 0.53, = 0.61]. We reserve a discussion of this failure to replicate Folk et al.’s (2009) finding for the General Discussion.

Finally, whether D2 and the target were from the same or different halves of the alphabet did not influence the results. There was no main effect of response congruency, and response congruency did not interact with any other factors (all p > 0.4).

Participants were not pressured to respond quickly. Nonetheless, we repeated the above analyses on the reaction time data (for correct responses only) to determine whether speed-accuracy tradeoffs contributed to our results. None of the latency analyses were significant (all p > 0.1). Thus, there was no evidence to suggest that speed-accuracy tradeoffs contributed to our results.

Discussion

Two sets of findings supported our hypothesis about reducing set-specific capture. First, D1 reduced set-specific capture arising from D2 when D1’s color matched a different attentional set than D2’s color: performance differed for BAA and BAB trials significantly less than it differed for _AA and _AB trials. This effect occurred because the addition of D1 led to improved performance in _AB trials containing set-specific capture: performance was better in BAB than in _AB trials. Second, D1 did not reduce set-specific capture arising from D2 when D1’s color matched the same attentional set as D2’s color: performance differed for AAA and AAB trials just as much as it differed for _AA and _AB trials. Additionally, performance was no better in AAB trials than in _AB trials. Together, these findings provide converging evidence that contingent attentional capture can be reduced under certain conditions (Folk et al., 2009). Next, we consider two hypotheses about how D1 might reduce set-specific capture arising from D2 when D1’s color matches a different attentional set than D2’s color.

The D2 blocking hypothesis posits that D1 reduces set-specific capture arising from D2 by preventing the attentional set corresponding to D2’s color from entering the focus of attention. Specifically, when D1 is detected, the attentional set corresponding to its color enters the focus of attention and still occupies it when D2 appears two items later (233 ms after D1 onset). Thus, when D2 is presented, the attentional set corresponding to its color is unable to enter the focus of attention. When the processing of D1 is completed, the attentional set corresponding to its color leaves the focus of attention, and this occurs before the target appears four items later (466 ms after D1 onset). Therefore, the focus of attention is unoccupied when the target appears, leading to equivalent performance in BAB and BAA trials.

The D2 blocking hypothesis also explains why performance is better in BAB than in _AB trials. In particular, since D1 is not presented in _AB trials, the attentional set corresponding to D2’s color is able to enter the focus of attention, where it still resides when the target appears two items later. Because the attentional set corresponding to D2’s color does not match the upcoming target’s color, target identification is impaired. In sum, the D2 blocking hypothesis explains our findings indicating that D1 reduces set-specific capture arising from D2 when the colors of D1 and D2 match different attentional sets.

The D1 enhancement hypothesis posits that D1 reduces set-specific capture arising from D2 by bringing the upcoming target’s attentional set into the focus of attention. As in the D2 blocking hypothesis, when D1 is detected, the attentional set corresponding to its color enters the focus of attention. Moreover, when a different-colored D2 is presented two items (233 ms) later, its corresponding attentional set is unable to enter the already-occupied focus of attention. Unlike the D2 blocking hypothesis, however, the attentional set corresponding to D1’s color still occupies the focus of attention when the target is presented four items (466 ms) later. Thus, in BAB trials, D1 reduces set-specific capture arising from D2 by (a) preventing D2’s corresponding attentional set from entering the focus of attention and (b) enhancing the attentional set corresponding to the upcoming target.

Can the D1 enhancement hypothesis account for the effects we have observed? On the one hand, it nicely explains why target identification accuracy is higher in BAB than in _AB trials. Specifically, the attentional set corresponding to the target’s color is already inside the focus of attention when the target appears in BAB (but not _AB) trials, thereby facilitating target identification. On the other hand, the D1 enhancement hypothesis appears to have difficulty explaining why accuracy is not significantly higher in BAB than in BAA trials. However, the absence of this effect may not be inconsistent with the D1 enhancement hypothesis. In some trials, the processing of D1 may be completed before D2 appears, meaning that the focus of attention is not occupied when D2 is presented⁴. Thus, the attentional set corresponding to D2’s color enters the focus of attention and is still there when the target appears. The effect of D2 occupying the focus of attention when the target appears is to impair target identification in BAB trials while enhancing it in BAA trials, potentially leading to no overall difference in performance between these trial types. For this reason, it is not clear whether our findings in Experiment 1 support the D2 blocking hypothesis or the D1 enhancement hypothesis. We therefore conducted a second experiment.

EXPERIMENT 2

In Experiment 2, we sought to distinguish between the D2 blocking and D1 enhancement hypotheses. The task was identical to that in Experiment 1, except that participants identified target letters appearing in any of three possible colors within the central RSVP stream. Using this task, we were able to include an additional trial type, BAC, in which D1, D2, and the target were all different target colors. We reasoned that even if the attentional set corresponding to D2’s color occasionally entered the focus of attention, it would impair target identification equally in BAB and BAC trials. Thus, contrasting performance in these two trial types would allow us to determine if it mattered whether D1’s color matched the target’s color and, consequently, to distinguish between the D2 blocking and D1 enhancement hypotheses. The D2 blocking hypothesis predicted equivalent performance in these trial types; in contrast, the D1 enhancement hypothesis predicted better performance in BAB trials than in BAC trials (see Figure 4).

Figure 4.

The D2 blocking and D1 enhancement hypotheses explored in Experiment 2. As in Figure 1, a dotted line around the focus of attention indicates it is empty, whereas a solid line indicates it is occupied. (A) The D2 blocking hypothesis. Performance is equivalent in BAB (left) and BAC (right) trials because the attentional set corresponding to D1’s color (e.g., “identify green letters”) prevents the attentional set corresponding to D2’s color (e.g., “identify orange letters”) from entering the focus of attention. However, D1’s attentional set leaves the focus of attention before the target appears. Thus, the focus of attention is available to enhance a different attentional set when the target appears in both BAB and BAC trials. (B) The D1 enhancement hypothesis. Performance is better in BAB (left) than in BAC (right) trials because the attentional set corresponding to D1's color still occupies the focus of attention when the target appears. The attentional set corresponding to D1’s color enhances target identification in BAB trials, but not in BAC trials.

Methods

Participants

Forty-four University of Michigan students (25 female) participated in exchange for course credit. All participants (age range: 18–25) reported normal or corrected vision and no history of neurological injury or disease. Participants gave written informed consent before the experiment in accordance with the University of Michigan Behavioral Sciences Institutional Review Board.

Procedure and Design

The task was the same as that in Experiment 1 except that participants were told to identify letters in the central RSVP stream that appeared in any of three possible target colors (Figure 5 provides example trials for the main trial types). Consequently, Experiment 2 employed the same trial types as Experiment 1 with the exception of one new trial type: BAC (see Figure 5A). In this trial type, D1’s color, D2’s color, and the target’s color (e.g., orange, green, and lavender) matched distinct attentional sets. The number of trials per condition differed slightly from that in Experiment 1 to accommodate the new trial type (see Table 1 for details).

Figure 5.

(A) Examples of the stimulus displays used in Experiment 2. Participants searched for target letters in a central RSVP stream that could appear in any of three possible colors (i.e., orange, green, and lavender) while ignoring occasional target-colored distractors that appeared in either of two peripheral RSVP streams. Using three target colors allowed us to include an additional trial type, BAC, in which the colors of D1, D2, and the target all matched different attentional sets. (B) A list of the trial types in Experiment 2, which includes an example stimulus sequence for each trial type.

Apparatus and Stimuli

The apparatus and stimuli in Experiment 2 were the same as those in Experiment 1 except that there were now three target colors (i.e., orange, green, and lavender: colors 1, 3, and 5 of the 6-color wheel shown in Figure 2C). Due to this change, the non-target color (i.e., D1’s color in _AA and _AB trials) was chosen randomly from colors 2, 4, and 6 in the color wheel (i.e., tan, turquoise, and magenta). Although each of these colors was not equally discriminable from each of the target colors, the main contrast of interest (i.e., BAB versus BAC) did not involve trial types in which a non-target color was presented. Thus, our choice was not problematic from the perspective of testing our hypothesis. It also allowed us to use the same color scheme as in Experiment 1, thereby making the stimulus displays in Experiments 1 and 2 comparable.

Results

Mean accuracy was the dependent measure in the main analyses. Before conducting the critical analyses, we excluded eleven participants (five female) whose performance indicated that they were unable to remember the target colors. These participants failed to correctly discriminate the target letter in more than 20% of target catch trials and/or produced false alarms in more than 20% of non-target catch trials.

As illustrated in Figure 6, the results of Experiment 2 replicated the main findings of Experiment 1. First, performance was worse in _AA (80.2%) than in target alone trials (82.7%), [t(33) = 2.71, p < 0.01], consistent with an attentional capture effect (Folk et al., 2009; Folk et al., 2002; Folk et al., 2008; Moore & Weissman, 2010; Serences et al., 2005). Second, performance was worse in _AB (74.0%) than in _AA (80.2%) trials [t(33) = 3.78, p < 0.001], in line with a set-specific capture effect (Moore & Weissman, 2010). Third, set-specific capture was reduced when D1’s color matched a different attentional set than D2’s color as indicated by statistically equivalent performance in BAB (77.8%) and BAA (76.4%) trials [t(33) = 1.15, p = 0.26]. A repeated-measures ANOVA confirmed that set-specific capture was reduced when D1’s color matched a different attentional set than D2’s color (BAA – BAB), relative to when D1 was absent (_AA - _AB_) [F(1,33) = 23.03, p < 0.0001]. As predicted, this effect was driven by better performance in BAB than in _AB trials [t(33) = 2.10, p < 0.043].

Figure 6.

Target identification accuracy plotted separately for each of the main trial types in Experiment 2. Experiment 2 replicated the three main findings of Experiment 1. First, there was an overall attentional capture effect as performance in all conditions involving distraction was worse than that in target alone trials. Second, there was a set-specific contingent attentional capture effect because performance in _AB trials was worse than performance in _AA trials. Third, set-specific capture was reduced when D1’s color matched a different attentional set than D2’s color. Specifically, performance in BAB trials did not differ from performance in BAA trials and was better than performance in _AB trials. Critically, in line with the D1 enhancement hypothesis, performance was better in BAB than in BAC trials. Error bars illustrate the standard error of the mean.

Surprisingly, set-specific capture also appeared to be reduced when D1’s color matched the same attentional set as D2’s color: performance did not differ in AAA (76.4%) and AAB (73.6%) trials [t(33) = 1.49, p = 0.146]. Moreover, a repeated-measures ANOVA revealed that, at trend levels of significance, set-specific capture was reduced when D1’s color matched the same attentional set as D2’s color (AAA – AAB), relative to when D1 was absent (_AA – _AB) [F(1,33) = 4.10, p = 0.051]. However, this reduction occurred because performance was unexpectedly worse in AAA than in _AA trials [t(33) = 2.16, p < 0.038], rather than because performance was better in AAB than in _AB trials [t(33) = 0.255, p = 0.80]. Thus, as in Experiment 1, we conclude that the addition of D1 did not facilitate performance in _AB trials when D1’s color matched the same attentional set as D2’s color.

As mentioned in the discussion of Experiment 1, the findings above do not allow us to distinguish between the D2 blocking and D1 enhancement hypotheses. We arbitrated between these hypotheses by comparing performance in BAB and BAC trials. As uniquely predicted by the D1 enhancement hypothesis, target identification accuracy was higher in BAB (77.8%) than in BAC trials (72.8%) [t(33) = 3.46, p < 0.002]. Also in line with the D1 enhancement hypothesis, set-specific capture was observed in BAC trials: performance was significantly worse in BAC (72.8%) than in BAA trials (76.4%) [t(33) = 2.39, p < 0.023].

As in Experiment 1, we also investigated whether focusing spatial attention on an upcoming target’s location reduces contingent attentional capture (Folk et al., 2009). Once again, no such effect was observed. As mentioned above, performance was worse in AAA than in _AA trials, [t(33) = 2.16, p < 0.038]. We reserve a discussion of this result for the General Discussion.

Whether D2 and the target were from the same or different halves of the alphabet did not influence the results. There was no main effect of response congruency, and response congruency did not interact with any other factors (all p > 0.3).

Finally, as in Experiment 1, we repeated all critical analyses using reaction time as the dependent measure to determine whether there was evidence of a speed-accuracy tradeoff. As in Experiment 1, the latency data yielded no significant results. Thus, there was no evidence to suggest a speed-accuracy tradeoff.

Discussion

In Experiment 2, we distinguished between the D2 blocking and D1 enhancement hypotheses. The D2 blocking hypothesis predicted equivalent performance in BAB and BAC trials. In contrast, the D1 enhancement hypothesis predicted better performance in BAB than in BAC trials. Our findings supported the D1 enhancement hypothesis. We conclude that D1 reduced set-specific capture by bringing the upcoming target’s attentional set into the focus of attention.

GENERAL DISCUSSION

In everyday life, visual search is often guided by multiple attentional sets for color. For example, in the produce section at the grocery store, one may simultaneously search for apples and bananas by maintaining the colors red and yellow in distinct attentional sets. Using a laboratory analog of this task in which participants searched for target letters appearing in either of two possible colors (e.g., orange and green), we recently reported a novel contingent attentional capture effect (Moore & Weissman, 2010). Specifically, approximately 100–300 milliseconds after a target-colored (e.g., green) peripheral distractor was presented, it was more difficult to identify a target if its color matched a different attentional set (e.g., identify orange letters) than if its color matched the same attentional set. Low-level factors (e.g., perceptual priming) could not account for this set-specific capture effect. We suggested that detecting a target-colored item leads the corresponding attentional set to briefly enter a limited-capacity focus of attention in working memory, which temporarily impairs the ability to attend and identify a subsequent item whose color matches a different attentional set (see Figure 1).

In the present study, we further tested our hypothesis by investigating whether it is possible to reduce set-specific capture. We reasoned that presenting a target-colored (e.g., orange) central distractor (D1) would lead the corresponding attentional set (e.g., identify orange letters) to enter the focus of attention. Therefore, if a different-target-colored (e.g., green) peripheral distractor (D2) was presented 233 ms afterward, the attentional set corresponding to its color (e.g., identify green letters) would be unable to enter the already-occupied focus of attention. The result, we predicted, would be a reduction of set-specific capture arising from D2.

In Experiment 1, we both replicated our prior findings of set-specific capture and observed initial support for our hypothesis. First, target identification accuracy was lower when D2’s color and the subsequent target’s color matched different attentional sets than when they matched the same attentional set, consistent with set-specific capture (Moore & Weissman, 2010). Second, this effect was reduced by presenting D1 before presenting D2, but only when D1’s color matched a different attentional set than D2’s color. In sum, the results of Experiment 1 supported our hypothesis that set-specific capture can be reduced by preemptively occupying the focus of attention.

In Experiment 2, we distinguished between two competing hypotheses about how D1 reduces set-specific capture arising from D2. The D2 blocking hypothesis posits that D1 prevents the attentional set corresponding to D2’s color from entering the focus of attention. In contrast, the D1 enhancement hypothesis posits that D1 prevents the attentional set corresponding to D2’s color from entering the focus of attention and brings the upcoming target’s attentional set into the focus of attention (see Figure 4). Consistent with the D1 enhancement hypothesis, set-specific capture was reduced in BAB trials, in which D1’s color and the target’s color matched the same attentional set, but not in BAC trials, in which D1’s color and the target’s color matched different attentional sets. Bringing the upcoming target’s attentional set into the focus of attention was thus crucial for reducing set-specific capture arising from D2.

One may wonder, however, whether a top-down enhancement of D1’s color and/or position (e.g., Maljkovic & Nakayama, 1994, 1996) led to better performance in BAB than in BAC trials without disrupting set-specific capture arising from D2. Weighing against this possibility, data from one of our prior studies indicate that top-down enhancement of the attentional set corresponding to a target-colored item disrupts set-specific capture arising from a subsequent target-colored item (Moore & Weissman, 2010, Experiment 3). In this study, participants searched for two colors (e.g., orange and green) in a central RSVP stream while ignoring irrelevant items in two peripheral RSVP streams. In some trials, the peripheral distractor appeared after a central target. Critically, in these trials, participants’ ability to identify a target in a particular color (e.g., orange) was the same, regardless of whether the subsequent distractor was different-target-colored (e.g., green) or non-target-colored (e.g. lavender). In other words, the different-target-colored distractor did not capture attention. We argued that detecting the initial target led the attentional set corresponding to its color to enter the limited-capacity focus of attention. As a result, the attentional set corresponding to the subsequent distractor’s color could not enter the already-occupied focus of attention. Therefore, the different-target-colored distractor was unable to capture attention. This conclusion supports our present argument that D1 undermined set-specific capture arising from D2.

Of course, a target-colored distractor may simply fail to capture attention when it is presented after (rather than before) a target. Ruling out this possibility, performance in our prior study above was worse when the distractor’s color (e.g., orange) matched the same attentional set as the previous target’s color (e.g., orange) than when it mismatched (e.g., green or non-target-colored). Consistent with our hypothesis, this result indicates that detecting the target led to an enhancement of the attentional set corresponding to its color, thereby permitting a similarly-colored-distractor to interfere with performance shortly afterward. For these reasons, our prior findings show that detecting a potential target both (a) undermines the identification of targets that are defined by different attentional sets and (b) aids the identification of targets that are defined by the same attentional set. Thus, our prior results bolster our present argument that, at least in part, performance was better in BAB than in BAC trials because D1 undermined set-specific capture arising from D2.

Although the present findings support the D1 enhancement hypothesis, whether a particular experiment provides support for the D1 enhancement or the D2 blocking hypothesis should depend on the amount of time separating D1 from the target. In our view, the attentional set corresponding to D1’s color remains in the focus of attention only as long as D1 is attended. Evidence to support the D1 enhancement hypothesis should therefore be obtained only when D1 is presented shortly before the target, such that the attentional set corresponding to D1’s color still occupies the focus of attention when the target appears. In contrast, evidence to support the D2 blocking hypothesis might be observed when more time separates D1 from the target. Finally, even when a constant amount of time separates D1 and D2 (e.g., 233 ms in the present study), D2 blocking and D1 enhancement might occur in different trials of the same trial type if there is variability across trials in the amount of time spent attending to D1. Thus, additional studies investigating the temporal parameters that give rise to D2 blocking and D1 enhancement will be needed to fully appreciate the conditions under which set-specific capture can be reduced.

Future studies investigating the time course of set-specific capture may also shed light on why the present results differ so markedly from prior findings indicating that attentional sets can be strategically primed across trials (Belopolsky, Schreij, & Theeuwes, 2010). In seeming contradiction to the present results, these findings indicate that target identification in trial n is worse if a target shares a feature with a distractor in trial n-1 than if it does not (Lleras, Kawahara, & Levinthal, 2009; Olivers & Humphreys, 2003). However, this discrepancy is likely caused because set-specific capture and inter-trial priming of attentional sets have strikingly different time courses. While set-specific capture lasts for only a few hundred milliseconds, inter-trial priming of attentional sets emerges only after a second or more has passed. Thus, we argue that set-specific capture reflects a short-lived enhancement of the attentional set corresponding to a distractor’s color, which serves to increase attention to a possible target (Dux & Marois, 2009). Consistent with this view, presenting a distractor shortly before presenting the second of two targets in an RSVP stream reduces the magnitude of the attentional blink most strongly when the distractor possesses the upcoming target’s color (Nieuwenstein, 2006; Nieuwenstein, Chun, van der Lubbe, & Hooge, 2005). In contrast, inter-trial priming of attentional sets may reflect a strategic modulation of attentional control settings that is influenced by whether attending to a particular target color was beneficial in the prior trial (Folk & Remington, 2008). In sum, although we have found that set-specific capture influences performance in ways that differ markedly from inter-trial priming of attentional sets, this result likely stems from the different temporal intervals over which the two phenomena operate.

We have argued that the present findings fit nicely with a limited-capacity focus of attention that maintains just a single item or representation (e.g. an attentional set), but other explanations are possible. For example, our findings could also arise from an unequal allocation of limited resources among multiple attentional sets, consistent with models in which attention is allocated to multiple representations in a graded fashion (McLeod, 1977; Vergauwe, Barrouillet, & Camos, 2009). According to this view, an attentional set might receive the most resources when a stimulus matching its color appears first, fewer resources when a stimulus matching its color appears second, and so on. In such a scenario, detecting D1’s color (e.g., orange) would lead the corresponding attentional set to receive the lion’s share of resources, but would not fully prevent the attentional set corresponding to D2’s color (e.g., green) from receiving some resources as well. Still, D1 would greatly reduce the resources allocated to D2. Thus, this “divided resource” account might also explain our finding that D1 reduced set-specific capture arising from D2.

To distinguish between focus of attention and divided resource accounts of our findings, one could use a variant of our task in which participants identify D1, D2, and the target at the end of each trial⁵. In this task, one could compare target identification accuracy in BAB and BAC trials given that both D1 and D2 were successfully identified. Selecting trials in which D1 and D2 were identified would be crucial because, according to our focus of attention account, simply detecting a target-colored item does not guarantee that the corresponding attentional set enters the focus of attention (i.e., it will not enter if the focus of attention is already occupied).

Critically, when D1 and D2 are both identified, the focus of attention and divided resource accounts make distinct predictions about relative performance in BAB and BAC trials. According to the focus of attention account, target identification accuracy in these conditions should not differ. Indeed, after D2 is identified, only the attentional set corresponding to its color (i.e., “A”) should occupy the focus of attention. Thus, the ability to identify a target whose color (i.e., “B” or “C”) matches a different attentional set should be uniformly poor. In contrast, the divided resource account predicts higher target identification accuracy in BAB than in BAC trials. This account posits that resources have been divided between the attentional sets corresponding to D1’s color (“B”) and D2’s color (“A”). Thus, a target should be identified more accurately if its color (“B”) matches one of these attentional sets than if its color (“C”) does not match either of these attentional sets. Future studies aimed at differentiating between these accounts may further our understanding of set-specific capture.

Studies that distinguish trials in which D2 is identified from those in which D2 is not identified might also be useful for determining whether task set inhibition influences the size of set-specific capture effects. Evidence for task set inhibition often comes from task switching studies, in which participants respond more slowly and less accurately when the task they are cued to perform in one trial mismatches (versus matches) the task they performed in the previous trial (Monsell, 2003; Rubenstein, Meyer, & Evans, 2001; Wylie & Allport, 2000). This effect is exacerbated when participants are required to switch to a task that was recently performed, suggesting that switching away from a task involves inhibiting the associated task set (Mayr & Keele, 2000). In the present study, we observed better performance in BAB than in BAC trials, suggesting that switching attention from D1 to D2 did not lead task set inhibition to be applied to D1. However, the application of task set inhibition to D1 may only have been necessary when the attentional set corresponding to D2’s color actually entered the focus of attention, which probably occurred infrequently in BAB and BAC trials. Thus, isolating trials in which D2 is identified, and therefore enters the focus of attention, could be helpful for revealing whether task set inhibition influences the magnitude of set-specific capture effects.

The present results fit conceptually with prior findings indicating that contingent attentional capture arising from a target-colored peripheral distractor can be reduced (Folk et al., 2009), but they do not replicate those findings because we did not observe better performance in AAA trials than in _AA trials. Our failure to replicate this effect may stem from any of several differences between our experimental design and that of Folk et al. (2009). These include requiring participants to maintain multiple attentional sets as compared to just one, embedding peripheral distractors in RSVP streams instead of presenting them in isolation, and using a target-colored digit as a central distractor instead of a letter surrounded by a target-colored square outline. Future studies will be required to determine more precisely why we did not replicate Folk and colleagues’ finding that performance was better in AAA than in _AA trials.

As we suggested earlier, the present findings may have important implications for everyday activities in which target selection is guided by multiple attentional sets (Most & Astur, 2007). For example, while driving on a winding highway, a driver may be searching for both a yellow warning sign indicating an upcoming curve in the road and a restaurant billboard that is printed in the same color as the warning sign (e.g., yellow, Waffle House) or in a different color (e.g., blue, International House of Pancakes). The existence of set-specific capture suggests that the driver would be more likely to miss the yellow warning sign when it is immediately preceded by a target-colored billboard appearing in a different (versus the same) color. However, the present findings suggest that such effects could be reduced by placing a salient yellow object on the side of the road shortly before the billboard appears, thereby bringing the color of the upcoming warning sign into the focus of attention. Because almost 80% of car accidents are preceded by a moment of driver inattention (Ranney, 2008), failing to minimize set-specific contingent attentional capture in real-world situations may lead to dire outcomes.

In conclusion, we have shown that set-specific capture can be reduced by bringing an upcoming target’s attentional set into a limited-capacity focus of attention. This finding fits with other data indicating that contingent attentional capture can be reduced (Folk et al., 2009). It also has important implications for everyday activities in which multiple attentional sets guide the selection of relevant stimuli. Future studies investigating whether and how reducing set-specific capture arising from a target-colored distractor depends on (a) the relative times and spatial positions at which target and distractor stimuli are presented, (b) conscious perception of the target-colored distractor, and (c) task set inhibition may reveal important new information about how to minimize distraction in both laboratory and real-world situations.