Filtering items of mass distraction: Top-down biases against distractors are necessary for the feature-based carry-over to occur more

Vision Research 47 (2007) 1570–1583 www.elsevier.com/locate/visres Filtering items of mass distraction: Top-down biases against distractors are necessary for the feature-based carry-over to occur Jason J. Braithwaite *, Glyn W. Humphreys Behavioural Brain Sciences Centre, School of Psychology, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK Received 20 January 2006; received in revised form 7 February 2007 Abstract In preview search a new target is difficult to detect if it carries a feature shared with the old distractors [Braithwaite, J. J., Humphreys, G. W., & Hodsoll, J. (2003). Color grouping in space and time: Evidence from negative color-based carry-over effects in preview search. Journal of Experimental Psychology: Human Perception and Performance, 29(4), 758–778.] Two experiments are presented which examined whether this negative color carry-over effect is dependent on an attentional-set to ignore old, irrelevant distractors. Consistent with this, the data show that the negative carry-over effect is greatly reduced if the attentional-set to ignore the old preview items is removed and replaced by a set to prioritize the old items instead. The findings demonstrate that preview search, and the carry-over effect, are at least partly determined by a top-down intentional bias against old, irrelevant information. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Visual search; Preview-search; Inhibition, Carry-over effects 1. Introduction Visual search performance can be greatly improved if observers receive an initial preview of half of the distractor items before presenting the additional distractors and the target (Watson & Humphreys, 1997; see Watson, Humphreys, & Olivers, 2003; for a review). Although, within this procedure, the initial distractors remain present in the visual field (the second items being added to unoccupied locations in the display), they do not compete strongly for selection. These findings demonstrate that, provided the interval between the first preview display and the second display is sufficient (see Humphreys, Jung-Stalmann, & Olivers, 2004; Humphreys, Olivers, & Braithwaite, in press), the first distractors can be effectively ignored. Performance in the preview condition is thus greatly facilitated relative to a baseline condition where all the items appear * Corresponding author. E-mail address: j.j.braithwaite@bham.ac.uk (J.J. Braithwaite). simultaneously (the full-set baseline; see Watson & Humphreys, 1997 for the original demonstrations). This advantage to search has become known as the ‘previewbenefit’ (Watson et al., 2003). The factors that lead to this preview benefit have been subject to considerable debate. In the original account, Watson and Humphreys (1997) argued that the benefit stemmed from top-down, goal-based inhibition applied to the locations of the old distractors. By means of this inhibition, old items were filtered from search (a process they termed ‘visual marking’), enabling new items to be prioritized for selection. Watson and Humphreys proposed that visual marking was under top-down control and took time to become manifest (see Humphreys et al., 2004, in press, for evidence on the time course of the effects). Central to this original account was that static old preview items were inhibited on the basis of their locations and not their featural attributes; in this sense inhibitory filtering was held to be ‘feature-blind’. This claim has received some further recent support. For example, Watson and Humphreys (2002) showed that, with static items, there is no impact 0042-6989/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.visres.2007.02.019 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1571 on search efficiency when color changes take place in preview displays as the search stimuli are added. If old items were ignored by inhibiting their color (e.g., Treisman & Sato, 1990), then changing the color of the old items should ‘release’ them from inhibition, decreasing the preview benefit. This was not observed (see Olivers, Watson, & Humphreys, 1999; for further evidence of location-based inhibition). In contrast to the notion of inhibitory filtering, Donk and Theeuwes (2001) have argued strongly that the preview benefit reflects nothing more than automatic attentional capture induced by the new onsets produced by the second, search display. Donk and Theeuwes (2001) examined preview search in both the presence and absence of abrupt onsets by presenting items that were or were not isoluminant to their background. Performance was assessed where either (i) both the preview and search displays were isoluminant (no onset signals at all), (ii) just the new items were isoluminant with their background (no onset signals associated with the arrival of new items), or (iii) the preview display was isoluminant while the search display arrived with an abrupt onset. Donk and Theeuwes (2001) found that preview benefits only emerged when the new items arrived with an abrupt onset. Based on this finding, Donk and Theeuwes argued that new onsets were necessary to establish a preview advantage (see also Donk & Theeuwes, 2003; for further recent argument). A similar non-inhibitory account was proposed by Jiang, Chun, and Marks (2002) who suggested that performance was based simply on the ability to temporally segment the preview and second search displays from each other over time. As long as sufficient time was allowed between presentations of the two sets of items—attention could be directed towards the relevant new display without any need to assume the presence of inhibition directed towards the irrelevant display. 1.1. Are non-inhibitory accounts of preview search sufficient? Although attractive, previous evidence indicates that non-inhibitory accounts of preview search are not sufficient to explain all the results. Two critical pieces of evidence come from (i) probe-detection studies, where probes are presented to assess where attention is allocated during preview search, and (ii) color-based carry-over effects, from old to new displays. When a probe falls on a preview item it is more difficult to detect compared to when the probe falls at the location of a new item (Braithwaite, Humphreys, & Hullleman, 2005; Olivers & Humphreys, 2002; Watson & Humphreys, 2000) and even relative to unoccupied (neutral) background locations (Humphreys et al., 2004). Importantly, these costs to probe-detection are particularly pronounced when participants are engaged in a search task where new items must be prioritized. Under these circumstances participants appear to use a goal-directed bias against old, irrelevant distractors. However, these costs are greatly reduced when probe-detection is the sole task being carried out (thus removing the negative bias against the old items). This suggests that the preview benefit is influenced by the intention of participants to prioritize the new stimuli and to actively ignore the old items. The evidence for probedetection being inhibited at the old locations is not consistent with either onset capture or temporal segmentation alone being critical. If those factors were singularly responsible then the cost to probes would not increase as a function of the goal-directed intention to ignore irrelevant items. Furthermore, the greater cost to probes falling at old locations relative to empty background locations cannot be explained by a temporal segmentation account, which would predict no differences between empty locations and those occupied by old items. On the other hand, evidence of worse probe detection on old items than on background locations is consistent with the old stimuli being inhibited. Alongside the studies on probe detection, support for inhibitory coding comes from the effects of color similarity between the first and second displays. When the new target carries the color of the old distractors, target selection is disrupted relative to when the target has a different color (Braithwaite & Humphreys, 2003; Braithwaite, Humphreys, & Hodsoll, 2003, 2004; Braithwaite et al., 2005; Olivers & Humphreys, 2003). This is the negative color carry-over effect, reflecting a form of sustained attentional blindness to new items with properties of items being ignored (see Braithwaite et al., 2003). The effect suggests that, in addition any process of location-based inhibition (Watson & Humphreys, 1997), there is also inhibition of the color of the old items (i.e., featural attributes). If this inhibition spreads and is applied to the new items carrying the same color, then these items will become difficult to detect. Note that, if either capture of attention by new onsets or temporal segmentation alone were critical, then all the new items should be selected equally and irrespective of their color. One counter explanation for the inhibitory carry-over effect might be that, rather than inhibition spreading to the same-colored new items making them more difficult to locate, attention is automatically captured by the differently colored new items. It has been typical in prior investigations of the carry-over effect to have the second search display contain items of two colors; one set carrying the color of the preview (i.e., red) and the other set carrying a new unique color (i.e., green). Given this, then the cost for new targets carrying the color of the preview items could reflect attention being drawn to the new distinctive color in the display. This too would predict a cost for new items carrying the color of old distractors (e.g., red, in this case)—but this would have nothing to do with inhibition. There are, however, a number of findings against this proposal. For example, there is a large advantage to be gained by providing observers with valid foreknowledge of the target’s color—even when that color is unique in the new search display (Braithwaite & Humphreys, 1572 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 2003). On the other hand, there should be no effect of cueing the target’s color if new items automatically capture attention. In addition, any automatic capture of attention by a new unique color should be relatively fast acting. Against this, Braithwaite et al. (2003) showed that both the overall preview benefit and the color carry-over to new items were abolished when the preview duration was reduced to 150 ms. The effects were re-established when the preview duration lengthened. Rather than a fast-acting capture process directing attention to a new unique color, these data are more consistent with the slow-acting inhibition of old items, that spreads to the new items carrying the color of the old stimuli (see Braithwaite & Humphreys, 2003; Braithwaite et al., 2003; for more extensive discussions). Braithwaite and colleagues (Braithwaite et al., 2005) have further shown that the negative color carry-over effect can be distinguished from effects of color grouping in the first display. Braithwaite et al. (2005) changed the color of the old items when the new stimuli appeared. Items in the preview display could be in one of two colors (say red or green), with one color carried by the majority of the stimuli (i.e., 66% of the items were red and 33% of the items were green). On the appearance of the second search display, the preview items were given new colors (i.e., blue and yellow). Braithwaite et al. (2005) found that new targets carrying the same color as the majority of the old distractors (red), remained difficult to detect (i.e., there was a negative color carry-over effect), even though these new targets could not group with the old distractors now on the basis of their color. In addition, a probe detection procedure was used to measure where attention was allocated in the search displays. Probes were difficult to detect if they fell on old items that carried the initial majority color (even though that color changed when the new items appeared). This is consistent with inhibition of the old distractors based on these items grouping by color, with the same items remaining grouped even when their common feature changed (from one color to another). Thus, there was one influence of color on the initial grouping, and another due to inhibition of that color, which could impact on subsequent search. These results are not predicted by either the onset capture or the temporal segmentation accounts (Donk & Theeuwes, 2003, 2001; Jiang et al., 2002). 2. Overview of the present study The account of the color carry-over effect, proposed by Braithwaite et al. (2005), assumes that it reflects featurebased inhibition of the majority color from old distractors. Moreover, the inhibition of old items is held to be an active process adopted to bias search against old distractors. As we have noted, evidence for active processes in preview search comes from studies of probe-dot detection, where the differences in detecting probes on old and new displays is greatly reduced when participants are not biased against old items for search (Braithwaite et al., 2005; see also Olivers & Humphreys, 2003). However, no previous studies have demonstrated that active biases against old items are necessary for the carry-over effects to occur. This was the aim of the present study. Does the color carry-over effect, like the evidence for inhibition of old items, depend crucially on the ‘set’ of the participants? To assess this question, we combined a preview search task (discriminate a target letter in the new display) and a probe detection task (discriminate a letter that is slightly brighter than the others which always fell in the old set of items only), while manipulating the relations between the colors of the stimuli in the old and new displays. We employed the color-biased displays devised by Braithwaite et al. (2005), where previews contained color groups that were unequal in size: 66% of the items were in one color (the old majority group) and 33% in another color (the old minority group). The search display contained the opposite color bias. Thus 33% of the new items carried the color of the old majority group (the new minority group) and 66% of the new items carried the color of the old minority group (the new majority group). In the final display the two biases cancelled each other out, so that there were equal numbers of items in each color. In Experiment 1, there was a ‘standard’ preview condition, where the task was to search for a target letter which always appeared in the new, search display (Target 100% new). This was complemented by a probe-detection condition, where the probe always fell on an old item in the preview display (Target 100% old). These conditions provide baselines to assess (i) whether there is a preview benefit in target search with the present displays, and (ii) the maximum efficiency of probe detection on old items (when resources were fully committed to probe detection). In a second (mixed) search condition the target search and probe detection tasks were combined. Probe detection was the primary task and there was a bias to prioritize the old items (on 80% of trials) for probe detection; search for a new target letter then occurred on just a minority (20%) of the trials. Here trials where there is search for a new target are like ‘catch trials’. Under these conditions, the old items should be prioritised for the probe detection task. Using these minority search trials, we then examined whether there remained color biases on the selection of the new items (in this case, when attention was biased towards (rather than away from) the old stimuli). If the negative color carry-over effect occurs in a passive manner (i.e., just due to the colors being used), then color biases with occasional new targets (20%) should be as strong as in the standard preview search condition (Target 100% new). This negative color carry-over would be shown by poorer detection of a target when it carried the same color as the old majority group (rather than the old minority group) of items in the preview display. Performance was also compared to a full-set condition (with all the items presented simultaneously), to provide a baseline measure of performance (see Fig. 1 for an illustration of the displays used). J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1573 I V V X I H X I H X H V H I H X H X I V and, in particular, for modulation of color carry-over effects. 3. Experiment 1 + N I V X H I X 3.1. Method 3.1.1. Participants Twenty participants (8 male, two left-handed) took part for course credit or small payment. The age of participants ranged from 18 to 42 years with a mean age of 25 years. All were undergraduate or postgraduate students at the University of Birmingham. All had self-reported normal (including normal color vision) or corrected-to-normal vision. 3.1.2. Stimuli and apparatus All the stimuli and the conditions were generated by computer programs written in Turbo Pascal (v7) and were run on a Pentium PC fitted with a 17-in. super VGA monitor. Viewing distance was not fixed but was approximately 60 cm. The stimuli consisted of colored (red & green— luminance-matched via a color fusion/flicker test) capital letters displayed on the plain black screen background. The letters had an approximate width of 6 mm, and height of 8 mm. These letters were randomly assigned to an invisible 48 cell, circular matrix consisting of three concentric circular ring grids. The distance from central fixation to the middle of the cells of the first ring measured approximately 20 mm (containing 8 cells), the second ring 40 mm, (containing 16 cells) and the third 60 mm (containing 24 cells). Distractors consisted of the upper case letters H, I, V, X, and the target letters (for the letter search task) were either a Z or N. Search displays were generated by randomly positioning each letter in the middle of individual matrix cells. Any distractor letter could repeatedly occur in multiple numbers in any display with the restriction that at least one distractor letter of each type was presented. The preview conditions involved the presentation of half (12 items) of the distractor letters first (in the first preview display) followed by the remaining half (12 items) in the second, search display. The full-set baseline consisted of a single presentation of both displays combined (consisting of 24 items in total). For target-letter search trials the target was a Z or N, 50% of the time (at random) and could be either color (red or green) equally often (see Fig. 1). On probe trials, a single probed old preview distractor underwent a marginal increase in luminance (when it fell on old preview items) or was presented as a marginally brighter item (in the full-set baseline). The luminance increment level was based on pilot studies and previous published research using this technique, where the increment was set so that the probe item did not immediately ‘pop out’, which would obscure differences between the critical search conditions (see Braithwaite et al., 2005). For preview conditions both the preview display and the second, search display contained unequal numbers of items in each + I X Preview display H I H X X H V Second display added Time (1000ms) Fig. 1. A schematic illustration of the preview paradigm used in the present study. The preview display was presented for a 1000 ms period. The items in the preview display were color-biased, with 66% being in one color (illustrated by the black letters), and 33% in a second color (illustrated by the light grey letters). After the 1000 ms preview period the second display was added to the existing items until a response was made or 10,000 ms had passed. The second display contained items with the opposite color-bias to the initial preview items (so the combined display had no overall ratio = 50/50). On probe trials, there was a temporary brightening of one of the items in the old majority or old minority group, concurrent with the onset of the second (search) display. Probe trials were signalled by an auditory cue occurring 20 ms before the arrival of the second display. On search trials, a target letter (Z/N) appeared in the new second display in either the new minority or new majority color. In this example trial the target is the letter ‘N’ and it occurs in the new minority group (i.e., the group carrying the colour of the majority of items in the preview). Under this condition, there is a strong negative color carry-over effect (see the text for details). Now, it could be argued that any contrast between standard preview search, and search in the mixed search condition could have been confounded by several differences between these conditions: for example, there was increased uncertainty in the mixed condition since either of two tasks could be required (probe detection as well as target search), whereas there was a single task (search) in the standard preview condition; also, in standard preview search a two-choice response was made whereas in the mixed condition participants made a simple reaction time response followed by a letter identification response (see below). Though it is unclear why such differences should differentially affect color carry-over effects, we nevertheless conducted Experiment 2 in which these differences were held constant across the critical conditions. Here we ran two identical preview conditions which both employed target letter search and probe-detection trials embedded in the same blocks of trials. The only difference was whether the bias was in favor of probe detection on the old preview items (80% old-item, 20% new-item: as with Experiment 1) or target letter search in the new items (80% new-item, 20% old-item). In the former condition, participants should prioritise the old items (for probe detection); in the latter condition they should prioritise the new items (for letter search). Any differences between these conditions would provide evidence for a role of top-down processes in search 1574 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 color (an old majority and old minority color group; i.e., 66% red vs 33% green in the preview display, versus a new minority 33% red, and new majority 66% green in the second search display). Color groups were balanced across the displays so that, overall, there were equal numbers of letters in each color (as was the case in the full-set baseline when all the items appeared together). The colors assigned to the majority or minority color groups were counter-balanced within blocks of trials (so that a given color was not reliably associated with a group size). 3.1.3. Design and procedure There were four experimental conditions, consisting of one full-set baseline condition and three preview conditions employed in a 4 · 2 (Condition · Group) within-subjects design. Each condition was run as a separate block of 200 trials. There were two forms of trial type; these were (i) target letter search (Z/N) or (ii) probe-detection (detect the luminance probe). The conditions differed as to whether they contained letter target search trials only, probe-detection trials only, or a mixture of both with a trial-by-trial bias favoring probe trials (80% of trials were probe trials in the combined conditions). For the preview search task, target-letters always fell at new locations and were presented with the second set of distractors. In contrast, for the probe-detection task, the probe always fell on old preview distractors and were signaled by an auditory cue occurring 20 ms before the presentation of the second display. There were three preview conditions all of which had an old majority and old minority color group in the preview, followed by items in a new minority and new majority group in the second display. Items in the old majority color were in the minority color in the new display and vice-versa for stimuli that were in the old majority group. In the standard preview condition, only target letter search trials occurred, the target being a new item in the second search display on all trials: the target 100% new condition. Participants were instructed to ignore the preview items, which were always irrelevant, and they had to discriminate the new target (Z or N?). In a second preview condition, only probe-trials occurred, with the probe always falling at the location of the one of the previewed items (100% of the time). We refer to this condition as target 100% old. Under these circumstances, observers were told that the old items were the relevant set and that a luminance-probe would always fall on one of the preview items, coincident with the presentation of the second display. Here participants may be expected to prioritize the old items in the preview. Table 1 The conditions and target types in Experiment 1 Condition Full-set New-item 100% Old-item 100% New-item 20% old item 80% Target search (new-item) (20% of the time) (100% of the time) (new min/new maj) — (20% of the time) (new min/new maj) In the third, mixed preview condition both trial types occurred. Participants were told that, on most occasions (80% of the time), the trial would be a probe-detection trial and the probe would always fall on an old item (which was valid). This was the primary task (the 80% old condition). However, on a minority of trials (the remaining 20%) there was no probe, and, instead, the task was to detect a targetletter in the new second display (i.e., target detection formed a type of catch trial): the target 20% new trials. Under these circumstances, we expected participants to prioritize the selection of the old rather than the new stimuli, as the old stimuli contained the target event most of the time. Coupled to these preview conditions was a full-set baseline where all the (24) search items were presented simultaneously (condition ‘full-set’). Note that, in this case the two color groups were equal in size. In this full-set baseline, either target letter search or probe detection could be carried out, with the probabilities of the two trial types matched to those in the target 80% old condition (i.e., there were probe targets on 80% of the trials and letter targets on 20%). Table 1 provides a summary of the conditions in Experiment 1. In the preview condition, for target letter search, there were equal numbers of trials for targets in the new minority and new majority groups. Similarly, the total number of probe trials, were divided equally across the two possible old distractor groups (old majority/old minority). The target was assigned to a group randomly across trials. The trial type (target search/probe detection) within each condition was also randomized within blocks and block order was randomized across participants. A general block of 40 practice trials, for each of the preview and the full-set baseline conditions, was completed at the beginning of the experiment. None of these practice trials were included in the analysis. Letter search trials took the following form. Each trial began with the presentation of a plain white fixation cross, which remained visible until the end of each trial. For the preview conditions, this was followed by the presentation of the preview display and participants were instructed to remain fixated and not to initiate search until the arrival of the second display (whether it was the relevant or irrelevant display). Reaction Times (RTs) were measured from the onset of the second display. For luminance probe trials a 20 ms long auditory beep of 1000 Hz occurred 20 ms before the presentation of the second search display, to signal this was a probe-detection trial. For conditions where both trial types could occur, prior to the cue, participants did not know the nature of the search task for that partic- Probe search (old-item) (80% of the time) — (100% of the time) (old min/old maj) (80% of the time) (old min/old maj) Mixed conditions Yes No No Yes J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1575 ular trial. Participants were instructed that, when the auditory cue occurred, one of the preview items would become slightly brighter and the task was to identify it as quickly and as accurately as possible. The probe could fall on any of the four distractor letter types (chosen randomly), and it fell equally often on either the old minority or old majority color. Participants responded to probes were made by pressing the space bar as soon as they located the probe. The display was then immediately cleared of all items and participants were presented with a new screen that asked ‘what was the brightest letter’, along with giving the four possible options (H I V X). Participants typed the identity of the probed letter from the four possible distractor letters and their accuracy was recorded. For letter search tasks participants made a button press (Z/N) to indicate whether the target was a letter Z or N. Note that the number of potential letter responses varied across the letter search (2 alternative responses) and probe trials (4 alternative letter responses). This may lead to general differences in reaction times across the tasks.1 However this general factor is orthogonal to the main variables of interest here—namely whether the target fell in a majority or minority color group. Any differential effects of color grouping on the two tasks, is unlikely to be due to variations in response uncertainty (though Experiment 2 was conducted as a further test of this). The procedure was the same for the full-set display except that all the items appeared together and on probe detection trials, the beep occurred at the onset (20 ms before) of the whole display. The experiment lasted approximately 50 min. 3.2. Results The reaction time (RT) data for both letter search and probe detection trials were trimmed for outliers (±2.5 standard deviations and any response faster than 200ms) and incorrect responses. All data were analyzed via a series of within-subjects ANOVAs carried out separately for each task. Further decompositions and planned comparisons were carried out where relevant. Three of the conditions, the full-set baseline, the target 100% new, and the target 80% old conditions, contained target-letter search trials. These were analyzed below. 3.2.1. Overall performance on letter search trials Overall RTs were compared (collapsed across both minority and majority groups) for letter search trials from An anonymous reviewer suggested that differences in target-letter (two choices) and probe-letter (four choices) might be a concern. However, based on this RTs would be expected to be faster on letter search than on probe detection trials. Contrary to this, RTs were faster in probe detection than letter search (compare Figs. 2 and 3 here with Fig. 4). This indicates that the choice of letter response had a much weaker effect on performance than the stimulus cues guiding attention to the target, which was more salient in probe detection (a luminance increment) than in letter search (where the target differed from distractors only in the arrangement of its features). 1 the full-set, target 100% new and target 80% old preview conditions in a one-way within-subjects ANOVA. This was highly significant, F(2, 38) = 40.69, p < .001. This effect was further explored by a series of planned comparisons. These revealed significant differences between the full-set and target 80% old (20% new) conditions, F(1, 19) = 27.60, p < .001, with RTs being quicker in the latter case. Even though observers were encouraged to prioritize the old items, they were quicker when a target came in the new display relative to the full-set baseline. There was also a significant difference between the full-set baseline and the target 100% new condition, F(1, 19) = 52.71, p < .001. This represents the standard preview benefit. Finally, RTs for new target letters were significantly faster in the target 100% new condition than in the target 80% old condition. F(1, 19) = 28.44, p < .001. RTs were fastest when participants always prioritized the new items. This suggests that there was an additional advantage to target letter search, when the new items were being actively prioritized. The overall RT data for new-item targets are shown in Fig. 2. 3.2.2. Color group effects on target letter search The effect of whether the target letter in the preview condition was in the new minority or new majority group was assessed across each pairing of letter search conditions using a 2 (search condition) · 2 (color group) ANOVAs. The data are presented in Fig. 3. 3.2.2.1. Full-set vs target 100% new. There were significant main effects of Condition, F(1, 19) = 53.23, p < .001, and Color group, F(1, 19) = 104.89, p < .001. The Condition·Color group interaction was also significant, F(1, 19) = 54.46, p < .001. RTs were significantly faster in the target 100% condition relative to the full-set baseline condition. Furthermore, there was a large effect of color group in the 100% new condition. In the target 100% new condition, RTs were slower when the target was in the new minority color in the second search display (carry2000 Mean correct RT (ms) 1750 1500 1250 1000 750 500 Full Target 20% new Target 100% new Disp 24 Condition Fig. 2. RTs (ms) for the target letter search tasks (to new items only). Performance is shown for the full-set baseline (left bar), the target 20% new condition (middle), and the target 100% new condition (right bar). 1576 2000 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1500 Mean correct RT (ms) Mean correct RT (ms) 1750 1500 1250 1000 750 500 Full Target 20% new Target 100% new Disp 24 1250 Min Maj 1000 750 500 Full-probe Target 80% old Target 100% old Condition Fig. 3. RTs (ms) for target letter search as a function of whether the target was in the new-minority (light grey) or new-majority (dark grey) color in the second display. Condition Fig. 4. RTs (ms) for the probe-detection task in Experiment 1. Performance for probe-detection is shown for the full-set baseline (left bar), the target 80% preview condition (middle bar) and the target 100% preview condition (right bar). ing the color of the old majority color in the preview display). This is the negative color-based carry-over effect. 3.2.2.2. Full-set vs target 80% old. Only the main effect of Condition was significant, F(1, 19) = 27.49, p < .001. Both the main effect of Color group, F(1, 19) = .85, p = .363, and the Condition · Color group, interaction were not significant, F(1, 19) = .66, p = .426. Although performance was improved in the preview condition where targets were 80% old (and thus 20% new), there were no effects of color group. This shows that the negative color carry-over is modulated if the negative bias towards the old items is removed (by having the target events fall there most of the time). 3.2.2.3. Target 100% new vs target 80% old. There were significant main effects of Condition, F(1, 19) = 22.25, p < .001, and Color group, F(1, 19) = 33.81, p < .001. The Condition · Color group interaction was also significant, F(1, 19) = 20.47, p < .001. RTs were faster in the target 100% new condition, and an effect of color group emerged only in this condition. Here the negative color carry-over for targets in the new minority was only present in the 100% new condition. 3.2.3. Probe trials RTs from the probe-detection trials were then compared from the Full-set, target 80% old and target 100% old conditions. There was a significant difference between the conditions, F(2, 38) = 25.50, p < .001. There were significant differences between the full-set and target 80% old conditions, F(1, 19) = 35.39, p < .001. RTs were quicker for probes falling on distractors in the full-set baseline condition relative to probes falling on old locations under preview conditions. There was also a significant difference between the full-set baseline and the target 100% old condition, F(1, 19) = 19.36, p < .001, with again probe RTs being faster in the full-set baseline. Finally, there was also a reliable difference between RTs for probes falling on old items in the target 80% old and the target 100% old conditions, F(1, 19) = 12.31, p < .01 (see Fig. 4).2 RTs to probes were faster when the probe fell on an old item 100% of the time relative to when it fell there 80% of the time. 3.2.4. Color group effects on probe-detection The effects of color grouping on probe detection were examined in a series of 2 (Condition) · 2 (Color group) ANOVAs conducted on each pairing of the conditions. 3.2.4.1. Full-set vs target 80% old. Both the main effect of Condition F(1, 19) = 37.21, p < .001, and of Color group, F(1, 19) = 10.97, p < .01, were significant. Probe detection was much faster in the full-set baseline than in the preview condition with 80% old targets. The Condition · Group interaction was also significant, F(1, 19) = 36.68, p < .001. Here there was a selective effect of grouping in the target 80% old condition. RTs were slower to probes to items in the old majority color than they were to probes on old minority color in the preview {F(1, 19) = 38.59, p < .001}. Note, this effect is in the opposite direction to that seen for the carry-over onto new-item search targets (see above). There was no effect of color in the full-set baseline {F(1, 19) = .48, p = .498}. The data are shown in Fig. 5. 3.2.4.2. Full-set vs target100% old. The main effect of Condition was significant, F(1, 19) = 18.74, p < .001. Probes were harder to detect in the condition where the target was 100% in the old set, relative to the baseline condition. The main effect of Color group was not significant, F(1, 19) = 2.38, p = .139. The Condition · Color group interaction was significant, F(1, 19) = 8.77, p < .01. There Note that the Y-axis for probe trials is fixed at 1500 ms maximum while for all target-search trials it is set at 2000 ms maximum (though incremental steps in RT are the same). This is due to the fact that some of the differences in the probe conditions are difficult to appreciate at 2000 ms because overall RTs to probes are much quicker. Nonetheless, RTs for trial type (probe vs letter search) are calibrated to be consistent within themselves. 2 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1500 1577 Min Maj a 20 18 16 14 12 10 8 6 4 2 Min Maj Mean correct RT (ms) 1250 1000 750 500 Full Target 80% old Target 100% old Error (%) 0 Target 20% new Target 100% new Condition Fig. 5. RTs (ms) for the probe detection task as a function of whether the probe fell on an item in the old-minority (light grey) or old-majority color (dark grey) in the preview display from Experiment 1. Condition b 20 18 16 14 Error (%) was an effect of color grouping in the target 100% old condition {F(1, 19) = 11.85, p < .01} but not the full set baseline. In the former condition probes were harder to detect if they fell in the old majority color in the preview display. 3.2.4.3. Target 100% old vs target 80% old. Both the main effect of Condition and Color group were significant, F(1, 19) = 15.78, p < .01, and F(1, 19) = 65.97, p < .001, respectively. Probe-detection was faster when observers prioritized the old display on 100% of the trials. The Condition · Color group interaction was also significant, F(1, 19) = 5.12, p < .05. Probes falling in the old minority group were detected faster than probes falling in the old majority group. However, the effect of color grouping on probe-detection was significantly reduced when observers prioritized those items on 100% of the trials. 3.2.5. Errors The total overall error rate was very low at 3.7%. There were 5.9% errors on letter search trials and 2.7% on probe detection trials. Thus probe-detection trials were not only on average faster, but also more accurate relative to target search trials. Errors on letter search trials were analyzed in a similar manner to RTs in a one-way within-subjects ANOVA, but here this failed to be significant, F(2, 38) = 2.67, p = .08. Errors for search targets from both the 100% new and 20% new preview conditions were compared as a function of group in a 2 · 2 (Condition · Group) ANOVA. Although the effects of group increased when target search was carried out 100% of the time (with more errors occurring for new minority targets) this was not significant. Indeed, only a significant main effect of Group was observed, F(1, 19) = 9.22, p < .01. The main effect of condition failed to reach significance, F(1, 19) = 3.66, p = .07, as did the Condition · Group interaction, F(1, 19) = 1.94, p = .180 (See Fig. 6). A similar analysis of probe-detection errors revealed a significant difference across the conditions, F(2, 38) = 14.73, p < .001. There were significantly fewer errors produced in the target 12 10 8 6 4 2 0 Target 80% old Target 100% old Min Maj Condition Fig. 6. Error rates (expressed as percentage) for new targets falling in the new minority/new majority color (a—top panel), and for probes falling in the old minority/old majority color groups (b—bottom panel) from preview conditions of Experiment 1. 100% old condition relative to both the target 80% old condition, F(1, 19) = 23.67, p < .001, and the full-set baseline, F(1, 19) = 11.07, p < .05. There were also fewer errors in the full-set baseline relative to the target 80% old condition, F(1, 19) = 7.48, p < .05. The effect of Group was explored by comparing old probe RTs from both preview conditions. This revealed only a significant effect of Condition, F(1, 19) = 23.67, p < .001. Errors were significantly fewer in number for the probe 100% old condition. The main effect of Group, F(1, 19) = .08, p = .774, and the Condition x Group interaction, F(1, 19) = .50, p = .487, were not significant. Errors generally followed the pattern of RT and there was no evidence of a speed-accuracy trade-off (see Fig. 6). 4. Discussion 4.1. Letter search There was an advantage for search in the two preview conditions—even when the old items should have been prioritized (target 80% old)—compared with the full-set baseline. The fact that there was an advantage for preview conditions even when the old items should 1578 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 have been prioritized (target 80% old) is suggestive of there being some relatively automatic component that benefits preview search compared to when all the items appeared together. This would be consistent with attention being captured automatically by the new items (Donk & Theeuwes, 2001) or there being automatic segmentation of the old and new items into separate temporal groups (Jiang et al., 2002). Alternatively, it may be that participants were not perfect at prioritizing the old items, and actively prioritized the new items on some trials. Whatever the case, the additional advantage for the target 100% new condition (i.e., the standard preview condition) indicates that, even if there is an automatic component in the preview benefit, there is also a substantial additional component due to active prioritization of the new items. Thus search was significantly improved when the new target was expected on 100% of the trials. In addition to differences in search, there was an effect on search of the color of the new targets, but only when new items were prioritized for search (in the target 100% new condition). In this condition, targets were easier to detect if they fell in the new majority color relative to when they fell in the new minority color. Note that this result is the opposite of the usual effects of color segregation and grouping that arise when all the search items appear simultaneously. With simultaneous presentations of the stimuli, targets are typically easier to find when they carry the minority color in the display (Bacon & Egeth, 1997; Egeth, Virzi, & Garbart, 1984; Kaptein, Theeuwes, & van der Heijiden, 1995; Moore & Egeth, 1998). Nevertheless, the pattern of data replicates that previously reported under preview conditions (Braithwaite et al., 2003, 2005) and fits the pattern of the negative color carry-over effect: targets were harder to detect if they carried the color of the majority of items in the preview. This reversal of the standard color grouping effect (with simultaneous items) is difficult to reconcile with effects due solely to automatic onset capture (Donk & Theeuwes, 2001, 2003), to automatic temporal segregation (Jiang et al., 2002) or to the feature-blind inhibition of locations (Watson & Humphreys, 2002, 1997; see Watson et al., 2003). For example, if there is automatic capture of attention by the new onsets, then targets in the new minority color should be easier to find. The results were opposite to this. Instead the data fit with the idea that there is inhibition of the old majority color carried by the preview items, which is transferred to the letters in the new search display carrying the same color (the new minority group). The new result here is that the carry-over effect was confined to when participants could actively prioritize the new items (in the target 100% new condition), and it did not occur when the old items were prioritised (target 80% old). This suggests that the effect is conditional on an active top-down bias against the old items, and their featural properties. 4.2. Probe detection Probes were generally more difficult to detect under preview conditions than they were in the full-set baseline. This is not surprising. In the preview conditions the probes appeared at the same time as the onset of the new search items, and so had to compete with multiple other new transients in the displays. Probes were also more difficult to detect when probes were prioritized on just 80% of the trials (under preview conditions) compared to when they were prioritized on 100% of the trials. This result is inconsistent with there being automatic capture of attention by the new onsets (Donk & Theeuwes, 2001). If there were automatic attentional capture alone, attention should have switched to the new onsets irrespective of whether the old items were being prioritized. Instead, the data suggest that participants can set themselves against new onset capture, at least to some degree. In addition to these overall effects, there was an effect of color grouping on probe detection, with probes being easier to detect if they fell in the old minority color in the preview, compared to when probes fell in the old majority color group. Note that this effect mimics the result from color grouping under standard search conditions, with all the items presented simultaneously (Bacon & Egeth, 1997; Egeth et al., 1984; Kaptein et al., 1995; Moore & Egeth, 1998). The effect of color grouping was greater in the target 80% old condition compared to the target 100% old condition. This color grouping effect may be partly due to automatic segmentation of the initial display into minority and majority color groups, with attention being drawn to the minority group. Even so, there was an enhancement of the effect in the target 80% old condition. This may reflect several factors. It could be that the effect of automatic color grouping is increased if attention is divided to some degree between the old and the new items, which may be the case in the target 80% old condition. It could also be that effects of automatic color grouping are reduced by attention being prioritized specifically to a luminance increment in the old items, in the target 100% old condition. It is important to note, however, that the present results dissociate the effects of grouping (which were evident on probe detection) from the color carry-over effects (which were absent on target search, when old stimuli were prioritised). The data suggest that, whereas the grouping effect may be relatively automatic, the color carry-over effect reflects an active process, produced when observers set a bias against the old items to prioritize the new (in the target 100% new condition only). The color carry-over effect is not an inevitable consequence of the display parameters used here. Experiment 1 contained conditions where both search and probe trials were embedded in the same block of trials (along with conditions containing only probe or only search trials). This intermixing of trial types in the embedded block meant that there was a period of uncertainty (until the auditory cue occurred) on any given trial as to J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1579 which display was relevant for the task (though the weighting of the two tasks induces an attentional bias towards the old items). This increased uncertainty might have lengthened RTs in the 80–20% condition compared to when there was certainly about the task being performed (in the 100% old and new conditions), and it could have contributed to the variations in the effects across the different conditions. In addition there was a contrast between two target letters being discriminated in the standard preview condition (100% target new) and a simple reaction time followed by a four-choice letter identification in the mixed (80–20%) condition.3 To test whether such factors were crucial, we conducted Experiment 2. In Experiment 2 there was always uncertainty about which task was to be performed (letter search or probe detection). As in Experiment 1 there was one condition in which there was an 80% bias in favor of probes falling on old preview items (with letter search in the second display on the remaining 20% of the trials). This should encourage a top-down bias in observers to prioritize the preview items on most trials. Coupled to this was an alternative preview condition where identical displays were employed but the bias between trial types was reversed. Here the task was to search for a new target letter on 80% of the trials and it was to search for a probe in the preview on 20% of the trials. Under this circumstance, observers should adopt an attentional set against the old items, since these stimuli are usually irrelevant. Note, however, that as both conditions contained search and probe trials within their respective trial blocks, any difference in performance between these conditions must be due to the relative top-down bias and not general uncertainty across the trial block. 5. Experiment 2 5.1. Method 5.1.1. Participants Eighteen participants (11 male, one left-handed) took part for course credit or small payment. The age of participants ranged from 18 to 35 years with a mean age of 21 years. All were undergraduate or postgraduate students at the University of Birmingham. All had self-reported normal (including normal color vision) or corrected-to-normal vision. 5.1.2. Stimuli and apparatus The stimuli were the same as those reported for Experiment 1. The only difference was the inclusion of a new preview condition where there was a bias towards target-letter search (see below). 5.1.3. Design and procedure There were two preview search conditions. Both conditions contained mixed trial types, with both target letter search (in the new items only) and probe-detection (in the old items only) being possible. The conditions differed in the relative probabilities of the two tasks. In one case we repeated the 80% probe detection (old-item) and 20% letter search (new-item) condition from Experiment 1 (the Target 80% old condition). Old items should be prioritized in this case. In addition, a second condition had 20% probe detection (old-item) trials and 80% letter search (new-item) trials (the Target 80% new condition). Here the new items should be prioritized most of the time. The remaining procedure matched that outlined for Experiment 1. 5.2. Results The data were made fit for analysis in the same way as for Experiment 1. We then analyzed RTs for targets falling in the second display followed by RTs to probes falling in the preview display. 5.2.1. Performance for new-item targets Performance was assessed for new item targets both when they occurred as the primary task (80% of the time) and when they occurred as a secondary task (20% of the time). A 2 (Condition: bias to new vs bias to old) · 2 (Color group: new minority vs new majority) within-subjects ANOVA revealed a marginally significant effect of Condition F(1, 17) = 4.30, p = .05. The main effect of Color group, F(1, 17) = 34.23, p < .001, and the Condition · Color group interaction, F(1,17) = 122.08, p < .001, were significant. RTs for new targets were faster when these items were prioritized (Target 80% new vs 20% new), and there was a larger negative carry-over effect on new minority group targets in this condition (see Fig. 7). We further examined individual effects of color group within each of the biased conditions. There were no significant effects of color group when the target was usually in the old display (Target 80% old), F(1, 17) = 2.32, p = .146. 2000 Mean correct RT (ms) 1750 1500 Min 1250 1000 750 500 Target 20% new Target 80% new Maj Condition Fig. 7. RTs (ms) for targets falling in the new displays from both conditions of Experiment 2. Performance is shown here as a function of the new group (new minority = light grey and new majority = dark grey). 3 We thank an anonymous reviewer for this suggestion. 1580 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 Error (%) In contrast, there was a significant effect of color group in the Target 80% new condition, when attention should be prioritized to the new items and against the old, F(1, 17) = 115.78, p < .001. Therefore, the overall effect of Color group is largely due to the carry-over present only in the target 80% new condition, where the old preview items had to be ignored most of the time. 5.2.2. Performance on probes in preview displays RTs for probes falling in the old preview display were analyzed in a 2 · 2 (Condition · Color group) within-subjects ANOVA. This revealed a significant effect of Condition, F(1, 17) = 14.33, p < .01. Probe-detection was slower when probes fell in the old items on 20% of the trials relative to when probes fell there on 80% of the occasions. There was also a significant main effect of Color group, F(1, 17) = 31.17, p < .001, and a significant Condition · Color group interaction, F(1, 17) = 15.72, p < .01. RTs for probes that fell in the old majority group were slower relative to probes that fell in the old minority group, and this effect was much greater when observers were biased against the old preview display (the new item 80% condition: see Fig. 8). We further assessed the grouping effects for each of the two bias conditions. In the target 20% new condition, there was a small but reliable effect of Color group. RTs for probes falling in the old minority group were faster than RTs for probes falling in the old majority group, F(1, 17) = 6.73, p < .05. This effect also held for trials where the target was usually in the new search display (target 80% new), RTs, F(1, 17) = 26.15, p < .001. The interaction arose because there were increased effects of color grouping on old-item probe RTs when previews were mostly irrelevant for the task. This is consistent with participants employing an inhibitory bias against the properties of the largest group of old preview items, when the new search displays are prioritized. 5.2.3. Errors The overall level of errors was low at 4.6%. There were 5.3% errors on letter search trials and 4.0% on probe detec1500 a 20 18 16 14 12 10 8 6 4 2 0 Target 20% new Target 80% new Min Maj Condition b 20 18 16 14 12 10 8 6 4 2 0 Target 80% old Target 20% old Min Maj Error (%) Condition Fig. 9. Error rates (expressed as percentage) for new targets falling in the new minority/new majority color (a—top panel), and for probes falling in the old minority/old majority color groups (b—bottom panel) from both preview conditions of Experiment 2. Mena correct RT (ms) 1250 Min Maj 1000 750 tion trials. Errors on new target letter search trials from both preview conditions were analyzed like RTs, in a 2 · 2 (Condition · Color group) within-subjects ANOVA. The main effect of Condition and the Condition · Color group interaction were not significant (all Fs < 1, p > .650). Although there was a trend for there to be more errors for targets in the new minority color relative to targets in the new majority color, this main effect of Color group failed to reach significance, F(1, 17) = 2.13, p = .163. A similar analysis was carried out for old-item probe RTs. Again the main effect of Condition and the Condition · Color group interaction was not significant, (all Fs < 1, p > .430). Although there was a trend for errors to be higher for probes falling in the old majority relative to the old minority color group, this main effect of Color group failed to be significant, F(1, 17) = 3.09, p = .09. There was no evidence of a speed-accuracy trade-off (see Fig. 9). 6. Discussion 500 Target 80% old Target 20% old Condition Fig. 8. RTs (ms) for probes in Experiment 2. Performance is shown here as a function of the old group (old minority = light grey and old majority = dark grey). The results from Experiment 2 are clear. When observers were biased against the old irrelevant items, in anticipation of a target event occurring in the second display, a strong negative color carry-over emerged. RTs for targets J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1581 falling in the new minority group were significantly slower relative to targets falling in the new majority group. This replicates the findings from Experiment 1 here and elsewhere (Braithwaite et al., 2003, 2005). However, when observers were biased towards the old preview items (probe on old on 80% of the trials), the carry-over effect to new targets was greatly reduced. Importantly, here both conditions contained mixed-trial types (target letter search and probe detection) and so these differences cannot be explained by effects of task uncertainty or the number of letters that had to be discriminated. The reduction in the degree of carry-over to the new items, when the bias against the old items is removed, suggests strongly that the carry-over itself is based on a top-down intention to ignore and filter a mass of irrelevant items on the basis of their color as well as their locations. In addition to finding that the color carry-over effect could be modulated by the bias to new or old items, we also found that the effect of color grouping (on probe detection) was affected. Probes were more difficult to detect when they fell on items carrying the majority color in preview displays (see also Experiment 1). This effect was significant even when there was a bias to attend to the old items (probe detection on 80% of the trials), suggesting some automatic component to the effect. Nevertheless, the magnitude of this effect increased when participants were set to prioritise the new items (target search on 80% of the trials). This last result indicates that, even if an attentional bias is automatically established against items in the larger of two color groups (Bacon & Egeth, 1997; Egeth et al., 1984; Kaptein et al., 1995; Moore & Egeth, 1998), this bias is augmented when attention is directed away from the old groups to prioritise the new stimuli. We suggest that this reflects intentional top-down inhibition, applied differentially as a function of the size of the old, to-be-ignored groups (Braithwaite et al., 2005). 7. General discussion The present study examined whether top-down biases are necessary for the color-based carry-over to occur in preview search. Several critical new results have been revealed. First, the findings from Experiment 1 showed that search was always advantaged to some degree when the target was a new item (target letter search), compared to a baseline with all the search stimuli presented simultaneously (the full-set baseline). This occurred even when the old items should have been prioritized (target 80% old), compared with the full-set baseline. This suggests that performance in the preview condition is at least partly mediated by an attentional-capture component directed towards the new items (even when they rarely contain the target: Donk & Theeuwes, 2001). However, this benefit varied further as a function of top-down bias towards the new items and against the old preview display (see below). Therefore, onset-capture mechanisms alone cannot provide a complete and sufficient explanation of preview search. Second, we replicated the negative color carry-over effect on preview search (when new targets carry the color of the old items) and showed for the first time that the carry-over effect is abolished when participants prioritise attention to the old items. Our data suggest that the carry-over effect is dependent on the top-down intention to ignore and filter irrelevant old distractors. These results go against the idea that the carry-over may be based in the statistical properties of the displays (the presence of uneven color groups), that they reflect passive grouping processes (e.g., between old and new items carrying the same color), and that the preview-benefit itself reflects only onset-capture processes. Finally, both experiments here found effects of color grouping on probe detection: probes were more difficult to detect when they fell on old items carrying the majority color in the preview display. However, this attentional bias was reduced when participants were set to prioritise the old display. Thus, even if grouping takes place automatically, the bias against the old majority group is augmented under conditions where participants are set against the old items. This is consistent with there being top-down inhibition of the old items when participants prioritise search for new stimuli, with this inhibition being stronger within the larger group. Braithwaite et al. (2005) discussed similar results in terms of Duncan and Humphreys (1989) proposals concerning grouping and selection. Duncan and Humphreys argued that distractors could be rejected on the basis of spreading suppression within distractor groups, with the strength of the suppression increasing with the size of the group. Here such spreading suppression processes are maximized by an intentional set against old stimuli. 7.1. Effects from bottom-up attentional capture In Experiment 1, new-item targets were always advantaged relative to the full-set baseline. This was the case even when observers were set to search the old preview items most of the time. These findings suggest that search was being captured and guided to some degree by the new onsets (even when they rarely contained the targets). This is consistent with both the onset-capture and temporal-segmentation accounts of preview search performance (Donk & Theeuwes, 2003, 2001; Jiang et al., 2002). However, the degree of this advantage to search was not matched across preview conditions which differed only in terms of topdown bias. The advantage for new-item targets was greatly improved when the target was always a new-item (target 100% new). If search were just based on capture mechanisms alone, performance should have been matched for new-item targets in both preview search conditions, irrespective of the degree of top-down bias. It was not. This was also the case for Experiment 2. Experiment 2 used the same displays but under two different bias conditions (where either the old or the new displays were prioritised). Despite the displays being matched for onsets and for the time available for temporal segmentation, preview search 1582 J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 differed markedly across the bias conditions. This result provides strong evidence for top-down effects contributing in addition to any effects of automatic attentional capture by new onsets. 7.2. Effects from top-down biases When observers knew that the target would always be in the new set (target 100% new, in Experiment 1), search improved substantially relative to when the target occurred on a minority (20%) of trials. This pattern was replicated in Experiment 2, where search was faster when the target was expected in the new display (target 80% new) relative to when probe detection was expected (probe 80% old). Taken together, these findings suggest that, in addition to any effects from bottom-up capture, preview search performance is further aided by top-down biases of the observer directed towards the relevant (and against the irrelevant) displays. This is directly in line with the original suggestion of the visual-marking mechanisms proposed by Watson and Humphreys (1997) where the degree of capture enjoyed by new items is further aided by a negative bias against irrelevant items (see Watson et al., 2003; for a recent review). 7.3. Filtering items of mass distraction: Top-down biases & the negative color carry-over effect Further evidence for top-down inhibitory effects on the irrelevant preview comes from the negative carry-over effect. Search for new target letters search was affected by the color relations between the old and new items only when the new displays were actively prioritized. In these conditions, targets were harder to detect if they fell in the new minority group (carrying the color of the old majority group) relative to when they fell in the new majority group. Crucially, this result is in the opposite direction to the effect of color grouping on attention, when all the search items appear simultaneously—where there is usually an advantage for targets in the minority group (Bacon & Egeth, 1997; Egeth et al., 1984; Kaptein et al., 1995; Moore & Egeth, 1998).4 This reversal of the standard color grouping effect (with simultaneous items) is difficult to reconcile with effects due to automatic onset capture (Donk & Theeuwes, 2001, 2003) or to automatic temporal segregation (Jiang et al., 2002). If these other factors were crucial, then attention should have been captured by all new items equally, or directed towards the new minority group. Clearly, this was not the case. These findings are consistent, however, with there being inhibition of the old majority color carried by the preview Note that when the preview items were being prioritised (80% old condition) there was a non significant trend for new item targets to display the standard minority-group advantage. This further suggests that switching attentional biases from being against old information in the visual field, to being directed towards them, recruits separate processes. 4 items, which is transferred to the letters in the new search display carrying the same color. Particularly interesting is the fact that the carry-over effect was confined to when participants could actively prioritize the new items (in the target 100% new condition of Experiment 1 and the 80% new condition of Experiment 2). In contrast, when the old preview display was the relevant display (thus removing any negative bias from them) the carry-over effect was abolished. It follows that a top-down negative bias against the irrelevant old stimuli is a necessary condition for the carry-over to emerge. Note also that the effects of color on search of the new items here cannot be attributed to an active bias towards the new items, since then the standard effect of color-grouping should have occurred (i.e., search should have been facilitated for targets in the new minority color, rather than the new majority color as we observed). The opposite direction of the color carry-over effects and the color grouping effect is of interest in its own right, since the data indicate that these two effects may reflect different underlying mechanisms. This proposal is supported by recent results reported by Braithwaite et al. (2005). Braithwaite et al. varied the duration of the preview and found evidence for fast color-based grouping of stimuli in preview displays, present even with short preview durations. In contrast, negative color carry-over effects were found only with long exposures of previews. Both these results, and the present data, are consistent with there being distinct stages of preview processing. Initially there appears to be rapid grouping of items in the preview based on their color, but with the locations of the old items then being coded together. When search is directed at the preview, these groups bias attention to the locations of the smaller set. However, once the locations are grouped, then the identity of the initial defining features can change. Thus the grouping effect is little affected when the colors of the initial groups are altered (Braithwaite et al., 2005). Nevertheless the groups as initially coded modulate a later-acting inhibitory process, that is applied when participants actively prioritize the new items. This inhibition is applied to the features of the to-be-ignored items, and greater inhibition is applied to the larger distractors groups. It is this feature-based inhibition that leads to the negative carry-over effect in preview search. 8. Conclusion The present results provide strong evidence that the negative color carry-over effect in preview search is based on the active prioritization of attention to new items and away from old distractors. In particular, the negative color carryover effect was eliminated when participants prioritized old rather than new stimuli. In contrast, there was evidence for color-based grouping of preview displays, which could be augmented by later inhibition against old items. The results point to a critical role of active processes in visual search J.J. Braithwaite, G.W. Humphreys / Vision Research 47 (2007) 1570–1583 1583 over time, and to color grouping being distinct from colorbased suppression of stimuli. Acknowledgments This research was supported by a Roberts Research Fellowship (RCUK) awarded to the first author and an MRC grant to the second author. We thank Marisa Carrasco, Jeremy Wolfe and an anonymous reviewer for helpful comments on earlier versions of this paper. References Bacon, W. F., & Egeth, H. E. (1997). Goal-directed guidance of attention: Evidence from conjunctive visual search. Journal of Experimental Psychology: Human Perception and Performance, 23(4), 948–961. Braithwaite, J. J., & Humphreys, G. W. (2003). Inhibition and anticipation in visual search: Evidence from effects of color foreknowledge on preview search. Perception & Psychophysics, 65(2), 213–237. Braithwaite, J. J., Humphreys, G. W., & Hodsoll, J. (2003). Color grouping in space and time: Evidence from negative color-based carryover effects in preview search. Journal of Experimental Psychology: Human Perception and Performance, 29(4), 758–778. Braithwaite, J. J., Humphreys, G. W., & Hodsoll, J. (2004). Effects of colour on preview search: Anticipatory and inhibitory biases for colour. Spatial Vision, 17(4–5), 389–425. Braithwaite, J. J., Humphreys, G. W., & Hullleman, J. (2005). Color-based grouping and inhibition in visual search: Evidence from a probe detection analysis of preview search. Perception & Psychophysics, 67(1), 81–101. Donk, M., & Theeuwes, J. (2001). Visual marking beside the mark: Prioritizing selection by abrupt onsets. Perception & Psychophysics, 93, 891–900. Donk, M., & Theeuwes, J. (2003). Prioritizing selection of new elements: Bottom-up versus top-down control. Perception & Psychophysics, 65(8), 1231–1242. Duncan, J., & Humphreys, G. W. (1989). Visual search and stimulus similarity. Psychological Review, 96, 433–458. Egeth, H. E., Virzi, R. A., & Garbart, H. (1984). Searching for conjunctively defined targets. Journal of Experimental Psychology: Human Perception and Performance, 10(1), 32–39. Humphreys, G. W., Jung-Stalmann, B., & Olivers, C. N. L. (2004). An analysis of the time course of visual marking using a probe-dot procedure. Perception & Psychophysics, 66, 713–730. Humphreys, G.W., Olivers, C.N.L., & Braithwaite, J.J. (in press). The time course of preview search with isoluminant shapes. Perception & Psychophysics. Jiang, Y., Chun, M. M., & Marks, L. E. (2002). Visual marking: Selective attention to asynchronous temporal groups. Journal of Experimental Psychology: Human Perception and Performance, 28, 717–730. Kaptein, N. A., Theeuwes, J., & van der Heijiden, A. H. C. (1995). Search for a conjunctively defined target can be selectively limited to a color defined subset of elements. Journal of Experimental Psychology: Human perception and Performance, 21, 1053–1069. Moore, C. M., & Egeth, H. (1998). How does feature-based attention affect visual processing? Journal of Experimental Psychology: Human Perception and Performance, 24(4), 1296–1310. Olivers, C. N. L., & Humphreys, G. W. (2003). Visual marking and singleton capture: Fractionating the unitary nature of visual selection. Cognitive Psychology, 47, 1–42. Olivers, C. N. L., & Humphreys, G. W. (2002). When visual marking meets the attentional blink: More evidence for top-down, limited capacity inhibition. Journal of Experimental Psychology: Human Perception and Performance, 28(1), 22–42. Olivers, C. N. L., Watson, D. G., & Humphreys, G. W. (1999). Visual marking of locations and feature maps: Evidence from withindimension defined conjunctions. The Quarterly Journal of Experimental Psychology, 52A(3), 679–715. Treisman, A., & Sato, S. (1990). Conjunction search revisited. Journal of Experimental Psychology: Human Perception and Performance, 16, 459–478. Watson, D. G., & Humphreys, G. W. (2002). Visual marking and visual change. Journal of Experimental Psychology: Human Perception and Performance, 28(2), 379–395. Watson, D. G., & Humphreys, G. W. (2000). Visual marking: Evidence for inhibition using a probe-dot detection paradigm. Perception & Psychophysics, 62, 471–481. Watson, D. G., & Humphreys, G. W. (1997). Visual marking: Prioritizing selection for new objects by top-down attentional inhibition of old objects. Psychological Review, 104(1), 90–122. Watson, D. G., Humphreys, G. W., & Olivers, C. N. L. (2003). Visual marking: Using time in visual selection. Trends in Cognitive Sciences, 7, 180–186.