Featural Guidance in Conjunction Search: The Contrast Between Orientation and Color by Giles Anderson | Papers by Giles

FEATURAL GUIDANCE IN CONJUNCTION SEARCH Featural Guidance in Conjunction Search: The Contrast Between Orientation and Color Giles M. Anderson, Dietmar Heinke, Glyn W. Humphreys University of Birmingham, UK Email: gma472@bham.ac.uk Address: School of Psychology University of Birmingham Edgbaston Birmingham B15 2TT 1 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Abstract Four experiments examined the effects of pre-cues on visual search for targets defined by a color-orientation conjunction. Experiment 1 showed that cueing the identity of targets enhanced the efficiency of search. Cueing effects were stronger with color than with orientation cues, but this advantage was additive across array size. Experiment 2 demonstrated that cueing effects interacted with bottom-up segmentation processes while Experiment 3 showed the stronger effects of color cues remained in a compound task. Experiment 4 confirmed the enhanced effect of color cueing even when verbal rather than visual cues were used. The targets used were balanced for search efficiency within both orientation and color dimensions. We suggest search benefits from the top-down cueing of color compared with orientation, because color cueing enhances the segmentation of displays into color groups more efficiently. This enables search to an appropriate color group to be initiated earlier. We discuss how top-down segmentation processes interact with differences in bottom-up segmentation to further improve target detection. 2 FEATURAL GUIDANCE IN CONJUNCTION SEARCH In everyday life we often have to search scenes to find targets on the basis of some known feature – for example their color or their shape (as when we look for a friend in a crowd). In a laboratory setting the processes underlying these behaviors have been studied through visual search tasks. Typically a search target may appear at a random location amidst varying numbers of distractors, and the time taken to decide that the target is present or absent is measured. Over almost half a century or so, research using visual search has unearthed a plethora of experimental evidence on the impact of physical properties of items (bottom-up factors) on the efficiency with which targets are selected (see Wolfe, 1998, and Krummenacher & Müller, 2006; for reviews). More recently, however, experimenters have also examined how knowledge about which items are being searched for influences the search process. For example, search for targets can be influenced by the action that participants perform (Bekkering & Neggers, 2002) or by defining the target in terms of its action rather than its visual properties (Forti & Humphreys, 2008; Humphreys & Riddoch, 2001). Even when targets are defined by their visual properties, differential effects may arise according to whether different features are expected – for example, as we elaborate below, an expectation for a target defined by its color may have a different effect than an expectation for a target defined by its orientation. Understanding how these different expectations influence search, and separating their effects from those of bottom-up factors, is important for developing complete accounts of the search process. There is a substantial literature demonstrating that search for a target defined by its color can be highly efficient, and elements defined by their color may be searched preferentially in relation to stimuli defined along other dimensions. For example, Williams (1966) investigated eye fixations involved in a search task where stimuli were defined by a conjunction of size, color or shape. He found that, even 3 FEATURAL GUIDANCE IN CONJUNCTION SEARCH when prompted to find a particular colored shape, participants still fixated more readily on stimuli defined by color than those defined by size or shape. More recently, moreover, Hannus, van den Berg, Bekkering, Roerdink, and Cornelissen (2006) used a pre-cue to indicate which stimulus in the following feature-defined or colororientation conjunction search participants should search for and fixate. The discriminability for both dimensions had been balanced so that in a feature-defined search there was no difference in the probability of a saccade landing on a stimulus with the cued color or orientation. However, in the conjunction search there was a large decrease in the likelihood of the fixation landing on a stimulus with the cued orientation, while there was no change for the color dimension between the two search types. Similarly in color and orientation conjunction searches, both Zohary and Hochstein (1988) and Poisson and Wilkinson (1992) found that stimuli defined by color were selected preferentially to those defined by orientation. By manipulating distractor ratio, both studies found that participants searched stimuli sharing the color of the target where a search based upon the target’s orientation would be a better strategy, although this preference was only evident in the target-absent data of Poisson and Wilkinson. The role of top-down and bottom-up contributions to this search advantage have not been clearly separated however. For example, targets defined by color differences relative to distractors may be found efficiently because there is rapid bottom-up organisation of elements into color-based groups (see Braithwaite, Humphreys, & Hulleman, 2005), with color differences between the groups serving to guide search to a target. Evidence consistent with this proposal comes from studies of so-called ‘subset search’. In subset search, participants may be presented with displays where, over trials, different numbers of distractors carry a 4 FEATURAL GUIDANCE IN CONJUNCTION SEARCH shared feature. Search is typically more efficient when the target falls within the smaller of two groups of distractors than when it falls within the larger of the groups, and this effect is pronounced when the subsets are formed through color-grouping (Egeth, Vrizi, & Garbart, 1984; Kaptein, Theeuwes, & Heijden, 1995; Bacon & Egeth, 1997). On the other hand, Sobel and Cave (2002) found no preference for searching a color-defined subset over one defined by the same orientation when both dimensions were highly discriminable, suggesting that search may be determined less by the dimension that a target is defined along than the discriminability of target and distractor elements within particular dimensions. Differences in discriminability within a dimension may also contribute to other findings where color appears dominant in the bottom-up capture of attention – one example being when attention is captured by a distractor that is a singleton along a dimension that is irrelevant to the search task. Attentional capture is typically stronger for singletons defined along the color dimension than for singletons defined by their orientation or shape (Theeuwes, 1992), but this may reflect variations in the feature values within the dimensions. Other evidence points to color having a strong effect on top-down guidance of search. One procedure used to examine this has involved giving participants one item to hold in working memory while they search for a different target. Search for the target is affected by the appearance of the working memory stimulus in the search display (see Downing, 2000; Olivers, Meijer, & Theeuwes, 2006; Soto, Heinke, Humphreys, & Blanco, 2005; Soto, Humphreys & Heinke, 2006), whereas there can be little effect of an initial cue that is not held in working memory (Soto et al., 2005, 2006). That is, the effect is due to top-down biases modulated through working memory rather than bottom-up priming from the initial appearance of a working memory cue. Soto et al. (2005) varied whether the color or shape/orientation of the 5 FEATURAL GUIDANCE IN CONJUNCTION SEARCH item in working memory matched one of the search stimuli and found stronger effects from color matching than from shape/orientation matching. Using a different procedure, Müller, Reimann, and Krummenacher (2003) cued participants on a trial-by-trial basis as to whether the target in a forthcoming display differed from distractors along a particular dimension (color or orientation), or particular values along each dimension (red vs. horizontal). Valid pre-cueing of the dimension that defined the target facilitated search, relative to when the dimensional cue was invalid. In addition to this, though, cueing a particular color value also influenced search, while, in contrast, there were only modest effects of cueing a particular orientation value. Müller et al. proposed that, within the color dimension, attention could be set for a particular feature value as well as for the dimension itself, while a top-down expectancy in the orientation domain could only be set for the dimension. On the other hand, Hodsoll and Humphreys (2005) provided evidence for top-down expectancies being set for targets whose orientations differed categorically from those of distractors. Search is more efficient when targets have a categorical difference in their orientation relative to distractors (e.g., when the target is steep and the distractors shallow), compared to when targets and distractors are categorically similar (e.g., both are shallow), even when the absolute difference in orientation between targets and distractors is matched in the categorical and non-categorical search tasks (Wolfe, Stewart, Friedman-Hill, & O’Connell, 1992). Hodsoll and Humphreys showed that this advantage for categorically-defined targets arose primarily when the targets were expected, with the effect being greatly diminished when participants searched in a pure bottom-up manner for the ‘odd one out’. This suggests that expectations can be set for at least coarse categories of orientation. 6 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Unfortunately, studies of top-down effects have rarely equated the stimuli for the saliency of targets along each dimension. As a consequence, any dominant effects of color could stem from the greater discriminability of targets along that dimension in a given experiment, when compared with targets defined along other dimensions. Effects may disappear once targets defined along the different dimensions are defined as being equally discriminable (cf. Sobel & Cave, 2002). In the present study, we set out to assess whether there is differential top-down guidance of attention from color relative to orientation cues when any bottom-up differences in saliency were eradicated. To our knowledge, no previous studies have looked at featural cueing in conjunction search. The distractors of the search displays were the same across all the experiments, comprising blue vertical and green horizontal bars. The targets were defined in terms of the conjunction of features not present in each distractor - green vertical or blue horizontal bars. Participants were cued on a trial-by-trial basis as to either the color or the orientation of the forthcoming target. To rule out bottom-up biases favoring color over orientation-defined targets, the saturation of the two colors was varied in an initial series of studies until there was no difference in search efficiency for a color-defined over an orientation-defined target1 (details in the Appendix). Subsequently, we examined whether cueing a color or orientation feature value for the target had any differential influence on selection (Experiment 1). In Experiment 2, we pitted the effects of the top-down cues for the target against bottom-up guidance produced by varying the numbers of each type of distractor to create different subsets for search. Experiment 3 tested for effects of The only difference between the color and orientation targets occurred with the items used in Experiment 3, where there was an overall RT advantage for orientation-defined targets (though there was no difference in search slopes). The items in Experiment 3 were designed for a compound search task and it is possible that participants weighted orientation information more strongly than color in this circumstance. The important point, however, is that this search advantage is in the opposite direction to the cueing effects we found, indicating that greater cueing effects cannot reflect easier bottom-up selection of particular targets. 1 7 FEATURAL GUIDANCE IN CONJUNCTION SEARCH cueing at a stage of response selection and Experiment 4 examined the generality of the results by contrasting verbal with visual cues. To foreshadow the results, there were differential effects of cueing attention in a top-down manner, with color cueing being more effective than orientation cueing. Top-down cueing also interacted with apparent bottom-up effects reflecting the ratio of distractors with particular attributes (color or orientation), with the costs from incorrect cueing being reduced if participants then had to shift attention to a small group containing the target. However, these effects were on the intercepts of the search functions rather than the slopes. Such intercept effects could reflect either a change in an initial stage of processing (e.g., parsing a display into two sets of items to be searched), prior to serial search, or they could reflect differences in the ease of assigning a response to a target, after the target has been selected (Cohen & Shoup, 1997, 2000; Cohen & Magen, 1999; Cohen & Feintuch, 2002; Theeuwes, Reimann, & Mortier, 2006). In either case, an overall difference in RT could arise, while the search process itself may be unaffected. To separate these two accounts, in Experiment 3 we used a compound search task, where the response required to the target was not linked to the cue. While cueing effects on initial processing should remain in a compound search task, differential effects on response selection may be eliminated. Finally, in Experiment 4, we presented a verbal cue rather than a visual representation of color or orientation. When the visual representation of the cue matches that of the following target, cueing effects could be due to priming (e.g., Kristjánsson, 2006). Using verbal cues (cf. Müller et al., 2003) allowed us to eliminate the possibility that bottom-up priming generated any color advantage, rather than top-down cueing. 8 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Experiment 1: Differential Top-down Cueing of Color and Orientation In Experiment 1 we used the conjunction search task from the pilot study (Experiment A, see Appendix for details), but preceded each display by a visual cue, either a color circle (indicating the likely color of the target), a white oriented line (indicating the likely orientation of the target) or a white circle (the neutral condition). The color and orientation cues were valid on 80% of the trials, to encourage participants to use the information provided. We assessed whether cueing participants about the feature value defining the target would facilitate search, and whether effects from cueing color would be the same as those from cueing orientation. Method Participants. Seventeen University of Birmingham students, six male, 11 female, aged between 18 and 24 (average age of 20.18) took part. All had selfreported normal or corrected-to-normal vision; and had normal color perception assessed using Ishihara's Tests for Color-Blindness (Ishihara, 1981). All participants were naïve as to the purpose of the experiment. Apparatus. Stimuli were presented at 1024 × 768 pixel resolution on a 17′′ color Samsung SyncMaster 793s monitor, driven by an Intel Pentium 4 PC with a Radeon 9000 AGP Pro video card. They were generated by an E-Prime programme (Schneider, Eschman & Zuccolotto, 2002) that recorded RTs and accuracy via a standard UK keyboard. Audio feedback was provided by stereo Cambridge Soundworks speakers. Stimulus luminance was independently measured using a Salford Electrical Instruments Exposure Photometer. Participants sat approximately 0.6m from the screen in a well-lit room. 9 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Stimuli. Prior to each trial, a white fixation cross was presented, 1mm thick (visual angle of 0.1º), 5mm wide (0.48º) by 5mm tall (0.48º). All the stimuli were presented on a black background and comprised either blue vertical, blue horizontal, green vertical or green horizontal bars. The dimensions of the bars were 6.5mm (visual angle of 0.62º) long by 1.5mm (0.14º) wide. The pre-cue2 was one of five stimuli: a blue patch, a green patch, a white horizontal bar, a white vertical bar and a white patch. Orientation cues were either horizontal or vertical white bars whose physical dimensions matched those of the stimuli used in the pilot study. Colored and neutral patch cues were filled circles, all with the diameter of 3.5mm (0.35º). The same color levels were used for cues and search stimuli and are outlined in Table 1. Color Blue Green White Black Light grey Hue 140 80 160 160 160 Saturation 40 40 0 0 0 Luminance 120 120 240 0 200 Photometer reading (cdm-2) 27.25 27.25 121.7 4.32 60.99 Table 1. Color levels used for the stimuli all experiments. Hue, saturation and luminance levels were adjusted using the Microsoft Paint computer programme. Photometer readings were taken using the Salford Electrical Instruments Exposure Photometer. Design. There were four main independent variables: cue validity (valid, neutral, invalid), cued dimension (color, orientation), array size (7, 15) and target type (green vertical, blue horizontal, absent). The main dependent variable was the reaction 2 We use this term to indicate that a cue was presented prior to each trial. 10 FEATURAL GUIDANCE IN CONJUNCTION SEARCH time (RT) taken to indicate the presence or absence of the target. Response accuracy was also measured. (a) Color cues Neutral cue (b) Orientation cues 1200ms 100ms Until response Cue ISI Search displays with possible targets Figure 1. Simplified timeline of a trial from Experiment 1. Only one cueing stimulus was presented prior to each search array, with the color and orientation cues predicting (80%) the feature of the following target. The two possible targets are shown (circled). The array size was either seven (Fig. 1a) or 15 (Fig. 1b). The stimuli were presented on a black background but for simplicity the background is shown here as white; blue stimuli are represented as black, green as grey and white as outlines. Procedure. As well as information about the nature of the targets, the instructions indicated the physical properties and predictive nature of the cue stimulus that was to be presented prior to each search. The cue was presented after the fixation cross for 1200ms and was followed by an ISI of 100ms and arrays of either seven and 15 stimuli (see Figure 1). The search stimuli were presented randomly within a rough, 11 FEATURAL GUIDANCE IN CONJUNCTION SEARCH invisible circle of diameter 5.5cm (2.62º) with 21 possible positions. The circle was positioned in the middle of the screen and stimuli positions were jittered (+/- 0.1º) vertically and horizontal to lessen spatial interactions between distractors. Due to high error rates during piloting, participants were informed about the nature of the targets prior to the experiment, with graphical representations of the target images presented adjacent to the monitor during the task. The target varied across trials (50% green vertical, 50% blue horizontal), with half the distractors being blue vertical bars and the other half being green horizontal bars. When the target was absent the distractor ratio remained the same but a randomly chosen distractor replaced the target. Participants were asked to indicate whether the search target was present or absent by pressing either ‘Z’ or ‘M’ on the computer keyboard (the key assignment was reversed for half the participants). Targets were present for 62.5% of the trials (the bias was introduced as only target-present trials were of interest; target-absent trials just functioned to regulate the rate of anticipation responses, cf. Müller et al., 2003). Feedback was provided. If the response was correct, participants heard a medium pitched sound and the word ‘Correct’ was displayed. If incorrect, a lower note was played and the word ‘Incorrect’ was displayed instead. The time until participants’ response was recorded (RTs), with the accuracy of the response also noted. On one third of the target-present trials the cue predicted the color of the target; on another third the cue predicted the target’s orientation, while the cue was non-informative (neutral) for the final third of the trials. A green or blue patch was used to predict the target’s color, a white vertical or horizontal line predicted the target’s orientation, while a white patch offered no prediction. The color and orientation cues were valid on 80% of the trials and invalid on 20% (see Müller et al., 2003). Participants were informed that the majority of the cues were valid. 12 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Participants took part in two one-hour sessions, each a minimum of 24 hours apart and consisting of four 96-trial blocks so that each participant undertook 768 experimental trials. All conditions were randomly interleaved in each identical block. At the beginning of the first session there was a block of 20 practice trials where data were not recorded, while at the beginning of second session there was a block of four practice trials. Results Several participants required breaks in the middle of the block of trials. For all analyses, these trials were discarded (RTs>5000ms). Less than 1% of data were removed as a consequence. All ANOVAs report the partial eta-squared (partial η2) statistic which describes the proportion of total variability attributable to the particular factor (see Olejnik & Algina, 2003). To control the family-wise error rate, the alpha level was adjusted when multiple ANOVAs were undertaken on the same data and Huynh-Feldt adjustments were used on probabilities where necessary. All pair-wise comparisons included Bonferroni adjustments and were measured as significant at the p<0.05 level. The data were pooled across the two orientation-cued conditions (horizontal and vertical) and, separately, across the two color-cued conditions (blue and green), as well as across target type in each case (blue horizontal, green vertical). As targetabsent trials were counted as catch trials, only target-present trials were analysed. We first report the basic RT analyses for color- and orientation-cued search, comparing valid, invalid and neutral trials in each case. We subsequently compare color and orientation cueing effects in cost-benefit analyses relative to the neutral condition. This is because the neutral condition served for both color and orientation cue, and so 13 FEATURAL GUIDANCE IN CONJUNCTION SEARCH could not enter into a full factorial design comparing across the cue types. This procedure was followed throughout the paper. RTs. Median RTs for each participant and each condition were calculated and the means across participants are shown in Figure 2. For the RT analysis, trials cued by target color and those cued by target orientation were analysed separately because the neutral condition could not be divided by dimension. Color-cued trials. A two-factor ANOVA (array size, cue validity) revealed significant main effects of array size (F(1,16)=75.355, p<0.001, partial η2=0.825) and cue validity (F(2,32)=44.173, p<0.001, partial η2=0.734) and a significant interaction (F(2,32)=5.361, p=0.046, partial η2=0.251). A priori pair-wise comparisons revealed significant effects of array size for all validity conditions (valid, neutral and invalid) with searches of 7-item trials shorter than those with 15-item (p<0.001 for all comparisons). A priori comparisons found that at both array size seven and 15, valid cues facilitated search compared to neutral cues (all ps<0.001), with RTs following invalid cues significantly longer than RTs in the other conditions (array size 7, p=0.032; array size 15, p<0.001). 14 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 2. Mean (+/- one standard error) of median RTs in Experiment 1 defined by the cue validity, cued dimension and array size. Orientation-cued trials. RT data from orientation-cued trials were analysed in a similar manner. The interaction failed to reach significance (F<1) but there was a main effect of array size (F(1,16)=89.168, p<0.001, partial η2=0.848). RTs with 7item arrays were shorter than those with 15-item trials. There was also a reliable effect of validity (F(2,32)=16.523, p<0.001, partial η2=0.508). A priori pair-wise comparisons found that valid cues facilitated search compared to neutral cues (p=0.001), while invalid cues slowed search compared to neutral trials (p=0.044). RT Cost-Benefit. The RT analysis showed that cue validity affected search, with RTs increasing from valid to neutral to invalid conditions. To assess the relative magnitudes of these cueing effects, median RTs in the valid and invalid conditions were subtracted from those for neutral trials to create RT Benefits (on valid trials) and Costs (on invalid trials). Mean values are shown in Figure 3. 15 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 3. Means (+/- one standard error) of RT Benefit/RT Cost in Experiment 1, defined by cue dimension, cue validity and array size. A positive value indicates a benefit relative to the neutral condition while a negative value indicates a cost. A three-factor ANOVA (array size, cue dimension and cue validity) revealed a reliable main effect of cue validity (F(1,16)=32.317, p<0.001, partial η2=0.669) and a significant interaction between cue dimension and cue validity (F(1,16)=22.069, p<0.001, partial η2=0.580). There was also a borderline significance array size x cue validity interaction (F(1,16)=5.130, p=0.076, partial η2=0.243). However, the threeway interaction between array size, cue validity and cue dimension was not significant (F(1,16)=1.45, p=0.492, partial η2=0.083). The array x validity interaction arose because the validity effects were larger at display size 15 than display size 7 (520 vs. 382ms, both ps<0.001). The cue dimension x cue validity interaction emerged because the cueing effect was larger for color cues (a validity effect of 548ms, p<0.001) than for orientation cues (validity effect of 367ms, p<0.001). This 16 FEATURAL GUIDANCE IN CONJUNCTION SEARCH was reflected in both increased benefits from color cues on valid trials (the validity benefit was enhanced by 104ms) and increased costs on invalid trials (the costs increased by 76ms). Accuracy. Accuracy effects followed RTs showing no speed-accuracy tradeoff. See Table 2 for details. Discussion The results demonstrate that valid cues facilitate search while invalid cues slow search. There was also a differential effect of the dimension that was cued – effects were larger following color relative to orientation cues. It is interesting that this differential effect of cueing arose even though it was possible for participants to generate an expectancy of the other property of the target from the cue (e.g., given the cue ‘blue’ then the target’s horizontal orientation could be predicted). However, if participants had generated an expectancy for both attributes then there should have been minimal differences between orientation and color cues. The results suggest otherwise. We presume that the cognitive load involved in generating an expectancy of the other dimension from the cue mitigated against participants adopting this strategy. Alternately, participants could prioritise the cued feature (e.g., blue) above the other property of the target (e.g., horizontal), even though cues to guide attention were equally salient in the search displays. Attention would then be directed to stimuli matching the cue in the cued dimension before search would continue within this group for a disparity within the alternate dimension (Friedman-Hill & Wolfe, 1995). Although both color and orientation cues improved search rates, and although color cues had a stronger overall effect on search, the improvements on search efficiency were equal for the two cue types. Thus the differential effect of color 17 FEATURAL GUIDANCE IN CONJUNCTION SEARCH cueing was on the intercepts rather than the slopes of the search functions. The effects of the cues generally on search efficiency can be explained if the cues bias participants to initiate search with a particular sub-group of distractors, while in neutral trials search operates randomly across all of the items present. Since the target will fall within the cued sub-group on valid trials, fewer items will on average need to be searched. In contrast, on invalid trials the target will fall within the sub-group searched second, slowing target detection relative to when a random search strategy is adopted (when the target will be detected on average after searching half of the items present). Since the type of cue affected the intercepts rather than the slope of the functions, it appears that the likelihood of search being biased to a given sub-group, following a cue, was equal for color and orientation cues. The differential effect of cue-type must thus come about for a different reason. We suggest at least two possibilities. One is that the two cue types have a differential impact on the time taken to segment the search display into color or orientation-coded sub-groups. If color cues exert a stronger influence on the initial segmentation process, then the time to start search of the cued sub-group will be faster following color relative to orientation cues. However, while this should lead to faster overall RTs, it is not clear why costs should be greater after color compared with orientation cues unless some other factor is involved. For example, it might be that the faster the assignment of attention to one sub-group, the slower the re-assignment of attention to the second sub-group when the cue is invalid – a type of ‘first-in, last-out’ process. A second possibility is that color and orientation cues have differential effects on a late stage of response selection, so the target-present response is assigned more rapidly to a color cued target compared to an orientation-cued target. For example, a decision criterion might be pushed more towards the expected target value after a color cue. This would speed responses when 18 FEATURAL GUIDANCE IN CONJUNCTION SEARCH the target matches the expectation, but it would also slow responses on trials when the target has the other color value (when the color cue is invalid) – on invalid trials, the decision criterion would be further away from the actual value of the target. These two possibilities were explored in Experiments 2 and 3 here. Experiment 2 contrasted the effects of directing attention from a cue with the effects of bottom-up segmentation, introduced by including in displays subsets of distractors with particular feature values. Search may be directed in a bottom-up manner to a subset of distractors having a minority feature in the display (cf. Bacon & Egeth, 1997). How does cueing the dimension and feature value of the target interact with this bottom-up directing of attention? For a late, response-selection account of the cueing effects, any effects due to bottom-up directing of attention to a subset of items should be additive with the cueing effect, since the cueing effect emerges only at a response assignment stage. On the other hand, if the cue itself affects display segmentation and directs attention to a relevant subset of items, then it may interact with bottom-up factors that generate subset search. For example, the normal direction of attention to a minority group of distractors (cued bottom-up) may be overruled when the cue directs attention to another sub-group of items. Experiment 3 introduced a compound search task, so response assignment could be divorced from attentional selection of the target. Effects of the cue on response selection should be eliminated when a compound search task is carried out. 19 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Experiment A Array size Search task Color-defined Orientation-defined Conjunction, target present Conjunction, target absent 7 97 97 92 95 11 96 98 92 96 15 98 97 89 96 7 Experiment B Array size 11 95 95 94 15 93 97 95 Experiment C Array size 7 96 96 96 11 96 97 95 15 95 96 97 95 97 94 Exp. 1 Array size Trial condition Neutral Valid color visual cue Invalid color visual cue Valid orientation visual cue Invalid orientation visual cue Valid color verbal cue Invalid color verbal cue Valid orientation verbal cue Invalid orientation verbal cue 7 94 95 88 97 90 15 93 94 85 94 87 Exp. 2 Distractor ratio 1 93 96 91 98 89 2 94 96 84 95 89 3 98 98 82 95 91 Exp. 3 Array size 7 95 96 91 96 88 15 97 96 93 97 89 Exp. 4 Array size 15 96 97 96 97 97 97 95 96 96 Table 2. The mean percentage accuracy across all experiments. 20 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Experiment 2: Cueing and Bottom-up Subset Search In Experiment 2 we varied the ratio of the different types of distractor in the conjunction search task, creating conditions that should induce subset search (e.g., Bacon & Egeth, 1997; Sobel & Cave, 2002). Using the same stimuli as Experiment 1 we introduced three ratio conditions, two with unequal numbers of each type of distractor and one where the ratios were balanced (as in Experiment 1). In the neutral condition, we expect to find subset search effects, with RTs being reduced when the target is in the minority set of distractors. (i.e., there should be a smaller-group bias, Sobel & Cave, 2002). In the cue conditions we assessed whether this bottom-bias to the smaller distractor group was modulated by the cue. Method The methodology of Experiment 2 was the same as that of Experiment 1 unless indicated below. Participants. There were twenty University of Birmingham students, 7 male, 13 female, aged between 18 and 26 (average 20.58). Two participants were removed due to an overall accuracy level of less than 90%. Design. There were four main independent variables: cue validity (valid, neutral, invalid), cue dimension (color, orientation), distractor ratio (see Procedure section) and target type (green vertical, blue horizontal, absent). Procedure. Participants took part in two one-hour sessions, each consisting of three 144-trial blocks so that each participant undertook 864 trials, with all conditions randomly interleaved in each block. At the beginning of the first session there were 20 practice trials where RTs and accuracy was not recorded, while at the beginning of the second session there was a block of four practice trials. 21 FEATURAL GUIDANCE IN CONJUNCTION SEARCH The procedure replicated that of Experiment 1 apart from two alterations: the array size was always 15 and the ratio of distractors was manipulated as follows: three blue vertical bars (BV) and 11 green horizontal bars (GH; see Figure 4i); 7GV and 7GH (Figure 4ii, ratio matches that in array size 15 for Experiment 1); and 11BV, 3GH (Figure 4iii). (i) (ii) (iii) Figure 4. Examples of the three conditions in Experiment 2. Stimuli were presented on a black background, but for clarity it is shown here as white; green stimuli are presented as grey, blue as black. The examples only have green vertical targets; however it was equally likely for the target to be a blue horizontal bar. Results The data were cleaned as in Experiment 2. Target-absent trials were considered catch trials so were not analysed. Therefore, median RTs from targetpresent trials in each condition for each participant were calculated as Experiment 1, although instead of array size we separated data by distractor ratio. RTs: Neutral trials. To assess if there was a bottom-up effect reflecting subset search, performance was examined in the neutral condition. Mean RTs across 22 FEATURAL GUIDANCE IN CONJUNCTION SEARCH participants are shown in Figure 5. A two-factor ANOVA was conducted with the factors being target type (blue horizontal vs. green vertical) and distractor ratio ((i) 3BV, 11GH; (ii) 7 BV, 7GH; (iii) 11BV, 3GH. See Figure 4). There was a main effect of target type (F(1,17)=14.441, p=0.002, partial η2=0.459), with longer RTs for the blue horizontal target. There was also a reliable effect of distractor ratio (F(2,34)=6.588, p=0.008, partial η2=0.279). The ratio analysis showed the U-shape typical of a ratio experiment, with RTs to displays with unequal ratios (i) and (iii) being shorter than those to displays with equal numbers of each type of distractor (ii) (p<0.001, p=0.046 respectively). There was also an interaction (F(2,34)=5.487, p=0.018, partial η2=0.244). The difference between RTs to the two targets were reduced for display type (iii) (a difference of 157ms, p=0.14) compared to that in both display type (i) (a difference of 320ms, p<0.001) and display type (ii) (a difference of 330ms, p=0.002). We also investigated the effect on behaviour of smaller subsets defined by color or orientation compared with when these factors were balanced. To analyse the effect of a smaller color subset, we calculated the difference in RTs between displays (i) and (ii) for trials with a blue vertical target and between (iii) and (ii) for trials containing a green horizontal target and vice versa for the orientation-defined subset. A two-factor ANOVA (subset dimension, target type) showed that there were borderline main effects of subset dimension (F(1,17)=5.793, p=0.066, partial η2=0.24) and target type (F(1,17)=5.793, p=0.056, partial η2=0.254), with the ‘subset effect’ larger for orientation compared to color (119ms vs. 37ms). 23 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 5. Mean (+/- one standard error) of median RTs from the neutral condition in Experiment 2, separated by distractor ratio (BV = blue vertical stimulus, GH = green horizontal stimulus) and target (BH = blue horizontal stimulus, GV = green vertical stimulus). These results indicate that there were bottom-up effects of subset search, with search being faster when there were unequal ratios of distractors than when the ratios were equal. Interestingly this bottom-up effect seemed to be strongest within the orientation rather than the color dimension. Note that search for the green vertical target was fastest when the smaller group contained 3BV distractors and search for the blue horizontal target was fastest when the smaller group contained 3GH distractors. RTs: Cued trials. For the cued data, we re-categorised the distractor ratio depending on the cue. Trials where the cued feature matched that defining the smaller distractor subset were coded as the Small cued group, while trials with the cue feature matching that shared by the larger distractor subset were included in the Large cued group condition (see Table 3). The coding for trials with 50:50 ratios was unchanged, 24 FEATURAL GUIDANCE IN CONJUNCTION SEARCH but they were labelled as the Equal cued condition. As in Experiment 2, the effects of color- and orientation-cues were analysed separately as the neutral condition served both dimensions, with the data being averaged across target color due to low numbers of trials in the invalid condition. Display type Target BH GV BH GV BH GV BH GV BH GV BH GV Cue Valid blue Valid green Invalid green Invalid blue Valid horizontal Valid vertical Invalid vertical Invalid horizontal Neutral I Neutral I Neutral II Neutral II Small group cued (i) 3BV, 11GH* (iii) 11BV, 3GH* (iii) 11BV, 3GH+ (i) 3BV, 11GH+ (iii) 11BV, 3GH+ (i) 3BV, 11GH+ (i) 3BV, 11GH* (iii) 11BV, 3GH* (i) 3BV, 11GH* (iii) 11BV, 3GH* (iii) 11BV, 3GH+ (i) 3BV, 11GH+ Equal group cued (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH (ii) 7BV, 7GH Large group cued (iii) 11BV, 3GH* (i) 3BV, 11GH* (i) 3BV, 11GH+ (iii) 11BV, 3GH+ (i) 3BV, 11GH+ (iii) 11BV, 3GH+ (iii) 11BV, 3GH* (i) 3BV, 11GH* (iii) 11BV, 3GH* (i) 3BV, 11GH* (i) 3BV, 11GH+ (iii) 11BV, 3GH+ Table 3. An outline of how the original displays (i), (ii) and (iii) (see Figure 4) were re-coded into display condition in Experiment 3. Displays were categorised depending on the cue, so that the Small group cued indicated a display where the smaller group of stimuli shared the cued feature; the Large group cued indicated that the larger group of stimuli shared the cued feature. The target and array stimuli in each condition are indicated as follows: BH = Blue Horizontal, GV = Green Vertical, BV = Blue Vertical, GV = Green Vertical. Symbols + and * indicate comparable cued and neutral displays; no such coding was necessary when the Equal-group of distractors was cued. 25 FEATURAL GUIDANCE IN CONJUNCTION SEARCH RTs: Color-cued trials. The original distractor ratios for valid and invalid color cues differed in both the Small and Large cued groups. For example, valid color cueing of the Small group occurred when the color of the smaller group of distractors matched with the target, while invalid color cueing of the Small group occurred when the color of the smaller subset was not the same as the target. To match performance with these conditions, two neutral conditions (Neutral I and II) were created from the same data, but with the extreme ratios switched over to match the cued ratios. As can be seen from Table 3, the ratios for valid color cueing matched that of Neutral I, while for invalid cueing they matched Neutral II. So for each display condition we compared the valid cueing RTs with one neutral condition and the invalid cueing with another. As both the Small and Large cued group conditions used the same neutral data, we analysed the Small and Large cue groups separately (since the same neutral data could not be used twice in a nested design involving the size of the cued group and cue validity). The Equal group was also analysed on its own, but only compared to one neutral condition. Means are shown in Figure 6. RTs: Color cueing of small group. For analysis, we split the data by cue validity (valid or invalid) and then cue condition, with the latter variable either being cued or neutral so that we compared valid cueing with Neutral I, and invalid cueing with Neutral II. A two-factor ANOVA found a reliable main effect of validity (F(1,17)=15.773, p=0.002, partial η2=0.481) and a borderline main effect of cue condition (F(1,17)=4.833, p=0.042, partial η2=0.221). There was also an interaction (F(1,17)=37.491, p<0.001, partial η2=0.668), with valid cueing showing a significant benefit (333ms, p<0.001) and invalid cueing a cost (201ms, p<0.001) relative to the neutral conditions. The benefit from valid priming was greater than the cost from invalid priming (hence the cue validity x cue condition interaction). 26 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 6. Means (+/- one standard error) of median RTs from color-cued trials from Experiment 2, separated by cue validity and display type. To match the corresponding displays, valid RTs were compared with the Neutral I condition, invalid RTs to the Neutral II (see Table 3). RTs: Color cueing of equal group. A one-factor ANOVA (validity) established that there was a significant main effect (F(2,34)=28.038, p<0.001, partial η2=0.623), with RTs increasing from valid to neutral (320ms, p<0.001) and neutral to invalid (322ms, p=0.002). RTs: Color cueing of large group. The data were analysed as for the Small group cued condition (valid cueing vs. Neutral I, invalid vs. Neutral II). There was a main effect of validity (F(1,17)=63.513, p<0.001, partial η2=0.789) and an interaction between validity and cue condition (F(1,17)=32.637, p<0.001, partial η2=0.658). As when the Small group was cued, the benefit from valid cueing (276ms, p<0.001) was greater than the cost from invalid cueing (235ms, p<0.001). 27 FEATURAL GUIDANCE IN CONJUNCTION SEARCH RTs: Orientation-cued trials. The analysis was similar to that for color-cued trials. Distractor ratios for Small and Large cued groups differed between valid and invalid trials, although the cued subsets were defined by the same orientation as the cue rather than the same color. As can be seen from Table 3, displays with valid orientation cueing were the same as those in the Neutral II condition, while those from invalid color cueing matched Neutral I. The cued conditions were therefore compared with the corresponding neutral conditions, with the Equal cued group involving only a single neutral condition. All three conditions were analysed separately. The mean RTs are shown in Figure 7. Figure 7. Means (+/- one standard error) of median RTs from orientation-cued trials from Experiment 2, separated by cue validity and display condition. To match the corresponding displays, valid RTs were compared with the Neutral II condition, invalid RTs to the Neutral I (see Table 3). RTs: Orientation cueing of small group. The analysis was similar to that for color-cued trials, although valid cueing was compared to Neutral II, invalid to Neutral I. A two-factor ANOVA found a main effect of validity (F(1,17)=21.438, p<0.001, 28 FEATURAL GUIDANCE IN CONJUNCTION SEARCH partial η2=0.558) and an interaction (F(1,17)=13.339, p=0.004, partial η2=0.44). The benefit from valid cueing (186ms, p=0.008) was larger than the cost from invalid cueing (123ms, p=0.058), measured relative to their respective neutral baselines. RTs: Orientation cueing of equal group. A one-factor ANOVA (validity) found a significant main effect (F(2,36)=25.478, p<0.001, partial η2=0.6), with RTs increasing from valid to neutral (231ms, p=0.014) and neutral to invalid trials (498ms, p<0.001). RTs: Orientation cueing of large group. A similar analysis was performed to that carried out with the Small group cued condition. Although no main effects were reliable, there was a significant interaction (F(1,17)=14.196, p=0.004, partial η2=0.455). The benefit from valid cueing (228ms, p=0.014) was stronger than the cost from invalid cueing (201ms, p=0.002). RT Cost-Benefit. The robust effects of validity in the RT data for both cue dimensions across all display conditions suggest that the bottom-up bias towards subset search (cf. Sobel & Cave, 2002) did not uniquely determine performance. The relative effects of cueing by color and by orientation were examined by subtracting RTs in each cueing condition with RTs on neutral trials, with the Small group and Large group cued defined as before, according to the assumed attribute of the target that would be used to guide search in the cueing condition. Means are shown in Figure 8. 29 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 8. Means (+/- one standard error) of RT Benefits/Costs from Experiment 2, separated by display type, cue validity and cue dimension. A three-factor ANOVA (display type, cue dimension, cue validity) revealed a main effect of validity (F(1,17)=35.051, p<0.001, partial η2=0.673) as well as significant interactions between cue dimension and cue validity (F(1,17)=13.449 p=0.004, partial η2=0.442) and display type and cue validity (F(2,34)=7.935, p=0.002, partial η2=0.318). The cue dimension x cue validity interaction was due to a larger modulation of performance by color cueing (571ms vs. 386ms); there were both larger RT Benefits (p=0.002) and larger RT Costs (p=0.048) on color-cued trials. The display type x cue validity interaction was unpacked by assessing the effect of cueing (RT Benefit-RT Cost) for each display type (the main effect of validity was reliable in all cases, all p<0.001). A two-factor ANOVA (cue dimension, display type) showed that there were main effects of cue dimension (F(2,34)=7.935, p=0.002, partial η2=0.318) and display type (F(1,17)=35.051, p<0.001, partial η2=0.673). However, the significant effect of display type showed that the original interaction was driven 30 FEATURAL GUIDANCE IN CONJUNCTION SEARCH by a significantly larger effect of validity at the Equal group cued condition compared to the Small group cued (p=0.008) and a trend towards significance compared to the Large group cue (p=0.17). However, there was no difference between the validity effects for the Small and Large group cued conditions (p=0.928). Overall, the data show that the larger modulation by color cueing was not affected by the number of stimuli matching the cued feature. While the effect of cue validity was largest at a 50:50 distractor ratio, there was no difference in effect size between cueing three or 11 stimuli. Accuracy. There was no speed-accuracy trade-off with accuracy largely following the pattern of RTs. See Table 2 for details. Discussion Experiment 2 replicated the effects of feature cueing on search, generating substantial costs and benefits on trials where the cues were valid or invalid. In this case, the effects occurred in the context of bottom-up cueing of attention to unequalsized distractor groups, enabling us to examine the interaction between the cues and bottom-up segmentation processes. In the neutral condition, there was evidence for bottom-up segmentation affecting performance, with performance being faster when there were unequal ratios of distractors relative to when there were equal numbers of each type of distractor. Interestingly, performance appeared to be more affected by segmentation into elements grouped by orientation, since the advantage for targets appearing in the minority group was particularly strong when group size was defined on the orientation rather than the color dimension. This occurred even though we matched the saliency of the individual items (Experiment A, see Appendix). The advantage for orientation 31 FEATURAL GUIDANCE IN CONJUNCTION SEARCH may arise because bottom-up grouping based on orientation was stronger than bottom-up grouping based on color here (cf. Sobel & Cave, 2002). There was also an interaction between the bottom-up bias for attention to be drawn to the smaller subset, and top-down cueing. This is revealed by there being larger effects of cue validity for displays with an equal ratio of distractors than for displays with unequal ratios of distractors (see the Cost-Benefit analysis). The change in the cue validity effect was most marked for invalid trials, with the cost of invalid cueing being reduced when there were unequal distractor ratios. This suggests that, although the cue guided search, participants were able to switch more rapidly to the uncued group when there was an unequal ratio of distractors. This may be illustrated most easily in relation to displays (i) and (iii) shown in Figure 4, when a green vertical (GV) target is present. On a valid color cue trial (green), the cue might relate to the small group (display iii) or to the large group (display i). On an invalid color cue (blue) trial the cue might relate to the large group (display iii) or the small group (display i). The cost of cueing might be reduced on invalid trials when attention is directed to the small group (display i) because this can be quickly rejected and search re-directed to the group containing the target. When attention is directed to the large color group (display iii), the presence of a small orientation group, segmented from bottom-up cues, allows attention to be switched more rapidly than when the distractor groups are equal in size (and note that the target, when present in this small group, will tend to have relatively high salience against the small number of distractors with a matching attribute). This also suggests that bottom-up segmentation might have operated in parallel with any segmentation induced by the top-down cue. For example, the top-down color cue might induce segmentation into color groups, but the parallel segmentation into a small and large orientation group allows attention then to 32 FEATURAL GUIDANCE IN CONJUNCTION SEARCH be switched to the small orientation group, when the target is a member of that group (cue green to the GV target in display i). As in Experiment 1, the effects of cueing color were stronger than those of cueing orientation. This may be because the color cue can initiate segmentation in a top-down manner more quickly than any top-down induction of segmentation by orientation. On the other hand, if the contrasting cue effects were due to differences at a response stage, then the cueing effects should have been additive with changes in the distractor ratio. There should only have been bottom-up guidance of search to the smaller group, followed by more rapid selection of the response to color. Critically, the costs and benefit of cueing should have been equal for displays with equal ratios of distractors and for displays with unequal distractor ratios. It was not. In Experiment 3 we sought to provide a further direct test of the response effect account by presenting participants with a compound search task. Experiment 3: Compound Search Recently, Theeuwes et al. (2006) presented evidence that intra-trial cueing of a target’s features can influence performance at the level of response selection. Using a verbal cue to predict the dimension defining the target of a feature-singleton search (cf. Müller et al., 2003), they found effects of cue validity when the response decision was present-absent, but none when the decision was search-irrelevant. They suggested, therefore, that cueing affected processes at the response level of visual selection, rather than early top-down modulation of attention. The stronger effects of color than orientation cueing, in Experiments 1 and 2, could be because color cues shift the decision criterion so that it is lowered for the expected color and raised for the unexpected color (on invalid cue trials). The net result would be that the costs and 33 FEATURAL GUIDANCE IN CONJUNCTION SEARCH benefits from color cueing are stronger than those from orientation cueing (see also Cohen & Feintuch, 2002; Cohen & Magen, 1999, Cohen & Shoup, 1997, for similar arguments from cueing effects on search). Experiment 3 used a compound task rather than having participants simply respond whether a target was present or absent. With a compound task the features of the target and the cue are independent from the response-defining feature. Hence, the cues cannot modulate the links between the selected feature and response selection. If the differential effect of color and orientation cueing was due to an effect on response selection, then we expect the effect to be eliminated here. Method The methodology was identical to that of Experiment 1 unless indicated below. Participants. Eighteen University of Birmingham students, four male, 16 female, aged between 18 and 33 took part (average age 24). One participant was removed due to unusually slow responses. Design. There were four main independent variables: cue validity (valid, neutral, invalid), cue dimension (color, orientation), array size (seven or 15 items) and target type (blue horizontal, green vertical). The main dependent variable was the RT taken to indicate which feature the target contained (+ or x). Accuracy was also measured. Stimuli. The search stimuli were similar to that of Experiment 1, except that small light grey symbols were added to the centre of all the items. These were either an ‘x’ or a ‘+’ and were distributed in a pseudo-random manner across distractors and targets, so that each feature was added to approximately half the stimuli. The choice 34 FEATURAL GUIDANCE IN CONJUNCTION SEARCH of symbols was planned to minimise interaction with the orientation of the stimuli. Both features were three-pixels wide, leading to dimensions of 1mm (0.05º) by 1mm (0.05º), thickness 0.3mm (0.01º). Color levels are outlined in Table 1. These symbols were also present on target stimuli, where they indicated to participants which response to make. Procedure. Participants took part in a one-hour session, consisting of five 80trial blocks, so that each participant undertook 400 trials. Trial conditions were randomly interleaved within each block with the first block regarded as practice where RTs and accuracy were not recorded. The procedure replicated that of Experiment 1 except that the participant’s response depended on the symbol on the target stimulus (whether an x or + was present), not whether the target was present or absent. Consequently there were no absent trials. Half the participants responded to the presence of ‘+’ with their dominant hand, the other half responded to ‘x’, pressing the ‘Z’ or ‘M’ keys depending on handedness. Signs were present on the computer monitor to remind participants of the targets and which key response corresponded to which symbol. Results RTs. The data were collated as in Experiment 1 except that RTs were also pooled across action-defining features (e.g., + and x). The mean of the participants’ median RTs from color- and orientation-cued trials are shown in Figure 9. The data were initially analysed separately for color and orientation cue conditions in comparison with the neutral condition, since the neutral condition was not nested into the design alongside effects of the two types of cue. 35 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 9. Mean of median RTs (+/- one standard error) from Experiment 3 defined by cue dimension, cue validity and array size. RTs: Color-cued trials. A two-way ANOVA (cue validity, array size) compared the valid color, invalid color and neutral conditions. Main effects of cue validity (F(2,32)=65.169, p<0.001, partial η2=0.803) and array size (F(1,16)=42.479 p<0.001, partial η2=0.726) were present. RTs increased across array size as well as across cue validity, with valid cueing facilitating search compared to the neutral condition and invalid search leading to longer RTs compared to the other two conditions (all ps<0.001). These effects were modulated by a significant interaction (F(2,32)=18.462, p<0.001, partial η2=0.536), with the effects of validity increasing at the larger display size. RTs: Orientation-cued trials. There were main effects of validity (F(2,32)=16.523, p<0.001, partial η2=0.508) and array size (F(1,16)=89.168, p<0.001, partial η2=0.848), RTs increased across array size as well as across cue condition, with valid cueing facilitating search compared to the neutral condition and invalid 36 FEATURAL GUIDANCE IN CONJUNCTION SEARCH cueing leading to longer RTs compared to the other two conditions (all ps<0.001). As with color-cued trials, there was also a significant interaction (F(2,32)=13.727, p<0.001, partial η2=0.462), indicating that validity effects increased at the larger array size. RT Cost-Benefit. RT Benefits and Costs were calculated as in Experiment 1 to allow effects with color and orientation cues to be compared directly. The resulting data are depicted in Figure 10. There was a main effect of validity (F(1,16)=59.579, p<0.001, partial η2=0.788) and a reliable interaction between array size and cue validity (F(1,16)=23.363, p<0.001, partial η2=0.595), and a borderline significant cue dimension x cue validity (F(1,16)=5.569, p=0.062, partial η2=0.258). The former interaction arose because the validity effects were smaller at display size 7 than at display size 15 (857ms vs. 1524ms, both p<0.001). The dimension x validity interaction was due to the validity effects being larger following color relative to orientation cues (1288ms vs. 1093ms, both p<0.001). This advantage for color cueing did not interact with the array size. Accuracy. The results showed no speed-accuracy trade-off. See Table 2. 37 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 10. Means (+/- one standard error) of RT Benefit/RT Cost from Experiment 3, defined by cue dimension, cue validity and array size. Discussion The results replicated the cueing costs and benefits found in Experiments 1 and 2 – there were strong effects of the cues, cue validity affected search rates, and cueing effects overall were greater for color than for orientation cues. The differential effects of color over orientation cues remained additive with the effects of the display size. These results contradict the proposal that the asymmetrical effects of color and orientation cues are due to color having a greater influence on shifts in response criteria. In a compound task, response assignment is independent of the dimension used to select the target, preventing the dimension from differentially influencing response selection. Despite this, greater effects of cueing emerged with color cues. 38 FEATURAL GUIDANCE IN CONJUNCTION SEARCH This result strongly points to the asymmetrical cueing effects reflecting an earlier process involved in parsing the display and guiding search to color- or orientationdefined groups. In Experiment 2 we presented evidence that, in the neutral baseline condition, bottom-up biases in segmentation favored orientation-defined targets. This bias appears to be overridden when a color cue is presented, with the result that the initial parsing is into color groups and search is directed into the color group indicated by the cue. The overall outcome is that a larger effect of color cueing emerges. In the final experiment, we assessed whether the differential effect of color cueing was indeed top-down in nature. When visual cues are used, it is not clear if cueing effects occur in a top-down or bottom-up manner, since bottom-up effects may arise on the basis of perceptual priming from the color or orientation values present in the cues. To eliminate bottom-up effects Experiment 4 was conducted using verbal cues. Top-down effects may survive this switch in cue type (see Soto & Humphreys, 2007), whereas bottom-up effects may not. Experiment 4: Visual vs. Verbal Cueing A number of studies (e.g., Maljkovic & Nakayama, 1994; Kristjánsson, Wang, & Nakayama, 2002; Olivers & Meeter, 2006; Kristjánsson, 2006) have found that the repetition of identical target stimuli, across trials, facilitates search compared to when the target changes. This perceptual priming has largely been attributed to be the consequence of bottom-up processes from stimulus repetition, processes that could be involved in the cueing effects found previously in our research. Indeed, Wolfe, Horowitz, Kenner, Hyle and Vasan (2004) replicated the facilitation of priming using an inter-trial stimulus cue that matched the following search target. Experiment 4 39 FEATURAL GUIDANCE IN CONJUNCTION SEARCH compared the effects of visual (color patch, line orientation) and verbal cues. With verbal cues, bottom-up priming within the same trial should be minimised. Method Experiment 4 used a similar methodology to Experiment 3, with the differences outlined below. Participants. Eighteen University of Birmingham students, five male, 13 female, aged between 19 and 41 (average age 24.28) took part. Stimuli. Response-decisions were identical to those used in Experiment 3, while the array stimuli differed only in size from Experiments 1-3. The bars were 0.8cm (0.77º) long by 0.2cm (0.19º) wide, symbols 0.2cm (0.19º) by 0.2cm (0.19º) and 0.025cm wide (0.024º). Color levels are outlined in Table 1. Trials were separated into visual-cued, verbal-cued and neutral blocks. In the visual-cued block, the cues were identical to those used in previous experiments (color patches, lines of a particular orientation). In the verbal-cued conditions, these stimuli cues were replaced with the words blue, green, horizontal, and vertical. The words were presented in capitals in white, Arial type, with a height of 7mm (0.67°) and width varying from 26mm (2.48°; ‘BLUE’) to 64mm (6.11°; ‘HORIZONTAL’). Design. There were four main independent variables: cue type (verbal, visual), cue validity (valid, neutral, invalid), cue dimension (color, orientation) and target type (blue horizontal, green vertical). The main dependent variable was the RT taken to indicate which feature the target contained (+ or x). Accuracy was also measured. Procedure. Participants took part in a single one-hour session, with three blocks of trials. Each block was made up of either neutral, visual-cued or verbal-cued trials with the order counter-balanced over participants. Neutral trials were different to 40 FEATURAL GUIDANCE IN CONJUNCTION SEARCH those of previous experiments, with no cue presented. Visual cues were unchanged from Experiment 4. The timeline of verbal-cued trials was identical to that of the visual-cued trials, except that the cues were words instead of stimuli. The blue patch stimulus was replaced with the word ‘BLUE’; the green patch with ‘GREEN’; the white horizontal line with ‘HORIZONTAL’; and the white vertical line with ‘VERTICAL’. Validity did not differ between cued blocks, but instructions were altered to indicate the differing natures of the cues. The neutral block was 60 trials long, with 10 practice trials preceding the block. Cued blocks each consisted of two blocks, one word and one stimulus cued, of 80 trials with 15 practice trials at the beginning of each block. Results RTs. Data was collected as in Experiment 4, except RTs were separated by cue type (e.g., visual or verbal) rather than array size. As there was a single neutral condition that could not be divided by cue type or cue dimension, we analyzed RTs from each cue type and dimension separately. Means of participants’ median RTs from visual and verbal cueing of color and orientation are shown in Figure 11. 41 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 11. Mean (+/- one standard error) of median RTs from Experiment 4, defined by the nature of the cue (visual or verbal), dimension cued (color or orientation) and cue validity (valid or invalid). RTs: Visual cues. Analysing the data from color visual cueing, a one-factor ANOVA found a main effect of validity (F(2,34)=66.898, p<0.001, partial η2=0.797), with RTs increasing from valid cueing to neutral to invalid cueing conditions (all comparisons at p<0.001). Similar analysis of data from visual orientation cueing trials again found a main effect of validity (F(2,34)=21.6, p<0.001, partial η2=0.56), with a priori comparisons showing that RTs increased from trials with valid cueing to neutral trials (p=0.001) to trials with invalid cueing (p=0.036). RTs: Verbal cues. Similar to the results with visual cues, there was a main effect of validity (F(2,34)=70.184, p<0.001, partial η2=0.805) with, compared to the neutral condition, valid verbal color cueing facilitating and invalid verbal cues slowing search (all ps <0.001). Analysis of verbal orientation cued trials also found a 42 FEATURAL GUIDANCE IN CONJUNCTION SEARCH main effect of validity (F(2,34)=70.184, p<0.001, partial η2=0.805), with RTs shorter in valid trials (p=0.002) compared to neutral trials and longer in invalid trials (p=0.003). RT Benefit-Cost. To compare the effect of verbal and visual cueing, RT Benefits and Costs were calculated by subtracting median valid and invalid RTs from the corresponding median RTs from the neutral block (as in Experiments 1-3). Mean values are shown in Figure 12. Figure 12. Mean RT Benefits/Costs (+/- one standard error) from Experiment 4, defined by type of cue (visual or verbal), dimension cued (Color or orientation) and cue validity (valid or invalid). A three-factor ANOVA (cue type, cue dimension, cue validity) showed no effects of cue type (whether the cue was visual or verbal; F(1,17)=1.048, p=0.64, partial η2=0.058). There was a main effect of cue validity (F(1,17)=81.828, p<0.001, partial η2=0.828) and an interaction between cue dimension and cue validity 43 FEATURAL GUIDANCE IN CONJUNCTION SEARCH (F(1,17)=9.861, p=0.012, partial η2=0.367), with larger modulation of behaviour by color (1294ms, p<0.001) compared to orientation cues (957ms, p<0.001). RTs: Quartile analysis. Although the analysis showed no difference between visual and verbal cues it is possible that this masks the presence of contrasting time courses to the effects. For example, visual cues might exert an earlier effect than verbal cues due to bottom-up priming of the perceptual system before the search display appears. To assess this, we separated early and late RTs based on fastest and slowest scores (e.g., the 25th and 75th RTs) for each participant in each condition. We then compared the cueing effects for the fast and slow search times (see also Soto et al., 2005, for a similar approach). Due to small number of trials with invalid cues, only valid RTs were used in the analysis. As the two quartile scores were not independent, they were analysed separately as were the validity conditions from the two types of cue, visual and verbal. Means across participants are shown in Figure 13, with the mean of median RTs included for visual comparison. 44 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 13. Means (+/- one standard error) of fastest (lower quartile), median and slowest (upper quartile) RTs from Experiment 4, defined by type of cue (visual or verbal) and dimension cued (color or orientation). RTs: Slowest quartile. The lack of an invalid cueing allowed direct comparison to be made between RTs in the valid color cueing, valid orientation cueing and the neutral conditions for each type of cueing3. Analysing data from visually cued trials, a one-factor ANOVA found a main effect (F(2,34)=24.203, p<0.001, partial η2=0.587), with search facilitated after color cues compared to orientation cues (difference of 106ms, p=0.002), and RTs from both valid conditions shorter than the neutral condition (difference of 494ms, p<0.001, for color, and 389ms, p=0.002, for orientation cues). Similar analysis of the verbal cueing also found a main effect (F(2,34)=22.8, p<0.001, partial η2=0.573), with RTs shortest following color compared to orientation cues (difference of 76ms, p=0.094). Search As a nested design is not being used we can include a single neutral condition along with single color and orientation conditions. 3 45 FEATURAL GUIDANCE IN CONJUNCTION SEARCH was facilitated in both types of valid conditions to the neutral condition (difference of 460ms for color and 384ms for orientation cues, both ps<0.001). RTs: Fastest quartile. Again, a one-factor ANOVA compared RTs from valid color, valid orientation and neutral trials. The data from visual cues showed a main effect (F(2,34)=26.612, p<0.001, partial η2=0.61), with RTs shortest following valid color cued compared to visual orientation cues (difference of 355ms, p=0.002), and RTs from both the valid conditions shorter than those in the neutral condition (difference of 934ms, p<0.001, for color, and 579ms, p=0.002, for orientation cues). The pattern was similar for RTs from verbal cueing conditions. Analysis showed a main effect (F(2,34)=24.625, p<0.001, partial η2=0.592), with faster search following verbal color cues compared to verbal orientation cues (difference of 255ms, p=0.004). RTs from both cued conditions were shorter than those in the neutral condition (differences of 864ms, p<0.001, for color, and 709ms, p=0.002, for orientation cues). RT Benefit: Quartile analysis. There were significant effects by both color and orientation cueing on both quartile measures. To directly compare the effect of visual and verbal cueing, the cueing effect was calculated for all conditions by subtracting the quartile score from valid trials the corresponding score from neutral, uncued trials. Again, analysis of the two quartile data was undertaken separately. Means are shown in Figure 14, with data from the median measure shown for visual comparison. 46 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Figure 14. Means (+/- one standard error) of RT Benefit on the fastest (lower quartile), median and slowest (upper quartile). RTs from valid cueing in Experiment 4, defined by type of cue (visual or verbal) and dimension cued (color or orientation). RT Benefit: Slowest quartile. A two-factor ANOVA (cue type, cue dimension) found a main effect of cue dimension (F(1,17)=11.736, p=0.006, partial η2=0.408) with a larger benefit from color compared to orientation cues. No other effects or interactions reached significance (Fs<1) RT Benefit: Fastest quartile. Analysis again found a main effect of cue dimension (F(1,17)=17.544, p=0.002, partial η2=0.508), with color cueing leading to a larger benefit than orientation cueing, but no main effect of cue type (F<1). There was a trend towards an interaction, but did not reach significance (F(1,17)=2.842, p=0.22, partial η2=0.143). Accuracy. There was no speed-accuracy trade-off. See Table 2 for details. 47 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Discussion Experiment 4 showed that the pattern of effects from stimulus cueing remained the same when the predictive information was delivered by verbal cues as when it was provided by a visual cue, with a larger modulation of selection by color than by orientation cues. Further to this, the pattern remained consistent across the RT distribution indicating that the underlying processes occur early on during search. These results go against the hypothesis that the advantage for color cueing found in Experiments 1-3 was merely a consequence of perceptual priming by the physical nature of the visual cues. Instead, the data are consistent with the processes involved in the dimensional difference being top-down in nature. Indeed, the RT quartile analyses indicated that effects of verbal cues affected both the fastest and slowest RTs similar to the visual cues. There was no evidence for bottom-up priming from visual cues. General Discussion We have reported four experiments showing robust effects on search from precueing the likely feature of the following target. In Experiment 1 we showed that the effects of cueing attention to the likely color of the target had a greater overall effect on performance than cueing attention to the likely orientation of the target. The enhanced effect of color cueing was additive across display size, although both color and orientation cueing effects increased at the larger display size. This additive effect of the cue type and display suggests that the type of cue either influenced an early segmentation process, prior to search being initiated, or a late process of response selection. In Experiment 2 the effects of the cue were examined in combination with bottom-up influences on segmentation and attentional guidance. In the neutral 48 FEATURAL GUIDANCE IN CONJUNCTION SEARCH baseline, search was facilitated when uneven ratios of distractors were presented consistent with bottom-up biases affecting performance. Despite the presence of these bottom-up biases when there were unequal distractor groups, cueing still affected performance. The effects of cueing were reduced, however, when there were unequal numbers of distractors in the displays, with the costs from invalid cueing reducing in magnitude. This result suggests that bottom-up segmentation took place in parallel with any top-down induced segmentation, making it easier to switch attention to a small group coded through bottom-up segmentation when invalid cues directed attention to the larger of the two groups present. Experiment 3 tested whether the effects of cue type arose due to differential effects on setting the response criteria. We found no evidence for this. In a compound search task, the cue used to select the target is separated from the property used to select the response. This should reduce cueing effects at a response level. Despite this there remained greater effects from cueing color than cueing orientation. Finally, Experiment 4 demonstrated that cueing effects were equally large from visual and verbal cues, and in both cases the effects were enhanced for color relative to orientation cues. This indicates that the greater effects from color cueing can arise from ‘pure’ top-down cueing4. Top-down Color Segmentation Both color and orientation cues reduced the slopes of the search functions (Experiments 1 and 3). We have suggested that this reflects participants starting search with the cued group, whereas on neutral trials they select color and orientation groups equally often. However, the advantage for color over orientation cueing was additive with the effects of display size. We conclude that there was no difference Pilot studies (see Appendix) had established color and orientation-defined targets were equally salient, for all of the experiments subsequently presented. 4 49 FEATURAL GUIDANCE IN CONJUNCTION SEARCH between color and orientation in terms of determining which group was searched first. Rather than this, we propose that color cues facilitated the segmentation of the displays into color groups more than orientation cues facilitated the segmentation of the displays into orientation groups. Thus, even though both forms of segmentation may operate in parallel (see above), color cueing allowed color segmentation to finish earlier in time, allowing search through the color groups to be initiated earlier. According to this account, color cues exert a strong priming effect on color segmentation (e.g., Hannus et al., 2006). This can operate even from verbal representations of the color cues, so the ‘set’ for color can be imposed from higherlevel representations that are abstracted from perceptual representations of color. Recently Soto and Humphreys (2007) have provided evidence that verbal information held in working memory can guide visual attention to a subsequent search display, even when the memory information is irrelevant to the search task. This is highly reminiscent of the verbal effects of the color cues here, except that we are proposing that the effects of the color cues are on an initial segmentation stage rather than guidance of search per se. One way to account for these cross-modal cueing effects is to suggest that information is held in working memory in a representation abstracted from the perceptual features of the stimulus, perhaps akin to the idea of an episodic buffer put forward by Baddeley (2000). This representation feeds back activation to earlier perceptual processes, enabling them to be completed more efficiently. Our data indicate that any feedback from higher-level processes facilitates segmentation of displays by color more than by orientation. However, we found in Experiment 2 that bottom-up segmentation and guidance of search to the smaller of two orientation groups was more effective than the bottom-up segmentation and guidance of search to the smaller of two color groups (in the neutral condition). The 50 FEATURAL GUIDANCE IN CONJUNCTION SEARCH stronger top-down effect of color, then, has to overcome the initial stronger bottom-up effects. This dominance for color segmentation has been found previously. Williams (1966) investigated eye fixations in a search task where stimuli were defined by a conjunction of size, color or shape. When participants were prompted to find a particular coloured shape they fixated more readily on stimuli defined by color than those defined by size or shape. More recently, dominance for color was evident in eye-tracking studies by Hannus and colleagues (Hannus, Cornelissen, Lindemann, & Bekkering, 2005; Hannus et al., 2006). In their 2006 paper, Hannus et al. used a pretrial stimulus that matched the target (which could be one of four different stimuli) in color- or orientation-defined singleton search and color-orientation conjunction search. Observers were required to make a saccade to the target as quickly as possible. Prior to this, the color and orientation of the stimuli were adjusted to determine individual thresholds for 70% discrimination for both dimensions. The data showed that, in feature search, the probability of the first saccade landing on a stimulus matching the orientation of the cue (and the target) was largely equated with it landing on a stimulus sharing the cue’s color. However, in a conjunction search with the same stimuli this ‘orientation probability’ was halved while no change was observed for color probability across search tasks. Hannus et al. also found a similar pattern when comparing color with the size dimension. In contrast, there were no differences between the dimensions in saccade latency, suggesting the color dominance was not due to differences in task difficulty. One proposal by Hannus et al. was that orientation and size discrimination suffer from “crowding”, e.g., the detrimental influence on feature discriminability generated by the presence of surrounding objects. Color discriminability, they proposed, is unaffected and they suggested that visual search theories should be amended to give color a preferential role in guiding 51 FEATURAL GUIDANCE IN CONJUNCTION SEARCH attention. This crowding explanation could easily be linked to our preferential color segmentation proposal, with the ease of segmentation of stimuli differentiated by color reducing the effects of adjacent stimuli on discrimination. Indeed, Mollon (1989) has previously suggested an evolution-driven advantage for color vision, with color-blind people finding it difficult to detect colored fruit surrounded by foliage of varied luminosity. He suggests that color vision facilitates the segregation of the visual field, offering advantages in identifying objects. These differential segmentation processes could reflect the different neural processes involved in segmentation by color and orientation. Color has been extensively documented as being processed in the ventral stream (e.g., Bartels & Zeki, 2000) and therefore may be more influenced by top-down processes (e.g., with more dorsal visual areas acting in a bottom-up manner to signal the presence of a strong stimulus for attention; see Corbetta & Shulman, 2002). Orientation, on the other hand, may in part be coded through more dorsal visual areas (Murata, Gallese, Luppino, Kaseda, & Sakata, 2000; Oliver & Thompson-Schill, 2003), although dorsal areas have also been found to be involved during color-discrimination judgements tasks (Claeys et al., 2004). Neglect data (Kristjánsson, Vuilleumier, Malhotra, Husain, & Driver, 2005) again point to color dominance, but in this case from implicit activation of color stimuli which facilitate search for targets on the next trial with the same color even when the first stimulus is neglected. Kristjánsson et al. looked at inter-trial priming, where a repetition of a feature trial-on-trial facilitates search, in two left visual neglect and extinction patients with damage in the right parietal lobe. When targets were presented in the patient’s contralesional visual field, repetition of target color and location facilitated search, however color priming did not depend on patients having 52 FEATURAL GUIDANCE IN CONJUNCTION SEARCH awareness of the preceding target. A level of color processing, therefore, was not impaired by the parietal damage, while processing of the target’s location was significantly reduced, suggesting some color processing in ventral areas could be intact. Theoretical Models What are the implications of these results for functional theories of visual search? Guided Search Theory (GST: Wolfe, 1994; Wolfe, Cave, & Franzel, 1989) proposes that there is an initial preattentive parallel stage of visual processing where basic visual features are coded independently in distinct retinotopic ‘feature maps’. Activation within the feature maps will reflect both the bottom-up saliency of the stimuli (e.g., generating by local differences between elements) and top-down cueing (pre-activation of the maps by foreknowledge of the target). Within this framework, pre-cueing in our studies should increase the activation of the stimuli sharing the cued feature, facilitating their selection over items not subject to top-down cueing. As a consequence, segmentation of the stimuli, and selection of one group of elements based on the cued feature, may be faster relative to when the cue is neutral. On a valid trial, search will be facilitated as the target will be a member of the selected group. This comes at the cost of performance on an invalid trial, when a distractor group will show speeded selection and attention may need to be dis-engaged in order for the target to be subsequently selected. This effect may emerge on the intercepts rather than the slopes of the search functions if search, even on neutral trials, is based on the segmentation into sub-groups of stimuli, but with this process occurring faster on trials where a feature cue is present. 53 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Now, when the ratio of distractors is manipulated (Experiment 2), the smaller group of distractors will benefit due to these items having increased local differences relative to the other distractors (cf. Sobel & Cave, 2002). These differences between distractor ratios would be reflected in the GST salience map, with increased activation for stimuli in the smaller subset. In addition, the bottom-up modulations based on the distractor ratio should occur in parallel to those induced top-down by the cue. Our data indicate that, at least with the current displays, top-down modulation was stronger than the bottom-up effects so that robust effects of cueing occurred even when there were uneven ratios of the different types of distractor. This occurred even when the top-down cue directed attention to the larger distractor group (there remained a cost to performance). Nevertheless, the switching of attention to the invalidly cued group was easier when the distractor ratio was small, compared with when there was an equal ratio of distractor types (see Figure 8). This in turn suggests that the bottom-up salience of the smaller set of distractors was maintained, even if the cued set of distractors was selected first. On winner-take-all accounts of visual selection, there would be a loss of any increased bottom-up activation for the smaller distractor group on trials where the larger top-down group ‘wins’ the competition for selection. The present results indicate that, even when this happens, either activation differences favoring the smaller group are maintained in the feature maps, or processing continues to take place after any initial top-down biased selection, with the bottom-up cues again favoring the smaller distractor group. Data consistent with the argument for early effects of top-down cueing comes from our analysis based on RT distributions in Experiment 4. Here we found a larger modulation of selection by color than orientation at the fast as well as the slow end of the response distributions. The same pattern of effects were present whether the target 54 FEATURAL GUIDANCE IN CONJUNCTION SEARCH was visual or verbal, indicating that the cueing effects on search did not reflect processes driven by the physical nature of the cue. Although GST can account for the relations between top-down and bottom-up guidance of attention, the model has little to say about our observed differences between the top-down cueing of color and orientation. Wolfe (1994) suggested “that color information guides attention more effectively than orientation” (page 208), without distinguishing between top-down and bottom-up guidance. Our results, however, indicate that there may be increased gain on the top-down input into the color feature map, thereby increasing the top-down biases towards segmentation into a winning and losing group. This facilitated segmentation from color cueing would generate an overall RT advantage even if there is subsequently serial selection within the ‘winning’ group of items. It is of interest that while top-down cueing of color appeared more effective than that of orientation, the data on bottom-up effects of distractor ratio generated the opposite picture. In Experiment 3, in the neutral, uncued condition search was fastest when the orientation of the smaller group of distractors matched that of the target. For trials with a green vertical target, RTs were shortest when there were three blue vertical distractors, both compared to when the ratio was balanced and when there were three green horizontal distractors. Likewise, for trials with a blue horizontal target search was fastest when there were three horizontal distractors. The smaller orientation-defined subset, therefore, guided attention more efficiently than a colordefined subset of the same size. This contrast between the different dimensions, even when overall saliency was matched (see Appendix) provides strong evidence for the two forms of saliency (bottom-up and top-down) being ‘driven’ independently in search, even if their outputs are subsequently pooled in the competition for selection. 55 FEATURAL GUIDANCE IN CONJUNCTION SEARCH A somewhat different account of the data can be formulated in terms of Attentional Engagement Theory (AET, Duncan & Humphreys, 1989, 1992). AET is a two-stage model in which there is preattentive grouping of stimuli followed by a matching of the representations to a template of the target. Pre-cueing the target will increase the ‘pertinence’ of matching distractors, enabling them to win the competition for selection (cf. Bundesen, 1990; Heinke & Humphreys, 2003, for similar, more formal accounts). For this account a distractor ratio effect may arise because spatial grouping between the larger set of distractors enables them to be rejected together, enabling the smaller set of distractors to be selected. Our data can be accounted for if any top-down cueing of the target’s template acts to off-set the rejection of the larger group of distractors, so that these items are selected first. The effect would emerge on the intercepts rather than the slopes of the search functions because of pre-attentive segmentation into distractor groups, and the cueing effects simply reflect which group is selected first. Whichever account of the data is maintained, the important result is that the top-down and bottom-up cueing effects appear to be functionally independent, and lead to opposite efficiencies for the different dimensions of orientation and color. Response-based Accounts of Performance Our data counter the proposal by Theeuwes et al. (2006) that top-down effects on selection (in their case, from intra-trial cueing) influence response selection. Theeuwes et al. used a verbal cue that predicted which dimension defined the target of a following feature-singleton search. Effects of cue validity were present when there was a present-absent response decision, but not when the decision was irrelevant to the search (see also Cohen & Feintuch, 2002; Cohen & Magen, 1999, Cohen & 56 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Shoup, 1997). However, contrary to Theeuwes et al., we cued the feature of a conjunction target rather than the dimension of the stimulus in an efficient featurebased search task. In our present-absent task (Experiment 1), the response-decision was consistent regardless of the cue and required information from both dimensions – color and orientation. This would likely reduce any direct link from the cued dimension to the response. Even stronger evidence comes from Experiment 4 here, where top-down cueing effects emerged in a compound search task, where the response was unrelated to the cued dimension. We conclude that the cueing effects here were not due to linkage of the cue to a response. Hard vs. Easy Search Previous investigators have assessed whether top-down effects influence easy or difficult conjunction search differently. In Sobel and Cave (2002), observers searched for a color-orientation conjunction target amongst two sets of distractors of varying ratios (see our Experiment 2). When both dimensions were easily discriminable, participants preferentially searched the smaller subsets of distractors, irrespective of whether the subsets were defined by color or orientation. Also, as the discrimination of the orientation dimension was more difficult, so participants were more likely to search the color subset for the target. This shows that there is a balance between top-down and bottom-up cues for selection, with bottom-up cues emerging more strongly when there was differential discriminability along the dimensions of the stimuli. Indeed in Experiment 2 here, bottom-up saliency effects seem less effective than top-down cues for selection. Heinke, Humphreys and Tweed (2006) compared the effect of pre-cueing of the target on easy – a V surrounded by rotated Ls – and hard – an upright L surrounded by rotated Ls. They found stronger cueing 57 FEATURAL GUIDANCE IN CONJUNCTION SEARCH effects when the search was easy compared with when it was more difficult and suggested that the target needed to have a sufficient level of discriminability in order to make contact with any (pre-activated) template for the target. The present set of pilot studies (see Appendix) showed that the target properties were relatively salient along each dimension (colour and orientation), and this may have been helpful in order to allow the top-down effects to emerge. In addition to this, we showed that the top-down cueing effects were equally strong from written as from visual representations of the target. It is possible that participants were able to translate the written cue into a visual image, and this acted in place of a visual template, guiding search. Alternatively, participants may extract semantic information from the visual displays which can be rapidly mapped onto a verbal representation for the target (see Soto & Humphreys, 2007). In either case, the data show that top-down cueing does not depend on a physical representation of the target being present (cf. Wolfe et al., 2004). Conclusion We conclude that there can be differential top-down cueing of attention based on the expected color rather than the orientation of a target, even under circumstances in which bottom-up segmentation based on orientation seems strong. This effect occurs with words as well as visual representations of stimuli and it occurs in compound as well as present-absent search tasks. That stronger guidance following color cues occurs in conjunction with a stimulus-driven bias towards orientation points to there being independent ‘drivers’ of top-down and bottom-up cueing in search. 58 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Acknowledgements. This work was supported by grants from the ESRC, BBSRC, EPSRC and MRC (UK). The work was completed in partial fulfilment of a PhD by the first author. 59 FEATURAL GUIDANCE IN CONJUNCTION SEARCH 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, 948–961. Baddeley, A. (2000). The Episodic Buffer: a New Component of Working Memory? Trends in Cognitive Sciences, 4 (11), 317-423. Bartels, A., & Zeki, S. (2000). The Architecture of the Color Centre in the Human Visual Brain: New Results and a Review, European Journal of Neuroscience, 12, 172-193. Bekkering, H., & Neggers, S. F. W. (2002). Visual Search is Modulated by Action Intentions. Psychological Science, 13 (4), 370-374. Braithwaite, J. J., Humphreys, G. W., & Hulleman, 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. Bundesen, C. (1990). A Theory of Visual Attention. Psychological Review, 97 (4), 523-547. Claeys, K. G., Dupont, P., Cornette, L., Sunaert, S., Van Hecke., P., De Schutter, E. et al. (2004). Color Discrimination Involves Ventral and Dorsal Stream Visual Areas. Cerebral Cortex, 14, 803-822. Cohen, A., & Shoup, R. (1997). Perceptual Dimensional Constraints in Response Selection Processes. Cognitive Psychology, 32, 128–181. Cohen, A., & Magen, H. (1999). Intra- and Cross-Dimensional Visual Search for SingleFeature Targets. Perception & Psychophysics, 61, 291–307. Cohen, A., & Shoup, R. (2000). Response Selection Processes for Conjunctive Targets. Journal of Experimental Psychology: Human Perception and Performance, 26(1), 391-411. 60 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Cohen, A., & Feintuch, U. (2002). The Dimensional-Action System: A Distinct Visual System (pp. 587-608). In V. Prinz and B Hommel (Eds.). Attention and Performance XIX. Oxford, England: Oxford University Press. Corbetta, M., & Shulman, G. L. (2002). Control of Goal-Directed and Stimulus-Driven Attention in the Brain. Nature Neuroscience, 3, 201-215. Duncan, J., & Humphreys, G. W. (1989). Attentional Engagement Theory: Visual Search and Stimulus Similarity, Psychological Review, 96 (3), 433-458 Duncan, J., & Humphreys, G. W. (1992). Beyond the Search Surface – Visual Search and Attentional Engagement. Journal of Experimental Psychology: Human Perception and Performance, 18 (2), 578-588. Downing, P. E (2000). Interactions Between Visual Working Memory and Selective Attention. Psychological Science, 11 (6), 467-473. Egeth, H. E, Virzi, R. A., & Garbart H. (1984). Searching for Conjunctively Defined Targets, Journal of Experimental Psychology: Human Perception and Performance, 10, 32-39. Friedman-Hill, S. R., & Wolfe, J. M. (1995). Second-order Parallel Processing: Visual Search for the Odd Item in a Subset. Journal of Experimental Psychology: Human Perception and Performance, 21 (3), 531-551. Forti, S., & Humphreys, G. W. (2008). Sensitivity to Object Viewpoint and Action Instructions During Search for Targets in the Lower Visual Field. Psychological Science, 19, 42-48. Hannus, A., Cornelissen, F. W., Lindemann, O., & Bekkering, H. (2005). Selection-forAction in Visual Search. Acta Psychologica, 118 (1-2), 171-191. Hannus, A., van den Berg, R., Bekkering, H., Roerdink, J. B. T. M., & Cornelissen, F. W. (2006). Visual Search Near Threshold: Some Features Are More Equal than Others. Journal of Vision, 6, 525-540. 61 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Heinke, D., & Humphreys, G. W. (2003). Attention, Spatial Representation, and Visual Neglect: Simulating Emergent Attention and Spatial Memory in the Selective Attention for Identification Model (SAIM). Psychological Review, 110 (1), 29-87. Heinke, D., Humphreys, G., & Tweed, C. (2006). Top-down Guidance of Visual Search: A Computational Account. Visual Cognition, 14, 4-8. Hodsoll J. P., & Humphreys G. W. (2005). The Effect of Target Foreknowledge on Visual Search for Categorically Separable Orientation Targets. Vision Research, 45 (18), 2346-2351. Humphreys, G. W., & Riddoch, M. J. (2001). Detection by Action: Neuropsychological Evidence for Action-Defined Templates in Search. Nature Neuroscience 4(1), 84-88. Ishihara, S. (1981). Ishihara's Tests for Colour-Blindness, 24 Plates Edition. Tokyo, Japan: Kenehara Trading. Kaptein, N. A., Theeuwes, J., & van der Heijden, 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 & Performance, 21, 10531069. Kristánsson, Á., Wang, D., & Nakayama, K. (2002). The Role of Priming in Conjunctive Visual Search. Cognition, 85, 37-52. Kristjánsson, A., Vuilleumier, P., Malhotra, P., Husain, M., & Driver, J. (2005). Priming of Color and Position During Visual Search in Unilateral Spatial Neglect. Journal of Cognitive Neuroscience, 17(6), 859-873. Kristjánsson, A. (2006). Simultaneous but Independent Priming of Different Features of a Single Object in Visual Search. Perception, 35, 164-164. Maljkovic, V., & Nakayama, K. (1994). Priming of Pop-Out. 1. Role of Features. Memory, & Cognition, 22 (6), 657-672. 62 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Mollon, J. D. (1989). “Tho’ she kneel'd in that place where they grew...” The Uses and Origins of Primate Colour Vision. Journal of Experimental Biology, 146 (1), 21-38. Müller, J. M., Reimann, B., & Krummenacher, J. (2003). Visual Search for Singleton Feature Targets Across Dimensions: Stimulus and Expectancy-Driven Effects in Dimensional Weighting. Journal of Experimental Psychology: Human Perception and Performance, 29 (5), 1021-1035. Müller, H. J, & Krummenacher, J. (2006). Visual Search and Selective Attention. Visual Cognition, 14 (4-8), 289-410. Murata, A., Gallese, V., Luppino, G., Kaseda, M., & Sakata, H. (2000). Selectivity for the Shape, Size and Orientation of Objects for Grasping in Neurons of Monkey Parietal Area AIP. Journal of Neurophysiology, 83(5), 2580-2601 Oliver, R. T., & Thompson-Schill, S. L. (2003). Dorsal Stream Activation During Retrieval of Object Size and Shape. Cognitive, Affective & Behavioural Neuroscience, 3(4), 309-322. Olivers, C. N. L., & Meeter, M. (2006). On the Dissociation Between Compound and Present/Absent Tasks in Visual Search: Intertrial Priming is Ambiguity Driven. Visual Cognition, 13 (1), 1-28. Olivers, C. N. L., Meijer, F., & Theeuwes, J. (2006). Feature-Based Memory-Driven Attentional Capture: Visual Working Memory Content Affects Visual Attention. Journal of Experimental Psychology: Human Perception and Performance, 32 (5), 1243-1265. Olejnik, S., Algina, J. (2003). Generalized Eta and Omega Squared Statistics: Measures of Effect Size for Some Common Research Designs. Psychology Methods, 8 (4), 434447. 63 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Poisson, M. E., & Wilkinson, F. (1992). Distractor Ratio and Grouping Processes in Visual Conjunction Search. Perception, 21, 21-38. Schneider, W., Eschman, A., & Zuccolotto, A. (2002). E-Prime Reference Guide. Pittsburgh: Psychology Software Tools Inc. Sobel, K. V., & Cave, K. R. (2002). Role of Salience and Strategy in Conjunction Search. Journal of Experiment Psychology: Human Perception and Performance, 28 (5), 1005-1070. Soto D., Heinke D., Humphreys, G. W., & Blanco M. J. (2005). Early, Involuntary TopDown Guidance of Attention From Working Memory, Journal of Experimental Psychology: Hyman Perception and Performance, 31(2), 248-261. Soto D., Humphreys G. W., & Heinke D. (2006). Working Memory Can Guide Pop-out Search. Vision Research. 46 (6-7), 1010-1018. Soto, D., & Humphreys, G. W. (2007). Automatic Guidance of Visual Attention From Verbal Working Memory. Journal of Experimental Psychology: Human Perception and Performance, 33 (3), 730-737. Theeuwes, J. (1992). Perceptual Selectivity for Color and Form. Perception & Psychophysics, 51, 599–606. Theeuwes, J., Reimann, B., & Mortier, K. (2006). Visual Search for Featural Singletons: No Top-Down Modulation, Only Bottom-up Priming. Visual Cognition, 14, 466-489. Williams, L. G. (1966). The Effect of Target Specification on Objects Fixated During Visual Search. Perception and Psychophysics, 1, 315-318. Wolfe, J. M., Cave, K. R., & Franzel, S. K. (1989). Guided Search: An Alternative To Feature Integration Model For Visual Search. Journal of Experimental Psychology: Human Perception and Performance, 15, 419-433. 64 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Wolfe, J. M., Stewart, M. I., Friedman-Hill, S. R., & O’Connell, K. M. (1992). The Role of Categorization in Visual Search for Orientation. Journal of Experimental Psychology: Human Perception and Performance, 18(1), 34-39. Wolfe, J. M. (1994). Guided Search 2.0: A Revised Model of Visual Search. Psychonomic Bulletin & Review, 1, 202-238. Wolfe J. M. (1998). Visual Search. In: Pashler H., editor. Attention. London UK: University College London Press. Wolfe, J. M., Horowitz, T. S., Kenner, N., Hyle, M., & Vasan, N. (2004). How Fast Can You Change Your Mind? The Speed of Top-Down Guidance in Visual Search. Vision Research, 44, 1411-1426. Zohary, E., & Hochstein, S. (1989). How Serial is Serial Processing in Vision? Perception, 18, 191-200. 65 FEATURAL GUIDANCE IN CONJUNCTION SEARCH APPENDIX: Equating Search Efficiencies The saturation levels of colors used in the experiment were reduced to control for bottom-up dimensional differences in search efficiency (see also Bacon & Egeth, 1997). This was conducted three times (Experiments A-C) in order to ensure that the stimuli used in the different subsequent experiments (Experiments 1-4) were also equated. Experiment A was based around the stimuli used in Experiments 1-2. Experiment B employed stimuli from Experiment 3 (which differed slightly from the earlier stimuli). Experiment C used stimuli from Experiment 4. Experiment A: Stimuli from Experiments 1 and 2 To avoid any differential top-down cueing effects, participants performed an ‘odd-one out’ task in which they searched for a feature singleton that could occur along either the color or the orientation dimension. If the saliencies of the targets along each dimension are equated, then there should be no search advantage for color over orientation-defined targets. Participants also performed a conjunction search task with the same stimuli to check whether search was then inefficient. Method Unless otherwise mentioned the Method was as Experiment 1. Participants. Ten University of Birmingham students, one male, nine female, aged between 18 and 56 (average age 24.9) took part in the experiment. Stimuli. We undertook exploratory experiments varying the color saturation finally settling on the color levels shown in Table 1, which were used for this study. For further details, see Experiment 1. 66 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Design. For the feature-singleton search there were three main independent variables: the defining dimension (color, orientation), array size (7, 11, 15) and target presence (present, absent), while there were only two main independent variables for the conjunction search task (array size, target presence). Procedure. Half the participants performed the feature-singleton task, followed by the conjunction task, while the order was reversed for the other participants. In the feature-singleton tasks, participants were instructed to search for a stimulus that was the odd-one out from the array. First, a fixation cross was present for 1000ms, then a 100ms inter-stimulus interval (ISI) before an array of stimuli with one target and either six or 10 or 14 distractors. The defining dimension and target for each array was varied trial-on-trial in equal numbers. For color-defined search, targets and distractors possessed different colors yet the same orientation (blue horizontal target vs. green horizontal distractors, or green vertical target vs. blue vertical distractors). For orientation-defined search, the target-distractor relationship differed only along the orientation dimension (blue horizontal target vs. blue vertical distractors; or green vertical target vs. green horizontal distractors). When the target was absent (50% of the time), it was replaced by a distractor. Participants undertook 24 practice trials followed by three blocks of 144 experimental trials. The methodology for the conjunction task was identical to that of Experiment 1, except for the following. No cues were presented prior to each trial and the target was present 50% of the time. There were 18 practice trials followed by a single block of 144. Results 67 FEATURAL GUIDANCE IN CONJUNCTION SEARCH RTs. Data were cleaned as in Experiment 1. For the feature singleton task, the data were pooled across trials within the target-defining dimension (color or orientation) and median RTs for each participant were calculated. Target-absent trials were treated as catch trials, so only target-present data were analysed. A two-factor ANOVA (task dimension, array size) was used to analyse the feature-singleton data. This showed a significant effect of array size (F(2,18)=5.523, p=0.013, partial η2=0.38), with RTs decreasing with increasing array size. There was no difference between the RTs for color- and orientation-defined targets (F<1), and no interaction with array size (F<1). Data from the conjunction task were pooled across target type and were analysed using a two-factor ANOVA (array size, target presence). The data were typical of an inefficient search task (Wolfe, 1998). There were significant main effects of array size (F(2,18)=25.95, p<0.001, partial η2=0.742) and target presence (F(1,9)=84.148, p<0.001, partial η2=0.903) and a significant interaction between target presence and array size (F(2,18)=9.179, p=0.002, partial η2=0.505). Targetpresent trials showed a search slope of 19ms/item while the slope for target-absent trials was 85ms/item. The means of median RTs for both feature-singleton and conjunction search are shown in Figure 15 (along with the data from Experiments B and C). Accuracy. In both search tasks performance was accurate and there was no speed-accuracy trade-off. Errors are presented in Table 2. Discussion The data confirm that (i) the orientation-defined targets used here were as salient as the color-defined targets (there were no differences in the slopes of the 68 FEATURAL GUIDANCE IN CONJUNCTION SEARCH search functions), and (ii) search for conjunction targets was relatively inefficient – demonstrating inefficient search is important in order to allow for the possibility that top-down cueing could alter the slope of the search function. Given these patterns of search, the same feature values were used in Experiments 1 and 2. Figure 15. Means (+/- one standard error) of median RTs from Experiments A, B and C, separated by search type (color-, orientation- or conjunction-defined) and presence/absence of target (note there was only target-absent condition in the conjunction task in Experiments A and B). Experiment B: Stimuli from Experiment 3 For Experiment 3, the response in the visual search task was changed from one based on the presence or absence of the target to a compound task (see Olivers & Meeters, 2006), where the response was determined by a search-irrelevant feature. This change to the physical make-up of the stimuli meant possible changes to the saliency of the targets along either the color or orientation dimensions. To ensure that 69 FEATURAL GUIDANCE IN CONJUNCTION SEARCH the stimuli were matched for this experiment, we replicated the odd-one out and conjunction search tasks of Experiment A but using a compound task procedure for the latter. Method As Experiment A, except where outlined below. Participants. Ten University of Birmingham students, three male, seven female, aged between 19 and 25 (average age 21.6) took part. Stimuli. The stimuli were identical to Experiment A, except that small light grey symbols were added to the centre of all the items. These were either an ‘x’ or a ‘+’ and were distributed across distractors and targets. See Experiment 3 for details. Procedure. As Experiment A, except for the following. The response to the conjunction task was indicated by the feature present on the target, with the nature of the response varied to control for handedness. To ensure that this change did not affect the efficiency of search, the number of trials for the conjunction task was increased, with 36 practise trials, followed by two blocks of 96. There were also only two blocks of 144 trials in the feature-singleton task. Results RTs. Cleaning and data analysis followed the procedure for Experiment A, although the conjunction task did not have target-absent data. Mean RTs across participants are shown in Figure 15. The analysis of the singleton feature task revealed a main effect of task dimension (F(1,9)=6.737, p=0.029, partial η2=0.428), with orientation-defined targets detected significantly faster than color-defined targets (685ms vs. 740ms). There was 70 FEATURAL GUIDANCE IN CONJUNCTION SEARCH no main effect of array size (F(2,18)=2.468, p=0.113, partial η2=0.215) or interaction with array size (F<1). A one-factor ANOVA assessed the effects of the array size on the conjunction data. The main effect of array size was reliable (F(2,18)=18.668, p<0.001, partial η2=0.663), with RTs from array size 7 shorter than those from both array size 11 and 15 (p=0.002 for both), although no difference was found between the latter array sizes (p=1). The overall slope of the search function was 33ms/item. Accuracy. In both search tasks performance was accurate and there was no speed-accuracy trade-off. Errors are presented in Table 2. Discussion The data show that (i) while RTs were quicker for orientation-defined targets with these amended stimuli, there were no differences in the slopes of the search functions, (ii) the conjunction search task was relatively inefficient. The similar search slopes for color and orientation-defined targets indicate that color- and orientation- defined targets were equally efficient in attracting attention. It is unclear why an overall RT advantage emerged for orientation targets, but we note that this would go against any tendency for color search to be more efficient under conditions of top-down cueing. Experiment C: Stimuli from Experiment 4 In Experiment 4, we use stimuli that differed slightly from the ones employed in Experiments 1-3. To ensure that saliency was again matched across the dimensions we replicated the ‘odd one out’ and conjunction search tasks with these stimuli, in Experiment C. 71 FEATURAL GUIDANCE IN CONJUNCTION SEARCH Method As Experiment B, except for that outlined below. Participants. Nine University of Birmingham students, two male, seven female, aged between 18 and 41 (average age of 24.3) took part. All participants had self-reported normal or corrected-to-normal vision. Stimuli. The make-up of the stimuli was exactly the same as for Experiment B, except the length and breadth of the bars and features were increased. For details see Experiment 4. Results RTs. Cleaning and data analysis procedures were the same as in Experiments A and B. Mean RTs across participants are shown in Figure 15. The analysis of the feature-singleton data showed no significant effects, with no difference between color- and orientation-defined search (F<1) and no interaction with array size (F<1). Analysis of the conjunction data showed a significant effect of array size (F(2,16)=7.381, p=0.011, partial η2=0.480), with no difference between RTs at array sizes seven and 11 (difference of 24ms, p=1), but RTs significantly longer from trials with 15 stimuli (difference of 294ms, p=0.006). Accuracy. In both search tasks performance was accurate and there was no speed-accuracy trade-off. Errors are presented in Table 2. Discussion The data show that the small changes made to the items here eliminated any small advantage in overall RTs for orientation-defined targets. Moreover, there were 72 FEATURAL GUIDANCE IN CONJUNCTION SEARCH again no differences in the search slopes for color and orientation-defined targets. As in the other control studies, conjunction search proved relatively inefficient. 73