Guidance of Eye Movements During Conjunctive Visual …...effect, Shen, Reingold, and Pomplun (2000) examined participants’ patterns of eye movements, the spatial dis-tribution of

Guidance of Eye Movements During Conjunctive Visual Search: The Distractor-Ratio Effect

Abstract The distractor-ratio effect refers to the find-ing that search performance in a conjunctive visualsearch task depends on the relative frequency of twotypes or subsets of distractors when the total number ofitems in a display is fixed. Previously, Shen, Reingold,and Pomplun (2000) examined participants’ patterns ofeye movements in a distractor-ratio paradigm anddemonstrated that on any given trial saccadic endpointswere biased towards the smaller subset of distractorsand participants flexibly switched between differentsubsets across trials. The current study explored theboundary conditions of this tendency to flexibly searchthrough a smaller subset of distractors by examiningthe influence of several manipulations known to modu-late search efficiency, including stimulus discriminabili-ty (Experiment 1), within-dimension versus cross-dimension conjunction search and distractor hetero-geneity (Experiment 2). The results indicated that theflexibility of visual guidance and saccadic bias exempli-fied by the distractor-ratio effect is a robust phenome-non that mediates search efficiency by adapting tochanges in the relative informativeness of stimulusdimensions and features.

Visual search is one of the dominant paradigms usedfor investigating visual attention. In a typical conjunc-tive search task, participants have to decide whether aprespecified search target is embedded in an array ofdistractors (nontarget elements), each of which sharesone or multiple features with the target item. Each trialcontains an equal number of distractors from each typewhereas the total number of items within a search dis-play (display size) is manipulated. Search efficiency isexamined by the change in response time and/or errorrate as a function of display size (see Treisman, 1988;Wolfe, 1998 for a review). Based on the response timeand error rate data, several models of visual searchhave been proposed to explain search efficiency acrossa variety of search tasks (e.g., the original feature-inte-

gration theory, Treisman, 1988; Treisman & Gelade,1980; the attentional-engagement theory by Duncan &Humphreys, 1989; the guided-search model by Wolfe,1994; Wolfe, Cave, & Franzel, 1989; see also the revisedfeature-integration theory by Treisman & Sato, 1990).

Several previous studies have shown that search per-formance in a conjunctive search task is also sensitiveto the distribution of distractor types, even when thetotal number of items in a display remains constant(e.g., Bacon & Egeth, 1997; Egeth, Virzi, & Garbart,1984; Kaptein, Theeuwes, & van der Heijden, 1995;Poisson & Wilkinson, 1992; Zohary & Hochstein, 1989).For example, Zohary and Hochstein (1989) adopted aColour x Orientation conjunction search task and askedparticipants to decide whether a red horizontal bar waspresent among an array of red vertical (same-colourdistractors) and green horizontal (same-orientation dis-tractors) bars. The search display was presented verybriefly (50 ms) and then, after a variable interval (stim-ulus onset asynchrony, SOA), masked. One criticalmanipulation in this study was the ratio between thetwo types of distractors (same-colour vs. same-orienta-tion) presented in a given array. Zohary and Hochsteinfound that the SOA required to reach a 70% correctresponse rate was a quadratic function of the numberof distractors sharing colour with the search target.Specifically, detection was relatively easy for displayswith extreme distractor ratios (i.e., either the same-colour or same-orientation distractors were rare) butrelatively difficult for displays in which the two types ofdistractors were equally represented. The finding thatvisual-search efficiency in a conjunctive search taskdepends on the relative frequency of the two types ofdistractors has been referred to as the distractor-ratioeffect (Bacon & Egeth, 1997).

Different interpretations have been proposed toexplain the distractor-ratio effect. Zohary andHochstein (1989) suggested that, upon the presentationof a search display, an a priori decision is made con-cerning the approximate ratio of the number of itemsin one subset (as defined by a specific feature shared

Jiye Shen and Eyal M. Reingold, University of Toronto

Marc Pomplun, University of Massachusetts at Boston

Canadian Journal of Experimental Psychology, 2003, 57:2, 76-96

DISTRACTOR-RATIO EFFECT 77

with the target item) relative to the number of items inthe other subset. Such information is then used to seg-ment the search display. Participants scanned a small-er subset of distractors and switched between differentsubsets from trial to trial (henceforth subset-switchingaccount). In contrast, Poisson and Wilkinson (1992)suggested that in a search task with distractor-ratiomanipulation, participants do not switch feature subsetsfrom trial to trial. Instead, participants restrict theirsearch to items belonging to the dominant feature sub-set (i.e., colour in their study). An item-by-item searchthrough a single feature set would produce responsetimes that increase linearly with increasing number ofitems in the selected subset. However, when most ofthe items in the display belong to one feature set,response times begin to decrease rather than continueto increase with increasing number of items in that sub-set. Poisson and Wilkinson argued that this decrease inresponse time is due to the operation of strong distrac-tor grouping because those search displays containlarge clusters of homogeneous distractors, which mayenable participants to reject them as a group (hence-forth distractor-grouping account).

To test between the subset-switching account andthe distractor-grouping account for the distractor-ratioeffect, Shen, Reingold, and Pomplun (2000) examinedparticipants’ patterns of eye movements, the spatial dis-tribution of saccadic endpoints in particular, during thesearch process. They employed a Colour x Shape con-junction search task and systematically manipulated theratio between the same-colour and same-shape distrac-tors in a display. They found a quadratic change insearch performance measures such as manual responsetime, number of fixations per trial, and initial saccadiclatency as a function of distractor ratio. Search perfor-mance was worse when the ratio between the same-colour and same-shape distractors approximated 1:1and gradually improved as the ratio deviated from 1:1,with performance being best at extreme distractorratios (i.e., very few distractors of one type). Moreimportantly, Shen et al. demonstrated that when therewere very few same-colour distractors, participants’saccadic end points were biased towards the colourdimension whereas when there were very few same-shape distractors, saccades were biased towards theshape dimension. Results from that study suggest thatin a distractor-ratio paradigm, participants take advan-tage of the display information and flexibly switchbetween different subsets of distractors on a trial-by-trial basis (see Zohary & Hochstein, 1989). This rulesout distractor grouping (e.g., Poisson & Wilkinson,1992) as a viable explanation for the distractor-ratioeffect, although this does not deny that distractorgrouping can occur.

Overview of the Current Study

The goal of the present study was to determine thegenerality of the flexible subset-switching effectdemonstrated by Shen et al. (2000) and to explore thepossible role played by this mechanism in mediatingsearch efficiency. Accordingly, the current studyapplied several manipulations known to modulatesearch efficiency to the distractor-ratio paradigm,including stimulus discriminability (e.g., Driver &McLeod, 1992; Duncan & Humphreys, 1989; Nagy &Sanchez, 1990; Neisser, 1967; Palmer, Verghese, &Pavel, 2000; Pashler, 1987; Rayner & Fisher, 1987;Theeuwes, 1992; D. E. Williams & Reingold, 2001;Wolfe, 1994), within-dimension versus cross-dimensionconjunction search (e.g., Carrasco, Ponte, Rechea, &Sampedro, 1998; Linnell & Humphreys, 2001; Wolfe,Friedman-Hill, & Bilsky, 1994; Wolfe, Yu, Stewart,Shorter, Friedman-Hill, & Cave, 1990) and distractorheterogeneity (e.g., Duncan & Humphreys, 1989;Kaptein et al., 1995; Palmer et al., 2000; Pashler, 1987).

Specifically, Experiment 1 was designed to examinethe effect of stimulus discriminability on search perfor-mance and saccadic selectivity in a search task withdistractor-ratio manipulation. This experimentemployed a Colour x Shape conjunction search taskand manipulated the discriminability along the shapedimension. Two search conditions, a high-discrim-inability condition and a low-discriminability condition,were included. If the distractor-ratio effect is merelydue to the unequal representation of the two subsets ofdistractors in a search display and participants consis-tently search through the smaller subset, very similarpatterns of search performance and saccadic biasshould be observed across the two conditions.However, if stimulus discriminability plays a role inmediating the distractor-ratio effect, lower discrim-inability along the shape dimension should lead togreater colour-shape asymmetry in search performanceand saccadic selectivity.

In Experiment 2, the distractor-ratio manipulationwas implemented in the context of within-dimensionconjunction searches. Previously mentioned studies onthe distractor-ratio effect (Bacon & Egeth, 1997; Poisson& Wilkinson, 1992; Shen et al., 2000; Zohary &Hochstein, 1989) examined cross-dimension conjunc-tion search tasks only. However, in a series of experi-ments, Wolfe et al. (1990) demonstrated that searchesfor a conjunction of two features from the same dimen-sion (e.g., Colour x Colour, Orientation x Orientation)were less efficient than searches for a conjunctionacross two dimensions (e.g., Colour x Orientation).Wolfe et al. 1990 (see also Wolfe, 1994, 1996; Wolfe etal., 1994, 1996) argued that whereas cross-dimension

78 Shen, Reingold, and Pomplun

conjunctions can be searched in a parallel fashion,within-dimension conjunctions are necessarily searchedin a serial self-terminating fashion. Given this argu-ment that within-dimension conjunction searches areprocessed in a qualitatively different fashion fromcross-dimension conjunction searches, the main goal ofExperiment 2 was to examine whether the distractor-ratio effect is also observed in within-dimension con-junction search tasks by employing a Colour x Colourwithin-dimension conjunction search task. In addition,Experiment 2 manipulated independently distractor-ratio and distractor heterogeneity. Although the influ-ence of distractor heterogeneity on visual search per-formance is well established (e.g., Duncan &Humphreys, 1989; Kaptein et al., 1995; Palmer et al.,2000; Pashler, 1987), no previous study so far hasexamined this variable in the context of the distractor-ratio effect.

As in Shen et al. (2000), the present investigationemployed eye-movement techniques in order to sup-plement the standard response-time and error rate dataand to obtain detailed spatio-temporal informationabout the search process. Thus search performancemeasures used included response time, accuracy, num-ber of fixations per trial, initial saccadic latency, fixationduration, and saccadic amplitude as well as saccadicselectivity (i.e., the bias in the distribution of saccadicend points).

Experiment 1

In Experiment 1, we examined the effect of stimulusdiscriminability on search performance and saccadicselectivity in a Colour x Shape conjunction search taskwith distractor-ratio manipulation. One finding fromShen et al. (see also Poisson & Wilkinson, 1992; Zohary& Hochstein, 1989) is that almost all of the search mea-sures examined (manual response time, fixation num-ber, saccadic latency, and saccadic bias) were asymmet-rical, suggesting that the colour dimension had greatersaliency or discriminability than the shape dimension.The importance of stimulus discriminability is wellestablished in the visual search literature and has beenincorporated into major theoretical frameworks of visu-al search (e.g., Duncan & Humphreys, 1989; Palmer,Verghese, & Pavel, 2000; Wolfe, 1994). Numerous pre-vious studies have demonstrated that detecting a searchtarget becomes more difficult with increased similarity(i.e., lower discriminability) between the target itemand the distractors (e.g., Nagy & Sanchez, 1990;Neisser, 1967; Pashler, 1987; Rayner & Fisher, 1987).

In conjunction search tasks, several previous studieshave reported that changing the stimulus discriminabili-ty along one dimension may drastically alter the search

performance (e.g., Driver & McLeod, 1992; Theeuwes,1992; D. E. Williams & Reingold, 2001). For example,D. E. Williams and Reingold (2001) found that inColour x Shape x Orientation triple conjunction searchtasks, changing the discriminability along the shapedimension (high-discriminability: C vs. T; low-discrim-inability: E vs. F) induced different patterns of eyemovements. Greater saccadic selectivity towards thosedistractors sharing shape with the target was observedwhen more discriminable shapes were employed.Results from this study pointed to the fact that searchperformance in conjunction search tasks was stronglyinfluenced by the relative discriminability between dif-ferent dimensions.

Thus, in addition to the distractor-ratio manipulation,the current experiment employed a Colour x Shapeconjunction search task and varied the discriminabilityalong the shape dimension. Two search conditions, ahigh-discriminability condition and a low-discriminabili-ty condition, were included. As in Shen et al. (2000),the shapes X versus O were used in the high-discrim-inability condition whereas in the low-discriminabilitycondition, the shapes were X versus K for half of theparticipants and O versus Q for the other half.

MethodParticipants. Eight undergraduate students at the

University of Toronto were tested individually in two 1-hour sessions. All participants had normal or correct-ed-to-normal visual acuity and normal colour vision.They were naïve with respect to the purpose of theexperiment and received $20 for their participation.

Apparatus. The eyetracker employed in the currentstudy was the SR Research Ltd. EyeLink system. Thissystem has high spatial resolution (0.005º) and highsampling rate (250 Hz). The EyeLink headband hasthree cameras, allowing simultaneous tracking of botheyes and of head position for head-motion compensa-tion. By default, only the participant’s dominant eyewas tracked in our studies. The EyeLink system usesan Ethernet link between the eyetracker and displaycomputers for real-time saccade and gaze-position datatransfer. In the present investigation, the configurableacceleration and velocity thresholds were set to detectsaccades of 0.5° or greater.

Stimulus displays were presented on two monitors,one for the participant (a 19-inch Samsung SyncMaster900P monitor with a refresh rate of 120 Hz and a screenresolution of 800 x 600 pixels) and the other for theexperimenter. The experimenter monitor was used togive feedback in real-time about the participant’s com-puted gaze position. This feedback was given in theform of a gaze cursor measuring 1° in diameter that


was overlaid on the same image being viewed by theparticipant. This allowed the experimenter to evaluatesystem accuracy and to initiate a recalibration if neces-sary. In general, the average error in the computationof gaze position was less than 0.5°.

Stimuli and design. Similar to Shen et al. (2000),four possible targets, a red X, a green X, a red O, and agreen O, were used. Participants searched for the tar-get item among distractors that shared either colour(same-colour distractor) or shape (same-shape distrac-tor) with the target. A high-discriminability conditionand a low-discriminability condition were included inthe current experiment. In the high-discriminabilitycondition, a pair of highly discriminable shapes, X ver-sus O, was used (see Figure 1A for an example). In thelow-discriminability condition, the same-colour distrac-tors were chosen to be visually similar to the searchtarget based on the interconfusability matrix reportedby van der Heijden, Malhas, and van den Roovart(1984). When the target shape was X, the same-colourdistractors were Ks and when the target shape was O,the same-shape distractors were Qs (see Figure 1B foran example). In both the high- and low-discriminabili-ty conditions, items of different colours (red, with CIE

xy-chromaticity coordinates of .582/.350 and green,with CIE xy-coordinates of .313/.545) were matched inluminance (20 cd/m2) and presented on a white back-

ground of 60.7 cd/m2. In any given display, the searchtarget could be present or absent with equal probabili-ty.

For both the high- and low-discriminability condi-tions, the total number of items presented in a displaywas fixed at 36. All display items were presented in a15.5º x 15.5º visual field at a viewing distance of 91 cm.Each individual item subtended 1.0º vertically and 0.8ºhorizontally. In target-absent trials, 3 to 33 (in multi-ples of 3) distractors shared colour with the targetwhereas the rest of the distractors shared shape withthe target, yielding 11 levels of distractor ratio betweenthe same-colour and same-shape distractors (3:33, 6:30,9:27, 12:24, 15:21, 18:18, 21:15, 24:12, 27:9, 30:6, and33:3). Target-present trials were created by randomlyreplacing one of the distractor items in a target-absentdisplay with the search target. Participants performed atotal of 1,320 trials in two individual sessions, whichamounted to 30 trials for each cell of the design (TargetPresence x Distractor Ratio x Search Condition). Thehigh- and low-discriminability conditions were tested inalternating blocks. At the beginning of each session,participants received a practice block of 44 trials, withone trial for each possible combination of target pres-ence, search condition, and distractor ratio.

Procedure. The experiment was run in a lightedroom, with the luminance of the walls being approxi-

Figure 1. Sample search displays used in Experiment 1. Red stimuli shown in white; green stimuli shown in black. Target was a green Xand the distractors were red Xs and green Os (in the high-discriminability condition; Panel A) or green Ks (in the low-discriminability condi-tion; Panel B). Both examples corresponded to a trial with a distractor ratio of 6:30 between the same-colour distractors and same-shape dis-tractors.

A B


mately 30 cd/m2. Participants were informed of theidentities of the target and distractor items before theexperiment started. They were asked to look for thesearch target and indicate whether it was in the displayor not by pressing an appropriate button as quicklyand accurately as possible. Participants were not givenany information about distractor ratio manipulation. A9-point calibration procedure was performed at thebeginning of the experiment, followed by a 9-point cal-ibration accuracy test. Calibration was repeated if anypoint was in error by more than 1º or if the averageerror for all points was greater than 0.5º. Each trialstarted with a drift correction in the gaze position.Participants were instructed to fixate on a black dot inthe centre of the computer screen and then press astart button to initiate a trial. The trial terminated ifparticipants pressed one of the response buttons or ifno response was made within 20 seconds. The timebetween display onset and the participant’s responsewas recorded as the response time.

ResultsTrials with an incorrect response (1.7% of the trials inthe high-discriminability condition and 3.3% in the low-discriminability condition) were excluded from furtheranalysis. In addition, those trials with a saccade orblink overlapping the onset of a search display (1.1%),or with a response time that was more than 3.0 stan-

dard deviations above or below the mean (1.1%) wereexcluded from further analysis. Following Shen et al.(2000), separate repeated-measures ANOVAs were con-ducted on response time, number of fixations per trial,initial saccadic latency, fixation duration, and saccadicamplitude with target presence (2: present vs. absent),search condition (2: high- vs. low-discriminability), anddistractor ratio (11 levels) as within-subject factors. Inaddition, the bias in the distribution of saccadic end-points as a function of search condition and distractorratio was examined.

Response time and number of fixations per trial.Figure 2 plots the mean and standard error of responsetime (Panel A) and number of fixations per trial (PanelB) as a function of target presence, search condition,and distractor ratio. The repeated-measures ANOVA

revealed that all of the main effects, two-way andthree-way interactions were significant (all Fs > 15.50,ps < .001 for response time and Fs > 15.73, ps < .001for number of fixations per trial). It is clear from thefigure that, in the high-discriminability condition,response time and fixation number varied quadraticallyas a function of distractor ratio. Those displays withequal number of same-colour and same-shape distrac-tors yielded longer response times and more fixationsthan did those displays with very extreme distractorratio (i.e., very few same-colour distractors or very few

Figure 2. Response times (in ms; Panel A) and number of fixations per trial (Panel B) as a function of target presence and the num-ber of same-colour distractors in both the high-discriminability condition and the low-discriminability condition in Experiment 1.For all figures, error bars represent the standard error of the mean calculated on an across-subject basis.

A B


same-shape distractors). In addition, longer responsetimes and more fixations were observed in those dis-plays with fewer same-shape distractors than in dis-plays with a comparable number of same-colour dis-tractors. A further trend analysis revealed that the qua-dratic trend accounted for 61.3% of the total variabilityin response time that was due to the distractor ratiomanipulation and 61.9% in fixation number, whereasthe linear trend accounted for 34.6% of the variabilityin response time and 30.9% in fixation number only.

In the low-discriminability condition, response timeand fixation number increased with increasing numberof same-colour distractors. Unlike the high-discrim-inability condition, for both target-absent and target-present trials, there was no drop in response time andfixation number even when there were very few same-colour distractors. Trend analysis shows that the lineartrend, which accounted for 96.3% of the variability inresponse time that was due to the distractor-ratiomanipulation and 93.9% in fixation number, was morepronounced than the quadratic trend, which accountedfor only 3.4% of the variability in response time and5.8% in fixation number.

Thus, the current analysis indicates that the distrac-tor-ratio effect was strongly influenced by the discrim-inability of stimulus dimensions. When the stimulusdimensions are highly discriminable, participants takeadvantage of the informativeness of the dimensionsand achieve greater search efficiency by flexibly

searching through the smaller subset. However, whenone of the dimensions becomes less discriminable, par-ticipants have to consistently search through the subsetof distractors belonging to the dominant dimension(colour in the present case).

Initial saccadic latency. Figure 3 shows initial sac-cadic latency as a function of search condition and dis-tractor ratio in both the target-absent trials (Panel A)and target-present trials (Panel B). The repeated-mea-sures ANOVA revealed a significant main effect forsearch condition, F(1, 7) = 9.18, p < .05, with the over-all initial saccadic latency shorter in the low-discrim-inability condition than in the high-discriminability con-dition. This difference was more pronounced in target-absent trials than in target-present trials. This was indi-cated by a significant interaction between search condi-tion and target presence, F(10, 70) = 11.23, p < .001.The figure also shows that initial saccadic latency var-ied as a function of distractor ratio, F(10, 70) = 33.46, p< .001. Similar to the above analyses on response timeand fixation number, the quadratic trend was more pro-nounced in the high-discriminability condition (lineartrend: 17.0%; quadratic trend: 79.1%) whereas the lin-ear trend was dominant in the low-discriminability con-dition (linear trend: 81.7%; quadratic trend: 10.9%).This was confirmed by a significant Search Condition xDistractor Ratio interaction, F(10, 70) = 4.32, p < .001.

Figure 3. Initial saccadic latency (in ms) as a function of the number of same-colour distractors in target-absent (Panel A) and target-present trials (Panel B) in both the high-discriminability condition and the low-discriminability condition in Experiment 1.

A B


Figure 4. Fixation duration (in ms) as a function of the number of same-colour distractors in target-absent (Panel A) and target-present trials(Panel B) in both the high-discriminability condition and the low-discriminability condition in Experiment 1.

A B

Figure 5. Saccadic amplitude (in degrees of visual angle) as a function of the number of same-colour distractors in target-absent (Panel A)and target-present trials (Panel B) in both the high-discriminability condition and the low-discriminability condition in Experiment 1.

A B


Fixation duration. Figure 4 plots fixation durationas a function of target presence, distractor ratio, andsearch condition. The repeated-measures ANOVA

revealed a significant main effect of search condition,F(1, 7) = 21.53, p < .01, with a shorter fixation durationin the low-discriminability condition than in the high-discriminability condition. There was also a significantmain effect of target presence, F(1, 7) = 22.13, p < .01,distractor ratio, F(10, 70) = 5.49, p < .001, as well as asignificant target presence x distractor ratio interaction,F(10, 70) = 2.02, p < .05. It is clear from the figure that,in target-present trials (Panel B), fixation durationremained roughly the same across the whole range ofdistractor ratio for both search conditions – neither thelinear nor the quadratic trend was significant, both Fs(1, 7) < 3.32, ps > .111. However, a U-shape patternwas observed in target-absent trials, with longer fixa-tion duration for displays with extreme distractor ratios(this is especially true for the low-discriminability con-dition). This was verified by a significant quadratictrend, F(1, 7) = 7.67, p < .05, for the distractor-ratiomanipulation.

Saccadic amplitude. Figure 5 presents saccadicamplitude as a function of target presence, distractorratio, and search condition. The repeated-measurerevealed a significant search condition x target pres-

ence x distractor ratio interaction, F(10, 70) = 6.93, p <.001. It is clear from the figure that in target-absent tri-als (Panel A), saccadic amplitude remained roughlyconstant in the high-discriminability condition. Neitherthe linear nor the quadratic trend was significant, bothFs(1, 7) < 4.42, ps > .073. In contrast, as the searchbecame progressively more difficult with increasingnumber of same-colour distractors, saccadic amplitudein the low-discriminability condition decreased steadily(about one degree). This was verified by a significantlinear trend, F(1, 7) = 54.08, p < .001, in the trendanalysis for the main effect of distractor-ratio. In target-present trials (Panel B), there was no differencebetween the high-discriminability condition and thelow-discriminability condition. Saccadic amplitudedecreased slightly with increasing number of same-shape distractors; this was verified by a significant lin-ear trend, F(1, 7) = 25.05, p < .01.

Saccadic selectivity. For each fixation, the distancebetween the fixation position and every display itemwas computed and the fixation was assigned to theclosest item. The number of fixations assigned to eachtype of distractors (same-colour vs. same-shape distrac-tors) was then summed to assess saccadic selectivityduring the search process. As pointed out by Zelinsky

Figure 6. Panel A: Relative frequency of saccades directed towards the same-colour distractors as a function of stimulus discriminability andthe number of same-colour distractors in Experiment 1. The diagonal line indicates chance performance. Panel B: Saccadic bias (the differ-ence between the observed frequency and chance performance) as a function of stimulus discriminability and the number of same-colourdistractors in Experiment 1.

A B


(1996), results from target-absent trials can be interpret-ed more clearly than those from target-present trialswhere the presence of the target item may influencesearch behaviour (participants tended to fixate on thesearch target before making a manual response).Therefore, only target-absent trials were included in thecurrent analysis. Figure 6A plots the observed frequen-cies of saccades directed towards the same-colour dis-tractors for both the high- and low-discriminability con-ditions. The diagonal line in the figure depicts theprobability of fixating same-colour distractors in theabsence of selectivity (i.e., chance performance).Figure 6B depicts saccadic bias towards the colourdimension (the difference between the observed fre-quency and chance performance) in both conditions.

A repeated-measures ANOVA on saccadic bias withsearch condition (2: high- vs. low-discriminability) anddistractor ratio (11 levels) as within-subject factorsrevealed a significant main effect of search condition,F(1, 7) = 159.39, p < .001, and distractor ratio, F(10, 70)= 85.62, p < .001, whereas the interaction between thetwo factors was not significant, F(10, 70) = 1.11, p =.365. It is clear from the figure that in the high-discrim-inability condition, when there were only very few dis-tractors sharing colour with the search target, saccadicend points were biased towards the colour dimension.On the other hand, when there were very few same-

shape distractors, a robust bias towards the shapedimension was observed. In the low-discriminabilitycondition, however, a consistent bias towards thecolour dimension was found, even in displays withvery few same-shape distractors. For both the high-and low-discriminability conditions, the change in sac-cadic bias was almost linear – the linear trend account-ed for 99.2% of the total variability in saccadic bias thatwas due to the distractor-ratio manipulation.

Following Shen et al. (2000), saccadic bias for boththe first and subsequent saccades was calculated toexamine the temporal dynamics of visual guidance. A2 (Saccade Sequence) x 2 (Search Condition) x 11(Distractor ratio) repeated-measures ANOVA revealed asignificant Saccade Sequence x Search Condition xDistractor Ratio interaction, F(10, 70) = 4.29, p < .001.Figure 7A shows that, in the high-discriminability con-dition, in those displays with extreme distractor ratios,saccade bias was stronger for the first saccades than forthe subsequent ones. Pairwise t-tests revealed a signifi-cant difference in saccadic bias for those displays with6, 9, and more than 21 same-colour distractors (i.e.,fewer than 15 same-shape distractors), all ts (7) > 2.92,ps < .05. In marked contrast, the spatial bias in thelow-discriminability condition was not influenced bythe saccade sequence in a trial across the whole rangeof distractor-ratio manipulation (see Figure 7B).

Figure 7. Saccadic bias of the first and subsequent saccades as a function of the number of same-colour distractors in the high-discriminabili-ty condition (Panel A) and the low-discriminability condition (Panel B) in Experiment 1.

A B


DiscussionThe current experiment examined the effect of stimulusdiscriminability on search performance and saccadicselectivity in a search task with distractor-ratio manipu-lation. Similar to Shen et al. (2000), response time,number of fixations per trial, and initial saccadic laten-cy varied quadratically as a function of distractor ratioin the high-discriminability condition. The saccadicselectivity analysis indicated that participants searchedthrough different subsets of distractors on the basis ofdistractor ratio (i.e., colour subset for displays with veryfew same-colour distractors and shape subset for dis-plays with very few same-shape distractors). In thelow-discriminability condition, although a quadraticchange in response time, number of fixations per trial,and initial saccadic latency was still observed, the lineartrend was much more pronounced and accounted forthe majority of the variability. Guidance analysisrevealed that saccades were consistently biasedtowards the colour dimension, irrespective of the dis-tractor-ratio manipulation. This indicates that, acrossall displays, participants searched through the subset ofdistractors that shared colour with the target item.

Neither a strict smaller-subset search strategy nor thedistractor-grouping account can account for the differ-ent responses to the distractor-ratio manipulationbetween the high- and low-discriminability conditions.This is because if a strict smaller-subset search strategyis adopted and participants consistently scannedthrough the subset with fewer items, the overall patternof results should not differ between the high- and low-discriminability conditions. Although a constant biastowards the colour subset in the low-discriminabilitycondition may be consistent with the distractor group-ing account, this model had difficulty in explaining theflexible subset-switching observed in the high-discrim-inability condition. Thus, the finding of a difference insearch performance and saccadic bias between thehigh- and low-discriminability conditions revealed thatthe distractor-ratio effect was due to the relative dis-criminability of the two stimulus dimensions and thatparticipants searched through the more informative, butnot necessarily smaller, subset of distractors. This find-ing is consistent with D. E. Williams and Reingold(2001), showing that guided search flexibly accommo-dates to changes in the informativeness of stimulusdimensions.

One interesting finding to emerge from the currentexperiment is that, for any given distractor ratio, sac-cadic bias towards the colour subset was stronger (byabout 20%) in the low-discriminability condition than inthe high-discriminability condition (see Figure 6B).This was the case even for those displays with very fewsame-colour distractors. It appears that in the low-dis-

criminability condition, given the poor utility of theshape dimension, saccadic selectivity was stronglybiased towards the colour dimension. Thus, the pre-sent experiment demonstrated that a minimal level ofdiscriminability in the nondominant stimulus dimensionis a prerequisite for obtaining the distractor-ratio effect.

Experiment 2

There were two goals in Experiment 2. Given theargument by Wolfe and his colleagues (Wolfe, 1994,1996; Wolfe et al., 1990) that within-dimension conjunc-tion searches are processed in a qualitatively differentfashion from cross-dimension conjunction searches, thefirst goal of the current experiment was to examinewhether the distractor-ratio effect is also observed inwithin-dimension conjunction search tasks. In the cur-rent experiment, a Colour x Colour within-dimensionconjunction search task was employed. Participantswere asked to search for a red/blue target among agroup of distractors that only shared the red colour andanother group of distractors that only shared the bluecolour. If the within-dimension conjunction search iscarried out in a serial self-terminating fashion acrossthe whole display, search performance and saccadicselectivity should not vary as a function of distractorratio.

The second goal of the current experiment was toexamine how distractor heterogeneity influences searchefficiency and saccadic selectivity. Numerous previousstudies have demonstrated that distractor heterogeneity,which decreases the grouping of distractors, has amajor impact on visual search efficiency (e.g., Duncan& Humphreys, 1989; Kaptein et al., 1995; Palmer et al.,2000; Pashler, 1987). Although the effect of distractorheterogeneity on visual search performance is wellestablished, no previous study so far has examinedhow distractor heterogeneity influences the patterns ofeye movements (fixation number, saccadic latency, sac-cadic amplitude, fixation duration, and saccadic selec-tivity) during the search process. Thus, in addition tothe distractor-ratio manipulation, the current experi-ment also manipulated independently the degree ofdistractor heterogeneity.

MethodParticipants. Eight participants from the same sub-

ject pool as in the previous experiment participated intwo 1-hour sessions. They were naïve with respect tothe purpose of the experiment and none of them hadparticipated in the previous experiment.

Stimuli and design. Participants performed a Colourx Colour conjunction search task. All display items


were a cross composed of two bars of different colours(red, blue, green, or yellow). For all participants, thesearch target was a conjunction of a red bar and a bluebar (red/blue), irrespective of their specific orienta-tions. The distractors were chosen in such a way that asubset of them shared the red colour with the target,which conjoined with another nonblue colour (i.e.,red/yellow or red/green), and the rest of the distractorsshared the blue colour with the search target, whichconjoined with a nonred colour (i.e., blue/yellow orblue/green).

The degree of heterogeneity of the distractors wasvaried in the current experiment. In one condition, thedistractor heterogeneity was relatively low (henceforththe homogeneous condition) – there were only twotypes of distractors in a single display: either thered/yellow and blue/green crosses in half of the trialsor the red/green and blue/yellow combinations in theother half of the trials (see Figure 8A for an example).In a second condition, the distractor heterogeneity wasmade relatively high by presenting all four types of dis-tractors in a single display (henceforth the heteroge-neous condition). Those items with a red bar weresplit evenly between the red/yellow and red/greencombinations whereas those items having a blue barwere divided evenly between the blue/green andblue/yellow combinations (see Figure 8B for an exam-ple). The CIE xy-coordinates for the colours were: red

(.497, .332), green (.309, .556), blue (.188, .175), andyellow (.439, .461). The items had an average lumi-nance of 23.7 cd/m2 and were presented on a whitebackground of 60.7 cd/m2.

Each search display consisted of 20 items, presentedin a 15.5º x 15.5º field at a viewing distance of 91 cm.Each individual item subtended 1.4º both horizontallyand vertically. The minimum distance between thecentres of neighbouring items was set at 2.3º. In target-absent trials, 2 to18 (in multiples of 2) distractorsshared the red colour with the search target and therest of the distractors shared the blue colour with thetarget. Therefore, there were nine possible ratios ofdistractors having a red bar to distractors having a bluebar (2:18, 4:16, 6:14, 8:12, 10:10, 12:8, 14:6, 16:4, and18:2).

The current experiment adopted a three-factor with-in-subject design. These factors were as follows: targetpresence (2: present vs. absent), search condition (2:homogeneous vs. heterogeneous), and distractor ratio(9 levels). All these factors were completely crossed.Participants performed 33 trials for each cell of thedesign, constituting a total of 1,188 experimental trials.They completed the experiment in two individual ses-sions. At the beginning of each session, participantsreceived 36 practice trials, with one trial for each cell ofthe design. The same apparatus and experimental pro-cedure as in the previous experiment were followed.

Figure 8. Sample search displays used in Experiment 2. Target was a red-blue conjunction ( = blue; = red; = green; =yellow). In the homogeneous condition (Panel A), only two types of distractors were used whereas in the heterogeneous condition (PanelB), four types of distractors were used. Panel A corresponded to a search display with a distractor ratio of 2:18 between the red and blueitems whereas Panel B illustrated a search display with a distractor ratio of 10:10.

A B


ResultsTrials with an incorrect response were excluded fromfurther analysis. This resulted in the exclusion of 2.1%of the trials in the homogeneous condition and 3.1% inthe heterogeneous condition. In addition, trials with asaccade or a blink overlapping the onset of a searchdisplay, or with an excessively long or short responsetime, were dropped from analysis. These exclusionsaccounted for 0.7% and 0.6% of all trials, respectively.Similar to the previous experiment, search performancemeasures (response time, number of fixations per trial,initial saccadic latency, fixation duration, and saccadicamplitude) were subjected to separate 2 (TargetPresence: present vs. absent) x 2 (Search Condition:homogenous vs. heterogeneous) x 9 (Distractor Ratio)repeated-measures ANOVAs. Following that, the bias inthe spatial distribution of saccadic endpoints was exam-ined as a function of search condition and distractorratio.

Response time and number of fixations per trial.Figure 9 plots response time (Panel A) and number offixations per trial (Panel B) as a function of target pres-ence and distractor ratio in both the homogeneous con-dition and the heterogeneous condition. The repeated-measures ANOVA revealed a significant main effect oftarget presence, F(1, 7) = 64.70, p < .001, for response

time and F(1, 7) = 51.44, p < .001, for number of fixa-tions per trial. This indicates that target-absent trialstypically had a longer response time and more fixa-tions. Figure 9 shows that response time and fixationnumber varied as a function of distractor ratio, withlonger RTs and more fixations for displays in the mid-dle range of distractor ratio than for displays withextreme distractor ratios. This is indicated by a signifi-cant main effect of distractor ratio, F(8, 56) = 83.50, p <.001, for response time, and F(8, 56) = 91.17, p < .001for number of fixations per trial. Unlike the previousexperiment on cross-dimensional searches, theresponse time, and fixation number curves in the cur-rent experiment were relatively symmetrical. This isconfirmed by a trend analysis, showing a strong qua-dratic trend (accounting for 97.9% of the total variabili-ty in response time that was due to the distractor-ratiomanipulation and 97.0% in fixation number) and aweak linear trend (only accounting for 0.4% of totalvariability in response time and 0.8% in fixation num-ber). In addition, the distractor-ratio effect wasstronger in target-absent trials than in target-present tri-als. This was indicated by a significant interactionbetween target presence and distractor ratio forresponse time, F(8, 56) = 37.38, p < .001, and for fixa-tion number, F(8, 56) = 14.23, p < .001.

Figure 9 also shows that visual search was less effi-

Figure 9. Response times (in ms; Panel A) and number of fixations per trial (Panel B) as a function of target presence and the number of reddistractors in the homogeneous condition and the heterogeneous condition in Experiment 2.

A B


cient in the heterogeneous condition than in the homo-geneous condition. This was indicated by a significantmain effect of search condition for both response time,F(1, 7) = 45.30, p < .001, and fixation number, F(1, 7) =34.86, p < .001. In addition, this effect was more pro-nounced in target-absent trials than in target-present tri-als as indicated by a significant Target Presence xSearch Condition interaction for both response time,F(1, 7) = 21.84, p < .001, and fixation number, F(1, 7) =14.35, p < .001. It is clear from the figure that theeffect of distractor heterogeneity was observed acrossthe whole range of distractor-ratio manipulation, all ts(7) > 2.70, ps < .05, except for those displays with fourred items, t < 1. In addition, there was a significantSearch Condition x Distractor Ratio interaction forresponse time, F(8, 56) = 3.03, p < .05, and for fixationnumber, F(8, 56) = 2.37, p < .05, showing that the effectof distractor heterogeneity was more pronounced indisplays with a distractor ratio approaching 1:1 than indisplays with extreme distractor ratios.

Initial saccadic latency. Figure 10 shows initial sac-cadic latency as a function of target presence, searchcondition, and distractor ratio. The repeated-measuresANOVA revealed a significant main effect of search con-dition, F(1, 7) = 13.65, p < .01, with longer latencies inthe heterogeneous condition than in the homogeneouscondition, and a significant main effect of target pres-ence, F(1, 7) = 7.86, p < .05, with longer latencies in

the target-present condition than in the target-absentcondition. Figure 10 also shows that initial saccadiclatency varied as a function of distractor ratio, F(8, 56)= 7.47, p < .001, with longer initial saccadic latency indisplays with a distractor ratio approaching 1:1.Further trend analysis revealed that the quadratic trendaccounted for 94.1% of the total variability in saccadiclatency that was due to the distractor ratio manipula-tion whereas the linear trend accounted for 0.2% only.

Fixation duration. Fixation duration as a function oftarget presence, search condition, and distractor ratio ispresented in Figure 11. Overall, fixation duration didnot differ between the homogeneous condition and theheterogeneous condition, F(1, 7) = 3.60, p = .10. Therewere a significant main effect of target presence, F(1, 7)= 25.35, p < .01, and distractor ratio, F(8, 56) = 5.79, p <.001, as well as a marginally significant Target Presencex Distractor Ratio interaction, F(8, 56) = 2.07, p = .054.Figure 11 shows that in target-present trials (see PanelB), there was a slight decrement in fixation durationwith increasing number of red items in a display. Thelinear trend, F(1, 7) = 5.59, p < .05, accounted for 45.3%of the total variability in fixation duration, whereas thequadratic trend, F(1, 7) = 13.70, p < .05, accounted for23.4% only. In contrast, in target-absent trials (seePanel A), a U-shaped fixation duration distribution wasobserved for both the homogeneous condition and theheterogeneous condition. Further trend analysis

Figure 10. Initial saccadic latency (in ms) as a function of target presence (Panel A: target-absent; Panel B: target-present) and the number ofred distractors in the homogeneous condition and the heterogeneous condition in Experiment 2.

A B


revealed that the quadratic trend, F(1, 7) = 5.13, p =.058, accounted for 55.0% of the total variability, where-as the linear trend, F < 1, accounted for 0.5% only.

Saccadic amplitude. Figure 12 plots the averagesaccadic amplitude as a function of target presence,

search condition, and distractor ratio. The repeated-measure ANOVA revealed a significant main effect oftarget presence, F(1, 7) = 204.19, p < .001, indicatinglonger saccadic size in target-absent trials than in tar-get-present trials. The main effect of search condition

Figure 12. Saccadic amplitude (in degrees of visual angle) as a function of target presence (Panel A: target-absent; Panel B: target-present) andthe number of red distractors in the homogeneous condition and the heterogeneous condition in Experiment 2.

A B

Figure 11. Fixation duration (in ms) as a function of target presence (Panel A: target-absent; Panel B: target-present) and the number of reddistractors in the homogeneous condition and the heterogeneous condition in Experiment 2.

A B


was not significant, F < 1, indicating that the two searchconditions (homogeneous vs. heterogeneous) hadroughly the same saccadic amplitude. Figure 12 alsoshows that the average saccadic amplitude varied as afunction of the ratio between the red and blue items, F(8, 56) = 18.82, p < .01, with a U-shaped distribution inboth target-absent and target-present trials. A furthertrend analysis indicated that the quadratic trend, F(1, 7)= 45.07, p < .001, accounted for 82.9% of the variabilityin saccadic amplitude, whereas the linear trendaccounted for only 11.5% of the variability, F(1, 7) =15.92, p < .01.

Saccade selectivity. To quantitatively examine sac-cadic selectivity in within-dimension conjunctionsearches, saccadic bias towards one particular colourfeature value (the red colour) at each level of the dis-tractor ratio manipulation was determined. Saccadicbias was calculated as the observed probability of fixa-tions on those items having a red bar (see Figure 13A)relative to the chance performance. Figure 13B plotssaccadic bias towards red items as a function of thenumber of red items in both the homogeneous condi-tion and the heterogeneous condition.

A 2 (Search Condition: homogeneous vs. heteroge-neous) x 9 (Distractor Ratio) repeated-measures ANOVA

revealed that saccadic bias varied as a function of dis-

tractor ratio, F(8, 56) = 255.11, p < .001. Figure 13Bshows that when there were very few red items in asearch display, participants’ saccadic end points werebiased towards the subset of red items. In contrast,when there were very few blue items, saccadic endpoints were biased towards the blue subset. Althoughthe main effect of search condition was not significant,F(1, 7) = 2.12, p = .189, there was a significant interac-tion between distractor ratio and search condition, F(8,56) = 7.92, p < .001. Figure 13B shows that the sac-cadic bias curves assumed a slanted 8-shape, indicatingstronger guidance in the homogeneous conditions thanin the heterogeneous condition. More specifically, forthose displays with 4, 6, or 8 red items, the bias in sac-cadic end points towards the red items was stronger inthe homogeneous condition than in the heterogeneouscondition, all ts (7) > 3.13, ps < .05. On the otherhand, for those displays with 12, 14, or 16 red items,the bias towards the blue items was stronger in thehomogeneous condition than in the heterogeneouscondition, all ts (7) > 2.15, ps < .069. These observa-tions were confirmed by a further trend analysis, show-ing that the linear trend, F(1, 7) = 13.74, p < .01, thequadratic trend, F < 1, and the cubic trend, F(1, 7) =7.45, p < .05, accounted for 44.2%, 0.1%, and 51.1% oftotal variability for this interaction, respectively.

Similar to the previous experiment, saccadic bias for

Figure 13. Panel A: Relative frequency of saccades directed towards the red distractors as a function of search condition and the number ofred distractors. The diagonal line indicates chance performance. Panel B: Saccadic bias (the difference between the observed frequencyand chance performance) as a function of search condition and the number of red distractors in Experiment 2.

A B


both the first and subsequent saccades within a trialwas calculated to examine the influence of saccadesequence on eye guidance. Figure 14 shows that,unlike the previous experiment (see also Shen et al.,2000), there was little evidence for an effect of saccadesequence on the selectivity in distribution of saccadicend points: there was neither a main effect of saccadesequence, F(1, 7) = 3.02, p = .126, nor any significantinteractions between saccade sequence and other fac-tors, all Fs (8, 56) < 1.87, ps > .082.

DiscussionThe current analyses showed that response time, num-ber of fixations per trial, initial saccadic latency, fixationduration, and saccadic amplitude varied as a functionof the ratio between the red and blue items. Searchingwas more efficient when there were either very few reditems or very few blue items than when the two typesof distractors were equally represented in a display.Saccadic selectivity analysis revealed that participantsconsistently searched through the smaller subset of dis-tractors. Thus, the flexibility in the guidance of visualattention occurs not only in the cross-dimension tasks(current Experiment 1, see also Bacon & Egeth, 1997;Poisson & Wilkinson, 1992; Shen et al., 2000; Zohary &

Hochstein, 1989) but also in within-dimension conjunc-tion search tasks. It is important to note that the over-all search performance of the current experiment wasstill very poor. This might be consistent with the argu-ment by Wolfe and his colleagues (Wolfe, 1994; Wolfeet al., 1990) that within-dimension conjunction searchesare carried out in a serial self-terminating fashion (butsee Carrasco et al., 1998; Linnell & Humphreys, 2001).However, one important implication of the currentexperiment is that this kind of search could be carriedout within a subset of distractors, instead of in an item-by-item fashion across the whole display as argued bythe original feature-integration theory (Treisman &Gelade, 1980) and the guided-search model (Wolfe,1994; Wolfe et al., 1990). This also raises a possibilitythat subset-selective processing could be one of thestrategies adopted in performing a standard within-dimension conjunction search task.

The current experiment also examined the effect ofdistractor heterogeneity on search performance andsaccadic selectivity. Longer response times and morefixations were observed in the heterogeneous conditionthan in the homogeneous condition whereas fixationduration, initial saccadic latency, and saccadic ampli-tude did not differ between the two search conditions.

Figure 14. Saccadic bias of the first and subsequent saccades as a function of the number of red distractors in the homogeneous condition(Panel A) and in the heterogeneous condition (Panel B) in Experiment 2.

A B


Guidance analysis revealed that saccadic bias wasstronger in the homogeneous condition than in the het-erogeneous condition, despite the fact that in bothsearch conditions, there were the same number of reditems and blue items in any display of a given distrac-tor ratio. Thus, it appears that distractor homogeneitymay enhance subset-selective processing, perhaps bymore easily segregating the relevant and irrelevant sub-sets.

General Discussion

Consistent with previous studies (e.g., Bacon & Egeth,1997; Egeth et al., 1984; Kaptein et al., 1995; Poisson &Wilkinson, 1992; Shen et al., 2000; Zohary & Hochstein,1989), the current study found that in a conjunctivesearch task, response time and patterns of eye move-ments were strongly influenced by the ratio betweentwo types of distractors. When the total number ofitems presented in a display was kept constant, fastermanual response time, shorter initial saccadic latency,and fewer fixations were observed when either type ofdistractors was rare than when the two types of distrac-tors were equally represented. The spatial distributionof saccadic end points also changed flexibly as a func-tion of distractor ratio. In the Colour x Shape conjunc-tion search tasks (Experiment 1; see also Shen et al.,2000), when there were only very few same-colour dis-tractors in a display, saccades were biased towards thecolour dimension. When most of the distractors sharedcolour with the search target, saccadic end points werebiased towards the shape dimension, unless the dis-criminability along the shape dimension was low. Thisshows that participants searched through the moreinformative, but not necessarily smaller, subset of dis-tractors. This finding also demonstrates that a mini-mum level of discriminability in the nondominant stim-ulus dimension is a prerequisite for obtaining such flex-ibility. In the Colour x Colour within-dimension con-junction search task (Experiment 2), saccades were reli-ably biased towards the smaller subset of distractors.Thus, the current analyses suggest that subset-selectivesearches can be flexibly applied to both cross-dimen-sion and within-dimension searches.

The distractor-ratio manipulation provides a differentperspective on examining visual search mechanismswithout the standard display size manipulation(Treisman, 1988; Wolfe, 1998). The observed change insearch performance measures and saccadic selectivityas a function of distractor ratio is not consistent withthe original feature-integration theory (Treisman, 1988;Treisman & Gelade, 1980). This is because the feature-integration theory argues that to search for a conjunc-tively defined target, attention is deployed serially toeach item in the display until the target is detected in

target-present trials, or an exhaustive search is per-formed in target-absent trials. Focal attention isrequired to integrate individual stimulus features (suchas colour, shape, size, orientation, etc.) into a unitaryobject and only then can that item be compared againstthe search target. However, the current experiments(see also Bacon & Egeth, 1997; Egeth et al., 1984;Kaptein et al., 1995; Poisson & Wilkinson, 1989; Zohary& Hochstein, 1989) clearly demonstrated that partici-pants could search through a subset of distractors (e.g.,red items or Xs), instead of conducting a serial self-ter-minating search over the whole search display. Uponseeing a search display, participants may first make aglobal assessment of all items in that display. In thosetrials with a strong disparity between the two types ofdistractors, a figure-ground segregation process mayoccur, with all of the items in a smaller or more salientsubset forming the figure and being examined seriallywhereas the rest of the distractors forming the groundand being rejected in parallel. Such a mechanism ofselecting a subset of distractors followed by a serialexamination of items within the selected subset couldbe a possible mechanism of efficient visual search,even for standard conjunction search tasks (i.e., with anequal number of distractors from each type). This isconsistent with several previous eye-movement studiesthat demonstrated a bias in the spatial distribution ofsaccadic end points in conjunctive search tasks by stim-ulus dimensions such as colour, shape, contrast polari-ty, and size (e.g., Findlay, 1997; Findlay, Brown, &Gilchrist, 2001; Findlay & Gilchrist, 1998; Hooge &Erkelens, 1999; Luria & Strauss, 1975; Motter & Belky,1998; Pomplun, Reingold, & Shen, 2001a, in press;Scialfa & Joffe, 1998; Shen, Reingold, Pomplun, D. E.Williams, 2003; D. E. Williams & Reingold, 2001; L. G.Williams, 1966; but see Zelinsky, 1996).

Traditional visual search theories have always main-tained an emphasis on covert attentional scanningrather than overt eye movement behaviour and accord-ingly focused on response time and accuracy as prima-ry and often exclusive measures of search performance.Such a practice has been questioned recently (e.g.,Binello, Mannan, & Ruddock, 1995; Findlay & Gilchrist,1998; Foster, Savage, Mannan, & Ruddock, 2000; Scialfa& Joffe, 1998; D. E. Williams, Reingold, Moscovitch, &Behrmann, 1997; Zelinsky, 1996; Zelinsky, Rao,Hayhoe, & Ballard, 1997; Zelinksy & Sheinberg, 1997;see Rayner, 1998 for a review). The employment ofeye-movement techniques in the present investigationhas yielded detailed spatiotemporal information aboutthe search process that supplements to the standardresponse-time and error rate data. For example, inExperiment 1, given that subtle shape discrimination (Xvs. K or O vs. Q) was involved in the low-discriminabili-


ty condition, visual span (the area from which informa-tion is extracted during a fixation) is small and partici-pants had to be very precise in targeting items(Pomplun, Reingold, & Shen, 2001a, 2001b, 2003;Rayner & Fisher, 1987). Participants tended to fixateclosely on the intended saccadic target (i.e., with high-er saccadic selectivity) and opted to move fast (with ashorter initial latency and fixation duration) but in smallsteps (i.e., shorter saccadic amplitude). This resemblesan item-by-item search within the selected subset. Incontrast, in the high-discriminability condition, partici-pants can process more information within a single fix-ation and make shape discrimination (X vs. O) withtheir parafoveal or peripheral vision. As a result, lessaccurate saccadic targeting was not as costly as in thelow-discriminability condition and participants tendedto make longer fixations and larger saccades.

In Experiment 2, the differences between the homo-geneous and heterogeneous conditions are mainlyattributable to the accuracy of selecting the relevantsubset of distractors, but not to other refined oculomo-tor measures such as initial saccadic latency, fixationduration, and saccadic amplitude. It seems that thegreater similarity among neighbouring distractors in thehomogeneous condition may have made the selectionof the task-relevant feature easier and thus necessitatedfewer fixations and yielded a faster manual responsethan in the heterogeneous condition. Taken together,the examination of eye-movement patterns accompany-ing the search process demonstrated that a differencein the global search performance measures likeresponse time could be subserved by very differentsearch mechanisms as reflected by those refined oculo-motor measures. Such a dynamic picture is very hardto capture by a study that focuses on the standard RTanalysis only.

The detailed spatio-temporal information suppliedby eye movements measures in the present investiga-tion provides additional implications for current visualsearch theories. Given the dominance of the guided-search model proposed by Wolfe and his colleagues(Cave & Wolfe, 1990; Wolfe, 1994; Wolfe & Gancarz,1996; Wolfe et al., 1989), and given that this model hasbeen specified in some detail, including a computation-al implementation, the present results are discussedwithin the framework of this model. According to theguided-search model, in a display with extreme distrac-tor ratio, the bottom-up activation would be relativelylarge at locations occupied by the smaller subset of dis-tractors, explaining the higher proportion of saccadeslanding on these items. To account for the finding thatmost search functions (i.e., RT, initial saccadic latency,fixation number, and saccadic bias) in the Colour xShape Conjunction search tasks were asymmetrical, the

guided-search model can have the activation due tocolour be larger than that due to shape (Cave & Wolfe,1990; Wolfe, 1994). In fact, the low-discriminabilitycondition in Experiment 1 can be viewed as oneextreme condition in which the bottom-up activationdue to the shape dimension is greatly diminished orturned off. The current Experiment 2 demonstratedthat the distractor-ratio effect is also observed in thewithin-dimension conjunction searches. The guided-search model can be amended to account for this find-ing by assuming that participants conduct a separateevaluation of the number of red items and blue itemsin a display and then search through the smaller subsetof distractors. This flexible change in search behav-iours can be seen as another exception (see also theefficient within-dimension conjunction search withpractice by Carrasco et al., 1998 and with part-wholeconfiguration by Wolfe et al., 1994) to the generalnature of inefficient within-dimension conjunctionsearches (Wolfe et al., 1990).

The current experiments also examined whethervisual guidance changes within a single trial. In thehigh-discriminability cross-dimension conjunctionsearch tasks (see also Shen et al., 2000), first saccadesin those displays with extreme distractor ratios pro-duced stronger bias than did the subsequent ones.However, in the low-discriminability cross-dimensionsearch task and within-dimension conjunction searchtask, saccadic bias did not differ between the first andsubsequent saccades. The guided-search model(Wolfe, 1994; Wolfe & Gancarz, 1996; Wolfe et al.,1989) can easily account for the existence of a sequen-tial effect by arguing that attention and saccades movefrom locations with highest activation values (i.e.,stronger saccadic bias) to lower ones on the activationand saccade maps. However, it is unclear why thesequential effect was not observed in the other twoconditions. Future studies may further investigate thisissue and examine whether the observed sequentialeffect is related to the phenomenon of concurrent pro-gramming of multiple saccades demonstrated in arecent work by McPeek, Skavenski, and Nakayama(2000).

In addition, the guided-search model argues that apreattentive parallel process guides the subsequent ser-ial shift of attention through display items. However,the current version of the model is somewhat vaguewith respect to the interface and the interactionbetween these two processing stages. For example,the current experiments (see also Shen et al., 2000)revealed a quadratic change in initial saccadic latencyas a function of distractor ratio, with longer initial sac-cadic latency in those displays with an approximately1:1 distractor ratio and shorter latencies in displays with


extreme distractor ratios. These findings can beexplained in several ways by the current version of theguided-search model. It may be speculated that thestronger activation peak associated with extreme dis-tractor ratios will result in shorter latency.Alternatively, it is possible that the time required forextracting an activation and saccade map varies withdistractor ratios. That is, useful preattentive informa-tion may be developed over time, instead of beingavailable immediately to participants irrespective ofspecific display composition (see also Friedman-Hill &Wolfe, 1995; Shen et al., 2000). Future developmentsof the guided-search model should be more explicitwith respect to the interaction between and the timecourse of the preattentive stage and the serial stage ofprocessing.

In conclusion, the present study provides anotherillustration of the utility of eye movement measures forproviding fine-grained indicators of the search processand for evaluating predictions derived from currentmodels of visual search. Furthermore, the presentinvestigation suggests that a subset selective searchmay be a fundamental aspect of efficient visual search.Similarly, the flexibility of visual guidance and saccadicbias exemplified by the distractor-ratio effect is a robustphenomenon that likely reflects a general tendency byparticipants to modify their search performance as afunction of changes in the relative informativeness ofstimulus dimensions and features.

Preparation of this manuscript was supported by a grant toEyal M. Reingold from the Natural Sciences and EngineeringResearch Council of Canada (NSERC) and a grant to MarcPomplun from the Deutsche Forschungsgemeinschaft(DFG). We wish to thank Peter Dixon and two anonymousreviewers for their helpful comments and suggestions.

Correspondence concerning this paper should be sent toJiye Shen or Eyal M. Reingold, Department of Psychology,University of Toronto, 100 St. George Street, Toronto,Ontario M5S 3G3. (E-mail: [email protected] or [email protected]).

References

Bacon, W. F., & Egeth, H. E. (1997). Goal-directed guid-ance of attention: Evidence from conjunctive visualsearch. Journal of Experimental Psychology: HumanPerception and Performance, 23, 948-961.

Binello, A., Mannan, S., & Ruddock, K. H. (1995). Thecharacteristics of eye movements made during visualsearch with multi-element stimuli. Spatial Vision, 9, 343-362.

Carrasco, M., Ponte, D., Rechea, C., & Sampedro, M. J.(1998). “Transient structures”: The effects of practice and

distractor grouping on within-dimension conjunctionsearches. Perception & Psychophysics, 60, 1243-1258.

Cave, K. R., & Wolfe, J. M. (1990). Modeling the role ofparallel processing in visual search. CognitivePsychology, 22, 225-271.

Driver, J., & McLeod, P. (1992). Reversing visual searchasymmetries with conjunctions of movement and orien-tation. Journal of Experimental Psychology: HumanPerception and Performance, 18, 22-33.

Duncan, J., & Humphreys, G. W. (1989). Visual search andstimulus similarity. Psychological Review, 96, 433-458.

Egeth, H. E., Virzi, R. A., & Garbart, H. (1984). Searchingfor conjunctively defined targets. Journal ofExperimental Psychology: Human Perception andPerformance, 10, 32-39.

Findlay, J. M. (1997). Saccade target selection during visualsearch. Vision Research, 37, 617-631.

Findlay, J. M., Brown, V., & Gilchrist, I. D. (2001). Saccadetarget selection in visual search: The effect of informa-tion from the previous fixation. Vision Research, 41, 87-95.

Findlay, J. M., & Gilchrist, I. D. (1998). Eye guidance andvisual search. In G. Underwood (Ed.), Eye guidance inreading, driving and scene perception (pp. 295-312).Oxford: Elsevier.

Foster, D. H., Savage, C. J., Mannan, S., & Ruddock, K. H.(2000). Asymmetries of saccadic eye movements in ori-ented-line-target search. Vision Research, 40, 65-70.

Friedman-Hill, S., & Wolfe, J. M. (1995). Second-order par-allel processing: Visual search for the odd item in a sub-set. Journal of Experimental Psychology: HumanPerception and Performance, 21, 531-551.

Hooge, I. T., & Erkelens, C. J. (1999). Peripheral vision andoculomotor control during visual search. VisionResearch, 39, 1567-1575.

Kaptein, N. A., Theeuwes, J., & van der Heijden, A. H. C.(1995). Search for a conjunctively defined target can beselectively limited to a colour-defined subset of ele-ments. Journal of Experimental Psychology: HumanPerception and Performance, 21, 1053-1069.

Linnell, K. J., & Humphreys, G. W. (2001). Spatially parallelprocessing of within-dimension conjunctions.Perception, 30, 49-60.

Luria, S. M., & Strauss, M. S. (1975). Eye movements duringsearch for coded and uncoded targets. Perception &Psychophysics, 17, 303-308.

McPeek, R. M., Skavenski, A. A., & Nakayama, K. (2000).Concurrent processing of saccades in visual search.Vision Research, 40, 2499-2516.

Motter, B. C., & Belky, E. J. (1998). The guidance of eyemovements during active visual search. Vision Research,38, 1805-1815.

Nagy, A. L., & Sanchez, R. R. (1990). Critical colour differ-ences determined with a visual search task. Journal of


the Optical Society of America A, 7, 1209-1217.Neisser, U. (1967). Cognitive psychology. New York:

Appleton, Century, Crofts.Palmer, J., Verghese, P., & Pavel, M. (2000). The psy-

chophysics of visual search. Vision Research, 40, 1227-1268.

Pashler, H. (1987). Target-distractor discriminability in visualsearch. Perception & Psychophysics, 41, 285-302.

Poisson, M. E., & Wilkinson, F. (1992). Distractor ratio andgrouping processes in visual conjunction search.Perception, 21, 21-38.

Pomplun, M., Reingold, E. M., & Shen, J. (2001a).Peripheral and parafoveal cueing and masking effects onsaccadic selectivity. Vision Research, 41, 2757-2769.

Pomplun, M., Reingold, E. M., & Shen, J. (2001b).Investigating the visual span in comparative search: Theeffects of task difficulty and divided attention. Cognition,81, B57-B67.

Pomplun, M., Reingold, E. M., & Shen, J. (2003). Area acti-vation: A computational model of saccadic selectivity invisual search. Cognitive Science, 27, 299-312.

Rayner, K. (1998). Eye movements in reading and informa-tion processing: 20 years of research. PsychologicalBulletin, 124, 372-422.

Rayner, K., & Fisher, D. L. (1987). Letter processing duringeye fixations in visual search. Perception &Psychophysics, 42, 87-100.

Scialfa, C. T., & Joffe, K. (1998). Response times and eyemovements in feature and conjunction search as a func-tion of target eccentricity. Perception & Psychophysics,60, 1067-1082.

Shen, J., Reingold, E. M., & Pomplun, M. (2000). Distractorratio influences patterns of eye movements during visualsearch. Perception, 29, 241-250.

Shen, J., Reingold, E. M., Pomplun, M., & Williams, D. E.(2003). Saccadic selectivity during visual search: Theinfluence of central processing difficulty. In J. Hyönä, R.Radach, & H. Deubel (Eds), The mind’s eyes: Cognitiveand applied aspects of eye movement research (pp. 65-88). Amsterdam: Elsevier Science Publishers.

Theeuwes, J. (1992). Perceptual selectivity for colour andform. Perception & Psychophysics, 51, 599-606.

Treisman, A. (1988). Features and objects: The fourteenthBartlett memorial lecture. The Quarterly Journal ofExperimental Psychology, 40A, 201-237.

Treisman, A., & Gelade, G. (1980). A feature integration the-ory of attention. Cognitive Psychology, 12, 97-136.

Treisman, A., & Sato, S. (1990). Conjunction search revisit-ed. Journal of Experimental Psychology: HumanPerception and Performance, 16, 459-478.

van der Heijden, A. H. C., Malhas, M. S. M., & van denRoovart, B. P. (1984). An empirical interletter confusionmatrix for continuous line capitals. Perception &

Psychophysics, 35, 85-88.Williams, D. E., & Reingold, E. M. (2001). Preattentive

guidance of eye movements during triple conjunctionsearch tasks: The effects of feature discriminability andsaccadic amplitude. Psychonomic Bulletin & Review, 8,476-488.

Williams, D. E., Reingold, E. M., Moscovitch, M., &Behrmann, M. (1997). Patterns of eye movements dur-ing parallel and serial visual search tasks. CanadianJournal of Experimental Psychology, 51, 151-164.

Williams, L. G. (1966). The effect of target specification onobjects fixated during visual search. Perception &Psychophysics, 1, 315-318.

Wolfe, J. M. (1994). Guided search 2.0: A revised model ofvisual search. Psychonomic Bulletin & Review, 1, 202-238.

Wolfe, J. M. (1996). Extending guided search: Why guidedsearch needs a preattentive “item map.” In A. F. Kramer,M. G. Coles et al. (Eds.), Converging operations in thestudy of visual selective attention (pp. 247-270).Washington, DC: American Psychological Association.

Wolfe, J. M. (1998). Visual search. In H. Pashler (Ed.),Attention (pp. 13-71). London: Psychology Press.

Wolfe, J. M., Cave, K. R., & Franzel, S. L. (1989). Guidedsearch: An alternative to the feature integration modelfor visual search. Journal of Experimental Psychology:Human Perception and Performance, 15, 419-433.

Wolfe, J. M., Friedman-Hill, S. R., & Bilsky, A. B. (1994).Parallel processing of part-whole information in visualsearch tasks. Perception & Psychophysics, 55, 537-550.

Wolfe, J. M., & Gancarz, G. (1996). Guided search 3.0. In V.Lakshminarayanan (Ed.), Basic and clinical applicationsof vision science (pp. 189-192). Dordrecht: Kluwer.

Wolfe, J. M., Yu, K. P., Stewart., M. I., Shorter, A. D.,Friedman-Hill, S. R., & Cave, K. R. (1990). Limitations onthe parallel guidance of visual search: Color x color andorientation x orientation conjunctions. Journal ofExperimental Psychology: Human Perception andPerformance, 16, 879-892.

Zelinsky, G. J. (1996). Using eye saccades to assess theselectivity of search movements. Vision Research, 36,2177-2187.

Zelinksy, G. J., Rao, R. P. N., Hayhoe, M. M., & Ballard, D.H. (1997). Eye movements reveal the spatiotemporaldynamics of visual search. Psychological Science, 8, 448-453.

Zelinsky, G. J., & Sheinberg, D. L. (1997). Eye movementsduring parallel-serial visual search. Journal ofExperimental Psychology: Human Perception andPerformance, 23, 244-262.

Zohary, E., & Hochstein, S. (1989). How serial is serial pro-cessing in vision? Perception, 18, 191-200.

Sommaire

Revue canadienne de psychologie expérimentale, 2003, 57:2, 96

L’effet distracteur-ratio renvoie aux résultats derecherche selon lesquels la performance de larecherche pendant une tâche de recherche visuelleconjonctive dépend de la fréquence relative de deuxtypes ou sous-ensembles de distracteurs, lorsque lenombre total des items présentés à l’écran est fixe.Dans une étude antérieure, Shen, Reingold et Pomplun(2000) ont examiné les modèles de mouvements ocu-laires des participants, sous l’angle du paradigme dis-tracteur-ratio, et ont démontré que, à un essai donné,les résultats à la fin de la saccade étaient biaisés auprofit d’un plus petit sous-ensemble de distracteurs etque les participants passaient de manière adaptatived’un essai à l’autre, entre les différents sous-ensembles.La présente étude examinait les conditions limites decette tendance à effectuer une recherche de manièreadaptative à partir d’un petit sous-ensemble de dis-tracteurs; pour ce faire, nous nous sommes penchés surl’influence de plusieurs manipulations, dont on connaîtl’efficacité relativement à la modulation de larecherche, y compris la discriminabilité du stimulus(expérience 1), la recherche conjonctive intra-dimen-sion et inter-dimension et l’hétérogénéité du distracteur(expérience 2).

La présente étude a démontré que, lors d’une tâchede recherche conjonctive, le temps de réponse et lesmodèles de mouvements oculaires sont fortement influ-encés par le rapport entre les deux types de dis-tracteurs. Lorsque le nombre total d’items présentés àl’écran demeurait constant, nous avons observé untemps de réponse manuelle plus rapide, un temps delatence initiale de la saccade plus court, et un moinsgrand nombre de fixations, lorsque l’un ou l’autre desdistracteurs était présenté peu fréquemment, par rap-port aux essais qui mettaient en scène les deux typesde distracteurs représentés en nombre égal. La distribu-tion spatiale des résultats à la fin de la saccade se mo-difie aussi de manière adaptative en fonction du ratiode distracteurs. Dans les tâches de recherche conjonc-

tive qui avaient pour condition expérimentale lacouleur x la forme (expérience 1), les saccades étaientbiaisées au profit de la dimension « couleur » lorsqu’iln’y avait que quelques distracteurs de même couleurprésentés à l’écran. En revanche, lorsque la plupart desdistracteurs avaient la même couleur que la cible derecherche, les résultats à la fin de la saccade étaientbiaisés au profit de la dimension « forme », sauf dansles cas où la discriminabilité à l’égard de la dimension « forme » était faible. Ces observations révèlent que lesparticipants exécutent la tâche de recherche à partir dusous-ensemble de distracteurs le plus informatif, maispas nécessairement à partir du plus petit. Cette constatation démontre également que le niveau mini-mal de discriminabilité, dans le cas de la dimensionreprésentée par un stimulus non dominant, est unprérequis à l’obtention de l’adaptabilité observée. Dansla tâche de recherche conjonctive intra-dimensioncouleur x couleur (expérience 2), les saccades étaient,de manière fidèle, baisées au profit des sous-ensemblesde distracteurs les plus petits. Ainsi, les analysesprésentées ici laissent croire que les recherches quifont appel à la sélection d’un sous-ensemble peuvents’appliquer de manière adaptative aux recherches inter-et intra-dimension.

Les présents résultats indiquent que l’adaptabilité del’orientation visuelle et du biais saccadique observé,reproduite par l’effet distracteur-ratio, est unphénomène robuste qui intervient dans l’efficacité de larecherche car il permet l’adaptation aux changementsdevant la présence relative d’éléments informatifs dansles caractéristiques et les dimensions du stimulus. Laprésente étude fournit une autre illustration du rôle desmesures du mouvement oculaire comme moyensd’obtenir des indicateurs très précis du processus derecherche et d’évaluer les prédictions découlant desmodèles actuels dans le domaine de la recherchevisuelle.


Documents

Guidance of Eye Movements During Conjunctive Visual …...effect, Shen, Reingold, and Pomplun (2000) examined participants’ patterns of eye movements, the spatial dis-tribution of