PAPER Why is visual search superior in autism spectrum disorder?

PAPER

Why is visual search superior in autism spectrum disorder?

Robert M. Joseph,1 Brandon Keehn,1 Christine Connolly,1 Jeremy M.Wolfe2 and Todd S. Horowitz2

1. Department of Anatomy and Neurobiology, Boston University School of Medicine, USA2. Brigham & Women’s Hospital and Harvard Medical School, USA

Abstract

This study investigated the possibility that enhanced memory for rejected distractor locations underlies the superior visual searchskills exhibited by individuals with autism spectrum disorder (ASD). We compared the performance of 21 children with ASDand 21 age- and IQ-matched typically developing (TD) children in a standard static search task and a dynamic search task, inwhich targets and distractors randomly changed locations every 500 ms, precluding the use of memory in search. Children withASD exhibited overall faster reaction time (RT) relative to TD children, and showed no disruption in search efficiency in thedynamic condition, discounting the possibility that memory for rejected distractors augments their visual search abilities.Analyses of RT x set size functions showed no group differences in slopes but lower intercepts for the ASD group in both staticand dynamic search, suggesting that the ASD advantage derived from non-search processes, such as an enhanced ability todiscriminate between targets and distractors at the locus of attention. Eye-movement analyses revealed that the ASD and TDgroups were similar in the number and spatial distribution of fixations across the search array, but that fixation duration wassignificantly shorter among children with ASD. Lower intercepts in static search were related to increased symptom severity inchildren with ASD. In summary, ASD search superiority did not derive from differences in the manner in which individuals withASD deployed their attention while searching, but from anomalously enhanced perception of stimulus features, which was in turnpositively associated with autism symptom severity.

Introduction

Autism is a complex neurodevelopmental disorderdiagnosed behaviorally on the basis of impairments andanomalies in three defining symptom domains:communication, reciprocal social interaction, andrepetitive and stereotyped interests and behaviors(APA, 1994). Although not an explicitly definingfeature of autism or autism spectrum disorder (ASD),abnormalities of visual attention or perception appearinherent in a number of behaviors observed in ASD. Ashas been well documented in several experimentalinvestigations, these include heightened attention tovisual detail and sensitivity to trivial changes in thevisual environment (Burack, Enns, Stauder, Mottron &Randolph, 1997; Plaisted, 2000). Particularly intriguingis that atypical visual processing in ASD is oftenexpressed in the form of enhanced performance relativeto non-ASD individuals. For example, individuals withASD have been found to excel in the Embedded FiguresTask (Joliffe & Baron-Cohen, 1997; Shah & Frith, 1983),the Wechsler (1994, 1997) Block Design subtest (Caron,Mottron, Berthiaume & Dawson, 2006; Shah & Frith,1993) and, the topic of the present study, visual search

(Kemner, van Ewijk, van Engeland & Hooge, 2008;O’Riordan & Plaisted, 2001; O’Riordan, Plaisted, Driver& Baron-Cohen, 2001).

In a typical visual search task, an observer looks for atarget item among an array of distractor items andresponds by indicating whether a target is present orabsent. Search tasks range in difficulty from efficientfeature searches, in which the observer’s attention isimmediately drawn to the target, to relatively inefficientsearches, in which the observer must make multiple shiftsof attentional focus, to search tasks which require theobserver to foveate each item in order to determinewhether it is a target or distractor. Search efficiency isestimated by varying the number of items in the searcharray (set size), measuring the time needed to completethe search and respond, and computing the slope of thereaction time (RT) · set size function. Highly efficientsearches yield slopes approaching 0 ms ⁄ item, whileinefficient searches that require multiple deploymentsof attention but that do not require eye movements haveslopes in the range of 20–60 ms ⁄ item.

Individuals with ASD have been found to excel relativeto non-ASD individuals in visual search (Kemner et al.,2008; O’Riordan, 2004; O’Riordan & Plaisted, 2001;

Address for correspondence: Robert M. Joseph, Department of Anatomy and Neurobiology, Boston University School of Medicine, 715 AlbanySt., L-814, Boston, MA 02118, USA; e-mail: [email protected]

� 2009 Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.

Developmental Science 12:6 (2009), pp 1083–1096 DOI: 10.1111/j.1467-7687.2009.00855.x

O’Riordan et al., 2001; Plaisted, O’Riordan & Baron-Cohen, 1998), particularly in relatively difficult searchesthat have moderately steep RT · set size slopes, and thuspresumably require multiple shifts of attention. Despitethe robustness of this finding, the differences in brainorganization and function that underlie superior searchin ASD are poorly understood. Investigation of thesedifferences can help to elucidate the neurofunctionalunderpinnings of ASD.

O’Riordan and colleagues (2001) proposed twoprocessing differences that could potentially explainsuperior visual search skills in ASD: (1) enhancedmemory for distractor locations already inspected, and(2) enhanced ability to discriminate between target anddistractor stimulus features. According to the firstexplanation, enhanced memory of the search history(e.g. via inhibitory tagging, see Klein, 1988; Takeda &Yagi, 2000; M�ller & von M�hlenen, 2000) would allowan observer with ASD to avoid re-attending to previouslyrejected distractors, thereby augmenting search efficiency.Although prior research indicates that memory forinspected locations plays at most a limited role in visualsearch among typical adults and children (Horowitz &Wolfe, 1998, 2003; Horowitz, Wolfe, Keehn, Connolly &Joseph, unpublished manuscript), memory may play adifferent role in the unusually developed visual searchskills of individuals with ASD. This possibility isconsistent with prior evidence that the ASD advantagein visual search becomes apparent only in relativelydifficult searches requiring multiple deployments ofattention through the search array. Enhanced memoryfor search locations, whether implicit or strategicallyapplied, could thus explain ASD search superiority, butits role has not yet been investigated.

The second possibility is that enhanced discriminationmight underlie search superiority in ASD. Supporting thispossibility, O’Riordan and Plaisted (2001) demonstratedthat, as with typical adults (Duncan & Humphreys, 1989;Wolfe, 1994), target–distractor discriminability was aprincipal rate-determining factor in visual search forchildren with ASD, and that their visual searchsuperiority relative to typically developing (TD) childrenincreased as a function of target–distractor similarity.They proposed that enhanced discrimination in ASD mayderive from (a) superior top-down modulation of stimulusactivation levels, resulting in increased excitation of targetstimuli and increased inhibition of distractor stimuli, or(b) enhanced bottom-up perception of dissimilar stimulusfeatures, either of which could lead to faster targetidentification in ASD. O’Riordan (2000) used positiveand negative priming paradigms to investigate the role oftop-down excitation and inhibition of stimulusrepresentations in children with ASD engaged in visualsearch. She found that children with ASD did not differfrom TD children in excitatory or inhibitory top-downcontrol of stimulus representations, leaving open thepossibility that the ASD advantage in search derived fromenhanced bottom-up perception of stimulus attributes.

Our goal in the present study was to examine thecontributions of both memory and enhanced perceptualprocessing to visual search superiority in ASD. To assessthe role of memory, we used the dynamic searchparadigm developed by Horowitz and Wolfe (1998,2003). In dynamic search, targets and distractors arere-plotted randomly in new locations every 500 mswithin a trial, rendering memory for the locations ofpreviously inspected items of little benefit to searchefficiency. If memory did play a significant role in search,search efficiency would be markedly diminished in thedynamic condition. More precisely, if there were perfectmemory for prior attentional deployments in standardstatic search and no memory in dynamic search, thendynamic search slopes would be expected to be doublethat in the static condition, based on the mathematics ofsampling. Under a sampling without replacement regime(perfect memory), finding a single target in n items willrequire an average of (n + 1) ⁄ 2 samples, while undersampling with replacement (no memory) the same searchwill require n + 1 samples (Horowitz & Wolfe, 2003,p. 261). Horowitz and Wolfe (1998) found that, althoughaccuracy was lower and RT was longer in dynamic thanin static search, search slopes were equivalent betweenthe two search conditions in typical adults. This findinghas been replicated in typical adults (Gibson, Li, Skow,Salvagni & Cooke, 2000; Horowitz & Wolfe, 2003; vonM�hlenen, M�ller & M�ller, 2003) and extended to TDchildren (Horowitz et al., unpublished manuscript),indicating that visual search normally proceeds viasampling with replacement and that memory for priorattentional deployments makes at most a limitedcontribution to search efficiency.

The dynamic search paradigm provides an ideal meansof testing the possibility that memory for prior searchlocations, despite its limited role in the searchperformance of typical individuals, might be a criticalfactor in the unusual search abilities of individuals withASD. Accordingly, we compared the performance ofchildren with ASD and age- and IQ-matched TDchildren in static and dynamic search. For bothconditions, we used a moderately difficult search taskin which participants were asked to look for a T amongLs that appeared in four orthogonal orientations. Thisspatial configuration task requires multiple shifts ofattention, has been found to yield slopes of25–30 ms ⁄ item in adults (Horowitz & Wolfe, 1998,2003) and slopes that are slightly steeper in children(Horowitz et al., unpublished manuscript), and isroughly comparable in difficulty to the search tasksused in prior research demonstrating ASD visual searchsuperiority (Kemner et al., 2008; O’Riordan, 2004,Experiment 2; O’Riordan et al., 2001, Experiment 2).If enhanced memory for prior attentional deploymentsunderlies superior search in ASD, children with ASDwould be expected to exhibit a diminution of searchefficiency in the dynamic condition relative to the staticcondition and to TD children. In addition to measuring

1084 Robert M. Joseph et al.

� 2009 Blackwell Publishing Ltd.

RT, we used eye-tracking to record participants’ point ofvisual regard throughout each search trial. This providedfurther information about possible group differences insearch strategies, whether based on differences in the useof memory or not.

We investigated the possibility that enhanced visualdiscrimination ability underlies superior search skills intwo ways. First, by varying the number of items in thesearch display from trial to trial, we were able to derive anRT · set size function for each participant and topartition this function into slope and intercept(Sternberg, 1966). As noted above, the slope is typicallytaken as a measure of search efficiency. Slope reflects theRT cost of each additional distractor and is determinedby the rate of attentional shifting between items as well asthe ability to filter out irrelevant items. The intercept isthe RT that would theoretically be observed if search wereeliminated from the task. The intercept is assumed to bedetermined by non-search task components, includingearly preattentive perceptual processing of the stimulusfeatures in the visual search display and speed ofperceptual processing once attention has been directedto a potential target. Although we do not mean to suggestthat slopes and intercepts divide neatly betweenattentional and perceptual processes in visual search, afinding of reduced intercepts for RT · set size functions inASD would be consistent with the idea that enhancedperceptual processing undergirds ASD search superiority.Second, we used the eye-tracking data to examine anygroup differences in the deployment and spatialallocation of attention throughout the search array ascompared to processing efficiency once a potential targetwas selected. Indications of increased speed of processingat the locus of selection would also be consistent with thehypothesis that enhanced perceptual discriminationabilities drive superior search skills in ASD.

Our final aim was to examine whether the informationprocessing biases underlying superior visual search mightbe related to the core symptoms of ASD. Consequently,we assessed individual differences in visual searchabilities, as measured by the slopes and intercepts ofRT · set size functions, in relation to variation incommunication, social, and repetitive behaviorsymptoms in participants with ASD. This allowed us todetermine if the component processes tapped by searchslopes and intercepts, and any neurofunctionaldifferences they potentially index, might be ofetiological significance with regard to the symptomsthat are defining of ASD.

Methods

Participants

Participants were 21 school-age children and adolescentswith ASD (17 males), all of whom were judged to meetDSM-IV (APA, 1994) criteria for autism or PDDNOS

by an expert clinician (first author), and an age-matchedcomparison group of 21 non-ASD children (17 males).Clinical diagnoses were confirmed using the AutismDiagnostic Interview-Revised (ADI-R; Rutter, LeCouteur & Lord, 2003) and the Autism DiagnosticObservation Schedule (ADOS; Lord, Rutter, DiLavore &Risi, 1999), both administered by specially trainedexaminers who had previously established inter-raterreliability. All children in the ASD group met diagnosticcriteria for autism on the ADI-R, with the exception oftwo children who were one point below the diagnosticthreshold in the repetitive behavior domain. On theADOS, 16 children met diagnostic criteria for autism,three met for a less severe diagnosis of autism spectrumdisorder, and two met ADOS criteria for autism in thesocial domain, but were below threshold in thecommunication domain. The latter two children metfull criteria for autism on the ADI-R. The sampleselection criteria conformed to guidelines for researchdiagnoses of ASD established within the NationalInstitutes of Health Collaborative Programs ofExcellence in Autism. Children with autism-relatedmedical conditions (e.g. Fragile-X syndrome, tuberoussclerosis) were not included in this study.

Comparison group participants were a subgroup of alarger sample of TD children (Horowitz et al., unpub-lished manuscript) who were selected to match the ASDgroup as closely as possible on age, sex, and IQ. Allcomparison group participants had no reported historyof autism and were confirmed to be free of autism-related symptoms and of any other neurological orpsychiatric conditions via parent report and expertclinical observation. Independent-samples t-tests con-firmed that the ASD and TD groups were matched onage, t(40) = 0.4, p = .69. The groups were matched onnonverbal IQ, t(40) = 0.0, p = .99, but not on verbal IQ,t(40) = 2.5, p < .02, as measured with the KaufmanBrief Intelligence Test-II (Kaufman & Kaufman, 2004;see Table 1).

Apparatus

The experiment was presented with E-Prime 1.1 softwareon a Pentium IV 3.2 GHz ⁄ 2 GB PC with a 19-inch LCD(refresh rate of 75 MHz). Participants were seated 57 cmfrom the monitor with eye level at center screen. Testresponses were registered with a Cedrus button box

Table 1 Participant characteristics

Autism (n = 21)M (SD)Range

Comparison (n = 21)M (SD)Range

Age 14;7 (2;8)10;6–19;3

14;2 (2;11)8;6–19;1

Verbal IQ 99 (19)70–136

112 (15)81–148

Nonverbal IQ 107 (10)85–121

107 (14)75–134

Visual search in ASD 1085


(model RB-420). Participants’ point of regard wasmonitored with an ISCAN Model ETL-500 head-mounted, pupil-corneal reflection tracking system thatallowed participants to move their heads freely duringthe test procedure. The eye-tracking computer registeredeach new screen image via a digital pulse programmed inE-Prime to coincide with the onset of each newexperimental stimulus or inter-trial interval.

Stimuli

The target was the letter ‘T’ and the distractor was theletter ‘L’. Letters were drawn in black Arial font on agray background and were randomly displayed in anyone of four orthogonal orientations. At the 57 cmviewing distance, the letters subtended 1.0 to 1.2degrees visual angle (�) in both dimensions, dependingon orientation. There were 60 possible stimulus locationsarranged on five concentric invisible circles. The circlessurrounded a central fixation cross at eccentricities of1.94�, 3.88�, 5.82�, 7.76�, and 9.70�. The innermost circlehad four possible stimulus locations, the second eight,the third 12, the fourth 16, and the fifth 20.

In the static condition, a single search frame wasgenerated for each trial, consisting of 15, 20, or 25 stimuliplaced at randomly selected locations. In the dynamiccondition, a new search frame was presented every500 ms during a trial. Within a trial, each frame was thesame with regard to the set size, target presence orabsence, and the set of occupied locations. Across frames,the locations of individual stimuli were shuffled amongthe set of occupied locations. Figure 1 provides asimplified illustration of static and dynamic (targetpresent) trials.

In the dynamic condition, when a target was present, itwas displayed randomly in one of only four locations(one per quadrant), with the restriction that it could notappear at the same location on two consecutive frames.This measure was taken to defeat a simple ‘sit-and-wait’strategy of attending to a single location (von M�hlenenet al., 2003). An observer who adopted this strategywould perform very poorly in the task because thelikelihood of choosing an actual target location was quitelow (i.e. 4 ⁄ set size).

Procedure

The experimental task was to indicate via the buttonbox whether the target stimulus was present (50% oftrials) or absent (50% of trials). A trial began with afixation cross (‘+’ drawn in the same font and size as theletter stimuli) presented alone for 1000 ms. With thefixation cross remaining on the screen, the first frameappeared. In the static condition this frame remained onthe screen until a response was made or until 7000 mshad elapsed. In the dynamic condition, successive frameswere presented every 500 ms until a response was madeor until 7000 ms had elapsed. The appearance of the

fixation cross served to alert observers to the start of thetrial, and provided a default initial fixation locus.However, observers were not instructed to maintainfixation and were free to move their eyes during thesearch task.

The experiment consisted of six 30-trial blocks. Theblocks alternated between the two search conditions,always beginning with static. Within each 30-trial block,target presence and set size were varied in pseudorandomorder. Before the first block of each condition, ademonstration was given and 18 practice trials wereadministered with corrective feedback. Participants wereinstructed to respond (with their dominant hand) asquickly as possible without making errors. Theseinstructions were repeated after the practice trials andbetween test blocks.

Results

Data were analyzed with SPSS 15.0 for Windows. Datawere not trimmed; instead, medians were used to reducethe influence of outliers. Unless otherwise noted, allstatistical analyses were mixed-model repeated measuresanalyses of variance. Partial eta-squared (gp

2) was used

Figure 1 A simplified illustration of a static (upper panel) anda dynamic (lower panel) trial. In the static condition, thestimuli remained stationary in a randomly selected set of fixedlocations for the duration of the trial. In the dynamic condition,a new search frame was presented every 500 ms. The stimuliappeared to move among the randomly selected set of fixedlocations by randomly re-plotting them in any one of fourorientations on each new frame.



as a measure of effect size. In all figures, error barsrepresent one standard error of the mean.

Search performance

Reaction time

Median RT data from correct responses are shown inFigure 2. We conducted an ANOVA with the between-subjects factor group (ASD, TD) and the within-subjectsfactors search type (static, dynamic), target presence(present, absent), and set size (15, 20, 25). There weremain effects of target presence, set size, and search type.As is typical for visual search studies, RTs were slowerfor target absent trials than target present trials, F(1,40) = 253.1, p < .001, gp

2 = .86; target present searchescan be terminated as soon as the target is detected, butwhen no target is found, the observer must search untilhe or she is convinced that a target is not likely to befound. As noted above, when a search task requires arelatively difficult discrimination (between Ts and Ls infour different orientations), search time increases as afunction of the number of distractors. Accordingly, wealso observed a set size effect, F(2, 80) = 112.2, p < .001,gp

2 = .74. Consistent with prior studies comparing staticand dynamic search, RT was faster in the staticcondition, F(1, 40) = 14.6, p < .001, gp

2 = .27.The most important findings from the RT analysis

pertained to group effects. There was a main effect ofgroup, F(1, 40) = 7.3, p < .01, gp

2 = .15, with faster RTin the ASD than in the TD group. In addition, there weretwo-way interactions between group and search type,F(1, 40) = 6.4, p < .02, gp

2 = .14, and between groupand target presence, F(1, 40) = 5.7, p < .02, gp

2 = .13.Follow-up ANOVAs revealed that the RT advantage ofthe ASD group over the TD group was larger in dynamicsearch, F(1, 40) = 9.7, p < .01, gp

2 = .19, than in staticsearch, F(1, 40) = 3.5, p < .07, gp

2 = .08, and larger fortarget absent trials, F(1, 40) = 7.8, p < .01, gp

2 = .16,than for target present trials, F(1, 40) = 2.9, p < .10,gp

2 = .07. Finally, there was a three-way interactionbetween group, search type, and target presence, F(1,40) = 8.1, p < .01, gp

2 = .17. This interaction wasanalyzed with a series of one-way, between-groupcomparisons. These analyses revealed a statistically

significant ASD advantage on dynamic absent trials,F(1, 40) = 11.2, p < .002, gp

2 = .22, a marginallysignificant ASD advantage on static absent trials, F(1,40) = 3.3, p < .07, gp

2 = .08, and non-significant groupdifferences on both dynamic present, F(1, 40), 2.6,p < .12, gp

2 = .06, and static present trials, F(1, 40),2.2, p < .15, gp

2 = .05.Slopes of the RT · set size functions for the target

present and target absent conditions of the static anddynamic searches were computed from the median RTdata reported above. As shown in Figure 3, slopes wereactually steeper for static than for dynamic search, F(1,40) = 15.6, p < .001, gp

2 = .28, and slopes were steeperfor target absent than for target present trials, F(1,40) = 31.4, p < .001, gp

2 = .44. There was no main effectof group on slopes, F(1, 40) = 0.4, or group interactioneffects, demonstrating that search efficiency wasconsistently similar for the ASD and TD groups acrosssearch conditions. These findings indicated that memoryfor rejected distractors contributes no more to the searchskills of children with ASD than to those of TD children.

Intercepts for the same RT · set size functions areillustrated in Figure 4. Intercepts were higher for dyna-mic than for static search, F(1, 40) = 35.7, p < .001,gp

2 = .47, and for target absent than for target presenttrials, F(1, 40) = 35.8, p < .001, gp

2 = .47. Interceptswere consistently lower in the ASD than in the TDgroup, F(1, 40) = 6.0, p < .02, gp

2 = .13, with nointeractions between group and other factors. Thisfinding is consistent with the possibility that non-search capacities, including enhanced perceptualprocessing of stimulus features, underlies ASD searchsuperiority.

Error

Figure 5 displays rate of error as a function of the samevariables analyzed for RT. An ANOVA on the raw errordata revealed main effects of search type, F(1, 40) = 79.0,p < .001, gp

2 = .66, target presence, F(1, 40) = 18.5,p < .001, gp

2 = .32, and set size, F(2, 80) = 17.7, p <.001, gp

2 = .31. The rate of error was higher for dynamic(9.2%) than for static (4.0%) search, was higher for targetpresent (8.5%) than for target absent (4.7%) trials, and

Static Search

900

1300

1700

2100

2500

2900

3300

3700

4100

15

RT

(m

s)

Set sizeASD Target Present ASD Target Absent

Dynamic Search

900

1300

1700

2100

2500

2900

3300

3700

4100

Set size

RT

(ms)

TD Target Present TD Target Absent

20 25 15 20 25

Figure 2 Median reaction time (RT) by group, search type, target presence, and set size.



increased as a function of set size. There was also asearch type · set size interaction, F(2, 80) = 18.8,p < .001, gp

2 = .32, reflecting a higher increase in errorwith set size in dynamic relative to static search. Therewas no main effect of group, F(1, 40) = 0.7, nor werethere any interaction effects between group and otherexperimental factors on error rate. Correlational analysesrevealed no speed–accuracy tradeoffs in any conditionfor either group.

Eye-movement behavior

Eye position was continuously monitored from the startto the finish of each trial. Parameters for collection ofraw point of regard (POR) data were set using ISCANsoftware. To count as a fixation, POR had to bemaintained for at least five continuous data samples(80–85 ms at a sample rate of 60 Hz) within an area of 1�of visual angle. Eye-tracking data were successfully

Static Search

400

800

1200

1600

2000

2400

Present Absent

Inte

rcep

t (m

s)

Dynamic Search

400

800

1200

1600

2000

2400

Present Absent

Intercept (ms)

ASD

TD

Figure 4 Median intercept of the RT · set size function by group, search type, and target presence.

Static Search

02468

101214161820

15Set size

% E

rror

ASD Target Present ASD Target Absent

Dynamic Search

02468101214161820

Set size

% E

rror


20 2515 20 25

Figure 5 Percent error by group, search type, target presence, and set size.

Static Search

20

40

60

80

100

Present Absent Present Absent

Slo

pe (

ms/

item

)

Dynamic Search

20

40

60

80

100

Slope (m

s/item)

ASDTD

Figure 3 Median slope of the RT · set size function by group, search type, and target presence.



collected for 18 of the 21 participants with ASD and allTD participants. The three ASD participants for whomeye-tracking data were not available did not differ fromthe other ASD participants in age, IQ, or searchperformance.

The eye-movement data were used to investigatepotential group differences in attentional deploymentand stimulus discrimination. Specifically, the measuresof fixation frequency, saccade amplitude, anddistribution of fixations by eccentricity and favoredquadrant were used to determine if there weredifferences, as reflected in overt shifts of attention, inthe way ASD and TD participants deployed theirattention in searching for a target. Fixation durationprovided an indication of processing efficiency at thelocus of attention, relevant to the question of enhancedvisual perception and discrimination in individuals withASD.

Fixation frequency

Number of fixations per trial (Figure 6) was higher inthe static (M = 4.1) than in the dynamic (M = 3.6)condition, F(1, 37) = 12.3, p < .001, gp

2 = .25, and washigher in the target absent (M = 5.2) than in the targetpresent (M = 2.5) condition, F(1, 37) = 363.4, p <.001, gp

2 = .91. Fixation frequency increased with setsize, F(2, 74) = 12.3, p < .001, gp

2 = .25. However, therewas no main effect of group on fixation frequency,

F(1, 37) = 1.1, nor were there any group interactioneffects.

Fixation duration

Fixation duration (Figure 7) was longer in the dynamicthan in the static search condition, F(1, 37) = 86.8,p < .001, gp

2 = .70, was longer for target present than fortarget absent trials, F(1, 37) = 92.5, p < .001, gp

2 = .71,and increased with set size, F(2, 74) = 17.6, p < .001,gp

2 = .32. ASD participants made shorter fixations thanTD participants, F(1, 37) = 7.8, p < .01, gp

2 = .17. Agroup · search type interaction, F(1, 37) = 6.3, p < .02,gp

2 = .15, reflected the increased disparity between groupsin fixation durations in the dynamic relative to the staticsearch condition. Separate analyses of the data by searchtype showed a larger group effect in dynamic search,F(1, 37) = 8.6, p < .01, gp

2 = .19, than in static search,F(1, 37) = 3.1, p < .09, gp

2 = .08.

Saccade amplitude

Saccade amplitude is the average distance betweenconsecutive fixations. Saccade amplitude was greater inthe static (M = 7.9�) than in the dynamic (M = 7.7�)search condition, F(1, 37) = 5.6, p < .05, gp

2 = .13, andfor target absent (M = 8.2�) than for target present(M = 7.4�) trials, F(1, 37) = 110.5, p < .001, gp

2 = .75.Saccade amplitude did not vary as a factor of set size,

Static Search

0

2

4

6

15 20 25Set size

Fix

atio

ns p

er T

rial


Dynamic Search

0

2

4

6

15 20 25Set size

Fixations per T

rial


Figure 6 Median fixation frequency by group, search type, target presence, and set size.

Static Search

200

300

400

500

600

15 20 25Set size

Fix

atio

n D

urat

ion

(ms)


Dynamic Search

200

300

400

500

600

15 20 25Set size

Fixation D

uration (ms)


Figure 7 Median fixation duration by group, search type, target presence, and set size.



F(1, 37) = 1.2. There was no effect of group on saccadeamplitude, F(1, 37) = 0.4, nor were there any groupinteraction effects.

Fixation distribution by eccentricity

Figure 8 plots the distribution of fixations for the ASDand TD groups by the five stimulus eccentricitiescomprising the search array. The distribution offixations for target present trials was nearly identicalbetween search types and between groups for the innerthree eccentricities, all fairly flat and peaking at 5.82degrees. However, whereas the distribution of fixations tothe outer two eccentricities declined similarly betweengroups in the dynamic condition, fixations to the outerregions of the display declined more markedly in theASD than in the TD group in the static condition. Thisdifference was reflected in a three-way interactionbetween group, search type, and eccentricity, F(4,148) = 2.4, p < .05, gp

2 = .06. The target absentdistributions were highly similar for the two groups.Across groups, the static and dynamic absentdistributions peaked at 5.82 degrees, and the dynamicdistribution declined more sharply for the outereccentricities, F(4, 148) = 10.6, p < .001, gp

2 = .22.

Fixation distribution by favored quadrant

This analysis was conducted to rule out the possibilitythat participants, whether ASD or TD, searched as

efficiently in the dynamic as in the static condition bymeans of a different search strategy, that is, by simplyattending to a single area of the dynamic display andwaiting for the target to appear there (von M�hlenenet al., 2003). For each trial, each quadrant of the searchdisplay was ranked according to the number offixations made within its boundaries. The number offixations was then averaged across trials for the mostfavored quadrant, second most favored quadrant, andso on. These data are displayed in Figure 9. Ifparticipants tended to use a ‘sit-and-wait’ strategy, thedynamic distribution would be expected to fall at asignificantly quicker rate than the static distributionand, in the extreme case, there would be no fixationsoutside of the first-ranked quadrant in the dynamicsearch condition. However, this was not the case.Although, as reported above, there were more fixa-tions in static than in dynamic search and for absentthan for present trials, the pattern was similar acrossconditions. The search type quadrant interaction wasnot significant in the ASD group, F(3, 51) = 1.6,p = .21, gp

2 = .09, or in the TD group, F(3, 51) = 1.6,p = .20, gp

2 = .08.

Visual search skills and autism symptom severity

Autism symptom severity in participants with ASD wasassessed with Module 3 of the Autism DiagnosticObservation Schedule (ADOS) for children with fluentspeech (Lord et al., 1999). The ADOS involves a series

Static Search

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.94 3.88 5.82 7.76 9.7Eccentricity (°)

Fix

atio

ns p

er T

rial


Dynamic Search

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.94 3.88 5.82 7.76 9.7Eccentricity (°)

Fixations per T

rial


Figure 8 Distribution of fixations across stimulus eccentricities by group, search type, and target presence.

Static Search

0

0.5

1

1.5

2

2.5

Ranked Quadrant

Fix

atio

ns p

er T

rial


Dynamic Search

0

0.5

1

1.5

2

2.5

1 2 3 4 1 2 3 4Ranked Quadrant

Fixations per T

rial


Figure 9 Distribution of fixations according to favored quadrant by group, search type, and target presence.



of experimenter-administered social events and pressesdesigned to elicit behaviors relevant to an ASDdiagnosis. Although the ADOS was designed as aclassificatory instrument, its use as a continuousmeasure of symptom severity is supported by itsability to discriminate between autism and the lesssevere diagnosis of PDD-NOS in validation studies(Lord, Risi, Lambtecht, Cook, Lenventhal, DiLavore,Pickles & Rutter, 2000), and it has been used as acontinuous measure of symptom severity in numerouspublished studies (e.g. Dapretto, Davies, Pfeifer, Scott,Sigman, Bookheimer & Iacoboni, 2005; Joseph, Tager-Flusberg & Lord, 2002). The ADOS symptom measuresused were the summary scores from the Module 3diagnostic algorithm for communication (possible scoreof 0–8), reciprocal social interaction (possible score of0–14) and repetitive behavior (possible score of 0–8)symptoms, with higher scores indicating increasedseverity.

Correlational analyses were conducted to assess thedegree of association between symptom severity and theslope and intercept components of the RT · set sizefunction for the present and absent conditions of staticand dynamic search (see Table 2). Intercepts for staticpresent search were negatively associated with ADOSsocial symptom scores, r(19) = ).56, p < .01, indicatingthat lower intercepts were associated with increasedautism symptoms (see Figure 10). There were no otherstatistically significant correlations between search andADOS measures. Partial correlations controlling for theeffects of age, r(18) = ).61, p < .01, verbal IQ, r(18) =).57, p < .01, and nonverbal IQ, r(18) = ).57, p < .01,showed that the negative association between ADOSsocial score and static present search intercept wasindependent of each of these factors. Because interceptscan vary inversely in relationship to slopes, a partialcorrelation was conducted to ensure that the negativecorrelation between social symptoms and static presentintercepts was independent of the sizeable, although non-significant, positive correlation between social symptomsand static present slopes (see Table 2). When variance

associated with static present slopes was partialled, thecorrelation between static present intercepts and ADOSsocial symptoms remained significant, r(18) = ).45,p < .05.

Discussion

We compared visual search performance in a group ofrigorously diagnosed children with ASD and acomparison group of TD children who were matchedon age and nonverbal IQ and found, consistent withprior research, that individuals with ASD significantlyoutperformed non-ASD individuals. Our primary goal,however, was to begin to identify the processingmechanisms that yield superior search performance inindividuals with ASD. We considered mechanismsdirectly related to the process of search, and specificallythe possibility that memory optimizes the deployment ofattention through the search array in ASD, and weconsidered non-search mechanisms, most notably thepossibility that facilitated perceptual processing resultingin enhanced discrimination of targets from distractorsmight boost visual search skills in ASD. Finally, weassessed the relationship between visual searchsuperiority, and the specific processing differencescontributing to it, and symptom severity in ASDparticipants. We outline our main findings below.

Search superiority in ASD is not due to memory

We examined the possibility that enhanced memory forthe location of rejected distractors allows children with

Table 2 Correlations between ADOS symptom severityscores and search slopes and intercepts

ADOS scores

Communication SocialRepetitivebehavior

Search slopeStatic present .25 .39 .28Static absent ).24 )15 ).08Dynamic present .04 .21 ).11Dynamic absent ).13 .01 .09

Search interceptStatic present ).23 ).56** ).30Static absent .27 ).02 .06Dynamic present ).05 ).18 .20Dynamic absent ).17 ).18 ).12

*p < .05; **p < .01.

Static Present Search Intercept150010005000

AD

OS

Soc

ial S

ympt

om S

core

14

12

10

8

6

4

r = .56

Figure 10 Negative correlation between the intercept of theRT · set size function for static present search and socialsymptom severity among ASD participants.



ASD to sample the search array without replacementand thereby to augment their search efficiency. To do so,we compared children’s search efficiency in a standardstatic search with search efficiency in a dynamic searchcondition, in which targets and distractors randomlychanged locations every 500 ms, enforcing sampling withreplacement and rendering memory for searchedlocations of little value. Like the TD comparisongroup, children with ASD showed no decrement insearch efficiency in the dynamic relative to the staticsearch condition, leading to the conclusion that memorydoes not play a special role in the visual search skills ofchildren with ASD.

A possible objection to this conclusion is thatparticipants may have used a radically differentstrategy in the dynamic condition, namely ‘sitting andwaiting’ in one region of the search display until thetarget came to them (Geyer, von M�hlenen & M�ller,2007; Horowitz & Wolfe, 2003; von M�hlenen et al.,2003). However, analyses of participants’ eye movementsindicated that they used similar search strategies in thetwo conditions. Participants from both groups appearedto search somewhat less actively in the dynamiccondition. In dynamic as compared to static search,they made fewer fixations per trial (3.6 versus 4.1) thatwere of longer duration (462 versus 365 ms), and theymade shorter saccades (7.7� versus 7.9�) and fewerfixations to the outer eccentricities of the search display.Nevertheless, the eye-movement data showed thatparticipants searched the display in the dynamiccondition rather than remaining fixated in one region,as would be expected if a sit-and-wait strategy were used.Furthermore, the distribution of fixations from most toleast favored quadrants was the same in the static anddynamic conditions for both the TD and ASD groups.Had either group used a sit-and-wait strategy in thedynamic condition, fixations would have beenconcentrated in the most favored quadrant. Thus, theeye-movement data seem to rule out a fundamentaldifference in search strategy between the static anddynamic conditions and support the conclusion thatindividuals with ASD as well as TD individuals samplewith replacement in visual search.

Our finding that participants’ eye-movement behaviorwas not markedly different between static and dynamicsearch is at odds with results recently reported by Geyeret al. (2007). These authors found that the number offixations and amplitude of saccades were several degreesof magnitude less in dynamic than in static search,showing that the two search conditions were associatedwith clear-cut differences in eye movements andsuggesting that observers adopted a sit-and-waitstrategy in the dynamic condition. The most likelyexplanation for these conflicting findings is thedifference between studies in the refresh-rate of thedynamic stimuli. Whereas Geyer et al. used a frameduration of 117 ms in their dynamic search task, we useda frame duration of 500 ms. We chose this longer

presentation time to reduce the number of errors thatoccur with the shorter presentation time (Horowitz &Wolfe, 2003) as well as to allow the possibility of overteye movements within a single frame, thus more closelyapproximating the conditions of static search, while atthe same time greatly reducing the effectiveness of anymemory-based search strategy.

Search superiority in ASD appears to derive fromnon-search processes

We further investigated the processing differencesunderlying ASD search superiority by comparing theslopes and intercepts of the RT · set size functions forstatic and dynamic search between groups. If searchsuperiority arose from the speed and efficiency withwhich participants with ASD shifted attention betweenitems in the search display, we would expect the ASDgroup to exhibit shallower slopes than the TD group.In fact, search slopes did not differ between groups ineither static or dynamic search, suggesting that chil-dren with ASD not only sampled the search displaywithout replacement like TD children, but that they wereotherwise similar to TD children in the manner in whichthey deployed their attention while searching.

While slopes were equivalent between groups,intercepts were uniformly lower in the ASD group inboth static and dynamic search. According to theadditive factors logic of Sternberg (1966), the interceptof the RT · set size function is the portion of RT requiredby non-search aspects of the search task. These non-search factors can include initial, preattentive processingof basic feature information (e.g. color, orientation, size)as well as focal processing of stimulus features onceattention has been deployed to a location of interest. Inthe present study, the critical configural informationneeded to distinguish a target (T) from a distractor (L)could not be perceived preattentively; rather, focalattention was required to discriminate between targetsand distractors (Wolfe, 1992). Thus, it is likely that theASD advantage reflected faster processing of individualitems after they had been selected by attention.Additional support for the conclusion that the ASDadvantage arose from non-search processes comes fromour eye-movement findings. The number and spatialdistribution of eye movements did not differ markedlybetween the two groups, with the exception that ASDparticipants made fewer saccades to the edges of thedisplay in the static target present condition. However,the fixations of the ASD participants were significantlybriefer, suggesting that they were able to make stimulusdiscriminations at the locus of attention more quicklythan the TD participants.

Our results differed from those previously reported inthe literature in that our ASD participants did notexhibit increased search efficiency (decreased RT · setsize slopes) relative to TD participants in either the staticor dynamic condition. In prior studies, shallower search



slopes for individuals with ASD have either beenexplicitly reported (O’Riordan et al., 2001) or can beinferred from group · set size interactions showing areduced increase of RT with set size in ASD (Kemneret al., 2008; O’Riordan, 2004; O’Riordan & Plaisted,2001; Plaisted et al., 1998). One explanation for thedifference between our and prior studies is the type ofsearch stimuli used. Prior studies have used tasks inwhich preattentive processing of basic features couldguide the deployment of attention. These have includedthe use of color in conjunction searches (O’Riordan,2004; O’Riordan & Plaisted, 2001; Plaisted et al., 1998)and orientation in searches for a vertical line amongtilted lines (Kemner et al., 2008; O’Riordan et al., 2001).An improvement in search efficiency in these tasks couldreflect an enhancement of preattentive processing inindividuals with ASD. This explanation is in line withKemner et al.’s (2008) finding that individuals with ASDmade significantly fewer fixations than TD participantsin searching for a vertical line target among tilted linedistractors and their observation that many of their ASDparticipants were able to localize targets without movingtheir eyes. No benefit of this sort would be a factor in ourstudy because no basic feature information differentiatedtargets, Ts, from distractors, Ls.

An ASD advantage in preattentive processing mayalso explain the difference in eye-movement behaviorsobserved in the present study and those reported byKemner et al. (2008). Kemner et al. argued that eitherfewer or shorter fixations in their ASD participantswould be consistent with enhanced discriminatoryabilities underlying ASD search superiority, whereaslonger fixations, which would allow increased extractionof information from the periphery of the fixation (Hooge& Erkelens, 1999), would be indicative of a differentsearch strategy. Using the same feature search taskadministered by O’Riordan et al. (2001, Experiment 2),Kemner et al. found that their ASD participants weresignificantly faster than comparison participants atsearch. Further, supporting their enhanced discrimi-nation hypothesis, they found that ASD participantsmade fewer fixations, although they did not differ fromcomparison participants in fixation duration. Incontrast, we found that ASD participants made thesame number of fixations as comparison participants,but the fixations were shorter. Both findings indicateenhanced discriminatory abilities, according to thelogic of Kemner et al. Nevertheless, the differenceis intriguing, and may reflect more efficient use ofpreattentive orientation information by ASD partici-pants in Kemner et al.’s study. A preattentive advantagewould not have been of much use with our stimuli, soASD participants may have relied on briefer fixations toobtain their faster RTs.

Consistent with virtually all prior research on visualsearch in ASD, we found that the ASD search advantagewas more prominent on target absent than on targetpresent trials. There are several possible explanations for

this robust finding. O’Riordan et al. (2001) suggestedthat, because determining the absence of a target is moredifficult than determining the presence of a target, thetarget absent condition may have simply provided a moresensitive measure of group differences in search skills.O’Riordan et al. (2001) also suggested that an enhancedability to discriminate among stimuli may giveindividuals with ASD increased confidence in theirjudgment that no target is present. In other words,individuals with ASD benefited from a lower quittingthreshold on absent trials that allowed them to terminatesearch more expediently (Chun & Wolfe, 1996). Such adecision-stage advantage would manifest itself in reducedintercepts, reflecting non-search components of taskperformance, which is precisely what we found. It mayseem equally plausible that individuals with ASDterminated target absent searches more quickly thancomparison participants simply because they were lessmotivated to keep looking and not because they had aperceptual advantage. However, if this were true, wewould have expected a higher rate of error in the ASDgroup, which we did not find.

Also notable was our finding that children with ASDwere less affected in their reaction time by the dynamicsearch manipulation than TD children. We propose twopossible explanations for this. First, the consistentlyshorter fixation durations of ASD participants suggeststhat they were more able than TD participants tocomplete an inspection of an area of the display withinthe 500 ms presentation frame, which would have theeffect of reducing median RT. Second, it is alsoconceivable that children with ASD were less distractedby the random re-plotting and reorienting of the searchelements in each new search frame. Although speculative,the idea that the dynamic manipulation may haveintroduced less noise into the visual display for ASDparticipants is supported by recent evidence of impairedattentional prioritization of abrupt onset stimuli in ASD(Greenaway & Plaisted, 2005; Keehn & Joseph, 2008). Inthe context of dynamic search, reduced sensitivity to therepeated re-plotting of the search elements could havecontributed to the RT advantage of ASD participants.

Participants with ASD not only exhibited consistentlylower search intercepts than did TD participants, butintercepts in static search varied inversely with ADOSsocial symptom severity in the ASD group. In otherwords, individuals who showed evidence of an enhancedability to discriminate between targets and distractorshad higher levels of social symptoms, suggesting thatenhanced visual perception, and the putative differencesin brain organization and function that this perceptualenhancement reflects, may be of etiological significancein the development of autistic social impairment.Evidence of a relationship between atypically enhancedvisual perceptual skills and increased symptom severityin ASD is in agreement with the findings of Joseph et al.(2002), who investigated IQ profiles in relation tosymptom severity in ASD using the ADOS. They



found that children with ASD whose visual-perceptualskills were discrepantly higher than their verbal skillsexhibited increased social symptom severity on theADOS, and that this relationship was independent ofchildren’s absolute levels of verbal and overall cognitiveability. Nonetheless, we must consider our presentfinding preliminary and in need of further investigationgiven that it was limited to static present search andextended to neither the absent nor the dynamic searchconditions.

Conclusions and implications

In conclusion, our findings suggest that non-searchfactors in general and enhanced visual perception inparticular account for the visual search advantageexhibited by individuals with ASD. These findings fitwith a growing corpus of literature indicating anabnormal enhancement of visual perceptual abilities inASD (see Dakin & Frith, 2005; Mottron, Dawson,Soulieres, Hubert & Burack, 2006, for reviews). Theneural bases of enhanced visual–perceptual discrimi-nation in ASD remain to be determined. Two types ofpossibility have been considered. One is that localanomalies at the site of early visual processing, such asmalformation of minicolumn microcircuitry (Casanova,Buxhoeveden & Gomez, 2003), result in an over-amplification of lower-level visual information (e.g.Bertone, Mottron, Jelenic & Faubert, 2005; Plaisted,Saksida, Alcantara & Weisblatt, 2003). The other is thatdisturbances in longer-range cortical connectivity, apossible outcome of early cerebral overgrowth (Lewis &Ellman, 2008), results in diminished top-down modu-lation or integration of lower-level perceptual processes(e.g. Courchesne & Pierce, 2005; Just, Cherkassky, Keller,Kana & Minshew, 2007). These possibilities are notmutually exclusive, and the underlying disturbances inbrain organization (regionally non-specific abnormalitiesin rate of cerebral growth and in cortical minicolumnformation) hypothesized to cause them are notnecessarily independent of one another. Innovations innon-invasive imaging methodologies for investigatingbrain structure and function at the macroscopic level, inconjunction with continued histological study of thebrain at the microscopic level, promise to advance ourunderstanding of the brain bases of differences in visualperception and attention in ASD.

Although preliminary, our finding of a link betweenenhanced visual perception and increased autismsymptom severity may shed light on neurofunctionaldifferences that underlie the development of the definingbehaviors of ASD. It could be argued that atypicalperceptual processing in children with ASD is asecondary effect of profound social-communicativedeficits that alter the nature of the child’s earlyenvironmental inputs and disturb the normal course ofbrain growth and organization. Studies showing thatrestricted interests (including unusual sensory interests)

become more pronounced over the preschool years intoddlers diagnosed with ASD (Charman & Baird, 2002;Cox, Klein, Charman, Baird, Baron-Cohen, Swetten-ham, Drew & Wheelwright, 1999) would seem to agreewith this view. However, other evidence indicates thatunusual visual sensory interests are present at the ageof 2 (Lord, 1995) and as early as 1 year of age(Zwaigenbaum, Bryson, Roberts, Rogers, Brian &Szatmari, 2005) in children who develop ASD, whichwould make this view less tenable. At least two otherdistinct possibilities remain. First, the same fundamentaldefects in the regulation of brain ontogenesis in ASDmay result in the co-occurrence of functionallydissociable abnormalities in visual perceptual processesand social-communicative competence. Second,atypicalities in visual perception and attention maycontribute directly to abnormal social-communicativedevelopment in ASD, regardless of whether or not thesefunctional anomalies arise from the same neuropatho-logical processes. This seems especially plausible giventhat visual information processing is so vital to social-communicative functioning. Longitudinal behavioralresearch with young children with or at risk for autismwill help to determine between these two possibilities andwill have important implications for the development ofmore effective remedial and preventive treatments forASD symptoms.

Acknowledgements

This research was funded by NIDCD grant U19 DC03610 (Project 1, PI: RMJ), which is part of theNICHD ⁄ NIDCD Collaborative Programs of Excellencein Autism, and by NIMH grant K01 MH 073944 (PI:RMJ). TSH and JMW were supported by AFOSRF49620-97-1-0045 (PI: JMW). We would like to thankour CPEA site PI, Helen Tager-Flusberg, for hersupport, Rhyannon Bemis, Danielle Delosh, andRebecca McNally for assistance in data collection, andthe children and families who generously gave their timeto participate.

References

American Psychiatric Association (1994). Diagnostic andstatistical manual of mental disorders (DSM-IV, 4th edn.).Washington, DC: APA.

Bertone, A., Mottron, L., Jelenic, P., & Faubert, J. (2005).Enhanced and diminished visuo-spatial information pro-cessing in autism depends on stimulus complexity. Brain, 128,2430–2441.

Burack, J.A., Enns, J.T., Stauder, J.E.A., Mottron, L., &

Randolph, B. (1997). Attention and autism: behavioraland electrophysiological evidence. In D.J. Cohen, & F.R.Volkmar (Eds.), Autism and pervasive developmental dis-orders: A handbook (pp. 226–247). New York: John Wileyand Sons.



Caron, M.J., Mottron, L., Berthiaume, C., & Dawson, M.(2006). Cognitive mechanisms, specificity and neural under-pinnings of visuospatial peaks in autism. Brain, 129,1789–1802.

Casanova, M.F., Buxhoeveden, D.P., & Gomez, J. (2003).Disruption in the inhibitory architecture of the cellminicolumn: implications for autism. Neuroscientist, 9,496–507.

Charman, T., & Baird, G. (2002). Practitioner review: diag-nosis of autism spectrum disorder in 2- and 3-year-oldchildren. Journal of Child Psychology and Psychiatry, 43,289–305.

Chun, M.M., & Wolfe, J.M. (1996). Just say no: how are visualsearches terminated when there is no target present? Cogni-tive Psychology, 30, 39–78.

Courchesne, E., & Pierce, K. (2005). Why the frontal cortex inautism might be talking only to itself: local over-connectivitybut long-distance disconnection. Current Opinion in Neuro-biology, 15, 225–230.

Cox, A., Klein, K., Charman, T., Baird, G., Baron-Cohen, S.,Swettenham, J., Drew, A., & Wheelwright, S. (1999). Autismspectrum disorders at 20 and 42 months of age: stability ofclinical and ADI-R diagnosis. Journal of Child Psychologyand Psychiatry, 40, 719–732.

Dakin, S., & Frith, U. (2005). Vagaries of visual perception inautism. Neuron, 48, 497–507.

Dapretto, M., Davies, M., Pfeifer, J.H., Scott, A., Sigman, M.,Bookheimer, S., & Iacoboni, M. (2005). Understandingemotions in others: mirror neuron dysfunction in childrenwith autism spectrum disorders. Nature Neuroscience, 9,28–30.

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

Geyer, T., von M�hlenen, A., & M�ller, H.J. (2007). What doeye movements reveal about the role of memory in visualsearch? The Quarterly Journal of Experimental Psychology,60, 924–935.

Gibson, B.S., Li, L., Skow, E., Salvagni, K., & Cooke, L.(2000). Memory-based tagging of targets during visual searchfor one versus two identical targets. Psychological Science, 11,324–328.

Greenway, R., & Plaisted, K. (2005). Top-down attentionalmodulation in autistic spectrum disorders is stimulus-specific. Psychological Science, 16, 987–994.

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

Horowitz, T.S., & Wolfe, J.M. (1998). Visual search has nomemory. Nature, 394, 575–577.

Horowitz, T.S., & Wolfe, J.M. (2003). Memory for rejecteddistractors in visual search? Visual Cognition, 10, 257–298.

Horowitz, T.S., Wolfe, J.M., Keehn, B., Connolly, C., & Jo-seph, R.M. (unpublished manuscript). The absent-mindedinspector: what can eye movements tell us about strategies indynamic search?

Jolliffe, T., & Baron-Cohen, S. (1997). Are people with autismand Asperger syndrome faster than normal on the EmbeddedFigures Test? Journal of Child Psychology and Psychiatry, 38,527–534.

Joseph, R.M., Tager-Flusberg, H., & Lord, C. (2002). Cognitiveprofiles and social-communicative functioning in childrenwith autism spectrum disorder. Journal of Child Psychologyand Psychiatry, 43, 807–822.

Just, M.A., Cherkassky, V.L., Keller, T.A., Kana, R.K., &Minshew, N.J. (2007). Functional and anatomical corticalunderconnectivity in autism: evidence from an fMRI study ofan executive function task and corpus callosum morpho-metry. Cerebral Cortex, 17, 951–961.

Kaufman, A.S., & Kaufman, N.L. (2004). Kaufman briefintelligence test (2nd edn.). Circle Pines, MN: AmericanGuidance Service.

Keehn, B., & Joseph, R.M. (2008). Impaired prioritization ofnovel onset stimuli in autism spectrum disorder. Journal ofChild Psychology and Psychiatry, 49, 1296–1303.

Kemner, C., van Ewijk, L., van Engeland, H., & Hooge, I.(2008). Brief report: eye movements during visual searchtasks indicate enhanced stimulus discriminability in subjectswith PDD. Journal of Autism and Developmental Disorders,38 (3), 553–557.

Klein, R.M. (1988). Inhibitory tagging system facilitates visualsearch. Nature, 334, 430–431.

Lewis, J.D., & Elman, J.L. (2008). Growth-related neuralreorganization and the autism phenotype: a test of thehypothesis that altered brain growth leads to altered con-nectivity. Developmental Science, 11, 135–155.

Lord, C. (1995). Follow-up of two-year-olds referred for pos-sible autism. Journal of Child Psychology and Psychiatry, 36,1365–1382.

Lord, C., Risi, S., Lambrecht, L., Cook, E.H., Lenventhal,B.L., DiLavore, P.S., Pickles, A., & Rutter, M. (2000). TheAutism Diagnostic Obervation Schedule-Generic: a standardmeasure of social and communication deficits associated withthe spectrum of autism. Journal of Autism and DevelopmentalDisorders, 30, 205–223.

Lord, C., Rutter, M., DiLavore, P.C., & Risi, S. (1999). AutismDiagnostic Observation Schedule-WPS (ADOS-WPS). LosAngeles, CA: Western Psychological Services.

Mottron, L., Dawson, M., Soulieres, I., Hubert, B., & Burack,J. (2006). Enhanced perceptual functioning in autism: anupdate, and eight principles of autistic perception. Journal ofAutism and Developmental Disorders, 36, 27–43.

M�ller, H.J., & von M�hlenen, A. (2000). Probing distractorinhibition in visual search: inhibition of return. Journal ofExperimental Psychology: Human Perception and Perfor-mance, 26 (5), 1591–1605.

O’Riordan, M. (2000). Superior modulation of activation levelsof stimulus representations does not underlie superior dis-crimination in autism. Cognition, 77, 81–96.

O’Riordan, M.A. (2004). Superior visual search in adults withautism. Autism, 8, 229–248.

O’Riordan, M., & Plaisted, K. (2001). Enhanced discrimi-nation in autism. Journal of Experimental Psychology, 54A,961–979.

O’Riordan, M.A., Plaisted, K.C., Driver, J., & Baron-Cohen, S. (2001). Superior visual search in autism. Journal ofExperimental Psychology: Human Perception and Perfor-mance, 27, 719–730.

Plaisted, K.C. (2000). Aspects of autism that theory of mindcannot easily explain. In S. Baron-Cohen, H. Tager-Flusberg,& D.J. Cohen (Eds.), Understanding other minds: Perspectivesfrom autism and cognitive neuroscience (2nd edn., pp.222–250). Oxford: Oxford University Press.

Plaisted, K., O’Riordan, M., & Baron-Cohen, S. (1998). En-hanced visual search for a conjunctive target in autism: aresearch note. Journal of Child Psychology and Psychiatry, 39,777–783.



Plaisted, K., Saksida, L., Alcantara, J., & Weisblatt, E. (2003).Towards an understanding of the mechanisms of weak cen-tral coherence effects: experiments in visual configurallearning and auditory perception. Philosophical Transactionsof the Royal Society: B (Biological Sciences), 358 (1430),375–386.

Rutter, M., Le Couteur, A., & Lord, C. (2003). Autism Diag-nostic Interview-Revised. Los Angeles, CA: Western Psycho-logical Services.

Shah, A., & Frith, U. (1983). An islet of ability in autisticchildren: a research note. Journal of Child Psychology andPsychiatry, 24, 613–620.

Shah, A., & Frith, U. (1993). Why do autistic individuals showsuperior performance on the block design task? Journal ofChild Psychology and Psychiatry, 34, 1351–1364.

Sternberg, S. (1966). High-speed scanning in human memory.Science, 153, 652–654.

Takeda, Y., & Yagi, A. (2000). Inhibitory tagging in visualsearch can be found if search stimuli remain visible. Percep-tion and Psychophysics, 62 (5), 927–934.

von M�hlenen, A., M�ller, H.J., & M�ller, D. (2003). Sit-and-wait strategies in dynamic visual search. PsychologicalScience, 14, 309–314.

Wolfe, J.M. (1992). Parallel guidance of visual attention.Current Directions in Psychological Science, 1, 124–128.

Wolfe, J.M. (1994). Guided Search 2.0: a revised modelof visual search. Psychonomic Bulletin and Review, 1,202–238.

Wolfe, J.M., Horowitz, T.S., Van Wert, M.J., Kenner, N.,Place, S.S., & Kibbi, N. (2007). Low target prevalence is astubborn source of errors in visual search tasks. Journal ofExperimental Psychology: General, 136, 244–262.

Zwaigenbaum, L., Bryson, S., Roberts, W., Rogers, T., Brian,J., & Szatmari, P. (2005). Behavioral markers of autism in thefirst year of life. International Journal of DevelopmentalNeurosciences, 23, 143–152.

Received: 30 April 2008Accepted: 15 October 2008



Documents

PAPER Why is visual search superior in autism spectrum disorder?