No-onset looming motion guides spatial attention

No-Onset Looming Motion Guides Spatial Attention

Adrian von MuhlenenUniversity of Warwick

Alejandro LlerasUniversity of Illinois

These 6 experiments explored the ability of moving random dot patterns to attract attention, as measuredby a simple probe-detection task. Each trial began with random motion (i.e., dots linearly moved inrandom directions). After 1 s motion in 1 hemifield became gradually coherent (i.e., all dots moved up-,down-, left-, or rightwards, or either towards or away from a vanishing point). The results show that onlylooming motion attracted attention, even when the task became a more demanding discrimination task.This effect is not due to an apparent magnification of stimuli presented in the focus of expansion. Whenthe coherent motion started abruptly, all types of motion attracted attention at a short stimulus onsetasynchrony. The looming motion effect only disappeared when attention was drawn to the target locationby an arrow. These results suggest that looming motion plays a unique role in guiding spatial attention.

Keywords: visual attention, motion perception, random dot motion, automatic processes

In the everyday environment, the visual system is flooded withmotion information from a variety of sources. People and otheranimals actively explore and sample their visual world with eye,head, and body movements. This active exploration createschanges in the retinal image just as the movement of things orbeings in the environment does. These changes often carry impor-tant information to the observer, such as time to impact both forself-motion and for approaching objects and direction of heading,and they also carry information about the behavior of other thingsand beings in the environment (Gibson, 1950). Such informationcan be critical for the survival of the observer. Thus, the questionarises whether the visual system is tuned to detect certain criticalchanges in motion information as an adaptive way to deal withdangerous events like the sudden appearance of a predator or thefast approach of an in-coming projectile.

Recent investigations have opened up a debate regarding howthis type of motion information can guide visual attention (e.g.,Abrams & Christ, 2003, 2005; Franconeri & Simons, 2003). Oureveryday experience suggests that motion, such as an approachingcar or a waving hand, may indeed attract our attention veryefficiently. However, empirical studies in the laboratory havecome up with mixed results. For example, Hillstrom and Yantis

(1994) found that objects exhibiting motion do not generally drawattention in a visual search task unless the motion is predictive ofthe target’s location. This result was obtained with several types ofmotion, including oscillation, looming motion, and nearby movingcontours. Based on these findings and previous studies on theuniqueness of abrupt visual onsets in capturing attention (Jonides& Yantis, 1988; Yantis & Jonides, 1984), Hillstrom and Yantisproposed that only events that signal the appearance of a newobject in our visual field (such as luminance and motion onsets)attract attention. They termed this proposal the new-object hypoth-esis (Hillstrom & Yantis, 1994; see also Jonides & Yantis, 1988).

Franconeri and Simons (2003) put forward an alternative to thenew-object hypothesis. They argued that, in the absence of com-peting goals, all stimuli capture attention when they signal an eventthat could require an urgent action. According to this behavioral-urgency hypothesis, not only the onset of new objects but alsosome types of motion could signal behaviorally urgent events. Ina series of experiments, they showed that some types of motioncapture attention, although the motion information did not predictthe target location, whereas other types of motion did not captureattention. They argued that some types of motion are behaviorallymore urgent than others. More specifically, they proposed thatlooming motion as opposed to receding motion, is more likely toreceive attentional priority because it signals a potentially behav-iorally urgent event (e.g., an approaching object).

Abrams and Christ (2003), in contrast to Franconeri and Simons(2003), argued that motion as such is far too common in oureveryday surroundings to be informative of behaviorally urgentevents. Instead, they claimed that the onset of motion (i.e., thesudden transition from stationary to moving) carries behavioralurgency, as it might, for example, be an indicator of animacy (i.e.,whether an object is alive or not). As a result, Abrams and Christattributed previous results on attentional capture by motion to thebehavioral relevance of the onset of motion and not to the motionper se. In a recent comment, Abrams and Christ (2005) claimedthat Franconeri and Simons observed attentional capture by motionbecause, in their displays, the onset of motion had occurred very

This research was supported by a grant from the Swiss National ScienceFoundation to Adrian von Muhlenen and by National Science FoundationGrants 0527361 and 0309998 to Alejandro Lleras. Some of the findingswere presented at the 2003 Psychonomic Society Meeting in Vancouver,British Columbia, Canada, and at the 2004 Vision Science Society Meetingin Sarasota, FL. We thank Azadeh Arjmandi Rafsanjani for her help withthe data collection and Ron Rensinck for discussions in the early stages ofthis project.

Correspondence concerning this article should be addressed to Adrianvon Muhlenen, Department of Psychology, University of Warwick, Cov-entry CV4 7AL, United Kingdom; or Alejandro Lleras, Department ofPsychology and Beckman Institute for Advanced Sciences, University ofIllinois at Urbana–Champaign, 603 East Daniel Street, Champaign, IL61820. E-mail: [email protected] or [email protected]

Journal of Experimental Psychology: Copyright 2007 by the American Psychological AssociationHuman Perception and Performance2007, Vol. 33, No. 6, 1297–1310

0096-1523/07/$12.00 DOI: 10.1037/0096-1523.33.6.1297

1297

shortly before the search items were revealed. As a result, anyattentional benefits observed at the moving item could be attrib-uted to the trailing effects of the item’s onset of motion rather thanto an attentional benefit caused by the specific motion of the target.However, this motion-onset hypothesis does not account for whyFranconeri and Simons found attentional capture only with certaintypes of motion (see Abrams & Christ, 2005, for a possibleexplanation for this dissociation).

A competing view of attentional capture exists and was first putforward by Folk, Remington, and Johnston (1992). They arguedthat attentional capture depends on the attentional control settingsinduced by task demands. Under this view, motion informationshould capture attention only if motion is relevant to the task, thatis, only if motion is part of the set of features people are using tosolve the task. In other words, motion should not draw attention ifit serves no purpose in the observer’s task. For example, a wavinghand in a crowded street might capture your attention if you arelooking for a friend in that crowded street and you expect yourfriend to wave her hand to call your attention. On the other hand,that same waving hand might fail to capture your attention if youare looking for a street address. The role of attentional controlsettings has been experimentally confirmed not only with motionbut also with other features, such as color and onsets (e.g., Folk etal., 1992; Folk, Remington, & Wright, 1994).

In most of the previously cited studies, motion was the attributeof a task-relevant stimulus. Yet Yantis and Egeth (1999) arguedthat a true test of attentional capture occurs only under conditionsin which the target and the capturing feature are uncorrelated. Itcould be argued that in these previous studies, the occasionalcoincidence of the target and the motion features within the sameobject might have triggered an attentional set for the moving object(e.g., Folk & Remington, 1998; Folk et al., 1992). If so, then theobserved attentional capture effect might have been due to theco-occurrence of motion and “targetness” rendering motion atask-relevant attribute rather than to the motion itself. Conse-quently, to better test the capturing properties of different motions,one might want to use a display where motion is not a property ofany task-relevant stimuli but rather a property of a totally task-irrelevant stimulus.

The goal of the experiments described here was to use a motionstimulus that could be manipulated independently of the task-relevant stimuli in order to offer a true test of whether motion itselfcan affect the deployment of spatial attention. To this end, we usedmoving random dot patterns (RDPs), which were presented in thebackground of the display. RDPs have a number of properties thatmake them an interesting stimulus to study in connection withvisual attention. First, RDPs are not tied to a specific location orobject, but they can be spread (or scattered) over a large area,providing a background surface to the task of interest. Second,moving RDPs can form very strong perceptual groups when thedots move in a specific coherent direction. For example, they areseen as surfaces when the dots move in the same direction (Qian,Anderson, & Adelson, 1994) or as optic flow when the dots movetoward a vanishing point (Gibson, 1950; Lappe, Bremmer, & vanden Berg, 1999). Third, moving RDPs are very easy-to-vary stim-uli in psychophysical studies. For example, by changing the totalnumber of dots or the percentage of dots moving in a certain way,

one can gradually modify the qualia of the perceived motion. Inother words, it allows the experimenter to subtley tune the wholepercept of motion.

The RDP stimulus used in the current study consisted of smalldots randomly spread over a large area of the display. The RDPscould have random motion, coherent motion, or a mix of bothrandom and coherent motion. With random motion, each dotmoved in a straight line, but its direction was randomly assigned(0–360°). With coherent motion, dots also moved in straight lines,but the moving directions were not random but determined inadvance to give rise to the percept of a coherently moving pattern.A typical trial began with random motion in the whole display,followed by the gradual change from random to coherent motion inone subregion of the display. The specific aim of the current studywas to investigate whether this subregion of coherent motion isable to draw attention to itself. Because the coherent motionstimulus was distributed over a large area in the background, itwould not be easily associated with the task stimuli, which areused to assess the deployment of attention. The location of coher-ent motion was completely task irrelevant. Finally, motion did notbelong to any task-defined attentional sets, as the target did notmove and the onset of the motion was ramped, unlike the targetonset, which was sudden.

To briefly preview, Experiment 1 tested two forms of coherentmotion, looming and receding motion, that had led to differentcapture results in Franconeri and Simons’ (2003) study. The resultsshowed that looming but not receding motion effectively guidedspatial attention. Experiment 2 tested unidirectional (i.e., parallel)motion in up, down, left, or rightward direction; and no attentionalguidance effect was observed for any of these directions. Experi-ment 3 replicated the loom-guidance effect with a discriminationtask rather than a detection task. Experiment 4 tested the hypoth-esis that the looming motion effect might be due to a magnificationin the apparent size of stimuli presented at the focus of expansion(FOE). However, the observed magnification effect was very smalland unlikely to be responsible for the entirety of the loom-guidance effect. Experiment 5 changed the onset of coherentmotion from gradual to abrupt and found a cueing effect at theshort stimulus onset asynchrony (SOA) for all types of motion,including unidirectional motion. Finally, Experiment 6 showedthat the effect of looming motion was abolished when a 100%valid arrow cue, which always pointed to the target location, wasadded to the display.

Experiment 1

In Experiment 1, a moving RDP was used in order to examinewhether a subregion of coherently moving dots can guide spatialattention. The display was horizontally split into two halves (orhemifields), and coherent motion occurred either in the left or inthe right half, while motion in the other half continued to berandom. To assess the deployment of attention, a simple luminanceincrement probe was presented in the center of either the left or theright hemifield. Two types of coherent motion were used, loomingand receding motion. Based on the findings of Franconeri andSimons (2003), we would expect that the looming RDP would actas an automatic cue and draw attention to the FOE. In contrast, andsomewhat counterintuitively, with a receding RDP, we would

1298 VON MUHLENEN AND LLERAS

expect no attentional effect, even though the dots moved towardsthe possible target location.

One important aspect of our experiment served to greatly dif-ferentiate it from previous studies on capture by moving stimuli.Thanks to the characteristics of RDP, we were able to graduallyintroduce the coherent (looming and receding) motion in thedisplay, avoiding a sudden onset of motion. That is, each trialstarted with random motion followed by a slow increase in motioncoherence until full motion coherence was reached. This manipu-lation was important in order to rule out the possibility that thesudden onset of the coherent motion attracted attention (Abrams &Christ, 2003, 2005). If the sudden onset of motion was indeedresponsible for the Franconeri and Simons (2003) results, noattentional guidance should be found here. However, if the loom-ing/receding dissociation of Franconeri and Simons was replicatedwith our stimuli, this would provide strong support in favor of thebehavioral-urgency hypothesis.

Method

Participants. Fourteen participants (eight women, age range20 to 28 years) took part in Experiment 1. All participants hadnormal or corrected-to-normal visual acuity and received one halfextra credit in a psychology course for participating in a 30-minsession. None of the participants was aware of the purpose of thestudy.

Apparatus. The stimuli were presented on a Dell 17-in. mon-itor running in VGA graphics mode (640 � 480 pixel) at 60 Hertzand controlled by a Pentium II-75 (running MS Dos 6.22). Partic-ipants viewed the monitor from a distance of approximately 57 cm,providing a total screen area of approximately 32° � 24° of visualangle. Participants’ responses were recorded using the right buttonof a serial Microsoft mouse, with track ball removed to improvetiming accuracy (Segalowitz & Graves, 1990). The laboratory wasdimly illuminated.

Stimuli. A fixation cross (size 0.6° of visual angle) was pre-sented in the center of the display. The fixation cross was white,luminance 72.21 candela per square meter (cd/m2) at the start ofthe trial and changed to dark gray (luminance 4.46 cd/m2) with thestart of cue motion. The random dot pattern consisted of 400 dotsrandomly placed within a given area of 560 pixels width and 280pixels height (28° � 14° of visual angle). The average dot densitywas 1.02 dots/deg2. The area had no visible borders and wasplaced in the center of the screen. A dot consisted of a single whitepixel (size � 0.05°, luminance 72.21 cd/m2) and moved at aconstant speed of 3.0°/sec in a randomly assigned direction (0–360°). Each dot had a lifespan of 500 ms. When a dot reached theend of its lifespan or when it moved outside the viewing area, itwas randomly relocated and its age reset to zero. In order to keepthe dot relocation rate balanced throughout the trial (at approxi-mately 13 dots/refresh frame), the initial age of each dot wasrandomly chosen between 0 and 500 ms. On target trials, a smalldetection probe, consisting of a small light gray square (size 0.3°,luminance 27.35 cd/m2), was presented 7° to the right or left of thefixation cross. At both possible target locations, there was aninvisible “protection zone” (1° � 1°) of blank screen. That is,when a dot appeared inside or moved into the protection zone, itwas randomly relocated.

Procedure. Each trial started with all dots moving in randomdirections (see stimuli section). After 100 ms, the white fixationcross was presented at the center of the display. After 650 ms, thefixation cross turned gray and coherent motion started gradually toappear in either the left or the right hemifield. Two kinds ofcoherent motion were used: looming motion and receding motion(see Figure 1). In the looming-motion condition, all dots in thecued hemifield started moving away from the vanishing point,which was at the center of the hemifield (7° to the left or right offixation). In the receding-motion condition, all dots in the cuedhemifield started moving towards the hemifield center. Dots in theuncued hemifield continued to move in random directions.

Dots in the cued hemifield did not change their movementdirection at once, but instead continued to move in their initialdirection until they reached the end of their lifespan (for anillustration of the course of a trial, see Figure 2). Only at the endof their lifespan were dots randomly relocated and their movementdirection changed according to the coherent motion condition (andtheir new location). This procedure had the effect that coherentmotion was only gradually introduced, linearly increasing at anaverage rate of 6.6 dots/frame. Full motion coherence was reached500 ms after cue onset. The cue was uninformative regarding thelocation of the target, which appeared either on the side withcoherent motion (50% valid cue) or on the side with randommotion (50% invalid cue). The target remained visible until theresponse button was pressed. The SOA between cue onset andtarget onset was either 500 or 1,000 ms.

Task and instruction. Participants were asked to keep theireyes fixated at the center cross throughout the whole trial. Theparticipant’s task was to detect the onset of the target stimulus asquickly as possible by pressing the right mouse button (go/no-gotask). In 14.3% of the trials, no target appeared (catch trials). Thetrial ended and the screen was cleared after a response was re-corded or 2 s had elapsed since the target onset, whichever hap-pened first. Trials with reaction times (RTs) shorter than 100 ms or

Figure 1. Schematic example displays for looming and receding cuemotion in Experiment 1. Arrows represent the direction of moving dots.

1299LOOMING MOTION GUIDES SPATIAL ATTENTION

longer than 2,000 ms were counted as wrong and were accompa-nied by acoustic error feedback. The intertrial interval was 1,000ms after correct responses and 2,500 ms after incorrect responses.Participants received written instructions, providing a general de-scription of the task and the procedure. They were explicitlyinstructed to concentrate on the target detection task and to ignorethe moving dots in the background.

Design. There were 24 trials for each combination of motion(looming, receding), cue validity (valid, invalid), and SOA (500,1,000 ms). In addition, there were 32 catch trials, giving a total of224 experimental trials, which were presented in random order.The experiment was divided into four blocks. At the beginning ofthe experiment, participants performed a block of 56 practicetrials. Each block began with three unrecorded warm-up trials.Between blocks, there were short mandatory breaks of at least 10 s.

Results

Correct RT trials were used to calculate mean RTs for each par-ticipant and condition, excluding outliers (two standard deviationsbelow or above mean). Of the total number of trials, 7.4% wereexcluded as errors or outliers. The false alarm rate was 4.9%. Table 1presents mean RTs and standard error means (SEMs) for the valid andinvalid trials and for the two SOA conditions. The values for loomingmotion are presented in the first row and for receding motion in thesecond row. The mean cueing effect (the difference between invalidand valid RTs) is plotted in Figure 3 as a function of SOA.

Individual RTs from each participant were subjected to a three-way analysis of variance (ANOVA) with the main variables beingmotion (looming, receding), cue validity (valid, invalid), and SOA(500, 1,000 ms). The main effects for cue validity, F(1, 13) � 9.14,p � .01, and SOA, F(1, 13) � 36.83, p � .001, were significant:RTs were 26 ms faster with valid than with invalid cues and 39 msfaster with long than with short SOAs. Furthermore, there was asignificant Cue Validity � SOA interaction, F(1, 13) � 4.98, p �.05, due to a larger cueing effect with long SOA than with shortSOA (32 vs. 20 ms, respectively), and a significant Cue Validity �Motion interaction, F(1, 13) � 10.91, p � .01. Two separatefollow-up ANOVAs, one for looming motion and one for recedingmotion, showed that the cue validity effect was significant withlooming motion, F(1, 13) � 29.33, p � .001, but not with reced-ing motion, F(1, 13) � 0.04, nonsignificant (ns, 46 vs. 5 ms,respectively).

Discussion

The results of Experiment 1 showed a clear advantage fordetecting targets when they appeared at the side with loomingmotion compared to when they appeared at the opposite side,where motion was random. This cueing effect was not significantlyaffected by SOA, indicating that looming motion had similareffects immediately after reaching full coherence and 500 ms later.This RT advantage is in accord with the attentional shifting ac-count (Posner, Snyder, & Davidson, 1980), where attention isdrawn to the cued side (here by the looming motion), facilitatingthe processing of stimuli presented at this side at the cost of stimulipresented at the other noncued location.

No comparable cueing effect was found with receding motion,although this type of motion is in many respects similar to loomingmotion. Importantly, the similarities in some of the low-levelfeatures between these two types of motion make the receding-motion condition an ideal control condition for the looming mo-tion. Obviously, having a vanishing point is not enough to drawattention. Some other aspect of looming motion must attract at-tention in a way that receding motion cannot. The behavioral-urgency hypothesis offers a possible explanation for this dissoci-ation: Looming motion draws attention because it signals the

Figure 2. Schematic course of a trial illustrating the ramped cue onset ofcoherent motion used in Experiment 1. At the beginning of a trial, dots inthe cued hemifield were moving in random directions (0% coherence).After 750 ms, coherent cue motion gradually started to come in, increasinglinearly for 500 ms until maximal motion coherence (100%) was reached.Full coherent motion was then continued until the end of the trial. Thestimulus onset asynchrony (SOA) between cue and target was measuredbetween the start of the coherent cue motion and the onset of the target.

Table 1Mean Reaction Time and Standard Error Means (inParentheses) for Experiment 1

Motion

SOA 500 ms SOA 1,000 ms

Invalid Valid Invalid Valid

Looming 477 (17) 438 (14) 451 (16) 397 (13)Receding 463 (17) 478 (20) 437 (16) 426 (18)

Note. SOA � stimulus onset asynchrony.

Figure 3. Mean cueing effect (the difference between invalid and validreaction times) and associated standard error means in ms for looming andreceding cue motion as a function of stimulus onset asynchrony (SOA) inExperiment 1.


presence of an approaching object (which is potentially dangerous)or an impending collision.

A second possibility is that our results might have been causedby the optokinetic nystagmus (OKN): a series of rhythmic reflex-ive eye movements evoked by a large moving visual field (seeCohen, Matsuo, & Raphan, 1977; Watanabe, 2001). In OKN, slowpursuit phases, where the eyes smoothly track the movement of thevisual field (in order to stabilize the retinal image) are interruptedby fast saccade phases, where the eyes jump in the oppositedirection of motion. Curiously, as the pursuit continues, the aver-age eye position also shifts in the opposite direction of motion.Recently, Watanabe (2001) argued that these fast saccades arepreceded by attentional shifts. Watanabe tested participants’ abilityto detect a brief flash in a field of unidirectional moving dots inone of two conditions: The flash could appear either in the in-coming field of dots or in the out-going field of dots, with respectto fixation. Thus, to detect the flash, participants would have tomove their attention either in the direction opposite to the motion(when the flash appeared in the in-coming field of motion) or inthe same direction as the motion (when the flash appeared in theout-going field of motion). Watanabe found an RT advantage fordetecting targets in the in-coming field condition over detectingtargets in the out-going field of motion. Importantly, this advan-tage occurred both when eye movements were allowed (consistentwith the OKN having helped move eyes and/or attention in thein-coming–motion condition) and where eye movements weresuppressed. Watanabe concluded that unidirectional motion tendsto automatically shift visual attention towards the field of in-coming motion. This in-coming field advantage is in some sensesimilar to the looming motion advantage because in both casesattention was drawn in the direction opposite to the motion.

Experiment 2

The findings of Experiment 1 suggest that there was somethingunique about the looming stimuli that drew attention in a mannerin which the receding stimuli could not. In Experiment 2, we testedthe possibility that not the looming stimulus as a whole but only asubset of the looming dots guided attention in Experiment 1. Forexample, in accord with Watanabe’s (2001) in-coming field ad-vantage, one could argue that only the dots moving towardsfixation were responsible for the attentional effect. Likewise, it ispossible that for some unforeseen reason only the downwardmoving part of the looming stimulus attracted attention. In short,before reaching a conclusion about the uniqueness of the loomingmotion pattern as a whole, one needs to rule out the possibility thatthe effect was not caused by specific component patterns of thelooming stimuli. Upwards, downwards, rightwards, and leftwardsfields of unidirectional motion were tested in Experiment 2.

Method

Fourteen participants (eight women, age range 20 to 33 years)took part in Experiment 2. Apparatus and stimuli were the same asin Experiment 1. The only difference from Experiment 1 was thetype of coherent motion: Dots in the cued hemifield moved eitherrightwards, downwards, leftwards, or upwards. Procedure, task,and instructions were the same as in Experiment 1. The design ofExperiment 2 had 24 trials for each combination of motion (right-

ward, downward, leftward, or upward), cue validity (valid, in-valid), and SOA (500, 1,000 ms). In addition, there were 64 catchtrials (with no target), giving a total of 448 experimental trials,which were divided into eight blocks.

Results

In Experiment 2, 6.0% of the total number of trials were ex-cluded from the analysis as errors or outliers. The false alarm ratewas 2.7%. Table 2 presents the mean RTs and SEMs for the validand invalid trials and for the two SOA conditions.

The correct RT data were analyzed by a three-way ANOVAwith main terms for motion (rightward, downward, leftward, up-ward), cue validity (valid, invalid), and SOA (500, 1,000 ms).Only the SOA main effect was significant, SOAs, F(1, 13) �12.08, p � .01: RTs were, on average, 17 ms slower with the shortthan with the long SOA. No other effects were significant. Indi-vidual t-tests showed that the cueing effect did not reach signifi-cance in any of the eight Motion � SOA combinations (all ps �.16).

Discussion

No reliable cueing effect was found, neither with upward, down-ward, rightward, nor leftward motion. Note, however, that hori-zontal motion was classified into left- and rightward motions,irrespective of the side of the cue (left or right hemifield). It couldwell be, for example, that leftward motion had a different effectwhen it was on the left hemifield, than when it was on the righthemifield. If it had opposite effects on each side, these effectswould cancel each other out when put in the same category. Thehorizontal motion data were therefore reanalyzed with two newcategories, the in-coming–motion condition, which grouped left-ward motion in the right hemifield with rightward motion in theleft hemifield, and the out-going–motion condition, which groupedleftward motion in the left hemifield and rightward motion in theright hemifield. The mean cueing effects (and SEM in parentheses)for the two SOA conditions, 500 and 1,000 ms, were 2 (11) and 1(9) ms with in-coming motion and 14 (12) and �6 (15) ms without-going motion, respectively. The corresponding three-wayANOVA showed no significant effects or interactions involvingcue validity (all ps � .38).

Table 2Mean Reaction Time and Standard Error Means (inParentheses) for Experiment 2

Motion

SOA 500 ms SOA 1,000 ms


Unidirectional 504 (17) 493 (17) 483 (17) 482 (16)Up 497 (16) 482 (20) 489 (17) 483 (15)Right 518 (16) 493 (16) 493 (16) 474 (16)Down 497 (16) 488 (18) 470 (18) 493 (14)Left 504 (17) 511 (14) 481 (15) 477 (18)

Note. The averaged values for the four unidirectional motion conditionsare summarized in Row 1, whereas the individual values for the fourmotion directions (upwards, rightwards, downwards, leftwards) are pre-sented in rows 2–5. SOA � stimulus onset asynchrony.


At first sight, the absence of an in-coming field advantage in thecurrent experiment stands in sharp contrast to Watanabe (2001),who showed a clear advantage for detecting targets appearing inthe in-coming field of motion. However, our RDP-stimulus dif-fered from Watanabe’s stimulus in several important ways. First,our RDP was smaller (14° � 14° vs. 32° � 32°, respectively) incomparison to Watanabe and the dots moved slower (3.0 vs.12.9°/sec, respectively). Maybe the overall motion signal in ourstudy was simply not strong or fast enough to yield the type ofattentional effect observed by Watanabe. Second, in our study theunidirectional motion was only present in one hemifield (the otherhemifield had random motion), whereas in Watanabe’s study thesame unidirectional motion was present in both hemifields. That is,in our study, the unidirectional motion ended at the middle of thescreen, whereas unidirectional motion was homogeneous acrossthe entire display (and perhaps crucially was homogeneous aroundfixation) in Watanabe’s displays. Thus, perhaps continuity ofmotion across hemifields (and around fixation) might be requiredfor an in-coming field advantage to occur.

Experiment 3

The results of Experiments 1 and 2 suggest that our visualsystem might process looming motion in a unique manner, so as toextract from this signal information useful in guiding spatial at-tention. However, it is possible that the result with looming motionin Experiment 1 was obtained not because of attentional guidanceto the FOE, but rather to a perceptual interaction between themotion pattern (dots moving away from the FOE) and the targetstimulus (a larger dot appearing inside the FOE). Maybe, in thisparticular task, the target dot might have played a privileged rolein the experiment because it might have been perceived as part ofthe looming pattern and its onset might have been encoded as achange in the looming pattern. Thus, it is possible that perceivingthat unique change in the looming pattern might be easier thanperceiving the target onset in the random motion hemifield.

In Experiment 3, we addressed this perceptual interaction hy-pothesis by changing the task to a more difficult letter-discrimination task. Furthermore, we made the task stimuli largerin order to make their integration with the looming motion evenmore difficult (and less likely). If the cueing effect observed inExperiment 1 was due to a perceptual interaction between thelooming pattern and the target dot, no such benefit should beobserved here. If, on the other hand, the looming motion doesguide spatial attention to its FOE, then not only target detection,but also target identification should be improved, which should

result in an even larger RT benefit. To briefly preview, largecueing effects were observed in the looming condition, whereas noeffect was found in the receding motion condition.

Method

Fourteen participants (nine women, age range 18 to 45 years)took part in Experiment 3. Apparatus and the RDP backgroundstimulus were the same as in Experiment 1, the only changeconcerned the stimuli used for the current discrimination task. Twogray letters (size 1°, thickness 0.1°, luminance 27.35 cd/m2) wererandomly rotated by 0, 90, 180, or 270°, and presented one 7° leftand the other 7° right of center fixation. The letters were largeenough so they could be discriminated in the periphery without theneed to make an eye movement. One letter, the target, was eithera “T” or an “L”; the other letter, the distractor, was always an “F.”The participants’ task was to press the left mouse button when theysaw a “T” or the right mouse button when they saw an “L”.Incorrect responses were accompanied by acoustic error feedback.Participants were instructed to maintain an overall error ratesmaller than 5%.

There were two changes to the experimental design. First, therewas a random condition, during which no cue was presented andall dots continued to move in random directions throughout theentire trial. Second, there were three levels of SOAs, 250, 500, and1,000 ms. There were 32 trials for each combination of motion(looming, receding), cue validity (valid, invalid), and SOA (250,500, 1,000). There were no catch trials, but 96 trials with randommotion in both hemifields (32 for each SOA level). Overall,participants completed eight blocks of 60 trials each.

Results

RT analysis. In Experiment 3, 3.0% of all trials were excludedfrom analyses as outliers. Table 3 presents the mean RTs and SEMfor Experiment 3. The mean cueing effect is plotted in Figure 4 asa function of SOA.

The valid and invalid RT data in Experiment 3 were analyzed bya three-way ANOVA examining motion (looming, receding), cuevalidity (valid, invalid), and SOA (250, 500, 1,000 ms). The maineffects for motion, F(1, 13) � 18.56, p � .001, and cue validity,F(1, 13) � 28.41, p � .001, and the Motion � Cue Validityinteraction, F(1, 13) � 22.26, p � .001, were significant. As canalso be seen from Figure 4, this was mainly due to the cueing effectbeing present only in the looming condition and not in the recedingcondition (53 vs. �2 ms, respectively). Separate follow-up

Table 3Mean Reaction Times and Standard Error Means for Experiment 3

Motion

SOA 250 ms SOA 500 ms SOA, 1,000 ms

Invalid Valid Neutral Invalid Valid Neutral Invalid Valid Neutral

Looming 750 (23) 729 (26) 769 (30) 697 (29) 778 (39) 713 (32)Receding 751 (29) 770 (29) 757 (31) 745 (26) 764 (29) 760 (30)Random 757 (32) 743 (26) 742 (32)

Note. Invalid/valid looming and receding conditions are shown in the first two rows and the (neutral) random condition in the third row. SOA � stimulusonset asynchrony.


ANOVAs for looming and receding motion confirmed this inter-pretation: The cue validity main effect was significant in thelooming ANOVA, F(1, 13) � 42.89, p � .01, but not in thereceding ANOVA, F(1, 13) � 0.07, ns. Furthermore, in the loom-ing ANOVA, the interaction between cue validity and SOA wassignificant, F(1, 13) � 7.35, p � .005, which was due to strongercueing effects with the two longer SOA conditions.

Random motion condition. To include the random condition inthe analysis, two separate ANOVAs, one with receding/randommotion and one with looming/random motion, were calculatedwith the factors cue validity (valid, invalid, neutral) and SOA. TheANOVA with receding/random motion showed no significant ef-fects. The ANOVA with looming/random motion showed a sig-nificant main effect for cue validity, F(2, 26) � 25.90, p � .001,and a significant Cue Validity � SOA interaction, F(4, 52) � 2.83,p � .05. Three split-up ANOVAs further examined the differencesbetween the three cue validity levels: Valid RTs were significantlyfaster than neutral RTs, F(1, 13) � 21.94, p � .001, and thaninvalid RTs, F(1, 13) � 27.34, p � .001, and neutral RTs weresignificantly faster than invalid RTs, F(1, 13) � 7.13, p � .05.

Error analysis. Because errors can be more informative indiscrimination tasks, the mean error rates of Experiment 3 (seeTable 4) were arcsine-transformed and analyzed by a three-wayANOVA with the factors motion (looming, receding), cue validity(valid, invalid), and SOA (250, 500, 1,000 ms). However, theANOVA revealed no significant effects (all ps � .13). Moreover,errors and RTs in this experiment were slightly positively corre-lated, rS(15) � .18, �2 � 0.89, p � .34, indicating that a speed-accuracy trading relationship is not a problem in this experiment.

Discussion

Looming motion enhances not only the detection but also thediscrimination of targets presented at the FOE. This effect built upvery fast and reached its maximum at the 500 ms SOA, when themotion cue had just reached full coherence. There was no furtheradvantage at the 1,000 ms SOA condition. The magnitude of thecueing effect at the 1,000 ms SOA condition in this experimentwas similar to that in Experiment 1 (65 vs. 54 ms, respectively),although overall RTs were considerably longer in Experiment 3than in Experiment 1 (750 vs. 450 ms, respectively), as was to beexpected given the increased difficulty of the discrimination task.In sum, the current finding replicated the cueing effect with loom-ing motion of Experiment 1 with different stimuli (letters ratherthan a detection probe), in a different task (discrimination ratherthan detection), and at considerably slower levels of performance.

The overall pattern of results, including the random condition,strongly argues against the perceptual interaction hypothesis whilebeing consistent with the attentional guidance account. First andforemost, having found a large cueing benefit using large letters astask stimuli suggests that the same cueing effect found in Exper-iment 1 was not due to a unique perceptual interaction between thelooming pattern and the target dot. Second, the RT costs in theinvalid condition compared to the neutral condition suggest that,on invalid trials, looming motion misguided attention first to thedistractor location. Again, if the perceptual interaction hypothesiswere true, invalid and neutral RTs would have been identical, as inboth conditions the target appeared surrounded by random motion.In sum, this overall pattern of results is much more consistent withattentional shifts accounts (e.g., Posner, Snyder, & Davidson,1980) where attention is drawn by the looming motion than with aperceptual interaction account.

It is worth mentioning that the results of Experiment 3 help tofurther differentiate our effects from those of Watanabe (2001)discussed earlier. Watanabe obtained an in-coming field advantagefor detection and localization tasks but failed to obtain one duringa shape-identification task (Watanabe, 2001, Experiment 2). Incontrast to his findings, Experiment 3 showed a cueing effect withlooming motion in a discrimination task. This provides furtherevidence that Watanabe’s in-coming field advantage and the loom-ing motion advantage might rely on different mechanisms.

It is also important to highlight that by presenting two letters inthe display (one target and one distractor), the task of Experiment3 became very similar to the visual search task (granted with a setsize of 2), which is often used in the study of attentional capture(e.g., Hillstrom & Yantis, 1994). We thus believe that Experiment3 successfully demonstrated that attention is drawn by looming

Figure 4. Mean cueing effect and standard error means in ms for loomingand receding cue motion as a function of stimulus onset asynchrony (SOA)in Experiment 3.

Table 4Mean Errors and Standard Error Means for Experiment 3

Motion

SOA 250 ms SOA 500 ms SOA 1,000 ms

Invalid Valid Neutral Invalid Valid Neutral Invalid Valid Neutral

Looming 7.4 (2.5) 8.5 (2.6) 8.3 (2.5) 7.8 (2.3) 6.9 (2.5) 4.9 (1.3)Receding 6.7 (2.3) 7.6 (2.2) 7.8 (2.2) 6.9 (1.4) 9.2 (2.8) 7.4 (1.8)Random 7.1 (2.7) 6.9 (2.7) 8.0 (2.7)

Note. SOA � stimulus onset asynchrony.


motion in the context of a visual search task, even when thelooming motion is totally irrelevant to the task and motion is nota part of the attentional set for the task.

It has been suggested that looming and receding motion mightprovide more exact information about the target’s location thanlinear motion and that looming and receding motion might differ insome other way.1 In order to investigate this location informationaccount, a control experiment similar to Experiment 3 was con-ducted. The only difference to Experiment 3 was that two squareplace holders, which had the same size as the letter stimuli, werepositioned at the two positions where the letter stimuli appeared.The results of 14 participants showed a similar pattern to Exper-iment 3: The cueing effect was more pronounced in the loomingcondition than in the receding condition (35 vs. 5 ms, respec-tively), Motion � Cue Validity interaction, F(1, 13) � 6.15, p �.05. The two separate ANOVAs showed that the cueing effect wassignificant in the looming condition, F(1, 13) � 26.48, p � .001,but not in the receding condition (F � 1). If location informationwas the crucial variable in the previous experiments, then thelooming motion effect should have disappeared because the place-holders provide the same information at both locations. However,that was not the case; and, even if location information hadcontributed to some degree to the effect of looming motion (assuggested by the reduced cueing effect of 35 ms in comparison tothe cueing effect of 53 ms in Experiment 3), it would not explainthe entire effect found in Experiments 1 and 3 nor the differentialeffects found for looming and receding motion.

Experiment 4

Experiment 3 showed that, even in a discrimination task with largeletters, a strong cueing effect by looming motion could be observed.However, it is still possible that yet another type of perceptual effectmight be responsible for this cueing effect: Perhaps the loomingstimulus in the background creates the impression of depth in thedisplay (with the FOE being the farthest point in the display). Fromoptical illusions such as the Ponzo illusion, it is well known thatstimuli presented closer to the vanishing point appear to be bigger insize. Hence, the looming motion could have a similar effect on thetarget stimulus, making it bigger, and thus easier to detect and dis-criminate. The goal of Experiment 4 was to estimate the size of sucha magnification effect using a simple size-discrimination task, so thatwe could estimate the impact that such magnification might have onthe detection of stimuli in our experiments.

In this experiment, two squares varying in size were presentedsimultaneously, one in each hemifield. As before, RDPs were inthe background, with random motion in one hemifield and loomingmotion in the other hemifield. If the square surrounded by loomingmotion appears bigger, the point of subjective equality would bereached for actual squares smaller than those of the square sur-rounded by random motion.

Method

Participants. Fourteen participants (11 women, age range 20to 36 years) took part in Experiment 4.

Apparatus, stimuli, and task. Apparatus and RDP stimuli werethe same as in Experiment 1, with gradual onset of looming motionin one hemifield and continuing random motion in the other

hemifield. The side of the looming motion was randomly deter-mined on each trial. The stimuli and task were new to this exper-iment. The stimuli relevant for the task consisted of two light grayoutline squares (thickness 0.1°, luminance 27.35 cd/m2). The sizeof the standard square was always 20 pixels in length (size 1°). Weused the same size as in Experiment 3, because the size used inExperiments 1 and 2 (4 pixels) was too small to allow subtlechanges in size on a conventional raster monitor. The size of thecomparison square was systematically varied between 17 and 23pixels (size 0.85–1.15°). The locations (center of left or righthemifield) of the two squares were determined randomly on eachtrial. Thus, in half the trials the comparison square appeared in thehemifield with looming motion, in the other half it appeared in thehemifield with random motion. Participants’ task was to decidewhich square was larger and press the corresponding left or rightmouse button.

Procedure and design. The trial sequence was similar to Ex-periment 1. The SOA between the onset of the cue and the twosquares was randomly chosen between 800 and 1,200 ms. Thesquares were only shown for 150 ms, but the RDP motion in thebackground continued until participants pressed a button. Partici-pants received no feedback throughout the experiment. The exper-iment had two independent variables, size of the comparisonstimulus (17, 18, 19, 20, 21, 22, 23 pixels), and motion in thebackground of the comparison stimulus (random, looming). Therewere no catch trials and 36 trials for each combination of size andmotion. Overall, participants completed 504 trials divided in nineblocks of 56 trials each. The dependent variable was the frequencyof reporting that the comparison stimulus was bigger. RTs are notreported as speed was not stressed in this task.

Results

Figure 5 plots the psychometric function of the percentage oftrials, where the comparison stimulus appeared bigger than thestandard stimulus, as a function of the size of the comparisonstimulus. The figure shows among other things, that when bothstimuli were equal in size (20 pixels), the stimulus presented in thelooming hemifield was judged to be bigger than the stimulus in therandom hemifield in 57.2% of the trials. The horizontal differenceof the two functions at the 50% chance level is 0.6 pixel. Thisdifference is statistically significant, t(13) � 2.91, p � .05. Thehalf of this value (0.3 pixel) provides a good measure for the sizeoverestimation at the looming hemifield. The difference must behalved because looming motion has an effect twice, on the com-parison stimulus when it was shown in the looming hemifield aswell as on the standard stimulus when it was shown in the loominghemifield.

Discussion

The enlarging effect of the looming motion, although statisti-cally significant, was only 0.3 pixel. This means that a square thathad a physical size of 20 pixels appeared to be 20.3 pixels widewhen it was surrounded by looming motion. This is only anincrease of 1.5% in terms of square length or 3.2% in terms of

1 This suggestion was made by an anonymous reviewer.


square area. We believe that such a small increase in apparent sizeis very unlikely to be the cause for RT differences in the range of50 to 80 ms, as was found in the previous experiments. In another,unreported experiment, we actually tested the RT benefit createdby larger targets in a detection task. We used the same loomingRDP as in Experiment 1 and systematically varied the target size(filled square with a size of 0.25–0.4°). Based on these results, weare able to estimate that an absolute increase in size of 0.3 pixelproduces an RT benefit of 6 ms (a relative increase of 1.5%produces even less benefit, 1.3 ms). This is insufficient to accountfor the overall looming effect reported in the previous experiments.We are therefore quite confident that the looming effect is not, oris to only a very small part, due to an increase in apparent size ofthe target stimulus. Rather, we believe it is consistent with theattentional shifting account (Posner et al., 1980) and thereforeprovides further support for the behavioral-urgency hypothesis(Franconeri & Simons, 2003).

Experiment 5

In all previous experiments, the coherent motion had a gradualonset; that is, it took 500 ms before the motion reached its fullcoherence. This was done in order to avoid attentional capture bythe sudden abrupt onset in one of the hemifields of the display. InExperiment 5, we tested the same motion conditions as in Exper-iments 1 and 2, but this time we used a sudden cue onset. That is,motion in the cued hemifield changed at once from random tocoherent. According to Abrams and Christ’s (2003, 2005) motion-onset hypothesis, the sudden onset of coherent motion in onehemifield should capture attention, regardless of motion type (thatis, if motion onsets play a unique role in guiding attention). If wewant to demonstrate that the guiding effect of looming motion isfundamentally different from that created by onsets, it is importantthat we show that our stimuli can indeed capture attention undersudden onset conditions that are more similar to those that havebeen previously studied. We expected that if the visual systemdoes indeed process looming motion in some unique fashion, we

should find that the guiding effects of looming motion go aboveand beyond the capture effects created by a sudden onset.

Method

Twenty new participants took part in Experiment 5, 8 men and12 women, aged between 20 and 28 years. Apparatus and stimuliwere the same as in Experiments 1 and 2, with six types ofcoherent motion: looming, receding, upwards, rightwards, down-wards, and leftwards. The only difference was the onset of cuemotion. At cue onset, all dots in the cued hemifield changed theirmovement direction at once, following their path of coherentmotion. There were two SOA conditions, 250 and 750 ms. Therewere 20 trials for each combination of motion type (looming,receding, upwards, rightwards, downwards, leftwards), cue valid-ity (valid, invalid), and SOA (250, 750). In addition, there were 80catch trials. The experiment was divided into 10 blocks of 56 trialseach. The number of participants was increased from previousexperiments to increase the stability of our measures because thenumber of trials in each cell of the design was lowered to fit theexperiment in one session.

Results

In Experiment 5, 5.6% of the total number of trials was excludedfrom the analysis as errors or outliers. The false alarm rate was5.0%. Table 5 presents the mean RTs and SEMs for Experiment 5.Figure 6 presents the cueing effect as a function of SOA forlooming and receding motion, as well as the mean of the unidi-rectional motion. As can be seen from Figure 6, the RTs for thelooming condition showed a strong cueing effect at both SOAs.The receding motion showed no effect at all and the linear motionsshowed only an effect at the short SOA.

The RT data were analyzed by a three-way ANOVA with themain variables being motion type (right, down, left, up, looming,receding), cue validity (valid, invalid), and SOA (250, 750 ms). Allthree main effects were significant: motion type, F(5, 95) � 8.22,p � .001, cue validity, F(1, 19) � 25.38, p � .001, and SOA, F(1,19) � 7.48, p � .05. Furthermore, the two-way interactionsbetween cue validity and SOA, F(1, 19) � 9.97, p � .01, and

Figure 5. Percentage of responses where the comparison square appearedbigger than the standard square, as a function of the size of the comparisonsquare in Experiment 4. The solid line represents the condition where thecomparison stimulus appeared in the hemifield with looming motion; thedotted line represents the condition where the comparison stimulus ap-peared in the hemifield with random motion.

Table 5Mean Reaction Time and Standard Error Means for Experiment5

Motion

SOA 250 ms SOA 750 ms


Unidirectional 519 (19) 483 (16) 501 (19) 487 (19)Up 534 (21) 489 (17) 508 (22) 480 (16)Right 520 (19) 477 (18) 502 (21) 485 (17)Down 512 (19) 477 (13) 505 (21) 496 (22)Left 508 (17) 489 (16) 487 (15) 486 (19)

Looming 541 (24) 451 (14) 511 (20) 434 (13)Receding 536 (18) 527 (19) 516 (20) 519 (15)

Note. The averaged values for the four unidirectional motion conditionsare summarized in Row 1, and their individual values are presented inRows 2–5. Rows 6 and 7 show the reaction times for looming and recedingmotion. SOA � stimulus onset asynchrony.


between motion type and cue validity, F(5, 95) � 7.92, p � .001,were significant. The former interaction indicates that the cueingeffect diminishes with the long SOA (45 vs. 29 ms, respectively).The latter Motion Type � Cue Validity interaction was furtherexplored by three separate ANOVAs. The first ANOVA looked atunidirectional motion and had the factors motion (right, down, left,up), cue validity and SOA. The main effect for cue validity, F(1,19) � 14.02, p � .001, and its interaction with SOA, F(1, 19) �7.79, p � .05, were significant. Valid RTs were on average 25 msfaster than invalid RTs, and this effect was stronger at the shortSOA than at the long SOA (36 vs. 14, respectively). Further testsshowed that this cueing effect was only significant at the shortSOA, F(1, 19) � 26.21, p � .001. Note that none of the effectsinvolving motion type reached significance (all ps � 0.21). Thesecond ANOVA looked at looming motion and had the factors cuevalidity and SOA. Only the two main effects were significant:Valid RTs were, on average, 83 ms faster than invalid RTs, F(1,19) � 34.38, p � .001, and RTs were 23 ms slower at the shortSOA than at the long SOA, F(1, 19) � 5.94, p � .05. The thirdANOVA looking at receding motion, with the factors cue validityand SOA, revealed no significant effects (all ps � .12).

Discussion

The method of Experiment 5 was almost identical to the methodof Experiments 1 and 2 except that coherent motion now had asudden onset (i.e., the dots in the coherent hemifield changed theirdirection at once). In the case of unidirectional motion, the suddenonset of coherent motion had an early facilitatory effect at the shortSOA, which diminished at the long SOA. Note that in Experiment2, unidirectional motion showed no cueing effects at the shortSOA. The fact that the sudden cue onset attracted attention only fora short time is not untypical for abrupt-onset stimuli (e.g., Yantis,1993). Such an early reflexive component is similar to that foundin the classical cueing paradigm with flashing cues (e.g., Muller &Rabbitt, 1989; Posner et al., 1980). This result is also similar toAbrams and Christ (2003), who showed that not motion per se butthe onset of motion captures attention. In some way the currentexperiments further qualify Abrams and Christ’s account, arguingthat the motion onset must be abrupt in order to capture attention.However, at the longer interval, they found the opposite effect

(longer discrimination times for the previously moving target),which they took as evidence for inhibition of return (IOR). In ourExperiment 5, there is no indication for IOR at the long SOA. Wecan only speculate that 750 ms was not long enough to produceIOR or, perhaps, that the continued presence of coherent motion inour displays might have prevented IOR. Indeed, other studies haveshown that not only the cue’s onset but also the cue’s offset isimportant to obtain IOR (e.g., Riggio, Scaramuzza, & Umilta,2000). We further discuss these issues in the General Discussionsection.

For looming motion, there was a strong cueing advantage atboth SOA conditions. The magnitude of the cueing effect waslarger than the one observed in Experiment 1, and it decreased withSOA, in contrast to Experiment 1, where it did not change withSOA. It is almost as if the magnitude of the cueing effect inExperiment 5 with looming motion was the added benefit of thesudden onset (only present at the short SOA, and similar inmagnitude to the effect observed here with unidirectional motion)and of the looming motion itself (present at both short and longSOAs, as evidenced in Experiments 1 and 3). This suggests thatthe cueing effects with looming motion observed in Experiments 1and 3 were produced by a different mechanism than the one(s)responsible for producing the cueing effect observed with suddenonsets.

Finally, the receding motion showed no significant cueing ad-vantage neither at the short nor at the long SOA. Although notsignificant ( p � .12), the cueing effect showed a similar decreasewith SOA as with unidirectional and looming motion. This mightbe an indication that the sudden onset still had an effect here, butthat it might have overlapped with a possibly negative effect of thereceding motion. The lack of a guiding effect for this type ofmotion, even in the presence of a sudden onset, deserves moreattention in future studies. Although the visual system appears tohave specific mechanisms for the processing of looming motion(because of its behavioral urgency) and for sudden onsets, thesemechanisms might fail to be recruited for the processing of a morecomplex and behaviorally nonurgent motion pattern, such as re-ceding motion. Note that the lack of a cueing effect here is by itselfnot surprising as motion in general had no relevance to the task andwas not predictive of target location.

Experiment 6

If the cueing advantage for looming motion was due to atten-tional capture, then the manipulation of attention by another cuethat is 100% valid should strongly interfere with, or even abolish,the attentional effect of looming motion. In order to test this idea,we carried out an experiment similar to Experiments 1 and 2, butadded a condition where the central cross was replaced by anarrow that always pointed to the side where the target was going toappear. Because the arrows point to a box or some other object inmost central cueing tasks and because it might be difficult to directattention to an empty space, a third condition was included thatwas like the arrow condition except that two box placeholders(similar to those in the control experiment conducted for Experi-ment 3) were added at the center of each hemifield. We expectedthat if looming motion was an attentional effect, then adding a100% valid arrow should abolish the effect of looming motion.

Figure 6. Mean cueing effect and standard error means in ms for loom-ing, receding and unidirectional cue motion as a function of stimulus onsetasynchrony (SOA) in Experiment 5.


Method

Fourteen participants took part in Experiment 6, 4 men and 10women, aged between 20 and 33 years. Apparatus and stimuliwere the same as in Experiments 1 and 2, with four types ofcoherent motion: looming, receding, upwards, and downwards.SOA was randomly varied between 650 and 850 ms. There werethree arrow conditions: The first condition had no arrow but afixation cross and was in all respects identical to the correspondingconditions of Experiments 1 and 2. The second condition, arrow-only, was like the no-arrow condition, except that the centralfixation cross was replaced by an arrow (length 0.8°, width 0.5°)pointing either to the left or to the right side. Importantly, thearrow always pointed to the side where the target was going toappear (i.e., it represents a 100% valid cue for the position of thetarget). The third condition, arrow-and-box, was like the arrow-only condition except that two boxes were added to the display,one at the center of each hemifield where the target stimuli couldappear. There were 24 trials for each combination of motion type(looming, receding, unidirectional), cue validity (valid, invalid),and arrow condition (no arrow, arrow only, arrow and box). Inaddition, there were 72 catch trials. The experiment had a total of504 trials and was divided into nine blocks of 56 trials each. Arrowcondition was kept constant within a block, but randomly variedbetween blocks.

Results

In Experiment 6, 6.4% of the trials were excluded from theanalysis as errors or outliers. The false alarm rate was 3.1%. Table6 presents the mean RTs and SEM for each arrow condition and foreach motion condition. Figure 7 presents the cueing effect as afunction of arrow condition, with separate bars for the looming, thereceding, and the unidirectional motion. As can be seen fromFigure 7, the cueing effect with looming motion in the standardcondition gets smaller when an arrow is added and vanishes whena box is added in addition to the arrow.

A three-way ANOVA on the RTs with the main variables—motion type (looming, receding, unidirectional), cue validity(valid, invalid), and arrow condition (no arrow, arrow only, arrowand box)—revealed, apart from the motion main effect, all mainand interaction effects to be highly significant (all ps � .01),especially the three-way interaction, F(2, 26) � 4.97, p � .01. Tofurther explore this interaction, three separate 3 � 2-wayANOVAs with motion type and cue validity were calculated, onefor each arrow condition. The no-arrow ANOVA and the arrow-only ANOVA both revealed significant interaction effects, F(2,26) � 6.39, p � .01, and F(2, 26) � 11.80, p � .001, respectively.Further split up t-tests revealed that, in the no-arrow condition,

only the cueing effect with looming motion was significant,t(13) � 5.70, p � .001, whereas in the arrow-only condition, thecueing effect for both looming motion, t(13) � 4.27, p � .01, andreceding motion, t(13) � 6.04, p � .01, were significant. Finally,there were no significant cueing effects in the arrow-and-boxcondition (all ts � 1).

Discussion

The results of Experiment 6 confirm our hypothesis: The cueingeffect is reduced when an arrow pointing to the target location isadded to the display. In the case of looming motion, the cueingeffect of 99 ms diminishes to 32 ms when a 100% valid arrow cueis added and almost completely disappears (1 ms) when the arrowis accompanied by two box place holders. The residual cueingeffect in the arrow-only condition might be because, on invalidtrials, participants had difficulties moving their attention to anempty location within a uniform background (i.e., when the arrowpointed to the hemifield with random motion). In contrast, on validtrials, these difficulties did not occur (or occurred less often)because the arrow pointed to the hemifield with looming motion,where the FOE provided a landmark for the deployment of atten-tion. In other words, the cueing effect in the arrow-only conditionwas due to an interaction of arrow and the positional informationof the looming motion. A similar logic applies to the effectobserved with receding motion in the arrow-only condition, wherea significant cueing effect of 32 ms was observed. In this condi-tion, the focus of contraction seemed to provide a useful landmarkthat facilitated attentional shifts on valid trials, when the arrow waspointing to the hemifield with receding motion. The results of thearrow and box condition seem to be consistent with the spatial-

Table 6Mean Reaction Time and Standard Error Means for Experiment 6

Motion

No Arrow Arrow Arrow & Box

Invalid Valid Invalid Valid Invalid Valid

Unidirectional 515 (22) 416 (13) 378 (14) 346 (12) 406 (19) 386 (14)Looming 505 (20) 469 (22) 372 (14) 340 (14) 398 (17) 387 (16)Receding 491 (18) 471 (18) 370 (12) 369 (12) 405 (15) 391 (15)

Figure 7. Mean cueing effect and standard error means in ms for loom-ing, receding, and unidirectional cue motion as a function of cue conditionin Experiment 6.


landmark interpretation of the arrow-only condition: once land-marks were added to both hemifields, participants seemed to beable to utilize the arrow information perfectly, and therefore noeffects of background motion were observed on RT.

Experiment 6 showed in the no-arrow condition a surprisinglylarge cueing effect of 36 ms with receding motion in comparisonto Experiments 1 and 5 (5 and 3 ms, respectively). This couldpossibly be due to carry over effects from blocks with arrows tono-arrow blocks. Unfortunately the number of trials within a blockwas too small to test this hypothesis. Also note that this effect wasonly marginally significant and it was to some extent due to twoparticipants with extraordinarily large cueing effects of 156 and147 ms; the remaining participants had a mean cueing effect of16 ms.

General Discussion

The principal question of the present research was whether amoving RDP with coherent motion could guide attention underconditions where the RDP is completely task irrelevant. The re-sults can be summarized as follows: First, RDPs with loomingmotion, but not with receding or unidirectional motion, signifi-cantly shortened detection times of targets presented at the hemi-field with coherent motion (Experiments 1 and 2). Second, theperformance improvement with looming-motion was not confinedto detection tasks but also occurred with discrimination tasks(Experiment 3). Third, the looming motion effect was not due to anincrease in apparent size of the stimulus presented at the FOE(Experiment 4). Fourth, when the onset of coherent motion wasabrupt, attention was briefly drawn to the coherent motion hemi-field by all types of motion (Experiment 5). Fifth, when an arrowis added to the display that always points to the side of the target,the cueing effect with looming motion disappears (Experiment 6).Thus, the answer to the above question is a resounding yes: RDPswith coherent motion can have a guidance effect on attention,either when they contain looming motion or when the coherentmotion has a sudden onset, even when the RDP stimulus is totallytask irrelevant. So, how do these findings inform current theoriesof attentional capture?

The new-object hypothesis (Jonides & Yantis, 1988) posits thata perceptual event must signal the appearance of a new object inorder to capture attention. The current finding with looming mo-tion is not easily explained by this account, as the coherent motionin the background is not tied to a specific object. One could arguethat the gradual onset of coherent motion represents the appear-ance of a new object (cf. Valdes-Sosa, Cobo, & Pinilla, 2000).However, if this were the case, it would not explain why a cueingeffect was absent with receding motion, and especially with uni-directional motion, which is perceived as a sliding surface. Thatbeing said, it may be the case that a weaker version of thenew-object hypothesis is correct, as it applies to all stimuli, butlooming motion. That is, looming motion might be the exceptionto the new-object rule.

The behavioral-urgency hypothesis (Franconeri & Simons,2003) proposes that events that might require an urgent actioncapture attention. Such events can be the sudden appearance ofnew objects, but they can also be the presence of certain types ofmotion in the environment. The current findings, especially thedissociation between looming and receding motion, fits well into

this account. Looming motion might signal danger, such as anobject approaching an observer. Receding motion on the otherhand might represent an object moving away from the observer,which is less urgent and does not require an urgent action. Thus,our results support the behavioral-urgency hypothesis and furtherconfirm the dissociation between looming and receding motioninitially reported by Franconeri and Simons (2003). It is importantto highlight however, that our findings are unique in that wesuccessfully dissociated motion effects from sudden-onset effects(compare Experiments 1, 2, and 5), whereas these two effects wereconfounded in Franconeri and Simons’s studies (all motions hadsudden onsets).

In contrast to Franconeri and Simons (2003), Abrams and Christ(2005) argued that it is not a specific kind of motion but the onsetof any motion that captures attention. On the one hand, our resultsare consistent with this view: In Experiment 5, the sudden onset ofcoherent motion had an early guiding effect at the short SOA onattention, for all but receding types of motion. Furthermore, whenthe same types of coherent motion were tested with gradual onsets,this early guiding effect was not observed (except for loomingmotion), as might be expected if motion onsets are required forattentional guidance. On the other hand, Abrams and Christ’sexplanation does not account for the strong looming effects ob-served with gradual-onset motion because their account does notdifferentiate between different types of motion. Thus, motion onsetmight be sufficient but not necessary to guide attention withmoving stimulus. It is important to remind the reader that Abramsand Christ did not directly compare looming and receding motion;however, they did obtain sudden-onset capture effects with reced-ing motion (which we failed to observe) when the motion wasproduced using stereo depth cues. This discrepancy regardingattentional guidance (or lack thereof) with receding motion stimulideserves careful scrutiny in future research.

Last, we address the role that attentional control settings mighthave played in our studies (Folk et al, 1992). Experiment 5provides evidence in line with the attentional-set hypothesis ofFolk and colleagues because the target was a sudden onset (whichcan be interpreted as a sharp discontinuity in the perceptual signal)and coherent motion also had a sudden onset. Thus, for all butreceding motion, the results from Experiment 5 can be accountedfor by having onsets or perceptual discontinuities as part of theattentional set for the task. However, the results of Experiments 1,2, and 3 are more difficult to reconcile with this view of attentionalcapture. The most stringent test came with Experiment 3 in whichno stimulus attributes were shared between the target stimulus andthe moving RDPs, and still an attentional effect was observed. Inaddition, because coherent motion had no predictive value regard-ing the location of the target, it is difficult to imagine why loomingmotion (but not receding motion) would be included in an atten-tional set for the task. Clearly, one could reconcile our results withtheir account if one were to assume that attentional control settingsare not solely determined by task factors. That is, it would bereasonable to argue that, regardless of the task, participants havesome hard-wired attentional control settings that are always active(such as a filter to detect their own name, a loud siren, or, in ourcase, looming motion). How such hard-wired settings becomeimplemented is a question for future research, but one can specu-late that repeated prior experience under varied tasks (as with


one’s name) may lead to the generalization of some task-specificsettings to be active during all tasks.

Cueing Effect and IOR

IOR refers to the behavioral cost to responding to a targetstimulus when the target appearance is somewhat delayed (morethan 300 ms) with respect to the presentation of an initial periph-eral signal even when the cue is totally uninformative as to thelocation (or identity) of the target (Posner & Cohen, 1984). It isinteresting that in our current experiments, which deal with atten-tional facilitation effects akin to those initially described by Pos-ner, Snyder, & Davidson (1980), we failed to observe any hints ofIOR when the SOA between motion onset and target onset wasincreased past 300 ms. As indicated by Klein (2000), the absenceof IOR in an experiment might be due to at least three straight-forward reasons. First, it is possible that the IOR effect waspresent, but that the task was insensitive to the magnitude of theeffect. Although possible, we believe it is unlikely that our taskslack sensitivity to detect an IOR effect, given that the tasks weused (detection of luminance onset and discrimination of targetidentity) have been used to successfully study IOR in the past (foran overview see Klein, 2000) and, further, because the tasks weresensitive enough to measure quite significant attentional facilita-tion effects. A second reason to fail to observe IOR is when theIOR effect is obscured by an accompanying effect in the oppositedirection. In the context of our experiments, it is unclear whatother effect could have obscured the IOR effect, though we do notethat the magnitude of our cueing effects with looming motion weresignificantly larger than those observed on typical Posner-cueingtype tasks (which tend to be 10–20 ms on average). So, it ispossible that the IOR effect might have been somehow obscuredby this much larger initial cueing advantage.

Third, it is also possible that an IOR effect was not found in ourexperiments simply because the effect was not there in the firstplace. Two possibilities could account for IOR’s true absence. (a)IOR requires attentional disengagement from the cued location,and such disengagement might not take place in our tasks becauseof the continued presence of the peripheral motion signal. Notethat this is an important difference between previous IOR studiesin cue-target experiments and ours. To our knowledge, in typicalIOR experiments the cue signal is always extinct prior to theappearance of the target. We are currently investigating this hy-pothesis in our labs. (b) Another possibility is that the attentionaladvantage produced by looming motion (as well as the lack ofIOR) might be intricately related to the uniqueness of loomingmotion itself. In other words, it is possible that the type of atten-tional benefits created by the looming motion signal are somewhatdifferent from those that are created by more transient stimuli(such as a brief flash) and, as such, might not give rise to IOR inthe same manner or on the same scale. This hypothesis is also thefocus of current investigations in our labs.

On the Uniqueness of Looming Motion

In addition to Franconeri and Simons’ (2003) and the currentstudy, other evidence supports the uniqueness of looming motionin vision. For example, Takeuchi (1997) used a visual searchparadigm to study the perception of looming and receding motion

in small squares. Takeuchi found that search for a looming targetsquare among receding distractor squares was very efficient andnot much affected by the number of distractors: the mean searchrates were smaller than 5 ms/item for both target-present andtarget-absent trials. In sharp contrast, search for a receding targetamong looming distractors was clearly inefficient: The meansearch rates were 30 ms/item for target-present and 60 ms/item fortarget-absent trials. This search asymmetry is again consistent withspatial attention having been guided by looming motion and not byreceding motion. Treisman and Gormican (1988) had previouslyproposed that such search asymmetries arise when the presence ofa feature is easier to discern than its absence; and, thus, Takeuchihypothesized the existence of specialized expanding-motion pro-cessing units in the human visual system. Our findings are con-sistent with the existence of such units and further suggest thatthese specialized units might actually be linked to the mechanismsin charge of deploying spatial attention. However, it is important tonote that alternative explanations for the looming–receding asym-metry in visual search have been proposed. For example, vonGrunau and Dube (1994) proposed that both units detecting loom-ing motion and units detecting receding motion exist, but that theformer should have higher sensitivity than the latter because of theecological factors constraining the two signals (such difference insensitivity would easily account for the search asymmetry). Still,the higher sensitivity of looming-motion detectors when added tothe behavioral importance of such motion signals might have led tothose detectors being preferentially linked to spatial attentionguiding mechanisms.

Further evidence for the uniqueness of looming motion comesfrom neurophysiological studies. Zeki (1974) was among the firstwho showed that the monkey visual cortex has single neurons thatare sensitive to changing size. More specifically, the medial supe-rior temporal area of the Macaque monkey has many directionallyselective cells, which respond either to straight parallel motion, torotation, or to expansion/contraction (Tanaka & Saito, 1989). Mostrelevant to our results here, cells responding to expansion weremuch more common than cells responding to contraction (76 vs.11 cells). Thus, it might be that both expansion and contractioncells are connected to spatial attention guiding mechanisms; but, inthe presence of competing signals, the larger presence of expan-sion signals might more easily affect the deployment of spatialattention than the relatively under-represented contraction signals.This finding might actually shed light on the discrepancy betweenour lack of attentional guidance by receding motion and thecapture effect with sudden-onset receding signals in the Abrams &Christ study (2003): It might be possible that having used stereo-depth cues in their study, the resulting receding signal might havebeen stronger than the one created by our RDPs; and, in particular,the receding signal might have been strong enough to drive theredeployment of spatial attention in a way our perhaps weakersignal could not.

Further evidence on the uniqueness of looming motion comesfrom the developmental literature. In classic studies by Jouen andhis colleagues (Jouen, 1990; Jouen, Lepecq, Gapenne, &Bertenthal, 2000), the sensitivity of 3-day old infants to loomingmotion has been tested. Although, these infants had obviously verylittle perceptual experience, they responded to the incoming mo-tion flow pattern by tilting their head backwards. The magnitude oftheir response was measured by changes in the pressure their heads


exerted onto supporting air pillows. Impressively, the incrementalhead pressure was linearly related to the velocity of the optic flow,thus demonstrating high levels of sensitivity to the looming motionstimulus and to fairly small changes in that motion signal. Thus,humans seem to have an innate special ability to detect (and reactto) looming motion.

Conclusion

After careful examination, we have found that the onset ofmotion most often captures attention regardless of the kind ofmotion that follows the onset (with the possible exception ofreceding motion) even when the motion stimulus is irrelevant tothe task. Although motion onsets capture attention, the presence ofcoherent motion in the display most often fails to capture attention.One important exception, however, is looming motion, whichseems to benefit from a unique standing in the visual system:looming motion, even under conditions when motion informationis irrelevant to the task and is not part of any task-defined atten-tional sets, produces an attentional advantage unlike that of anyother motion or that created by sudden onsets. Whether spatialattention is drawn to the FOE as soon as looming motion isdetected in the environment or whether the deployment of spatialattention to the location of the FOE is expedited by the presence ofthe looming motion we cannot tell yet. However, current experi-ments in our laboratories are aimed at investigating these and othermechanisms that might be involved in this attentional effect bylooming motion.

References

Abrams, R. A., & Christ, S. E. (2003). Motion onset captures attention.Psychological Science, 14, 427–432.

Abrams, R. A., & Christ, S. E. (2005). The onset of receding motioncaptures attention: Comment on Franconeri & Simons (2003). Percep-tion & Psychophysics, 67, 219–223.

Cohen, B., Matsuo, V., & Raphan, T. (1977). Quantitative analysis of thevelocity characteristics of optokinetic nystagmus and optokinetic after-nystagmus. Journal of Physiology, 270, 321–344.

Folk, C. L., & Remington, R. W. (1998). Selectivity in distraction byirrelevant featural singletons: Evidence for two forms of attentionalcapture. Journal of Experimental Psychology: Human Perception &Performance, 24, 847–858.

Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntarycovert orienting is contingent on attentional control settings. Journal ofExperimental Psychology: Human Perception & Performance, 18,1030–1044.

Folk, C. L., Remington, R. W., & Wright, J. H. (1994). The structure ofattentional control: Contingent attentional capture by apparent motion,abrupt onset, and color. Journal of Experimental Psychology: HumanPerception & Performance, 20, 317–329.

Franconeri, S. L., & Simons, D. J. (2003). Moving and looming stimulicapture attention. Perception & Psychophysics, 65 (7), 999–1010.

Gibson, J. J. (1950). The perception of the visual world. Boston: HoughtonMifflin.

Hillstrom, A. P., & Yantis, S. (1994). Visual motion and attentionalcapture. Perception & Psychophysics, 55, 399–411.

Jonides, J., & Yantis, S. (1988). Uniqueness of abrupt visual onset incapturing attention. Perception & Psychophysics, 43, 346–354.

Jouen, F. (1990). Early visual-vestibular interactions and postural devel-

opment. In H. Bloch & B. I. Bertenthal (Eds.), Sensory-motor organi-zation and development in infancy and early childhood (pp. 199–215).Dordrecht, The Netherlands: Kluwer.

Jouen, F., Lepecq, J-C., Gapenne, O., & Bertenthal, B. I. (2000). Opticflow sensitivity in neonates. Infant Behavior & Development, 23 (3–4),271–284.

Klein, R. M. (2000). Inhibition of return. Trends in Cognitive Sciences, 4,138–146.

Lappe, M. Bremmer, F., & van den Berg, A. V. (1999). Perception ofself-motion from visual flow. Trends in Cognitive Sciences, 3 (9),329–336.

Muller, H. J., & Rabbitt, P. M. A. (1989). Reflexive and voluntaryorienting of visual attention: Time course of activation and resistance tointerruption. Journal of Experimental Psychology: Human Perception &Performance, 15, 315–330.

Posner, M. I., & Cohen, Y. (1984). Components of visual orienting. In H.Bouma & D. Bonwhuis (Eds.), Attention and Performance X: Control ofLanguage Processes (pp. 551–556). Hillsdale, NJ: Erlbaum.

Posner, M. I., Snyder, C. R. R., Davidson, B. J. (1980). Attention and thedetection of signals. Journal of Experimental Psychology: General, 109,160–174.

Qian, N., Andersen, R. A., & Adelson, E. H. (1994). Transparent motionperception as detection of unbalanced motion signals I: Psychophysics.Journal of Neuroscience, 14, 7357–7366.

Riggio, L., Scaramuzza, E., & Umilta C. (2000). Modulation of inhibitionof return by type and number of dynamic changes of the cues. Psycho-logical Research, 64, 56–65.

Segalowitz, S. J., & Graves, R. E. (1990). Suitability of IBM XT, AT, andPS/2 keyboard, mouse, and game port as response devices in reactiontime paradigms. Behavior Research Methods, Instruments, & Comput-ers, 22, 283–298.

Takeuchi, T. (1997). Visual search of expansion and contraction. VisionResearch, 37, 2083–2090.

Tanaka, K., & Saito, H. (1989). Analysis of motion of the visual field bydirection, expansion/contraction, and rotation cells clustered in the dor-sal part of the medial superior temporal area of the macaque monkey.Journal of Neurophysiology, 62, 626–641.

Treisman, A., & Gormican, S. (1988). Feature analysis in early vision:Evidence from search asymmetries. Psychological Review, 95, 15–48.

Valdes-Sosa, M., Cobo, & A., Pinilla, T. (2000). Attention to object filesdefined by transparent motion. Journal of Experimental Psychology:Human Perception & Performance, 26 (2), 448–505.

von Grunau, M., & Dube, S. (1994). Visual search asymmetry for viewingdirection. Perception & Psychophysics, 56, 211–220.

Watanabe, K. (2001). Modulation of spatial attention with unidirectionalfield motion: An implication for the shift of the OKN beating field.Vision Research, 41, 801–814.

Yantis, S. (1993). Stimulus-driven attentional capture. Current Directionsin Psychological Science, 2, 156–161.

Yantis, S., & Egeth, H. E. (1999). On the distinction between visualsalience and stimulus-driven attentional capture. Journal of Experimen-tal Psychology: Human Perception & Performance, 25, 661–676.

Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective atten-tion: Evidence from visual search. Journal of Experimental Psychology:Human Perception & Performance, 10, 601–621.

Zeki, S. M. (1974). Functional organization of a visual area in the posteriorbank of the superior temporal sulcus of the rhesus monkey. Journal ofPhysiology, 236, 549–573.

Received February 13, 2005Revision received February 23, 2007

Accepted May 1, 2007 �


No-onset looming motion guides spatial attention

Documents