Migration and fusion of tactile sensation—premorbid susceptibility to allochiria, neglect and extinction?

Neuropsychologia 42 (2004) 1749–1767

Migration and fusion of tactile sensation—premorbidsusceptibility to allochiria, neglect and extinction?

Anthony Marcela,∗, Peggy Postmaa, Helge Gillmeisterb, Sylvia Coxa,Christopher Rordenc, Ian Nimmo-Smitha, Bundy Mackintosha,d

a Medical Research Council Cognition and Brain Sciences Unit, 15 Chaucer Road, Cambridge CB2 2EF, UKb Brain & Behaviour Laboratory, Birkbeck College, University of London, Malet Street, London WC1E 7HX, UK

c School of Psychology, University of Nottingham, Nottingham NG7 2RD, UKd Psychology Department, The Open University, Milton Keynes MK7 6AA, UK

Received 25 June 2003; received in revised form 27 April 2004; accepted 30 April 2004

Abstract

Migration of tactile sensation was found to occur very frequently in about 25% of normal people (High Error subjects) and veryinfrequently in others. When synchronous stimuli touched the two hands, if the unattended stimulus was modulated when the attended onewas not (and was thus more salient) it “migrated” to the attended hand and fused with or replaced the attended stimulus. However, latenciesreflecting congruence and incongruence of simultaneous stimuli showed that their identities on each hand had been (nonconsciously)registered veridically. Subsequent experiments, involving Focused and Divided Attention without speed pressure showed that mislocationerrors in these subjects (i) were not due to confusion about location of otherwise perceptually distinct stimuli, (ii) nor to speed demand,(iii) nor to relative salience per se, (iv) were immune to attentional manipulation and practice in most subjects, (v) required a stimuluson the attended hand, and (vi) reflected a changed experience. Finally the same subjects rated similarity of the attended stimulus whenaccompanied by a distractor to each stimulus alone. Scaling distributions tested against a sampling model showed that most High Errorsubjects experienced a fusion of the stimuli, some experienced a replacement, and Low Error subjects experienced neither. The individualdifference appears to be in attentional separability and spatial binding of tactile stimuli. Mislocation to the focus of spatial attention iscommon in healthy people, especially when binding is prevented. The present phenomenon appears equivalent to allochiria, but alsoaccounts for phenomena in neglect and extinction, and suggests a premorbid susceptibility to spatial migration and integration that can beexaggerated by brain damage.© 2004 Elsevier Ltd. All rights reserved.

Keywords: Touch; Space; Attention; Neglect; Extinction; Binding; Allochiria

1. Introduction

This paper reports a type of spatial migration of tactilesensation in normal people not reported so far. It consistsin the tendency of some individuals to mislocate dispropor-tionately the stimulus at an unattended location to the at-tended location when two synchronous stimuli are presentedand the unattended one is more temporally modulated. Thephenomenon appears very similar to, and may constitutean example of, spatial mislocation that has been assumedto be pathological and the result of neurological damage.This has implications not only for accounts of normal andpathological spatial attention and experience, but also forthe relationship between neuropathology and normality, and

∗ Corresponding author. Tel.:+44 1223 355294x650.E-mail address: [email protected] (A. Marcel).

for accounts of allochiria, neglect and extinction. Allochiriaconsists in the displacement of the experienced location ofstimuli or sensations, usually from the contralesional sideof space to or toward the ipsilesional side, often to a ho-mologous location. It can occur in various sensory modal-ities, but principally in somatosensation (Bisiach & Berti,1995; Kawamura, Hirayama, Shinohara, Watanabe, &Sugishata, 1987; Meador, Allen, Adams, & Loring, 1991). Itmanifests under various neurological conditions (Kawamuraet al., 1987) but is particularly associated with parietal lobedamage and neglect (Bisiach & Berti, 1995). In fact, spatialmigration of perceptual experience is counted as a char-acteristic of neglect itself, as in ipsilesional crowding ofcontralesional stimuli such as clockface numerals. IndeedManly, Woldt, Watson and Warburton’s (2002)recent worksuggests that in neglect patients perseveration of cancellationof ipsilesional stimuli in cancellation tasks may be due to

0028-3932/$ – see front matter © 2004 Elsevier Ltd. All rights reserved.doi:10.1016/j.neuropsychologia.2004.04.020

1750 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

migration of contralesional targets to the loci of ipsilesionalones.

In research on normal subjects there are a number ofphenomena where spatial attention appears to affect expe-rienced location. Indeed, many experiments conform to theessential features of the present paradigm, namely presen-tation of stimuli bilaterally with attention directed to oneside. In the Ventriloquism effect (Driver, 1996) experienceof an auditory stimulus migrates to the location of a vi-sual stimulus placed elsewhere, particularly when one is at-tending to the latter or there is reason for their commonsource, e.g. a voice migrating to a mouth moving withroughly the same rhythm. In principle little distinguishesthis from the present effect. Since some stage ventriloquistsdo not use a doll, merely directing the audience’s atten-tion to an empty area, mislocation in normal people neednot depend on a stimulus in the target location. In touch,Kilgard and Merzenich (1995)showed that tactile stimuliappear to move in the direction of an attended location. Inexperiments where visual stimuli are presented briefly andsubjects have to attend to and report those in certain loca-tions, either whole stimuli or parts of them migrate to theattended location (Shallice & McGill, 1978; Mozer, 1983).In Treisman & Schmidt’s (1982)experiments where sub-jects perceive false conjunctions, features migrate from unat-tended locations to those subjects have to report (where,even if attention is object-based, objects are spatially lo-cated). Other phenomena are also interpretable in terms ofspatial attention producing migration of experience. In shad-owing, two spatially separated auditory messages are pre-sented simultaneously, binaurally or dichotically, and sub-jects try to report one and ignore the other. The to-be-ignoredmessage is often reported when it is more salient or as-sociated to the attended message in meaning or prosody(Treisman, 1969). This has usually been treated only as unat-tended stimuli gaining attention or access to report. Exper-imenters have rarely examined the experience of subjectsmaking such errors (though it is relevant that subjects sel-dom realise they have made an error). Plausibly, in at leastsome cases the report error is due to to-be-ignored stim-uli being experienced in the location of attended stimuli.Thus mislocation can be shown in various attentional pro-cedures. However, no study has identified a subgroup ofnormal subjects who show such effects at high enough lev-els that they might be considered pathological. This studydoes so.

The relevant behaviour in normal people reported here isthe erroneous identification on the attended hand of a tactilestimulus delivered on the unattended hand. The conditionswhich produce this were set up for clinical reasons that arenot immediately relevant. The technique has been used byDriver and Grossenbacher (1996)to measure facilitation andinterference in touch, and was carried out in Experiment 1with the aim of validating the procedure in normal peoplefor obtaining indirect effects. It consisted of speeded iden-tification of binary tactile stimuli on the attended hand with

a congruent, incongruent or no stimulus on the other hand.Experiment 1 exposed the phenomenon of interest and thesubsequent experiments investigated its nature. This paperwill focus on mislocation errors produced by normal sub-jects. However, we will return at the end to discussion of therelation of the errors reported here to pathologies of spatialattention.

2. Experiment 1

2.1. Method

2.1.1. SubjectsTwenty-five right handed subjects (14 male, 11 female)

were recruited from the CBU volunteer panel and were paidfor participation. Their ages ranged from 23 to 73 years,with a mean of 42.4 years.

2.1.2. Apparatus, stimuli and set-upCustom software was written in XGen version .882

(http://www.mricro.com) and run on a portable Dell Pen-tium III computer. Vibrotactile stimuli were presented viatwo solenoid-operated “tactors” placed on the back of theright and the left hand, about 1 cm below the first knucklebetween first and second fingerbones, and held in placewith Velcro around the hand. Whenever a current passedthrough it, each solenoid drove a metal rod with a bluntconical tip (tactor) on to the hand with a force of about1 N.

Two different vibrotactile stimuli were programmed us-ing XGen. One consisted of the rod tip contacting the handfor 250 ms. The other stimulus was also 250 ms long, butconsisted of three 50 ms contacts separated by two gaps of50 ms. The stimuli were called “Tap” and “Drum” respec-tively: these two names were used as vocal responses be-cause they not only described the stimuli but both had initialplosives suitable for activating a voice key.2.1.2.1. Stimulus salience. It is important to remark at thispoint that the Drum stimulus is experienced by everyoneas more salient than the Tap. This was difficult to changewith the software at the time. However, the relevant pointis that the relative salience of the Drum may play somepart in the phenomenon in question, as will become clearbelow.

Subjects sat at a table with their hands resting on twosupports, whose internal edges were separated by a 34 cmbar. The distance between the tactors on the two hands was55 cm. Subjects looked at a central fixation point throughout,slightly below eye level and just beyond arm length, in orderto maintain central visual orientation and to discourage themfrom looking at their hands. Reaction times from onset ofstimuli were recorded via a microphone-operated voice key,attached to a pair of ear mufflers (AIWA R model HP–17E).The ear mufflers were worn to eliminate any audible cuesto stimuli.

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1751

Table 1Combinations of stimuli delivered to attended and unattended hand

Condition

1 2 3 4 5 6

Attended hand–unattended hand Tap–Nil Tap–Tap Tap–Drum Drum–Nil Drum–Drum Drum–Tap

2.1.3. DesignThe task was a speeded two-choice discrimination be-

tween vibrotactile stimuli on one hand (the attended hand),with possible distractors on the other hand. The attendedhand (left/right) remained constant throughout a block, butvaried across blocks in an ABBA order counterbalancedacross subjects. Four blocks of 30 trials each were presented.On one third of the trials in each block (10) there was nodistractor: these served as control baseline trials. The targetstimulus on half of these was Tap, on the other half Drum.On the remaining 20 trials a distractor stimulus was pre-sented on the unattended hand, simultaneous with the targetstimulus on the attended hand. Half of these bimanual trials(10) were congruent (i.e., the distractor was the same as thetarget) and the other half were incongruent (i.e., the distrac-tor was different from the target). This yielded six differentstimulus combinations, as shown inTable 1. All six condi-tions were present in equal proportions within each block,presented in random order. In total there were 120 trials.

2.1.4. ProcedureThe experiment was explained and the tactors were then

placed on the back of each hand and secured with Velcro(see above). After the ear mufflers with the attached micro-phone were put on, the position of the microphone and thesensitivity of the voice key were adjusted for each subject forequal registration of the two vocal responses. The two dif-ferent stimuli were demonstrated on the hand that was to beattended first, until the subject was comfortable with the dif-ference between them. Subjects were instructed to respondorally, saying either ‘tap’ or ‘drum’ depending on which tac-tile stimulus they felt on the attended hand, and to ignorethe unattended hand. They were told to respond as quicklyand accurately as possible. Subjects next received 30 prac-tice trials, consisting of equal numbers of all six conditionsin random order. The first block of trials was then run. The

Table 2Percent errors per condition in Experiment 1

Condition

1 2 3 4 5 6

Attended–unattended Tap–Nil Tap–Tap Tap–Drum Drum–Nil Drum–Drum Drum–TapAll subjects 0.8 0.6 29.4 1.4 1.4 5.8KST 0 0 65 0 0 10RW 5 0 95 5 0 0SG 0 0 100 0 0 0JS 0 0 85 0 0 5LS 0 0 85 0 0 0JB 10 0 70 0 0 0Other subjects (range) 0–10 0–10 0–40 0–30 0–10 0–40

familiarisation and practice procedure was repeated for thenext block of trials when targets occurred on the other hand.Following that, the second, third and fourth blocks of trialswere run with breaks between but without further practice,in the fourth block targets again occurring on the originalhand. After the experimental sessions, subjects were askedabout their experience of the task, with increasingly specificquestions: whether they found some stimulus combinationsmore difficult to perceive or respond to than others, whetherthey thought they may have made mistakes, and if so whatproportion on particular conditions.

Each block of trials was initiated with a key press bythe experimenter, generating a delay of 500 ms and thendelivery of the stimuli. The subject’s oral response triggeredthe voice key, which led to registration of the reaction time.The experimenter recorded the response (Tap or Drum) bypressing one of two assigned keys on the PC. This key pressinitiated the next trial with an interval of 500 ms. Therewas no warning interval. Since the experimenter’s responselatency became quite constant, the intertrial interval fromthe subject’s response was about 1 s. The whole experimentlasted about 30 min.

2.2. Results

Although the original aim of the experiment was to ex-amine the effect of distractors on RTs, the error rates weresuch as to necessitate dealing with errors before appropriateanalyses of RTs could be performed.

2.2.1. ErrorsTwo subjects made no errors, and therefore play no part

in the following error analysis. Errors occurred primarilyin the two incongruent conditions, 3 and 6 (Tap–Drum andDrum–Tap; seeTable 2). A generalised linear mixed modelanalysis (seeBreslow & Clayton, 1993) was performed on

1752 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

the error data. Due to the disproportional distribution of er-rors over conditions and to the low number of errors in thefour non-incongruent conditions, error rates were pooledover these four conditions and compared with those of con-ditions Tap–Drum and Drum–Tap, using Wald’s�2. Theanalysis showed a significant effect of condition (Wald’s�2

= 45.7, d.f. = 2,P < 0.001). Separate pairwise comparisonsof error rates were all significant (P < 0.001, Bonferronicorrected), i.e. each of the incongruent conditions differedfrom the other conditions and from each other.

However, there was evidence in the presence of outly-ing residuals of significant heterogeneity between subjects(Wald’s χ2 = 196.8, d.f. = 44,P < 0.001). Six subjectsmade an extremely high number of errors on Condition 3(Tap–Drum) in particular. These will be individuated by ini-tials. Their error rates and the range of those of other subjectsare shown inTable 2. Since there was no effect of attendedhand, data are collapsed across hands.

On the basis of this observed difference in error rate be-tween subjects, they were divided into those with error ratesfor Condition 3 above and below 50%. In fact the lowesterror rate for the former was 65% and the highest rate forthe latter was 40%. Separate comparisons of error rates forConditions 3 and 6 with all other conditions were carried outfor each subject singly, using Fisher’s exact test. For all sixHigh Error subjects, Condition 3 differed significantly fromthe other conditions (P < 0.001 for all). For only one (KST)Condition 6 differed significantly from the other conditions(P < 0.05). Regarding Low Error subjects, for three of them,both Conditions 3 and 6 differed significantly from Condi-tions 1, 2, 4 and 5 (P < 0.05 for all 3); for six, only Condition3 differed significantly from the other conditions (P < 0.001for 4; P < 0.05 for 2); for one, only Condition 6 differedsignificantly from the other conditions (P < 0.05); sevenshowed no difference between either Condition 3 or Condi-tion 6 and the other conditions; two made no errors at all.

Thus, although both incongruent conditions tend to pro-duce errors more than other conditions, the majority of sub-jects were accurate on most trials. However, this is not truefor High Error subjects, and selectively so in Condition 3(Tap–Drum).

Fig. 1. Experiment 1. Effect of Distractor on reaction times for Tap and Drum targets on attended hand for High Error and Low Error subjects.

2.2.2. Reaction timesError RTs were excluded from the data set, eliminating

197 of 3000 data points (6.6%). Correct RTs above and be-low 2 S.D.s for each subject in each condition were alsoexcluded, eliminating 143 of 2803 data points (5.1%), fromfurther analyses. For the remaining valid RTs, means andS.D.s were computed for each subject in each of the six con-ditions. Where there were insufficient correct RTs in a givencondition (in Condition 3 SG made 100% errors, RW made95% errors), the mean RTs for this condition were substi-tuted using iterative missing values replacement (Healy &Westmacott, 1956). These data were then divided into thosefor the High Error subjects and those for the Low Error sub-jects. SeeFig. 1.

The partition of subjects on the basis of widely divergenterror rates made an initial overall ANOVA inappropriate.Moreover further preliminary analysis suggested that het-erogeneity of covariance and the relatively small number ofHigh Error subjects would compromise the sensitivity of amixed model ANOVA to apparent within-subject differencesin Fig. 1. Following preliminary treatment above, the meansfor each group were subjected to a paired samplet-test anda Repeated Measures ANOVA.

There was no significant main effect of attended hand(left versus right) for either group. In the High Error groupthe main effect of Target (Tap versus Drum) was significant(F(l,5) = 14.8,P < 0.025). This was not true for the LowError group. The main effect of Distractor (congruent ver-sus incongruent versus none) was significant in both groups(High Error group:F(1.2,6) = 25.4,P < 0.01; Low Errorgroup:F(1.4,24.6) = 74.0,P < 0.001). For each group pair-wise comparisons were carried out between the three levelsof Distractor. In the High Error group Incongruent Distractor(Conditions 3 and 6) was significantly different from Con-gruent Distractor (Conditions 2 and 5) (P < 0.025, Bonfer-roni corrected) and from No Distractor (Conditions 1 and4) (P < 0.005, Bonferroni corrected). However, CongruentDistractor and No Distractor were not significantly differentfrom each other. In the Low Error group each pairwise com-parison was highly significant (P < 0.001). The interactionbetween Target and Distractor was significant for the Low

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1753

Error group (F(1.3,22.7) = 16.4,P < 0.001) but nonsignifi-cant for the High Error group.

To obtain a more detailed account of the effect of Dis-tractor, Paired Samplet-tests were conducted between thelevels of Distractor for each of the two levels of Target.This was done separately for each group. For the High Errorgroup, Condition 1 (Tap–Nil) was not significantly differ-ent from Condition 2 (Tap–Tap) but it was from Condition3 (Tap–Drum) (t(5) = −3.7, P < 0.025). Also, Condition 2was significantly different from Condition 3 (t(5) = −3.2,P< 0.05). Condition 4 (Drum-Nil) was not significantly dif-ferent from either Condition 5 (Drum–Drum) or Condition6 (Drum–Tap). Condition 5 was not significantly differentfrom Condition 6. For the Low Error group, Condition 1was significantly different from Condition 2 (t(18) =−4.6,P < 0.001) and from Condition 3 (t(18) =−7.6,P < 0.001).Condition 2 was also significantly different from Condition3 (t(18) =−7.7, P < 0.001). Condition 4 was significantlydifferent from Conditions 5 (t = −5.1, P < 0.001) and 6 (t= −5.9,P < 0.001). Condition 5 was also significantly dif-ferent from Condition 6 (t = −2.8, P < 0.025).

2.2.3. Subjects’ commentsSeeAppendix A.

2.3. Discussion

Although the experiment was designed to examine theeffect of unattended stimuli on response latency to attendedstimuli, the main concern of this paper is the extremely higherror rates made by some subjects in certain conditions.These subjects had disproportionately high error rates onlyin the Tap–Drum condition, and thus could not be discardedas unable to perform the task. Before discussing the errors,the latency data will be summarised.

Inspection ofFig. 1 shows that for both groups of sub-jects the presence of a distractor and its relation to the stim-ulus on the attended hand had an effect on reaction time.Interestingly, a congruent distractor did not provide an ad-vantage over no distractor, and in fact slowed RT somewhat.This may be due to opposing effects: distraction by a secondstimulus and facilitation from a congruent stimulus. How-ever the low number of observations per subject makes anyanalysis of predicted bimodality or greater variance of RTstoo insensitive. It is also clear that in both groups the ef-fect of an incongruent Drum on the response latency to anattended Tap was much greater than the effect of an incon-gruent Tap on latency to an attended Drum. (The formercondition is also the one producing most errors in the HighError subjects, see below.) As we have remarked, the Drumstimulus is universally experienced as more salient than theTap. This difference between distractors itself differed forthe two groups of subjects. For the Low Error subjects anunattended Tap significantly lengthened response latency toboth attended Tap and attended Drum. In the High Errorsubjects an unattended Tap had a small but nonsignificant

effect whether there was a Tap or a Drum on the attendedhand. One might presuppose that the interference effect onRT (as opposed to that on errors) is automatic or derivesfrom a nonconscious representation of the unattended stim-ulus. Indeed the experimental procedure was designed (i) onthe supposition that in Neglect patients unattended stimulithat are apparently nonconscious would nonetheless affectlatency to report attended stimuli, and (ii) on the basis of theliterature on normal people indicating that unattended stim-uli that are outside of awareness influence response times torelated attended stimuli. It is difficult to decide how muchweight to place on the differential effects of the two dis-tractors on RT for the two subject groups (especially giventhe variance for the small number of High Error subjects).To the extent that the difference is valid, it seems to implythat whatever produces the errors is not independent of whatmodulates response latency. In fact we shall suggest laterthat these are separate.

Turning to the errors, all subjects, as might be expected,had higher error rates in both incongruent Conditions 3 and6. Although the Low Error subjects also made more errors inCondition 3 (Tap–Drum), it is difficult to determine whetherthere is really a continuum of subjects in the rate of errorson this condition or whether subjects differ qualitatively.However, those subjects making at least 65% errors espe-cially merit attention. What is the nature of these errors?Were they confusions? And if so, why were these subjectsconfused much more than other subjects? If subjects wereindeed confused this would account, at least in part, for thelonger latencies in Condition 3. Two kinds of confusion arepossible: (a) confusion as to which stimulus was on whichhand, and (b) confusion as to which hand one is attendingto. However, the question arises as to why this should ob-tain when the Tap is on the attended hand and the Drum ison the unattended hand, but not when it is the other wayround. One possible answer is that since the Drum is moresalient it attracts attention to itself and to its location. Inthis case the subjects in question may indeed be reportingthe stimulus to which they are (now) attending but not thestimulus on the hand to which they are supposed to attend,i.e. the locus of the focus of their attention has been shifted.This raises the question of whether they know which loca-tion they are attending to when they erroneously report themore salient stimulus. On the other hand, it is possible thatsubjects maintain their attention to the correct hand but re-port the wrong stimulus as being on that hand. At the timeeither they suspect/know that they have made an error, orthey do not. If they do not, this amounts to experiencing thewrong, more salient stimulus on the attended hand. Thus ei-ther (a) subjects maintain attention to the attended hand butare confused as to which stimulus occurred on it, and areaware of such confusion, or (b) their attention is drawn tothe unattended hand, but they are unaware of this, or (c) theymaintain attention to the intended hand and experience onit the stimulus at the to-be-ignored location. These are notmutually exclusive.

1754 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

Certain observations bear on these possibilities. After thewhole experiment had been run subjects were asked forcomments; they were then asked if they had found any con-ditions more difficult than others, and, if so, which (seeAppendix A). Since we did not know about the numbersof errors until after the experiment, we did not specificallyask about confusions. However, while two of the six HighError subjects spontaneously reported some confusion ofTap–Drum trials, the others did not. Although one of thesetwo (LS) thought that she may have made errors on that con-dition, four of them reported that they were confident thatthey had made no errors. In addition, the first author wasrun on the experiment after it was completed. Being awareof the issue, he paid attention to his experience. He reportedthat he had been confused on about three trials of Condi-tion 3 and said he expected that he had made about threeerrors on that condition. In fact he had made nine errors butwas quite unaware of any uncertainty. In other words he hadthought that he was reporting the stimulus on the attendedhand. Therefore, as far as one can tell, it appears that whensubjects make errors on Condition 3, although they may beaware of some difficulty (possibly due to self-observationof their own relative slowness), they show no evidence ofawareness of attending to the wrong hand; and althoughthey may be aware of the presence of different simultaneousstimuli, they are confident of correctly reporting that on thedesignated attended hand. Furthermore, if the High Errorsubjects were confused on Tap–Drum trials, why did theynot report Tap and Drum on those trials about 50% each?Something must have made them more confident that Drumwas the stimulus in the attended location.

If indeed a substantial proportion of errors on Condition3 were the result of mislocation of the salient stimulus tothe attended from the unattended hand, this would be veryinteresting since it would constitute a failure of selectivityrestricted to certain individuals. If mislocations occur, theirnature and source is important. However, what is immedi-ately most relevant is to throw more light on whether theerrors in question reflect genuine mislocations of sensationor confusions or something else. This was the aim of fur-ther experiments that were conducted on subjects with higherror rates.

3. Experiments 2(a) and (b) and repeat of Experiment 1

In designing these experiments there were several con-siderations. (1) Experiment 1 required speeded discrimina-tion, where time pressure may have exacerbated confusion.Removing time pressure might reduce confusion and allowsubjects to be more sure of attentional locus and stimuli inthat location. (2) Some subjects may have had difficulty fo-cussing on the attended hand. (3) It is relevant to assess towhat extent subjects know that there are two stimuli on therelevant trials, and, if so, to see whether they know whichstimulus is on which hand. (4) Relative salience may play a

significant role in the errors either by drawing attention tothe nominally unattended hand or by determining selectionof a stimulus that, via accurate spatial focus, is mislocated.(5) In pathological allochiria mislocation of stimuli can oc-cur when there is no objective stimulus at the attended lo-cation (Meador et al., 1991). In Experiment 1 there had al-ways been a stimulus at the attended location. It is of interestwhether mislocation would occur in the normal High Errorsubjects when there is no stimulus on the attended hand.(6) Conditions where there is no stimulus on the attendedhand could reduce confusion by differentiating the stimulion each hand more clearly. Alternatively they could increasethe relative salience of the unattended stimulus; this mightproduce more errors when no stimulus on the attended handis paired with a Tap on the unattended hand, and even moreerrors when it is paired with a Drum on the unattended hand(in subjects not already at ceiling). (7) The question arisesof the stability or persistence of the error rate in High Errorsubjects. They might reduce their error rate either by prac-tice or by experiencing a different task set (e.g. by focus-ing or dividing attention or by abandoning speed). This canbe assessed by balancing over subjects the order of experi-ments meeting the above criteria and rerunning Experiment1 subsequently.

These considerations were addressed as follows. First,time pressure was removed and accuracy of report of theattended hand was emphasised. Second, in one experimentsubjects were asked to focus exclusively on the attendedhand; in another experiment they were asked to prioritiseaccuracy for the attended hand and report it first, but thento try to report the stimulus on the unattended hand. Third,conditions where no stimulus occurs on the attended handwere added, and those without a stimulus on the unattendedhand were removed. Requiring subjects to report what is onboth hands informs on what they know about stimuli in theunattended location and on whether they are confused aboutthe location of stimuli. In addition, comparison of the effectsof attention to a single side with those of attention to twosides should illuminate whether differences in error rates aredue to individual differences in ability to focus attention.These experiments were balanced over subjects and afterthey were performed Experiment 1 was rerun.

3.1. Method

3.1.1. SubjectsA second sample of High Error subjects. Shortly after Ex-

periment 1 an almost identical experiment was conducted(which will not be reported here) on a separate sample ofsubjects, also from the CBU volunteer panel. It was equiv-alent to Experiment 1 except that subjects visually orientedtoward either the attended or unattended hand. In this Ex-periment 6 out of 21 subjects again showed very high errorrates in Condition 3 (Tap–Drum) of the same kind as theoriginal subjects, the other subjects making such errors ata maximum of 35%. It is of note that about the same pro-

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1755

portion of subjects showed this effect in the two samples.These subjects were then tested on the original Experiment1. One of them could not return, and the errors in Condition3 of another dropped to 25%. The error rates of the remain-ing four were: CG: 85%; EG: 75%; MK: 100%; RN: 75%.These four subjects were added to those carrying out thefurther experiments.

One of the original sample (JS) could not return for test-ing, leaving five subjects in the first group. Since the ex-perimental histories of the first 5 and the second 4 subjectswere slightly different, they were treated as separate groups.However, subjects were also treated as individuals and areidentified by initials. Sex and age are given in parentheses.First 5: KST (F 73), RW (M 25), SG (M 25), LS (F 50), JB (F44); second 4: CG (F 46), EG (F 21), MK (F 35), RN (F 26).

3.1.2. DesignThe design for the Focused and Divided Attention ex-

periments differed from Experiment 1 in only two respects.(i) They were bothnon-speeded discrimination tasks. (ii)The possible stimuli on the attended hand were Tap, Drumand Nothing; there was always a stimulus on the unattendedhand, Tap or Drum.Table 3shows the conditions. Compar-ison of Conditions 3 and 6 with Conditions 1 and 4 assesses(a) whether a stimulus is necessary on the attended hand forthe unattended stimulus to be misreported there, and (b) theeffect of relative salience on mislocation errors. Four blocksof 30 trials were presented, with distributions of stimuluscombinations identical to Experiment 1.

3.1.3. ProcedureThe apparatus and set-up were the same as in Experiment

1. The only differences in procedure were as follows. Forboth experiments subjects were not instructed to respond asfast as possible, but were told to take the time they needed todetermine the identity of the stimulus on the attended handas accurately as possible. They were instructed to respondby saying “tap”, “drum” or “nothing” according to what theyfelt on the attended hand.

(a) Focused Attention: Subjects were instructed to attendas well as they could to the designated hand and toignore the other hand as much as possible.

(b) Divided Attention: Subjects were instructed to attend tothe designated hand and to give priority to reporting thestimulus on that hand first and accurately, but also to tryafter that to report the stimulus on the unattended hand.They were asked to report in the form “tap, drum”,“nothing, drum”, etc. They were told to say if theywere unsure or unaware of the unattended stimulus.

Table 3Combinations of stimuli delivered to attended and unattended hand in Experiments 2(a) and (b)

Condition

1 2 3 4 5 6

Attended hand–unattended hand Nil–Tap Tap–Tap Tap–Drum Nil–Drum Drum–Drum Drum–Tap

The experimenter wrote down responses, since in theDivided Attention experiment reports were of both at-tended and unattended hands. The intertrial interval fromthe subject’s response was about 1100 ms in the FocusedAttention experiment and about 1300 ms in the DividedAttention experiment. Five subjects performed the FocusedAttention experiment before the Divided Attention exper-iment, four performed them in the reverse order. Subjectsthen performed Experiment 1 again.

3.2. Results

(1) The first main aspect of the results is the comparisonof error rates for the attended hand over conditions betweenthe initial Experiment 1 and the Focused and Divided At-tention experiments. In all these experiments such errors oc-curred mostly on Tap–Drum and less on Drum–Tap trials.The other conditions produced extremely low rates of errors(0–10%). Therefore errors in all the other, non-incongruent,conditions were pooled. The error rates for each subject ineach experiment are shown inTable 4. (Error rates are basedon erroneous reports of the stimulus on the attended hand.)

Tap–Drum was the only condition producing high errorrates on the attended hand. This is also the main conditionof interest to be compared across experiments, and is the fo-cus of the following analyses. These analyses first treat thetwo subject groups separately, given their slightly differentexperimental histories, and compare them treating the sec-ond 4 subjects as a replication of the first 5; and second,deal with each subject separately, treating them as a seriesof individual case studies.

In order to assess whether there was an effect of exper-iment on the Tap–Drum error rates between the initial Ex-periment 1 and the Focused and Divided Attention exper-iments for each group of subjects, a log linear regressionanalysis of error rates was performed, allowing account tobe taken of subject-specific differences. Over all subjects ineach group, the effect of experiment was significant (first 5subjects:χ2 = 19.91, d.f. = 2,P < 0.001; second 4 subjects:χ2 = 27.5, d.f.= 2,P < 0.001), showing that the Tap–Drumerror rates were present at differing levels over the three ex-periments for each group. However this is not true for allsubjects individually. To assess the effect of experiment onthe Tap–Drum error rates for each subject, single subjectanalyses were conducted using Fisher’s exact test. (1) Sub-jects JB, KST and CG had comparable error rates across thethree experiments; (2) Subject SG showed a significant dropin errors from Experiment 1 to Focused Attention (100%→ 70%, P < 0.01); (3) Subjects LS, EG and MK showed

1756 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

Table 4Error rates (%) on attended hand for each subject in Experiments 1, 2(a) and 2(b) and l(b) (repeat) for conditions Tap–Drum, Drum–Tap and otherconditions pooled

ExperimentCondition

Experiment 1, first run Focused Attention Divided Attention Experiment 1 second run

T–D D–T Other T–D D–T Other T–D D–T Other T–D D–T Other

KST 65 10 80 80 1.25 95RW 95 2.5 10 5 5 5SG 100 70 35 80 20 10 1.25LS 85 85 50 90 5JB 70 2.5 70 5 60 55

Mean 83 2 1 63 9 0 55 0 0.25 53 3 0.25

CG 85 90 70 100EG 75 45 35 1.25 65MK 100 1.25 90 70 100RN 75 20 5 5 20 1.25

Mean 83.75 0 0.31 61.25 0 0 45 1.25 0.31 71.25 0.31 0

a significant drop in errors from Experiment 1 to DividedAttention (LS: 85%→ 50%, P < 0.05; EG: 75%≥ 35%,P < 0.025: MK: 100%→ 70%, P < 0.025); (4) SubjectsRW and RN showed a significant drop in errors from Ex-periment 1 both to Focused Attention (RW: 95%→ 10%.P < 0.001; RN: 75%→ 50%, P < 0.001) and to DividedAttention (RW: 95%→ 5%,P < 0.001; RN: 75%→ 5%,P< 0.001). No subject showed a difference in error rates be-tween Focused and Divided Attention, except for LS (85%versus 50%,P = 0.04). Thus for six out of these nine sub-jects, while they showed some variation in error rates fromthe initial Experiment 1 to the Focused and Divided Atten-tion experiments, their Tap–Drum error rates did not fallbelow the entry criterion of 50%. RW’s and RN’s initiallyhigh error rates proved unstable. EG’s error rate did not fallas low but reduced to 45% and 35% in the two experiments.In order to assess whether there was a difference in the ef-fect of experiment between the two subject groups, a loglinear regression analysis was performed on the Tap–Drumerror rates. This was not significant (χ2 = 1.00, d.f. = 2,P= 0.61). Therefore the two groups behaved in the same waywith respect to Experiments 1, 2(a) and 2(b).

The second main aspect of the results is the effect of per-formance of Experiments 2(a) and (b) on persistence of er-rors on Tap–Drum trials under the conditions of Experiment1. Given that there was a small but inconsistent effect ofexperiment on Tap–Drum error rates over all subjects, thequestion arises as to whether this was restricted to the na-ture of specific experimental procedures or due to a practiceeffect. Thus Tap–Drum error rates were compared betweenthe first and second runs of Experiment 1 (seeTable 4).

The error rates of each group of subjects were comparedusing log linear regression analysis. Over all the first 5 sub-jects there was a significant effect of run of Experiment1 (χ2 = 21.3, d.f. = 1,P < 0.001); over all the second 4subjects the effect of run of experiment was not significant(χ2 = 3.62, d.f. = 1,P = 0.06). Treating the subjects singly,

Fisher’s exact test showed (1) subjects JB, LS, CG, EG, andMK had comparable error rates across the two runs; (2) RW,SG and RN showed a significant drop in errors from firstto second run (RW: 95%→ 5%, P < 0.001; SG: 100%→20%, P < 0.001; RN: 75%→ 20%, P < 0.001); (3) sub-ject KST showed a significant increase in errors from first tosecond run (65%→ 95%,P < 0.05). For the two subjects(RW and RN) who showed a significant drop in error ratesfrom Experiment 1 to the Focused and Divided Attentionexperiments, this effect was maintained when they repeatedExperiment 1. Therefore the instability in their error ratesappears to be due to practice rather than to specific experi-mental procedures.

Log linear regression analysis showed that there was nodifference between the two groups of subjects in the rela-tionship between first and second runs of Experiment 1 (χ2

= 2.02, d.f. = 1,P = 0.15). Nor was there any consistent ef-fect of kind or order of experiment on errors across individ-uals

(3) The errors dealt with so far consist only of reportingDrum on the attended hand in the Tap–Drum condition. Thethird main aspect of the results is the nature of reports forboth attended and unattended hands and the distribution oferrors over conditions in the Divided Attention experiment.Table 5shows the latter.

No errors were produced in the non-incongruent condi-tions (except for a single error on two such conditions byKST). No errors were made when there was no stimuluson the attended hand (except a single one by KST). In theDrum–Tap condition, although the attended hand was al-ways reported correctly (except on one trial by RN), re-ports of the unattended hand now tended to be incorrect,i.e. Drum–Tap was reported as Drum–Drum. Averaging oversubjects, this tendency (45%) was as great as that for misre-porting Tap–Drum as Drum–Drum (48%). Pooling the dataand using a log-odds ratio confirmed no significant differ-ence in the tendency to report Drum–Drum on the two kinds

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1757

Table 5Percent of trials on each stimulus condition on which different kinds of error were made by each subject in the Divided Attention experiment

Subjects Conditions: stimuli presented on attended and unattended hand

N–T T–T T–D N–D D–D D–T

Error % Trials Error % Trials Error % Trials Error % Trials Error % Trials Error % Trials

KST N–D 5 D–D 80 N–D 5 D–D 50D-N 45

RW D–D 5 D–D 5?–D 15 D–? 15N–D 10

SG D–D 80 D–D 90T–T 10

LS D–D 30D–T 20

JB D–D 50 D–D 50D–T 10

CG D–D 70 D–D 90EG D–D 35 D–D 15MK D–D 70 D–D 80RN D–D 5 D–D 30

T–D 5

T: Tap, D: Drum, N: Nothing; first and second letter: stimulus on/report of attended hand and unattended hand.

of trial (Z = 0.48, P = 0.62). Three subjects did show asignificant difference (Fisher exact: KST,P = 0.048; LS,P = 0.010; RN,P = 0.046), but not in a consistent direc-tion. In both conditions the error was to report the Tap asDrum. In the two conditions producing almost all the errors,Tap–Drum and Drum–Tap, there was no tendency to misre-port both stimuli, i.e. for reports to switch which hand thetwo stimuli were on. The overall error rates on these trials,regardless of kind of error, were the same for all subjectsexcept for LS and RN (Fisher exact:P = 0.001 and 0.022,respectively) whose differences were in different directions.Relative frequencies of Drum–Drum reports versus all otherkinds of error were the same on Tap–Drum and Drum–Taptrials for all subjects except KST (Fisher exact:P = 0.001).

3.2.1. Subjects’ commentsSeeAppendix A

3.2.2. SummaryThe relevant results of these experiments can be sum-

marised as follows. (a) Of the nine High Error subjects,six maintained a high error rate on Tap–Drum trials withno speed pressure in both the Focused Attention experi-ment (70–90% errors) and the Divided Attention experiment(50–80% errors). (b) In neither experiment were errors pro-duced when there was no stimulus on the attended hand. (c)In the Divided Attention experiment the six subjects main-taining a high error rate made a roughly equal proportionof errors on Tap–Drum and Drum–Tap trials, in both casesreporting Drum for Tap irrespective of whether it was onthe attended or unattended hand. Drum was always correctly

reported. (d) There was no overall or consistent effect of re-peating Experiment 1 after the other two experiments. Threesubjects reduced their error rates on the Tap–Drum condi-tion to 20%, 20% and 5%, but this was also evident in thepreceding experiments. The other six still produced errorrates ranging from 55% to 100%.

3.3. Discussion

The results bear on several issues.

(1) Three things taken together imply that the errors onTap–Drum trials in Experiment 1 were not due to con-fusion as to which of two different stimuli was on whichhand. (a) The major type of error in the Divided At-tention experiment was to report Drum–Drum for bothTap–Drum and Drum–Tap. (b) Errors switching whichhand the stimuli were on were almost absent. (c) Theerror rate on Tap–Drum trials was not reduced by re-moval of speed pressure. If anything, there appears tobe a “reduplication” of the Drum from one hand to theother.

(2) Only two subjects (RW and RN) showed a significantdrop in errors from Experiment 1 to both the Focusedand Divided Attention experiments; they also continuedto make few errors in the repeat of Experiment 1. Thisindicates that it was not a speed constraint which isresponsible for the high error rates.

(3) Subjects’ comments were consistent with the first twopoints above in that, although they indicate that sub-jects were often aware of the difficulty of the conditions

1758 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

producing most errors and that they were sometimesconfused, they suggest that confusion does not accountfor the majority of such errors.

(4) Five subjects showed comparable error rates in the tworuns of Experiment 1, one showed an increase, and threeshowed a decrease. There was no consistent effect onerror rate across subjects of the type of task or constraint.Although three subjects’ immediate drop in errors fromExperiment 1 to all the other experiments suggests thatthey were affected by practice, the error rates of theother six were stable and unaffected by practice, taskset or attentional mode. Therefore in the majority ofHigh Error subjects their errors appear a stable featureof them as individuals.

(5) The absence of relevant errors when there was no stim-ulus on the attended hand bears on several issues. First,relative salience alone is insufficient to produce the er-rors. This rules out an explanation in terms of spatialattention being drawn to the hand with the more salientstimulus, which is then reported. Second, since confu-sion is ruled out as an explanation (see 2, above), itwould appear that the presence of a stimulus on the at-tended hand plays some part in inducing mislocationof the stimulus on the unattended hand to the attendedhand. This effect may obtain irrespective of the locationof attention (see General Discussion).

(6) In the Drum–Tap condition of the Divided Attentionexperiment Drum was reported on both hands, i.e.Drum was also mislocated to the supposedly unattendedhand at rates as high as the original converse error onTap–Drum trials. This bears on both the nature of theerrors and the role of attention.

3.3.1. The nature of the errorsSubjects’ comments indicate that although some were

aware of the relative difficulty of certain types of trial, theysignificantly underestimated the proportion of such trials onwhich they made errors. That is, they were confident on themajority of their errors. Given that confusion is ruled out, itdoes seem therefore that the High Error subjects genuinelyexperienced the Drum stimulus in the wrong location andthat the experimental procedure produces a tendency for amislocation in the spatial experience of the stimuli, espe-cially in these subjects. The term “spatial experience” is ap-propriate for the following reason. Even though subjects’reports on many incongruent trials were of congruent stim-uli (Drum + Drum), their RTs indicated that at some levelthe incongruence of the stimuli had an effect. This last pointsuggests that what produces the errors is to some extent sep-arate from what influences RTs. The subjects’ confidence intheir erroneous reports indicates that they consciously expe-rienced congruent stimuli. If the RT differences reflect (ac-curate) representation of incongruent stimuli, this suggeststhat such representation is nonconscious.

What kind of phenomenon do the errors represent andwhat underlies it? (1) Are the mislocations due to the rel-

ative salience of the stimuli or are they due to the struc-tural and temporal characteristics of the stimuli? (2) Whatis the role of spatial attention? One hypothesis is that theTap is replaced by the Drum, and that this is due to thecombination of the Drum being more salient and the lo-cation of spatial attention. That is, what one experiencesconsciously is affected by salience, but where one experi-ences it is at the location of attention. On the other hand,(mis)perception of the attended Tap as Drum may be dueto perceptual grouping or fusion. There may be groupingprinciples in touch analogous to those in audition. Tempo-ral synchrony is known to be a powerful factor (Bregman,1990). The onset and offset of the Drum coincided withthose of the Tap, and could easily be integrated or fusedwith it; if it did so no part of either would be left tem-porally extraneous. The quality of the fused percept wouldprobably take on that of the internally modulated stimulus(Drum) or be somewhere between the two stimuli. Subjects’comments suggest that some of them did experience Drumstimuli of differing qualities, consistent with the idea thaton Tap–Drum trials the Drum was fused with the Tap. Ifthe only response alternatives were Tap and Drum, subjectswould have to respond “Drum”. The lack of errors on tri-als with no stimulus on the attended hand indicates thatrelative salience per se is insufficient for the errors. How-ever, it does not rule out the possibility that salience doesplay a part but only if there is a stimulus in the attendedlocation.

Drum was reported for Tap not only when Tap was on theattended hand but also when it was on the unattended hand(and such errors were of equal frequency in each case). Thismight seem to weaken any hypothesis that gives a majorrole to spatial attention, e.g. that the error occurs in the di-rection of the focus of attention. However, there is a logicaland methodological problem that forestalls such a conclu-sion. The task demands of the Divided Attention experimentwere to place emphasis on and report first the stimuli onthe attended hand but also to report the stimuli on the unat-tended hand if possible. Therefore subjects attempted ei-ther to rapidly switch or to divide attention. The unattendedhand becomes, to at least some extent, attended. Logicallyone cannot tell what is experienced on the unattended handwhen subjects are instructed to focus on one hand alone;any manipulation that informs on experience of the unat-tended hand necessarily induces attention to it (seeDuncan,1987).

Given the above, Experiment 3 was designed to assess,for High Error and Low Error subjects, the quality of thepercept experienced on the attended hand in Tap–Drum,Drum–Tap and other trials. If subjects are allowed to rate theapparent stimulus quality between a single Tap and a sin-gle Drum, their response distributions could reveal to whatextent they experience an attended Tap on Tap–Drum tri-als (and vice versa) as a Tap, as a Drum (produced by re-placement), or as something between the two (produced byfusion).

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1759

4. Experiment 3

The same presentation procedure as in Experiment 1 wasused; but subjects were asked to rate attended stimuli ac-cording to a scale representing a continuum between Tap andDrum, and speed was not a demand (see below). Subjectswere exposed to the original target stimuli in isolation priorto each experimental session, and could indicate in their rat-ings how similar to them attended stimuli felt. On Tap–Drumand Drum–Tap trials, several experiential outcomes relativeto those of Tap and Drum in isolation are possible. (i) Allsubjects may experience the Tap (or the Drum) as an amal-gam of or something between the two stimuli, and the dif-ference may lie in their response criterion, i.e. when to calla Drum–like Tap “tap” and vice versa (an attended Drum onDrum–Tap trials may seem more Drum–like than the con-verse); (ii) Only the High Error subjects may have such ex-perience; (iii) The Tap may be experienced as a Drum (im-plying replacement of the Tap) by the High Error to a muchgreater extent than by the Low Error subjects; (iv) Individ-uals may be subject to differing proportional combinationsof such experiences. Given the above, data were analysedfor each subject singly prior to group comparison.

4.1. Method

4.1.1. SubjectsOut of the 6 subjects who had maintained a high error rate

on both runs of Experiment 1, five were available for thisexperiment (CG, MK, EG, JB, LS). Five Low Error subjectsas identified by the previous experiments were matched withthe High Error subjects for sex and as closely as possiblefor age (mean age: High Error subjects = 40.2, S.D.: 12.07;Low Error subjects 38.2, S.D.: 15.83), with no age differencebetween groups on an independentt-test (t(8) = 0.225,P >0.05).

4.1.2. Apparatus, stimuli and set-upThe apparatus and experimental set-up were the same (ex-

cept where specified) as in the previous experiments. Thecustom software was written in MEL version 2.0 (Schneider,1988). A rating scale was displayed vertically with its centre18 cm below the fixation point, and subjects were asked torespond according to it. Since they were familiarised withthis rating scale before the experiment proper, its presenceserved as a reminder if needed. It consisted of a verticalline of 21 cm, intersected at 3 cm intervals. The eight letters,A–H, were displayed next to each interval point such thatA appeared at the top of the scale and H at the bottom. Theword TAP appeared at the top of the scale next to A, andthe word DRUM appeared at the bottom next to H.

For the duration of the experiment subjects were played acontinuous stream of pink noise (3 dB/octave slope in powerspectrum, low pass filtered at 19 kHz, 70 dB SPL) through aset of AIWA headphones (model HP-17E) via a Panasonic

portable CD player (SL-S210). This was to exclude the in-fluence of any audible cues from the tactile stimuli.

4.1.3. DesignEach subject provided a total of 20 rating responses on the

eight-point scale for each of the six trial types. In 4 blocksof 30 trials the hand to be attended remained constant in ablock, but varied across blocks in an ABBA order. The taskwas to rate each stimulus on the attended hand accordingto the scale while ignoring any simultaneous stimuli on theother hand. The scale enabled subjects to express any varia-tions in the perceived sensations. Stimulus conditions wereidentical to those in Experiment 1 (seeTable 1).

4.1.4. ProcedureThe procedure was the same as in Experiment 1 apart

from the following. Subjects were told to report what thetactile stimulus on the attended hand felt like, and to ig-nore the other hand. It was explained that many of the stim-uli during the experiment might feel exactly like the Tapsand Drums experienced during familiarisation, in which casethey should be rated as A and H respectively. It was pointedout that some stimuli might feel different, in which case theother letters in the scale should be used. Subjects were toldthat this was not a reaction-time task; nonetheless they wereasked to respond reasonably quickly because the purposewas to record their first impressions of the stimuli. It wasemphasized that there was no ‘right’ or ‘wrong’ response,and that the aim was simply to establish how stimuli wereexperienced. The experiment then proceeded as in Experi-ment 1. The experimenter entered each response (A to H)on the corresponding key on the PC. This started the nexttrial after 500 ms.

4.2. Results

Fig. 2shows the mean ratings for Tap and for Drum in thedifferent conditions by High Error and Low Error subjects. Itis clear that the ratings for Tap on Tap–Drum trials by HighError subjects were consistent with results from the previousexperiments, where they reported it as Drum on such trials.However, it is also clear that they used responses other thanthe extreme values of A and H, suggesting that a degree offusion was occurring. For these subjects, this is also appar-ent but to a much smaller degree on Drum–Tap trials. Thedata for each subject were tested to determine whether rat-ing distributions on Tap–Drum and Drum–Tap trials couldbe accounted for by a sampling model based on the distribu-tions of ratings observed on Tap–Nil and Drum–Nil trials.The possible outcomes of this correspond to the possibleexperiential outcomes outlined above: (a) rating distribu-tions on Tap–Drum trials fitted by a sampling model basedon Tap–Nil trials; (b) the opposite, where they are fitted bya sampling model based on Drum–Nil trials; (c) they arefitted by a sampling model based on a combination of (a)

1760 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

Fig. 2. Mean ratings by High and Low Error subjects of the attended stimulus in conditions (a) with Tap and (b) with Drum on the attended hand.

and (b); (d) rating distributions that are not fitted by any ofthese.

In the sampling model, the data in all four criticalconditions (Tap–Nil, Drum–Nil, Tap–Drum, Drum–Tap)are accounted for by the response probability distribu-tions for the Tap–Nil and Drum–Nil conditions togetherwith a sampling parameterλ. If t(k) and d(k) representthe probabilities of responsek (k = A, B, . . . , H) in theTap–Nil and Drum–Nil conditions, then the probabilityof responsek in the Tap–Drum condition,td(k), is givenby td(k) = λt(k) + (l − λ)d(k). Thus λ and l − λ rep-resent the probability of sampling from the Tap–Nil andDrum–Nil distributions respectively. The model was fittedby means of Monte Carlo Markov Chain (MCMC) methods(Carlin & Louis, 2000) using the WinBugs (Spiegelhalter,Thomas, & Best, 1999) program. Results are given interms of (a) the most probable values ofλ, and (b) thegoodness of fit of the model, in terms of the posteriorexpectation of the deviance statistic (seeCarlin & Louis,2000). Table 6shows how well the rating distributions onTap–Drum and Drum–Tap trials are fitted by the samplingmodel.

For all Low Error subjects the rating distributions onTap–Drum and Drum–Tap trials were accounted for by sam-pling from those on Tap–Nil trials and Drum–Nil trials re-spectively. That is, none of these subjects showed evidenceof fusion. For four of them there were no significant devia-tions in rating responses between the Tap–Nil and Tap–Drumconditions nor between Drum–Nil and Drum–Tap (forPvalues seeTable 6). However one subject (KS), while pre-dominantly responding on Tap–Drum trials according to thesame distribution as on Tap–Nil trials, also rated a small

number of these trials consistently with Drum–Nil trials,thus showing a small frequency of replacement of Tap byDrum.

The High Error subjects showed two distinct profiles onTap–Drum trials. MK, LS and EG showed a response distri-bution on Tap–Drum trials different from those on Tap–Niland Drum–Nil trials (forP values seeTable 6). The par-ticular profiles suggest that in these cases the stimulus onthe to-be-attended hand is perceived as a fused version ofTap and Drum. The distributions on Tap–Drum trials of thetwo other subjects, JB and CG, were accounted for by sam-pling from those on Drum–Nil trials. (Although this doesnot show fusion, it is in contrast with the ratings of Tapon Tap–Drum trials by Low Error subjects, which were ac-counted for by sampling from those on Tap–Nil trials.) JBrated the majority (75%) of Tap–Drum trials within the samerange as Drum–Nil trials, while also rating a small numberof such trials (25%) according to the Tap–Nil distribution.Although CG consistently rated all Tap–Drum trials withinthe same range as Drum–Nil trials, closer inspection of herdata shows that, while she only used the extreme value of Hon 25% of Drum–Nil trials, shenever used this value dur-ing Tap–Drum trials. This raises the question of whether adegree of fusion was occurring here. Therefore these twosubjects, at least on the majority of Tap–Drum trials, seemto experience the target stimulus as a Drum.

The ratings of Drum on Drum–Tap trials were accountedfor by sampling from the distributions on Drum–Nil trialsfor all High Error subjects except one. For MK the samplingmodel was narrowly rejected (P = 0.047), indicating that ona small proportion of these trials she experienced the Drumas a fusion of Drum and Tap.

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1761

Table 6Assessment of the sampling model and its goodness of fit to ratings of attended Tap on Tap–Drum trials and attended Drum on Drum–Tap trials

Subject Tap on Tap–Drum trials Drum on Drum–Tap trials

Outcome of fitting asampling model

Goodness of fit measuredby posterior expectationof deviance

Outcome of fitting asampling model

Goodness of fit measuredby posterior expectationof deviance

Low Error subjectsKS Tap–Drum→ Tap 26.5 (P = 0.231) Drum–Tap→ Drum 26.5 (P = 0.231)KAM Tap–Drum → Tap 28.6 (P = 0.156) Drum–Tap→ Drum 26.5 (P = 0.231)RB Tap–Drum→ Tap 27.8 (P = 0.182) 15:85 sampling with Drum–Tap

→ Drum predominating28.0 (P = 0.176)

TD Tap–Drum→ Tap 26.5 (P = 0.231) Drum–Tap→ Drum 26.5 (P = 0.231)MF 60:40 sampling with Tap–Drum

→ Tap predominating28.6 (P = 0.157) 7:93 sampling with Drum–Tap

→ Drum predominating27.4 (P = 0.196)

High Error SubjectsMK Sampling model rejected 53.4 (P < 0.001) Sampling model rejected 35.9 (P = 0.031)LS Sampling model rejected 38.1 (P = 0.018) Drum–Tap→ Drum 26.5 (P = 0.231)JB 25:75 sampling; Tap–Drum

→ Drum predominating24.9 (P = 0.302) Drum–Tap→ Drum 27.0 (P = 0.211)

CG Tap–Drum→ Drum 30.8 (P = 0.100) Drum–Tap→ Drum 23.3 (P = 0.385)EG Sampling model rejected 49.4 (P < 0.001) Drum–Tap→ Drum 29.2 (P = 0.139)

4.2.1. Subjects’ commentsSeeAppendix A.

4.3. Discussion

Inferences from these results about subjects’ experiencerequire caution. First, using subjective rating scales has lim-itations, especially when similarity to “extreme” stimuli israted without intermediate stimuli being presented. How-ever, since there is no guarantee that subjects’ experience ofthe affected stimuli varies only unidimensionally, it is diffi-cult to know how to approximate their “intermediate” expe-riences. In addition, no subject in their performance or com-ments appeared to find using the rating scale implausible ordifficult. Second, the observed discrepancy between Highand Low Error subjects might be due merely to a differencein uncertainty or in response criterion, e.g. reluctance to useextreme ratings or to guess. But this is implausible herebecause the High Error subjects’ ratings were asymmetric,on Tap–Drum trials reliably clustering toward and includ-ing “Drum”. Indeed if they were at all uncertain, it seemsfrom their comments to be precisely when and because theyexperienced amalgams.

The scaling responses of each subject were tested forwhether they conform to or deviate from a sampling modelbased on responses to Tap and Drum alone, in order to iden-tify whether each subject’s experience of Tap on the attendedhand on Tap–Drum trials reflects (a) accurate perception, (b)perception of it as Drum, (c) perception of neither of thesebut some fusion of the two, or (d) a distribution derivingfrom a mixture of more than one of these. This procedureindicates that of the 5 available Low Error subjects, 4 ex-perienced an unalloyed Tap and one experienced mostly anunalloyed Tap with an unalloyed Drum on a few trials. Itindicates that of the 5 available High Error subjects, 3 expe-

rienced a fusion of unattended Drum with the attended Tap,and 2 mostly experienced an unalloyed Drum (replacement),one of whom experienced fusion on some trials. That is, ontrials with bilateral stimuli Low Error subjects on the wholeare able to keep the tactile experience on each hand sepa-rate, with one showing evidence of occasional replacementof the attended Tap by the unattended Drum. On such trialsHigh Error subjects on the whole cannot keep the tactile ex-perience on each hand separate, at least not for the attendedhand; in most cases unattended Drum fused with attendedTap, in some cases it replaced it.

To what extent are these conclusions from the scalingresponses consistent with comments made by each of therelevant subjects? The High Error subjects were asked forcomments after each experiment, but unfortunately the LowError subjects only participated in Experiment 1 and the fi-nal scaling experiment. Comments are listed inAppendix A.Of the Low Error subjects, TD and KS reported that bilat-eral stimuli were kept separate and that they were percep-tually distinct; KAM and RB reported that they were awareof indistinctness and confusion, their awareness plausiblyenabling them to cope; MF made no relevant comments.Of the High Error subjects, the four subjects whose scal-ing responses indicated fusion all made comments reflect-ing indistinctness of the experience both in the separationof stimuli on the two hands and in what was on the at-tended hand; JB’s comments, whose scaling responses in-dicated replacement of attended Tap by unattended Drum,convey that she was not aware of making errors and thatshe was not aware of differential difficulty among the con-ditions, which is what would be expected if Drum replacedTap. Overall then, the implications of the scaling responsesare consistent with the subjects’ comments. Of course, it isdifficult to assess how much weight can be placed on suchcomments.

1762 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

Regarding ratings of Drum on Drum–Tap trials, no sub-ject in either group showed any effect of the unattendedTap. The absence of effect of unattended Tap indicates thatmigration and fusion does not occur irrespective of the na-ture of the simultaneous bilateral stimuli. Rather, in thoseindividuals susceptible to it, it only occurs in the directionfrom Drum to Tap. This may be due to the relative salienceof Drum or to the internal modulation of Drum but not ofTap. Experiment 3 does not address whether this dependson the location of spatial attention, and testing this remainsmethodologically problematic for the reasons given in dis-cussion of Experiment 2.

In summary, the results indicate that for most High Er-ror subjects their experience of Tap stimuli on the attendedhand is an amalgam of the Tap on that hand and the Drumon the unattended hand, though for some the Drum experi-entially replaces the Tap. What determines whether the twoperceptual representations compete or integrate is unclearand none of the data bears on this. However, it is certainlythe case that both of these result in such subjects from afailure of the tactile representations on the two hands to bekept separate by spatial attention to one hand. While thismay occur on a few occasions in Low Error subjects, theirspatial attention is successfully selective. (Interestingly, us-ing the same rating procedure in further experiments, theHigh Error subjects are also much more susceptible thanthe Low Error subjects to auditory dichotic fusion in at-tending to one ear, indicating that the individual differenceis supramodal).

5. General discussion

The present studies demonstrate in about 25% of normalsubjects high rates of migration and fusion of tactile sensa-tion so far unreported. The original samples (25 and 21) aretoo small to warrant inferring two distinct sets of subjects.Although subjects appear bimodal in error rates (>65% and<40%) nothing that follows would be affected or invalidatedif they lay on a continuum. As far as we know, apart fromvon Békésy (1960, 1967), none has reported tactile fusionacross discontinuous areas of skin. In von Bekesy’s exam-ples (especially 1967) vibratory stimuli on two adjacent fin-gers were felt as a single vibration in the empty space be-tween the fingers. In our case the fused stimulus is felt ona hand separated by space from the other hand. In the dis-placements of tactile sensation reported byBender (1952)one stimulus caused another to migrate toward it but notnecessarily to fuse, and this only occurred across continuousareas of skin.

5.1. Is the present phenomenon equivalent topathological mislocation?

The relation between the current errors and pathologicalmislocation bears on two issues. (i) If apparently neurolog-

ically intact people behave as do those with brain damage,it raises the question of normality and pathology. (ii) If thepresent errors are equivalent to those in patients with a prob-lem in spatial attention, it reinforces the role of spatial at-tention in the former. Although mislocation is a feature ofneglect, the most analogous specific pathology is allochiria.Unfortunately there is an inconsistency in the usage of theterms allochiria and alloaesthesia.Meador et al. (1991)dis-tinguish them on the basis of history and derivation as fol-lows. Allochiria denotes mislocation of a sensation or exter-nal stimulus to the corresponding opposite half of the bodyor space (not necessarily from contra- to ipsilesional side)without apparent stimulation or sensation at the latter lo-cation. Alloaesthesia denotes mislocation of a sensation orexternal stimulus to a remote position, sometimes inducedby a sensation or stimulus in that location. Although classi-cally allochiria was described as occurring without a stimu-lus where the mislocation occurs (Janet, 1890; Obersteiner,1881), whether or not migration is induced by another stim-ulus is neither consensually criterial nor an empirically re-liable distinction. The essential difference, andBender’s(1952) usage for allochiria, appears to be whether migra-tion is to the opposite side of the body or space. We shalltherefore stipulatively follow this usage. Allochiria occursin a range of brain damage, mainly right parietal (Bisiach &Berti, 1995). HoweverKawamura et al. (1987)also foundallochiria (called by them alloaesthesia, but see above) inpatients with anterolateral spinal cord lesions, whose symp-toms did not differ from those with cerebral haemorrhage.Given this, several points are relevant to the relation of thepresent errors to allochiria.

(a) For the majority of the mislocation errors, subjects didexperience the stimulus from the to-be-ignored side atthe to-be-attended location (fused with or replacing thatat the latter location). They were generally unaware ofthe extent to which they made such errors.

(b) Whether a stimulus in the migratory location is neces-sary or not is debatable. In pathology, migration with-out any stimulus in the experienced location covers onlysome cases of such migration. First, in Bender’s mono-graph (1952) all somatic migrations in patients were in-duced by a stimulus on the sensorily “stronger” or intactside of the body. Second, the attention of most patientswho exhibit somatic allochiria, especially in neglect, ispathologically biased to the ipsilesional side of theirbody and its appropriately located sensations. Arguablytherefore, although there may be no evident stimulus,such patients are plausibly super-aware of any ipsile-sional sensations, which may constitute stimuli. Third,on the same grounds, it is plausible that in normal peoplevoluntary attention alone is not enough to achieve thesame effect, needing a stimulus to induce the necessarydegree of attentional bias and focus. Fourth,HalliganMarshall and Wade (1992)reported what they termedallochiria in a patient with visual neglect. In copying

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1763

drawings, she transferred what was on the left side ofobjects to the right side, but only when the drawing wassymmetrical in having features on both left and rightsides of its central vertical axis. Finally, even if somewriters define allochiria as migration without any stim-ulus at the migratory location, it may be more theoreti-cally profitable to relax strict diagnostic categories andto envisage cases of migration requiring and not requir-ing an inducing stimulus as on a continuum or as casesof a more general phenomenon.

(c) In this study high mislocation rates occurred in about25% of normal subjects. Although we know of no for-mal calculation, in relevant reports over the last centurythe occurrence of allochiria and alloaesthesia in parietalpatients is low.Kawamura et al. (1987)report what wecall allochiria in 20 of 122 acute cases of right hemi-sphere cerebral haemorrhage, 17 of 35 cases of rightputaminal haemorrhage, and only 1 of 30 cases of rightthalamic haemorrhage. Compared to such statistics therate of occurrence in a normal population, given oursample sizes, is not low, and is remarkably similar. It ispossible that a proportion of the population is premor-bidly susceptible to allochiria, perhaps requiring partic-ular neurological impairment to provoke it in strongerform.

(d) In the present High Error subjects the effect can be frag-ile. It can be reduced by attentional manipulation and inabout a third of these subjects it seems to reduce greatlywith practice per se. However, while both neglect andallochiria are most associated with right parietal dam-age, in fact they occur in a minority of such patients,and, more relevantly, are fragile even within the acutestage but especially after.

These considerations suggest that the present effect maywell be equivalent to allochiria. Given what we said aboutindividual differences in normal subjects and that allochiriais associated with pathology of spatial attention, what dis-tinguishes normal and brain damaged individuals who showit from those who do not is likely to be an attentionalcharacteristic. However, if subjects making many misloca-tion errors are in some way no different from neurologi-cal patients showing allochiria, then either allochiria is notpurely pathological and restricted to brain damage or our“allochiric” subjects are incipiently pathological. Our spec-ulation above is that phenomena shown by some patientsresult from a premorbid predisposition that can be revealedby appropriate experimental conditions; i.e if those with ahigh mislocation rate suffer appropriate neurological insult,they will show full allochiria. In one respect the present phe-nomenon differs from allochiria. In the Divided Attentionexperiment High Error subjects reported a stimulus on theunattended hand when there had been one, even if its iden-tity was wrong. In clinical allochiria this would not happenfor the contralesional side. Therefore the extra effect of rel-evant neurological damage is either to produce a migration

that leaves no representational trace of the stimulus in theoriginal location or to preclude the patient attending to theoriginal location such that experience of a stimulus there isrestored.

5.2. Spatial attention, experiencedlocation and pathology

The errors in question represent a failure of at least oneaspect of attention, namely the ability to be selective inkeeping the stimuli in an ignored location separate fromthose in an attended location. This is only a failure to preventone kind of unattended stimulus (Drum) from interactingwith another kind of attended stimulus (Tap). Whether onlystimuli in an ignored location migrate to and affect those inan attended location or whether it is bi-directional is difficultto decide, both on the present evidence and on principle,for reasons given earlier. However, several considerationssuggest that the direction of mislocations is indeed linkedwith spatial attention. First, as discussed above, the errorsseem restricted to conscious representation. On incongruenttrials in the Divided Attention experiment Drum–Drum wasreported, and subjects’ comments expressing confidence inthese reports indicate that they were their conscious expe-rience. But the longer RTs on these trials in Experiment 1imply that the stimuli were accurately represented as differ-ent at some level, presumably nonconsciously. Whether ornot consciousness is restricted to the focus of attention, nor-mally what is attended is conscious. It is therefore probablethat spatial attention plays some role in the errors. Sec-ond, this is exactly the tenor of the studies sketched in theIntroduction—that spatial attention is widely found to affectexperienced location. But if the effect of spatial attentionon experienced location is as widespread as those studiessuggest, why has it not been reported more often? First,researchers usually require subjects to perform at a crite-rial level, and subjects making many errors are discarded,without the nature of the errors being considered. Second,even when such subjects are included, experimenters maynot have dwelt upon the experiential aspect of the errors,treating them as capture of attention or confusions. Third,we suggest that these errors have in fact been reported inequivalent paradigms, but not as such, and their significanceand relation to pathological phenomena has not been recog-nised. Thus the present phenomenon may be more frequentthan supposed.

The difference between High and Low Error subjectsclearly lies in ability to keep separate the tactile representa-tions on the two hands. This could operate in two aspects ofspatial attention: (a) determination of the experienced spa-tial location of what is perceived, (b) selection of candidatesfor control, consciousness, or report. Spatial focus of at-tention is often treated primarily as a selection criterion for(b), and experienced location is treated as either subsidiaryor tightly coupled to it. Recent research, especially thaton false conjunctions (Treisman & Schmidt, 1982), distin-

1764 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

guishes these, and emphasises that candidates for attentionthat are objectively spatially separate may not remain soin conscious experience but be integrated. Thus High Errorsubjects may be equally or more spatially focused in whatdetermines experienced location, but may be less spatiallyselective; i.e. they are more susceptible to integrate or bindat the attended locus stimuli from another location (whenthey can be integrated with what is there) which compete forattention.

Degree of spatial focus may be what affects binding atthe migratory destination. But what permits or preventsmigration itself? The cause of full allochiria must lie inthat aspect of spatial attention underlying spatial selectivity.Normally, when one is attending to one place and thereis a stimulus at another as yet unattended place, attentionto the first locus does not cause the stimulus at the latterto migrate experientially to the former. If it did, all spatialperception would be mistaken, one could not attend selec-tively to space, and one would never be able to be awareof stimuli at locations other than where one is presentlyattending. However, it may be that special circumstances(e.g. abnormal temporal or spatial pressure, narrowed at-tention, strong unattended competition, uncertainty) permitspatial focus to produce experiential mislocation. For thisto happen it must be the case that the representations ofstimuli that migrate are not tightly bound to the repre-sentation of their objective location (at least as regardsconsciousness). This is plausible if the representation ofspatial source is compromised, as is suspected for the leftside with right parietal damage (Bisiach & Vallar, 2000). Itis also supported byBaylis, Gore, Rodriguez, and Shisler’s(2001) finding that extinction (unawareness of one of twosimultaneous stimuli) is reduced when patients do nothave to combine (bind) information about both identityand location. It has been conjectured that conscious per-ception involves binding features constituting perceptualidentity with those that are “episodic”, especially loca-tion, and that focal attention produces experiential binding(Treisman & Gelade, 1980; Treisman, 1998). Interventionsthat forestall adequate focal attention in normal people,such as central visual masking, would prevent this. Migra-tion exemplifies awareness of a stimulus without it beingbound to its veridical location. The stronger cases in ne-glect and extinction when stimuli on the contralesional sideare not consciously experienced at all but show indirecteffects may reflect more severe damage to representationof space. The present migration with fusion or replace-ment seems to be an example of the general case andsuggests that normal individuals differ in that aspect of at-tention which binds and prevents false binding. However,perhaps susceptibility to migration of an unattended stim-ulus and its interaction with an attended one should notbe viewed only as a deficiency, but rather as one aspectof a more integrative style of attention which nonethe-less renders one more vulnerable to effects of parietaldamage.

5.3. Spatial migration in allochiria, neglect and extinction

Although our account has centred on allochiria, it alsopredicts and explains conscious experience in neglect andextinction. More relevantly, in so doing, the account (a) high-lights that it is not clear whether, if High Error subjects suf-fer appropriate amplifying damage, their primary symptomswould conform to allochiria, neglect or extinction, and (b)renders this clinical distinction somewhat spurious and po-tentially a matter of severity of damage. If neglect resultsfrom narrowly focused and laterally biased attention and ifrelevant (parietal?) damage impairs representation of con-tralesional spatial source, neglect would often involve al-lochiric mislocation. There is reason to believe that this istrue.Manly et al. (2002)tested patients with left visual ne-glect on a cancellation task, who started from the right, andat a certain point, leaving targets on the left uncancelled,re-cancelled targets on the right already cancelled. How-ever, the greater the part of the test-field on the left that wasempty of targets, the less did patients re-cancel targets on theright. Manly et al. surmised that uncancelled targets on theleft were experientially transposed to the right, and that theywere misperceived at the locations of already cancelled tar-gets still within the focus of attention, but were seen (veridi-cally) as uncancelled. This raises the more general possibil-ity that, in neglect, stimuli in contralesional space are notentirely ignored or out of awareness, but as a result of spa-tial attention are misperceived as being on the other side—ineither homologous locations or those of appropriate stimuli.Even if stimuli do not migrate to the exact location of ip-silesional items, this proposal would also account for caseswhere patients experience or draw contralesional items ip-silesionally, e.g. where they crowd all clockface numeralson the right or are satisfied that homologous right-sided fea-tures alone represent all that is to be accounted for. Thisallochiric account of rightward crowding without replace-ment obviates the need to explain it in terms of horizontalanisometric distortion of spatial representation (see:Milner,1987; Bisiach, Pizzamiglio, Nico, & Antonucci, 1996). Wesuggest that a main difference between our High Error sub-jects and patients with neglect is that the latter would beunable to report stimuli in an unattended lateral location,which the former were able to do in the Divided Attentionversion of Experiment 2.

The essential form of extinction is that when two stimuliare presented simultaneously, laterally aligned and briefly,there is unawareness of one stimulus. It occurs in vision,somatosensation, audition and olfaction, and is most oftenassociated with right parietal damage (Heilman, Watson, &Valenstein, 1993), where the extinguished stimulus is on thecontralesional side. Extinction is most obtained when thestimuli are identical; the more different they are the more itis reduced (Baylis, Driver & Rafal, 1993). Putative accountsare that attention is biased to the ipsilesional side and is ei-ther spatially narrowed (where space is defined by the figuralstimuli) or its capacity is reduced to a single perceptual ob-

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1765

ject or in rate of gain of information (quasi-simultanagnosia).In both cases, what is outside of attention is outside of aware-ness. However, an alternative account of at least some casesof extinction is as allochiria, where the contralesional stimu-lus is experientially drawn to the location of the ipsilesionalor attended one when spatial binding occurs. If it is identi-cal to, structurally compatible with or more salient than theattended stimulus it will experientially replace or fuse withit and only one stimulus is experienced—at that location.This account does not involve unawareness of the contrale-sional stimulus per se, but rather unawareness of it as lo-cated at its objective location and as separate from one inthe attended location. This proposal gains plausibility fromthe features governing probability of extinction (see above)and from other properties. First, extinction, allochiria andunawareness of the content of contralesional space tend toco-occur. Second, even when stimuli are extinguished theyhave (nonconscious) effects (Volpe, LeDoux & Gazzaniga,1979). Third, extinction is most probable when binding isinvolved (Baylis et al., 2001). Our present proposals wouldmake allochiric migration central to at least some cases ofneglect and extinction. Of course cases of either without in-dependent allochiria would weaken the case for allochiriaasgenerally basic.

In sum, our account suggests that High Error subjects willbe susceptible to neglect and extinction as well as allochiria.What differentiates these is proposed to be severity of dam-age to spatial representation. More basically, the account ex-plains what underlies the present errors and allochiria, butalso leads to neurological predictions about differences be-tween High and Low Error normal subjects, e.g. in parietalfunctioning.

Acknowledgements

We are grateful to Helen Keane for first piloting the pro-cedure, to Bob Carlyon and Rhodri Cusack for engenderingExperiment 3, and to Edoardo Bisiach for advice and dis-cussion. We also thank the editor and anonymous refereesfor help in improving the manuscript.

Appendix A. Subjects’ comments

(Experimenter’s clarification, interpretation or remarksare within brackets.)

A.1. Experiment 1

A.1.1. High Error subjectsKST: Difficulty with conditions Tap–Drum and Drum–

Tap. No mention of errors.RW: Found Tap–Drum more difficult because of salience

on unattended hand. Unaware of any errors.

SG: Found Tap–Drum most difficult because of salience(on unattended hand). Unaware of any errors.

LS: Some confusion on certain trials (Tap–Drum) as towhat was on the attended hand. But no confusion aboutwhich hand stimuli were on.

JB: No difference in difficulty between conditions exceptcondition Tap–Drum slightly more difficult. (No mention oferrors implies unaware of errors?).

JS: No comments.

A.1.2. Low Error subjects(Out of 19 Low Error subjects, these 5 were those who

were later matched for Experiment 3. See Experiment 3.)RB: Found Tap–Drum most difficult. No special com-

ments otherwise.KAM: Tap–Drum hardest, especially if Tap on the right

hand.MF, KS and TD: No relevant comments.

A.2. Experiments 2(a) and (b) and repeat of Experiment 1

KST: When Drum was on attended hand uncertain whetherTap or Nothing was on other hand. Felt Drum on other hand,possibly because Drum “cut across” to the other hand; it wasmore that when Drum was on the attended hand she did notcatch what was on the other hand. Certain that this did nothappen when Tap was on attended hand. She thought sheperformed well.

Experiment 1 rerun. No differential difficulty noticed be-cause she was concentrating on attended hand so much thatshe was unaware of what was on unattended hand. Thoughtshe had not made many errors. RW. Found conditions withNil easiest. Found Drum–Tap hardest.

Experiment 1 rerun. Easier to concentrate on right hand.Thought that he had made some errors on Tap–Drum. Noconfusion.

SG: Felt difference in strength of stimulation on the twohands. Sometimes unsure; on those trials he then decidedon Drum on attended hand because he felt Drum there ingeneral.

Experiment 1 rerun. Thought some errors made inTap–Drum. Sometimes he realised this after the trial. Usu-ally unaware of errors.

LS: Found it easier to focus on one hand (than in Exper-iment 1). Found Tap–Drum easier.

Experiment 1 rerun. Confused as to real identity ofTap. (This was referring to Tap–Drum trials.) Thought someerrors made on Tap–Drum.

JB: No difference in difficulty between conditions.Experiment 1 rerun. Did not think she made any er-

rors. Some confusion sometimes about stimulus (on attendedhand).

A.2.1. Second sample of subjectsCG: Experiment 1. Did not realise that on some trials

stimuli on attended and unattended hands were different.Noticed a few errors made.

1766 A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767

Other experiments. Often said “Drum” then thought itmight have been a Tap. Not many trials with different stim-uli. Realised that some trials must have had. In Tap–Drumcondition Tap felt like a different kind of Drum. Neverclearly felt Tap on attended hand when Drum on unattendedhand.

Experiment 1 rerun. No comment.EG: Experiment 1. Drum more distinct than Tap. Thought

some errors made, saying Drum for Tap.Other Experiments. Found that trying to ignore stim-

uli made them more apparent. Did not think she madeany errors in Divided Attention; some made in FocusedAttention.

Experiment 1 rerun. Only comment: Easier than first run.MK: Experiment 1. Tap–Drum difficult. It made hands

feel similar; had to concentrate to identify stimuli.She thought she made three or four errors altogether. Only

noticed errors in Drum–Tap condition.Other Experiments. Some Drums feel harder than others.

In Tap–Drum often said Drum+ Drum, then thought shehad made a mistake. “On unattended hand Drum can feelfaint, so that it can be confused with Tap.” This made herwonder if they were in fact Drums. She thought some errorsin Divided Attention; sometimes noticed after responding.

Experiment 1 rerun. When Drum on unattended hand,Drum on attended hand is sometimes strong, sometimes soft.Thought some but few errors made.

RN: Experiment 1. Easier to identify Drum when no stim-ulus on other hand. (These were in fact Taps.). Some Drumsfelt harder, some softer; softer Drums appeared on attendedhand when Drum was on unattended hand. Made few errors:one noticed (in Tap–Drum).

Other Experiments. Sometimes not sure; when not sureprobably the stimuli were different.

Experiment 1 rerun. Thought less errors than in Experi-ment 1.

A.3. Experiment 3

A.3.1. High Error subjectsLS: “Not aware of any gradations, so I took it as how

convinced I was of whether it was a Tap or Drum. I mighthave made a mistake on one occasion when I said H whenI should have rated it as a Tap.”

JB: “Trials with a Tap on the attended hand and Drum onthe other were more difficult as it was hard to tell which waswhich. I think I made mistakes but not sure when. I had nodifficulty keeping the two stimuli or locations separate as itwas only for a short period.”

CG: “I did not seem to notice extremes; the stimuli didnot feel like A or H. I noticed no difference in difficultyover the different combinations, as I don’t feel consciouslyaware of what is happening on the other hand”. Asked if shethought she had made mistakes, “There were times whenI was wondering if I should have taken extremes, e.g. Hinstead of G”.

EG: Reported feeling strong Taps and strong Drums, butnot much in between. She found it was hard to “feel the tap”on Tap–Drum trials.

MK: “At times it was difficult to concentrate on the at-tended hand, especially when there was a Drum on the otherhand and a Tap on the attended hand. I think I might havemade mistakes on Tap–Drum trials, where I would haverated it as Drum.”

A.3.2. Low Error subjectsRB: “On Tap–Drum trials it was hard to tell whether there

was a Tap or a Drum on the attended hand. I am sure Imade mistakes in this condition.” She said she had somedifficulty keeping the two stimuli or locations separate, butthe confusion waswhat it was on which side.

KAM: She thought lots of taps felt like drums, especiallyas she got used to the sensation. She thought she mademistakes, “taps feeling like drums”.

MF: No relevant comments.KS: “Nothing felt ‘in between’ A and H as the stimuli

were very distinct. Tap–Drum trials were most difficult.” Hethought he had made one or two mistakes on trials with twostimuli, but was not sure on which conditions.

TD: He found Tap–Drum trials the most difficult, but hadno difficulty in keeping the two stimuli or locations sepa-rate.

References

Baylis, G. C., Driver, J., & Rafal, R. (1993). Visual extinction and stimulusrepetition.Journal of Cognitive Neuroscience, 5, 453–466.

Baylis, G. C., Gore, C. L., Rodriguez, P. D., & Shisler, R. J. (2001).Visual extincton and awareness: the importance of binding dorsal andventral pathways.Visual Cognition, 8, 359–379.

Bender, M. B. (1952).Disorders of perception (p. 111). Springfield Ill:Charles C. Thomas Publishers.

Bisiach, E., & Berti, A. (1995). Consciousness in dyschiria. In M.S. Gazzaniga (Ed.),The cognitive neurosciences (pp. 1331–1340).London: MIT Press.

Bisiach, E., Pizzamiglio, L., Nico, D., & Antonucci, G. (1996). Beyondunilateral neglect.Brain, 119, 851–857.

Bisiach, E., & Vallar, G. (2000). Unilateral neglect in humans. In F. Boller,J. Grafman, & G. Rizzolatti (Eds.),Handbook of neuropsychology (vol.1, pp. 459–502). Amsterdam: Elsevier.

Bregman, A. S. (1990).Auditory scene analysis. Cambridge, MA: MITPress.

Breslow, N. E., & Clayton, D. G. (1993). Approximate inferencein generalized mixed models.Journal of the American StatisticalAssociation, 88, 9–25.

Carlin, B. P., & Louis, T. A. (2000).Bayes and empirical Bayes methodsfor data analysis (2nd ed.). Boca Raton: Chapman & Hall.

Driver, J. (1996). Enhancement of selective listening by illusorymislocation of speech sounds due to lip-reading.Nature, 381(6577),66–68.

Driver, J., & Grossenbacher, P. (1996). Multimodal constraints on tactilespatial attention. In Innui, T., & McClelland, J. (Eds.),Attention andperformance (vol. XVI, pp. 209–235). Cambridge MA: MIT Press.

Duncan, J. (1987). Attention and reading: wholes and parts in shaperecognition—a tutorial review. In M. Coltheart (Ed.),Attention andperformance (vol. XII, pp. 39–61). Hillsdale NJ: Lawrence ErlbaumAssociates.

A. Marcel et al. / Neuropsychologia 42 (2004) 1749–1767 1767

Halligan, P. W., Marshall, J. C., & Wade, D. T. (1992). Left on right:allochiria in a case of left visuo-spatial neglect.Journal of Neurology,Neurosurgery and Psychiatry, 5, 717–719.

Healy, M. J. R., & Westmacott, M. H. (1956). Missing values inexperiments analysed on automatic computers.Applied Statistics, 5,203–206.

Heilman, K. M., Watson, R. T., & Valenstein, E. (1993). Neglect andrelated disorders. In K. M. Heilman, & E. Valenstein (Eds.),Clinicalneuropsychology (3rd ed., pp. 279–336). New York: Oxford UniversityPress.

Janet, P. (1890). Une alteration de la faculté de localiser les sensations.Société de psychologie physiologique, Revue de Philosophie, 1, 659.

Kawamura, M., Hirayama, K., Shinohara, Y., Watanabe, Y., & Sugishita,M. (1987). Alloaesthesia.Brain, 110, 225–236.

Kilgard, M. P., & Merzenich, M. M. (1995). Anticipated stimuli acrossskin. Nature, 373, 663.

Manly, T., Woldt, K., Watson, P., & Warburton, E. (2002). Is motorperseveration in unilateral neglect ’driven’ by the presence of neglectedleft-sided stimuli?Neuropsychologia, 40, 1794–1803.

Meador, K. J., Allen, M. E., Adams, R. J., & Loring, D. W. (1991).Allochiria vs allesthesia. Is there a misperception?Archives ofNeurology, 48, 546–549.

Milner, A. D. (1987). Animal models for the syndrome of spatial neglect.In M. Jeannerod (Ed.),Neurophysiological and neuropsychologicalaspects of spatial neglect (pp. 259–288). Amsterdam: Elsevier Science.

Mozer, M. C. (1983). Letter migration in word perception.Journal ofExperimental Psychology: Human Perception and Performance, 9(4),531–546.

Obersteiner, H. (1881). On allochiria: a peculiar sensory disorder.Brain,4, 153–163.

Schneider, W. (1988). Micro experimental laboratory: an integrated systemfor IBM-PC compatibles.Behaviour Research Methods, Instrumentsand Computers, 20, 206–217.

Shallice, T., & McGill, J., (1978). The origins of mixed errors. In J. Requin(Ed.), Attention and performance (vol. VII, pp. 193–208). HillsdaleNJ: Erlbaum.

Spiegelhalter, D. J., Thomas, A., Best, & N. G. (1999). WinBUGS Version1.2 User Manual. Cambridge, UK: MRC Biostatistics Unit.

Treisman, A. M. (1969). Strategies and models of selective attention.Psychological Review, 76, 282–299.

Treisman, A. M. (1998). Feature binding, attention and object perception.Philosophical Transactions of the Royal Society of London B, 353,1295–1306.

Treisman, A. M., & Gelade, G. (1980). A feature-integration theory ofattention.Cognitive Psychology, 12, 97–136.

Treisman, A. M., & Schmidt, H. (1982). Illusory conjunctionsin the perception of objects.Cognitive Psychology, 14, 107–141.

Volpe, B. T., LeDoux, J. E., & Gazzaniga, M. S. (1979). Informationprocessing of visual stimuli in an “extinguished” field.Nature,282(5740), 722–724.

von Békésy, G. (1960).Experiments in Hearing (translated and editedby E.G. Weaver). London: McGraw-Hill. 1960. (See Chap 13,esp. pp. 568–634).

von Békésy, G. (1967).Sensory inhibition (pp. 220–227). Princeton NJ:Princeton University Press.