SPECIAL REPORT

A componential analysis of the ERP elicited bynovel events using a dense electrode array

KEVIN M. SPENCER,a JOSEPH DIEN,b and EMANUEL DONCHINa

aDepartment of Psychology and Beckman Institute, University of Illinois at Urbana-Champaign, USAbCenter for Neuroscience, University of California, Davis, USA

Abstract

In this study, we examined the relationship between the novelty P3 and the P300 components of the brain event-relatedpotential~ERP!. Fifteen subjects responded manually to the rare stimuli embedded either in a classical auditory oddballseries or in a series in which “novel” stimuli were inserted. The electroencephalogram~EEG! was recorded with a densearray of 129 electrodes. The data were analyzed by using spatial Principal Components Analysis~PCA! to identify aset of orthogonal scalp distributions, “virtual electrodes” that account for the spatial variance. The data were thenexpressed as ERPs measured at each of the virtual electrodes. These ERPs were analyzed using temporal PCA, yieldinga set of “virtual epochs.” Most of the temporal variance of the rare events was associated with a virtual electrode witha posterior topography, that is, with a classical P300, which was active during the virtual epoch associated with the P300.The novel stimuli were found to elicit both a classical P300 and a component focused on a virtual electrode with afrontal topography. We propose that the termNovelty P3should be restricted to this frontal component.

Descriptors: P300, Novelty P3, Dense electrode arrays, PCA

This report attempts to elucidate the different event-related poten-tial ~ERP! components that are elicited by rare and “novel” eventsembedded in an “oddball” sequence. In keeping with the definitionof ERP components proposed by Donchin, Ritter, and McCallum~1978!, a component is defined by its morphology, scalp distribu-tion, and response to experimental variables. It is important toemphasize that the termcomponent, as used in discussions ofERPs, refers to a hypothetical construct whose existence is inferredfrom patterns of variance in the experimental data. A component ofthe ERP is the manifestation at the scalp of the activity of a specificbrain circuitry that is invoked to accomplish a specific informationprocessing task.

The study reported here was motivated by the fact that for thepast two decades, studies of “late positive components” of thebrain ERP elicited by deviant events, presented in the context of

the oddball paradigm, have been plagued by a tendency to con-found and conflate several, possibly distinct, ERP components.Investigators often write as if components such as the P300~Sut-ton, Braren, Zubin, & John, 1965; also referred to as “P3” and“P3b”!, “P3a” ~Squires, Squires, & Hillyard, 1975!, and “NoveltyP3” ~Courchesne, Hillyard, & Galambos, 1975! are members of acategory of P300 components that are related to each other by acommon morphology, with the termP300merely indicating that apositive-going deflection with a 300-ms latency has been ob-served. It is crucial to understand that the termcomponentin thesense proposed by Donchin et al.~1978! is not synonymous withthe termdeflection. Deflections with a common latency provideone element of the definition of a component, its morphology.However, similar deflections associated with different scalp dis-tributions and responding differentially to experimental variablesconstitute different ERP components. Similarly, deflections withthe same scalp distribution may be associated with different com-ponents because they occur with different latencies, and responddifferentially to experimental variables.

In the literature there is the further implication that novel eventsdo not elicit a classical centro-parietal P300 but rather elicit solelya fronto-central novelty P3. In this study, we examined the scalpdistributions of ERPs elicited by rare and novel events under dif-ferent circumstances, using the high spatial resolution provided bya dense electrode array. The results established that the P300 andnovelty P3 are distinct components, characterized by different scalpdistributions and different responses to experimental manipula-tions. Presenting this substantive result allows us to introduce anew approach to the analysis of data obtained by means of a denseelectrode array, the spatiotemporal Principal Components Analysis

K. Spencer was supported in part by National Research Service AwardMH11516 and by predoctoral training program MH19554 from the PublicHealth Service. J. Dien was supported by postdoctoral training programMH19554 from the Public Health Service. Support for establishing theDense Electrode Array facility was provided by the Beckman Institute andthe University Research Board at the UIUC.

Some of the research reported here was previously reported at the 37thAnnual Meeting of the Society for Psychophysiological Research~1997!.

We thank Monica Fabiani and Dave Friedman for providing us withtheir auditory stimuli. We acknowledge the helpful comments of Art Kramer,Clay Holroyd, Marten Scheffers, Greg Miller, and the three anonymousreviewers.

Address reprint requests to: Dr. Emanuel Donchin, Cognitive Psy-chophysiology Laboratory, University of Illinois at Urbana-Champaign,603 East Daniel Street, Champaign, IL 61820, USA. E-mail: edonchin@s.psych.uiuc.edu.

Psychophysiology, 36~1999!, 409–414. Cambridge University Press. Printed in the USA.Copyright © 1999 Society for Psychophysiological Research

409

~PCA!. This analysis allows a major reduction in the dimension-ality of the data set and provides measures of ERP components thatfit naturally with the definition of components proposed by Donchinet al. ~1978!.

The Novelty P3 was first described by Courchesne et al.~1975!who reported that rare, nontarget stimuli that were more complexthan the other stimuli in the series~“novels”! elicited a componentdistinguishable from the P300 by its fronto-central scalp topogra-phy. Knight ~1984; see also Knight, 1997!, studying ERPs in pa-tients with brain lesions, reported that whereas the novel stimulielicited a Novelty P3 in control subjects, the frontal part of thiscomponent was abolished in the patients, while the parietal portionwas unaffected. These results, taken together with the findings ofCourchesne~1978!, suggest that novel stimuli elicit both a P300and a separate component with a frontal topography. Lacking,however, are data obtained from normal, healthy individuals thatwould allow determining whether novel events indeed elicit twodistinct ERP components.

By acquiring our data with a dense electrode array to providehigh spatial resolution, in addition to the high temporal resolutionthat is characteristic of ERPs, and by applying spatiotemporal PCAto the data set, we were able to dissociate the ERPs elicited by rareand novel events into two components, the P300 and a frontalcomponent most responsive to novel stimuli. We demonstrate thatboth components are elicited to different degrees by rare and novelevents, and we propose that the termNovelty P3be restricted to thefrontal component.

Methods

SubjectsFifteen right-handed students~average age: 19.5 years; 2 men; 13white, 1 Asian, 1 African-American, as described by self-report! atthe University of Illinois at Urbana-Champaign were recruited forthe study. The subjects, who gave informed consent, received ei-ther credit toward meeting introductory psychology requirementsor monetary compensation.

Data AcquisitionThe electroencephalogram~EEG! was recorded using a 129-electrode Geodesic Sensor Net~Tucker, 1993!, with 12-bit digi-tizing at 250 Hz. Electrode impedances were kept below 30 kV ~anacceptable level for this system!, and each electrode was refer-enced to the Cz site. Amplifier bandpass was 0.1–50 Hz~3-dBattenuation at 50 Hz!. Single-trial epochs of EEG began at 200 msprior to stimulus onset and lasted until 804 ms after stimulus onset.Subjects responded by pressing a button on a hand-held box.

StimuliThe auditory stimuli were those described in Fabiani and Friedman~1995!. There were three classes: two tones~350 and 500 Hz, 336ms duration, 10 ms rise and fall times! served as the “rare” and“frequent” events in the oddball sequences, and complex environ-mental sound samples1 were the “novel” events in the noveltyoddball. The novels’ duration did not exceed 400 ms. All stimuliwere presented binaurally at 65 dB.

Experimental Design and ProceduresSubjects were presented with two oddball sequences, the classicoddball and novelty oddball. The classic oddball sequences con-sisted of two tones~P@rare# 5 0.12 andP@frequent# 5 0.88!. Thenovelty oddball sequences consisted of the same two tones, towhich were added “novel” sounds~P@novel# 5 0.12, P@rare# 50.12, andP@frequent# 5 0.76!. Subjects were presented with 300trials per sequence, with an interstimulus interval of 1,000 ms.Subjects were instructed to press a button in response to the raretones as quickly as possible while maintaining a high level ofaccuracy. Both responding hand and which of the tones was rarerwere counterbalanced across subjects. To assure that the novelevents would be equally novel for all subjects, the classic oddballtask was always performed before the novelty oddball. Subjectswere not told that the novel sounds would be presented. The datafrom two “passive” experimental conditions will not be discussedhere~see Donchin, Spencer, & Dien, 1997!.

Data ProcessingThe vertical electrooculogram~EOG! was derived by subtractingthe activity measured at an electrode positioned below the left eyefrom an electrode positioned above it. The horizontal EOG wasderived by subtracting the activity measured between electrodesplaced at the outer canthi of the eyes. The EEG channels were thencorrected for vertical and horizontal eye movements using themethod of Gratton, Coles, and Donchin~1983!. Next, the EEGchannels were re-referenced to an averaged mastoid reference, andERP averages were derived for each stimulus condition in eachtask. The averages were digitally filtered~0–20 Hz! and baseline-corrected using the2150 to24 ms period as the baseline. Topo-graphic maps of the voltage data and spatial factor loadings wereplotted using spherical spline interpolation.

Spatiotemporal AnalysesFirst a “spatial,” then a “temporal,” PCA~cf. Donchin & Heffley,1978! was performed on the dataset to reduce its dimensionalityand disentangle overlapping ERP components. The input to thespatial PCA consisted of all the averages for each electrode andstimulus type~rare and frequent from the classic oddball; rare,novel, and frequent from the novelty oddball! for each subject.These averages were resampled at 125 Hz, and only the 0–752-msportion of the epoch was used. For the spatial PCA, the covarianceamong electrode sites was analyzed across the timepoints of theaverages for each of the stimulus types and subjects, yielding a setof spatial factors representing characteristic spatial patterns in thedata. We refer to these factors asvirtual electrodesbecause eachfactor allowed the computation of a linear combination~i.e. aweighted sum! of the voltages measured at the 129 electrodes, eachof which was given a weight proportional to the factor’s loading atthe electrode. The loadings for a given spatial factor were thecorrelations between the data recorded at each electrode site andthe factor.

Thus, for any given factor, these weighted sums measured theactivity contributed by the factor to the original observations at anyspecific time point, for any specific subject and experimental con-dition. These factor scores, obtained for each time point, for eachERP, could be plotted as a function of time. Each of the originalERPs could therefore be represented by plots of virtual ERPswhich display the scores as a function of time, one plot for each ofthe virtual electrodes.

In the next step of our analysis, the set of virtual ERPs wassubmitted to a temporal PCA, analyzing the covariance among

1There was a total of 36 sounds, consisting of 6 sounds drawn fromeach of 6 categories: bird calls, animal sounds, machine noises, music,human noises, and electronic sounds.

410 K.M. Spencer, J. Dien, and E. Donchin

time points for all of the spatial factors, stimulus types, and sub-jects, yielding characteristic temporal patterns represented by thetemporal factors orvirtual epochs. The resulting temporal factorscores for each spatial factor were used to measure the activity inthe ERP with the morphology and scalp distributions of interest.For both spatial and temporal PCAs, as many factors as wererequired to account for 95% of the variance in the input data setwere retained for Varimax rotation. Analyses of variance~ANOVAs!were performed on the temporal factor scores with the Greenhouse–Geisser correction reflected in the reportedp values.

Results

Figure 1 displays the grand-average waveforms~at midline sites!and topographic maps~at the time point of maximum voltage! forthe stimuli in the classic and novel oddball tasks. The rare events,in both tasks, elicited a classic, centro-parietal P300, whereas thenovel events elicited the more fronto-central pattern described byCourchesne et al.~1975!. Our task, in analyzing the data, was todetermine the componential structure underlying these ERPs.

The analysis began with spatial PCA, which yielded 18 spatialfactors~SFs; see Figure 2!. The spatial factors are represented in

Figure 2 by maps of the loadings of each of the 129 electrodes oneach of the factors. These factors, collectively, allow us to repre-sent the data that were originally recorded in a “space” of 129dimensions~each dimension corresponding to one of the originalelectrodes! in a new, reduced, space of 18 dimensions, each factorconstituting a dimension. The set of spatial factors thus representeda set of scalp distributions that accounted for the spatial variancein the data, each factor representing one scalp distribution.

Of interest to the present report is SF1, which was weightedheavily around the posterior electrodes, and SF3, which had afrontal distribution.2 The contribution of each SF at each timepoint for each of the ERPs in our database is represented by thefactor scores or virtual ERPs. The virtual ERPs averaged acrosssubjects for SF1 and SF3 are shown in Figure 3. Figure 3 shows

2The two spatial factors examined in the present report are the onlyfactors, of the 18 we obtained, whose variance is affected by the experi-mental variables of concern in the present report, which focuses strictly onthe differences between the ERPs elicited by attended rare and novel events.Other factors, such as SF2, although accounting for significant variance inthe EEG, were not affected by the experimental manipulations of interesthere. For more detailed information about all aspects of this study, seeSpencer, Dien, and Donchin~in preparation!.

Figure 1. Grand-average event-related potentials for the classic and novelty oddballs. Left and center: Time series at selected midlinesites. Right: Topographic maps of rare and novel stimuli.

Componential analysis of the ERP 411

that at SF1, the virtual ERP has a waveform in which the commonpattern of P300 was elicited by both rare and novel stimuli, albeitwith different amplitudes across the tasks. The pattern elicited atSF3 by the novel stimuli displayed a large amplitude positive wavethat also peaked at around 350 ms. The rare events in both tasksalso elicited some activity at SF3 in this time range, but of con-siderably smaller amplitude. In the later portion of the epoch, thevirtual ERP at SF3 reversed polarity for the rare stimuli. Thefrequents elicited little activity in either SF. Thus, each virtualelectrode accounted for ERP activity with a distinct topography,each with different polarities and time courses. These virtual elec-trodes represent scalp distributions and not conventional compo-nents of the ERP. As is illustrated by our data, it is entirely possiblefor the same virtual electrode to be associated with more than onecomponent if activity with that particular scalp distribution oper-ates at different latencies and responds differentially to experimen-tal variables.

To disentangle temporally overlapping ERP activity, a temporalPCA was performed on the whole set of virtual ERPs, yielding thesix temporal factors~TFs! or virtual epochs shown in Figure 4~leftpanel!. As can be seen, TF1 had high loadings at the 200–400-ms

portion of the epoch, corresponding to the waveforms of the P300and Novelty P3 in the voltage data, and thus was selected forfurther analysis in this report. The factor scores for TF1 at SF1 andSF3 are shown in Figure 4~right panel!. In the TF1 virtual epoch,rare and novel stimuli elicited activity of approximately the sameamplitude at SF1, although they differed with respect to the amountof activity that was elicited at SF3.

These observations were confirmed by statistical analyses. TheTF1 factor scores for SFs 1 and 3 were submitted to separateANOVAs for each task and SF with stimulus type~rare0frequentfor the classic oddball, rare0novel0frequent for the novelty odd-ball! as the factor. Significant effects of stimulus type were foundfor both SFs in the classic oddball, SF1:F~1,14! 5 33.64,p ,.001; SF3:F~1,14! 5 5.39,p , 0.05, and novelty oddball, SF1:F~2,28! 5 19.35, p , .001; SF3:F~2,28! 5 15.77, p , .001.Within the novelty oddball data, rares and novels did not elicitdifferent amounts of activity for SF1,F~1,14! 5 3.62,p . 0.05,but for SF3, novels elicited more activity than rares,F~1,14! 530.30,p , .001. Comparing across the tasks, rare stimuli in theclassic oddball elicited more activity than did rare stimuli in thenovelty oddball at SF1,F~1,14! 5 6.34,p , 0.05, but not at SF3,

Figure 2. Topographic maps of the spatial factor loadings~virtual electrodes!. Percentage of variance accounted for by each factor afterrotation is shown next to the factor number.

412 K.M. Spencer, J. Dien, and E. Donchin

F~1,14! 5 0.87, p . 0.05. Rares in the classic oddball did notdiffer from novels in the amount of activity elicited at SF1,F~1,14! 5 3.47, p . 0.05, but novel stimuli elicited more SF3activity than the classic rares,F~1,14! 5 45.83,p , 0.001.

We conclude from these results that the responses to rare andnovel events consist of at least two ERP components, one of whichis the P300 component described by Sutton et al.~1965!. The otheris a positive frontal component that is most responsive to novelstimuli. We believe it would make sense to restrict the use of thelabel Novelty P3 to the positive frontal component represented inour data by SF3 and TF1.

Discussion

In this report, we examined the late positive ERP componentselicited by rare and novel stimuli. In keeping with the definition of

components proposed by Donchin et al.~1978!, defining compo-nents requires disentangling the variance in space, across elec-trodes, and the variance in time across the time points at which theERPs are measured. By applying PCA first to the spatial distribu-tion of the observations, followed by a PCA of the temporal be-havior of the data at the virtual electrodes that emerged from thespatial PCA, we were able to examine the effect of the experimen-tal manipulations on distinct spatiotemporal elements of the data.The data presented here suggest that the rare events in an oddballseries and the novel events that have been used to elicit the so-called novelty P3 differ mainly in the relative amplitude of thecomponents they elicit. Both classes of stimuli elicit a frontalcomponent, the Novelty P3, and a posterior component, the P300.Rare events, especially if they are directly associated with thesubject’s task, elicit primarily a large P300. The novel events,although they do~being rare! elicit a P300, also elicit a relatively

Figure 3. Grand-average spatial factor scores~virtual event-related potentials! for spatial factors 1 and 3.

Figure 4. Left: Temporal factor loadings~virtual epochs!. Percentage of variance accounted for by each factor after rotation is shownnext to the factor number. Right: Temporal factor scores for each stimulus type at spatial factors 1 and 3.

Componential analysis of the ERP 413

large amplitude activity at frontal electrodes. Thus, the Novelty P3represents a component that is distinctly different from the P300 asit responds to different aspects of the pattern of stimulation. Elu-cidation of the functional significance of the Novelty P3 requiresa detailed identification of its antecedent conditions, and a methodto measure this component independently. In this study, we haveprovided such a method, spatiotemporal PCA, which can be ap-plied to numerous problems in the ERP literature.

Scalp distribution has always served as a defining attribute ofERP components~Donchin et al., 1978!. Whereas the improvedcharacterization of scalp distribution by the use of dense electrodearrays provides greater power to disentangle components, the shift

to high-density recordings requires a shift in analysis technique aswell. It is difficult to reap the benefit of a dense electrode array withconventional analyses, because these analyses make too apparent themultiplicity of overlapping components that defy simple-mindedwindowed measures. Spatiotemporal decomposition of the ERP byspatial and temporal PCAs provides a method suited for such analy-ses. The use of densely spaced electrodes provides information aboutthe variance in topography at the level of detail dictated by thenature of the data, but this method yields a database of overwhelm-ing dimensionality. PCA provides for a systematic reduction of thedimensionality, capturing the richness of the database in a man-ageable number of “virtual electrodes” and “virtual epochs.”

REFERENCES

Courchesne, E.~1978!. Changes in P3 waves with event repetition: Long-term effects on scalp distribution and amplitude.Electroencephalogra-phy and Clinical Neurophysiology, 45, 754–766.

Courchesne, E., Hillyard, S. A., & Galambos, R.~1975!. Stimulus novelty,task relevance, and the visual evoked potential in man.Electroenceph-alography and Clinical Neurophysiology, 39, 131–143.

Donchin, E., & Heffley, E.~1978!. Multivariate analysis of event-relatedpotential data: A tutorial review. In D. Otto~Ed.!, Multidisciplinaryperspectives in event-related potential research~pp. 555–572!. EPA-60009-77-043. Washington, DC: U.S. Government Printing Office.

Donchin, E., Ritter, W., & McCallum, W. C.~1978!. Cognitive psycho-physiology: The endogenous components of the ERP. In E. Callaway,P. Tueting, & S. H. Koslow~Eds.!, Brain event-related potentials inman. New York: Academic Press.

Donchin, E., Spencer, K. M., & Dien, J.~1997!. The varieties of deviantexperience: ERP manifestations of deviance processors. In G. J. M. vanBoxtel & K. B. E. Bocker~Eds.!, Brain and behavior: Past, present,and future~pp. 67–91!. Tilburg, The Netherlands: Tilburg UniversityPress.

Fabiani, M., & Friedman, D.~1995!. Changes in brain activity patterns inaging: The novelty oddball.Psychophysiology, 32, 579–594.

Gratton, G., Coles, M. G. H., & Donchin, E.~1983!. A new method for

off-line removal of ocular artifact.Electroencephalography and Clin-ical Neurophysiology, 55, 468–484.

Knight, R. T. ~1984!. Decreased response to novel stimuli after prefrontallesions in man.Electroencephalography and Clinical Neurophysiology,59, 9–20.

Knight, R. T. ~1997!. Distributed cortical network for visual attention.Journal of Cognitive Neuroscience, 9, 75–91.

Spencer, K. M., Dien, J., & Donchin, E.~in preparation!. Spatiotemporalanalysis of the late ERP responses to deviant stimuli.

Squires, N. K., Squires, K. C., & Hillyard, S. A.~1975!. Two varieties oflong-latency positive waves evoked by unpredictable auditory stimuliin man. Electroencephalography and Clinical Neurophysiology, 38,387–401.

Sutton, S., Braren, M., Zubin, J., & John, E. R.~1965!. Evoked-potentialcorrelates of stimulus uncertainty.Science, 150, 1187–1188.

Tucker, D. M.~1993!. Spatial sampling of head electrical fields: The geo-desic sensor net.Electroencephalography and Clinical Neurophysiol-ogy, 87, 154–163.

~Received July 2, 1998;Accepted September 14, 1998!

414 K.M. Spencer, J. Dien, and E. Donchin