EEG EVENT-RELATED POTENTIALS AND SIGNAL …cognitrn.psych.indiana.edu/busey/eegseminar/pdfs...EEG EVENT-RELATED POTENTIALS AND SIGNAL DETECTION Robert T WILKINSON and David M SEALES

Biological Psychology 7 (1978) 13-28 © North-Holland Pubhshmg Company

EEG EVENT-RELATED POTENTIALS AND SIGNAL DETECTION

R o b e r t T WILKINSON and David M SEALES

Naval Health Research Center, San Dwgo, Cahforma 92152, U S A

Accepted for pubhcatxon 22 June 1978

The purpose of this study was to determine whether certain previously reported relation- ships between event-related potentials and measures of signal detection performance occur in vigilance as well as psychophyslcal settings In the course of the study, evidence was found which challenges previously proposed psychological correlates of 'P3o0' EEG was recorded while 15 subjects carried out a 40-mm signal detection test. CNV was measured between a warning chck and the brief offset of a dim hght denoting the possible (/9 = 0 5) occurrence of the signal, a faint tone m the constant background noise P3oo was measured at a point 300 msec following offset of the dim light As reported previously m psychophyslcal settmgs, P3oo amphtude ~as positively related to signal intensity and response confidence, and was larger for correct detections (Hits) than tot correct rejections, misses or false alarms From first to second half of the test both Hits and false alarms fell, response criterion 13 rose, and the amphtude of both CNV and P3oo tell The latter negative relationship between 13 andP30o contrasted with a positive one when subjects rated their signal reports at three levels of confidence, here the most confident ratings (high /3) were assoctated with the highest amphtude of P3oo CNV reflected mdwldual ablhty to sustain performance Results are explained in terms of a two- factor version IWflkmson, 1976) of the prior state/reactive change hypothesis (Karhn, 1970) It is suggested that time on task constitutes a prior state influence on P3oo, whereas other vari- ables influence P3oo by means of either reactive change or 'real' change m an endogenous P3oo component

1 Introduction

This IS pr imar i ly a s tudy of the re la t ionship b e t w e e n signal d e t e c t i o n perfor-

mance and event - re la ted po ten t i a l s (ERPs) In the EEG The re levant pa rame te r s o f

signal de t ec t i on are d ' , the index o f a pe r son ' s a b l h t y to d iscr iminate fa in t signals in

noise , and/3 , the c r i te r ion which he adop t s m deciding h o w risky to be m repor t ing

a given d i s tu rbance as a signal ( T a n n e r and Swets , 1954) The two ERP measures

are the c o n t i n g e n t negat ive var ia t ion (CNV), a negat ive-going wave first r epo r t ed by

Walter , Cooper , Aldr idge, McCal lum and Winter ( 1 9 6 4 ) and t h o u g h t to reflect the

level o f e x p e c t a n c y wi th wh ich a pe r son awai ts the onse t o f an impera t ive s t imulus ,

and 'P3oo' , a pos t - s t imulus , posi t ive wave which has b e e n he ld to ref lect a var ie ty o f

psychologica l pa rame te r s associated wi th decis ion processes , a t t e n t i o n , and dlscrun-

m a t l o n (see Tue t l ng and S u t t o n , 1973, Tue t ing , in press, for recen t reviews) I t

13

14 R T h'tll~mson, D M Seales/ Lvol, cd potenttals

may be noted here that the cognomen 'P~oo" will be used %1 thl,, phenomenon throughout this paper purely as a lnatter of ~onvemence and without prejudice to the vmws of the present autho)s, o~ othels, as to the latency of this event or its quahflcatmns to be considered one of the components of the stnnulus-ew~ked response

The aim of the study is to examine tire relationship of P~oo to performance in a vigilance, as opposed to a psychophyslcal, setting of signal detectmn in ordel to learn more about the psychological s~gnlflcance of thePa0o wave In vigilance tasks the stgnal detect ion palameters d' and 3 have been shown to vary with time at work (Broadbent and G~egory, 1963, see Broadbent, 1971, pp 6 6 - 1 1 l , for review) The inaln questmn here is whethei PJoo covarles with these changes m d' and 3, and if

so whether this corresponds to the covarlation found between PJo0 and d' and 3 when the parameters are signal strength (Hlllyard, Sqmres, Bauer and Lmdsay. 1971), signal probabil i ty, response confidence (Paul and Sutton, 1972. K Squires, N Squires and Hfllyard. 1975a, 1975b), and payoff schedule (Paul and Sutton, 1972) As regards CNV, the questmn is whether it covanes with perfm- mance as would be predicted by theories of vigilance based upon psychological prmmples of expectancy (Deese, 1955), and whether therefore It may provide a guide to individual differences in the abihty to lnamtam prolonged concentration during long, boring tasks

2 Method

2 ! Sublects and procedure

24 enlisted men between the ages of 20 and 30 at tended the laboratory one at a time for a single sessmn of about 3-hour duration to carry out the test of signal detection Dmmg tlus time electrodes were placed to record EEG and EOG (dec- trooculoglam) The subject was then given instructions on the performance task, followed by a prehmlna)y practice run o f 4 mm and a final one of 10 mm There followed a control run of 5 mm in which the subject heard and saw the task stimuh passively, hawng been instructed to ignore them and relax, but not fall asleep After a break of about 5 mm the mare 40-rain task was camed out This was followed by a further 5-mln control period passively ignoring the stlmuh Subjects were reformed befolehand that they would be given a summary of thmr scores on the performance test and an indication of how they related to those of othel subjects

2 2 The perfi)rmance task

The task was one of prolonged signal detection in which the subject had to detect faint tones in the plesence of continuous white noise at a level 70 dBA above thleshold The tones were of frequency 1 KHz and duration 50 msec The test con- slsted of 240 trials The lntertrlal interval vaned quasi-randomly from 3 to 13 sec

R T Wllkmson, D M Seales / Evoked potenttals 15

(mean 8 sec). A trial comprised three stlmuh coming at 1-sec intervals. The first ($1) was a short, high-pitched chck presented as a warning for the subject to listen for a possible signal 1 sec later The second ($2) was the signal, when it occurred (p = 0 5), combined with the brief offset of a hght-emlttmg diode (LED) The LED blink came whether a signal was present or not and acted as a time marker so that the subject could know precisely when the signal was absent as well as present (Sqmres et al , 1975a) Finally, 1 sec later, a low-pitched click ($3) served as a response cue directing the subject to report whether he had heard a signal. This he did by pressing one of four buttons on a response panel to indicate, respectively, no signal, signal reported at high confidence, signal at medium confidence, and signal at low confidence level. The chcks, tones, and white noise were presented through a loudspeakel located 30 cm behind the subject's head The LED hght was mounted 50 cm m front of the subject at eye level He was Instructed to fixate this hght as continuously as possible throughout the test. The LED provided a small, red light of moderate intensity, the bhnk of which was clearly discernible

Signals occurred on half of the 240 trials. The order of signal and non-signal events was random over the first quarter of the test and this order was repeated m subsequent quarters An attempt was made to divide the signals into three levels of intensity. In fact the adjustment of signal strength was unsatisfactory m that the 'High' and the 'Medium' intensity signals were about the same and allowed a high probablhty of detection whereas the 'Low' antenslty ones were much more difficult to hear Unfortunately the time available for the experiment precluded a restart wath signal strengths adjusted, so it was decided to continue wath essentially two signal strengths, High for ] of the signals and Low fm the rest, randomized on the same basis as for the signal and non-signal stlmuh

The performance records were scored m terms of Correct Detections (Hits) and Errors of Omlssmn (Misses) in the case of trials containing signals, and False Alarms (FAs) and Correct Rejections (CRs) for the non-signal-bearing trads In addition, measures of d' and/3 were obtained from the observed overall probabilities of Hit and FA by entering Freeman's 'Table of d' and/3' (1973) with the appropriate fig- ures Unfortunately these assessments must be heated with some clrcumspectmn because of the two intensities of signal used It is not clear what degree of normahty is expected of the neurological 'noise' and 'noise plus signal' dlstnbutmns as assumed by the Theory of Sagnal Detection for calculating d' and 13, nor can actual distributions be measured in the brain Nevertheless, the present assessments of d' and 13 should be legarded as more approximate than usual, for with two signal strengths rather than one there is a greater probablhty of departure from normahty. However, there is one change m/3 which is of particular importance to the conclu- sions drawn from this study, namely an increase in 13 from first to second half of the test In this case it is assumed that if both Hits and Fas fall with tame on task it is reasonable to conclude that cntermn has moved in the direction of greater caution, that is, a change slmflm to a rise in 13 has occurred

16 R T h'tlkmson, D M Scales 1.) oAed potcnttala

2 3 E lec t rophys to log tca l measures and the ir analysis

Sllvm disc electrodes wele emplo)ed m recording EEG between Ce and hnked mastoids (impedance <5 kg2), and EOG from snes located &agonall? on the upper and lower canthl of the left eye Electrode sites were cleansed u n h acetone Fur- ther, to reduce skin potentials at the EEG recording sites, the outer layers of skin were removed by abrasmn w)th pmmce paste EEG and EOG we)e amphfled and wmten out by means of a Beckman Type R Dynoglaph (response 3 dB down at 0 02 and 32 Hz) The wrlteout also contained malkers coded m relation to the nature of the signal tone and the subject 's response Data were stored on analogue tape for offlme analysis using a PDP-12 computer EEG and EOG were digmeed for each trial from a point 500 msec before the onset of the warning chck (S~) for a period of 3 sec (256 dlgmzatmn points), that is to a point 500 msec after the onset of the response-cue chck ($3) Polygraph records were examined visually for reals with eye movements clea)ly influencing the morphology of the EEG and also t~mls with muscle altefact These trials (mean 25(~, mtersubject range 1 5 - 3 8 # ) were deleted from further electrophystologmal analysis, though not from analyses of performance Acceptable trials were combined into ave)aged event-related poten- tmls (ERPs) m relation to var)ous criteria (e.g tone mtensny confidence of response, quarter of task), and a tutther check for the influence of eTye movements was made by exammg the ave,aged EOG obtained wnh each ERP. No further reJec- tion of ERPs was wmlanted

The electrophys)olog)cal measures a,e dlustrated m fig 1 and were as follows (1) CNV The amphtude of the EEG averaged over the period 50 msec before the point of signal onset and related to the EEG level averaged over the 50 msec p~mr to the onset of the warning d ick (2)P~o0 The posltlvny of the EEG tiace 300 msec aftra the s~gnal onset related to the average EEG level over the 50 msec prior to the s)gnal onset point

. ~ _ CNV ~" ~--'L~ I See" .... ~= ~ I / |$e(:" ~I

Worn S=g. ~ Resp. Fig 1 Schematic ot an averaged event-related potential to illustrate the method o1 quantifying CNV and P3o0 amphtudes Blocks indicate where a mean value `has derived from 50 msec of EEG The P300 trough ,has measured at a single point 300 msec lol lo'alng signal onset Warn = ~ a r m n g cl ick, Slg = signal onset and brmt LED offset (50 msec) , Resp = response-cue chck

R T Wtlkmson, D M Seales / Evoked potentials 17

2 4 Stattsttcal analysts

Non-parametric statastlcs (Siegel, 1956) were used throughout to avoid assump- tions being made about the normality of the dxstrlbutaons of the data Spearman's rho was used for correlations and exther a simple Binomial Test or Wllcoxon's Matcheo Pairs, Signed Ranks p~ocedure (T) for comparing pairs of scores o~ electro- physiological levels

3. Results

Fig 2 dlustrates for a typical subject the 3-sec, computer-averaged examples of the EEG derwed for the various combinations of trials reported below The figure also shows the averaged traces (combined) for the passwe control runs which were

Hit Miss FA CR

Category ~ ~ ~ ~'~

No rone]

Category ~ ~ [~/ ~

H Cot rol ]

Confidence Level

Quarter 't ~ ~ ~J ~/

We i l ~ p . Fig 2 Three-second averages of EEG, covering a whole trial, from one typical subject The last trace, thtrd row is from the pre- and post-test passwe control runs Otherwise in each row the total of 240 trials (less about 25% lost through eye, muscle, or movement artifact) are sub- dwlded along different dmaenslons category of behavloural response, signal intensity, response confidence, and quarter of the test The number of samples m each average trace as shown to the right of the trace Events during each trml are denoted as follows Warn = warning clack (St), Slg = signal onset point ($2), Resp = response-cue chck ($2)

l 8 R F Wdkmson. D M Scales / 1- vo~ed potenttals

Table 1 Measures derived from averaged EEG traces samples as a lunctmn ot behavloural response, signal intensity, and response confidence

P3 o o CNV

Behavloural Hits 14 3 5 4 response FA 7 4 a) 6 4

CR 6 8 a) 4 8 Miss 6 2 a) 3 6

Signal High 14 6 - intensity Medium 15 1 -

Low 7 7 b) _

Confidence High 15 3 5 9 level Medium 6 l c) 5 2

Low 6 9 c) 5 1

a) Amphtude less than that of Hits, p < 0 01 b) Amphtude less than that of High or Medium intensity signals, p < 0 01 c) Amplitude less than that of High confidence, p < 0 01

presented before and after the main 40-ram performance test Table 1 shows the

measures of P3oo and CNV averaged over all subjects and derived from the EEG

traces averaged with respect either to Behavloural Response, Signal Intensity, or Response Confidence

3 1 Behavloural response

The amplitude of P3oo associated with Hits was lalger than that to CRs m all 15 subjects ( p < 0 001), and larger than that to Misses and FAs in 14 out of 15 sub-

jects in each case (p < 0 001) There were no slgmficant differences among the lat- ter three categories

3 2 Stgnal mtens t ty

As would be expected d' was higher for the High (d ' = 5.00) than for the Low intensity (d' = 0.77) signals in all subjects (p < 0 00 l ) The same was true for the amphtude of P3oo (P < 0.001) Thus, when the intensity of the signal was varied from trial to trial durmg a long test, a large P3oo was associated with high slgnal &s- crmnnablhty (d')

3.3. Confidence level

For comparisons revolving confidence level the numbers of subjects considered was hmlted to 12 This was due to some subjects fading to make any responses in

R T Wtlkmson, D,M Seales / Evoked potentials 19

the Medxum or Low categories of confidence Signals (or FAs) which were reported at a High level of confidence were associated with higher P3o0 waves than those reported at Medium or Low levels of confidence in 12 out of 12 (p < 0 001) and 11 out of 12 (t9 < 0 01) subjects respectwely High confidence is normally associated with a high level of/3 (more Hits m relation to the number of FAs) Thus when con- fidence level was the parameter hxgh /3 was assocmted with a high amplitude of

PJoo

3 4 Changes with ttme on task

Changes m performance and EEG as a function of time on task were examined by dividing the test into halves, and assessing the difference between first half and second half score For the purpose of correlating time-related changes of the various measures within each subject scores were taken over each quarter of the test Where d' and/3 were calculated only 12 subjects could be included in the analysts involving halves of the test. This was because three subjects faded to make any FAs m at least one of thear halves, for the same reason only nine subjects could be included m the within-subject correlations, considered later, of d' and/3 with CNV and P3oo across quarters of the test

Table 2 shows Hits, FAs, P3oo and CNV averaged for each quarter of the test The measures of P30o and CNV were derived from all artefact-free trials m each quarter A freer analys~s within quarters was impossible because the numbers of samples per average trace would have been too small for clear measurement of the P3oo and CNV amplitudes Values of d' and 13 were averaged in the same way for

Table 2 Scores m the performance and EEG measures In each quarter of the test, or half of the test m the case of d' and

Quarters of the test

1 2 3 4

Hits (%) 81 8 77 3 75 1 70 7 False alarms (%) 18 0 17 1 15 6 14.9 P3oo (,uV) 11 1 10 8 8 1 9 2 CNV (~V) 5 2 7.0 3 6 5 4

Halves of the test

1 2

d' (N = 12) 1 87 1 79 (N = 12) 1 47 2 08

20 R T Wilkinson, D M 5ealcs / Lroked porenttals

each half of the test for the 12 subjects for whom these parameters could be cal- culated on a half-test basis As regards performance the results confirm an general those of Deaton, Tobias, and Wilkinson (1971) for changes with time oil task in a prolonged test of signal detection From first to last half of the test l-hts felt in 14 out of 15 s u b j e c t s ( p < 0 0 0 1 ) Fa l seAlamls fell a l s o ( T = 1 9 ~ , N = 1 3 , p < 0 0 5 . one-tail) Taken together these ~esults suggest that the subjects' criterion becanle more cautious from first to second half of the test, and m some confirmation of this the measure 3 (though an approxnnat lon, see section 2 2 above) rose in 11 out of 12 subjects (p < 0 0 l ) Regarding sensltlVlty, there was no significant change in d' with time on task, although the downward direction of change was again the same as that found by Deaton et al (1971) Assessing the physiological measures on the same basis, there was a dechne in the amplitude o fbo thP30o , (T = 10, N = 15, p < 0 01), and CNV (T = 18, N = 14, p < 0 02) from first to second half of the test

In order to see whether the average confidence level of ~eports changed during the course of the test an arbitrary measure was detlved by counting 3, 2, and 1 respectively for High, Medmm, and Low confidence reports, summing, and dividing the total by the numbers of such reports for each subject and for each quarter of the test On this basis the change m response confidence was small during the test and insignificant statlstically, the figures being 2 67, 2 65, 2 64 and 2 64 for the first through fourth quarters Eight subjects showed a fall and six an increase m confidence with time

3 5 Correlattons, wtthtn sublects, as a function o f t t m e on task

For each subject the scores of Hits In each quarter were correlated with the cor- responding levels of P30o using Spearman's rho These correlations were then aver- aged over all subjects and the significance of the difference of this average from zero assessed by the Wllcoxon procedure For the colrelatlon between Hits and P3oo rho = +0 24, N = 15 and between Hits and CNV rho = +0 12, N = 15 Neither correlation was significant Similar correlations were performed for d' and/3 in rela- tmn to the P3oo and CNV for the nine subjects for whom d' and 3 could be calcu- lated in every quarter of the test (i e had no zero FA scores m an)' quarter) Both of the correlations involving P3oo were significant P3oo vs d ' , rho = +0 36, N = 9 ( p < 0 0 5 ) , P3oo vs 3, rho= 0 4 4 , N = 9 ( p < 0 0 5 ) Thus correlation of the present measure of 3 with P3oo confirms what is seen when concurrent changes in P3oo, Hits, and FAs are observed f iom first to second half of the test When P3oo falls, criterion rises This negative relatmnship is in contrast to the positive one ob- served in the present study (and prewous ones, see Discussion) when P3oo (to Hits only) was related to crltermn in terms of the confidence level with which subjects reported signals. Finally the CNV failed to correlate significantly with either d' or /3 d ' vs CNV, rho = +0 1, N = 9 ,3 vs CNV, rho = - 0 29, N = 9, and there was a sub- stantlal positive correlation between the two physmloglcal variables P3oo and CNV, rho = +0.50, N =15 (p < 0 01)

R T Wtlktnson, D M Seales / ~ yoked potentlals 21

3 6. Correlattons, between sublects, as a function o f time o f task

There were no significant relationships between overall individual levels of either CNV or P30o and individual levels of performance. However, when decrement m performance with time was the parameter (the Index being again first minus second half of the test) it was found that subjects with the largest overall CNV suffered least decrement m Hits rho = - 0 59, N = 15 (p < 0 02) The corresponding corre- latmns for CNV with d' and/3 failed to reach significance CNV vs d', rho = - 0 34, N = 12, CNV vs/3, rho = +0 31, N = 12. None of the corresponding correlations of P3o0 with decrement in Hits, d' or t3 were slgmficant, but there was a substantial negative correlatmn of P3oo with decrement in FAs rho = -0.56, N = 15 (p < 0 05) Subjects wRh the largest overall P3oo showed the smallest dechne in FAs during the test

4 Discussion

We have studied P300 as a functmn of signal detection performance in a setting of prolonged vigilance rather than the more usual one of a series of bnef psycho- physical assessments, the question being whether the relationship of P300 to perfor- mance and other psychological factors remains the same. It does not, and this is taken as an ln&catmn that two largely independent factors may contribute to changes in the amplitude of what is recorded as Paoo in behavloural settings

4 1 Relatlonshtp between P3oo and 13

Regarding Paoo and performance, the only correlations of any appreciable mag- nitude across the four quarters of the test were those of P3oo with/3 and with FAs In previous studies of P3oo and/3 the independent variable was either the confidence rating of the detection (Hdlyard et al, 1971, Squires, Hlllyard and Llndsay, 1973, Squires et al, 1975a, 1975b) or the payoff basis (Paul and Sutton, 1972) Large P3oo waves to Hits were associated with the high criterion level as implied by high confidence Hits or a payoff schedule dictating caution m responding 'signal'. We ob- tained the same result when response confidence was the independent vanable, but when the parameter was the passage of time during the relatively long and boring task the correlatmn between P3oo and cntenon was in the opposite d~rectlon A rise in c~ltenon by the second half of the test was associated with a fall m P3oo This provides the first hint that two factors may be operating to Influence the amphtude of P3oo One factor, which produces a large P3oo when cmerlon is high, may contri- bute maximally when levels of endeavour are great, for example in bnef tests when Instructions demand high confidence and certainty before a signal is reported as such The other factor, which makes for a small P3oo when criterion is high, may predominate when high criterion reflects a general dlsInchnatlon to make reports,

22 R T Wilkinson, D M Seales / Evoked potentials

whether Hits or FAs, towards the end of a prolonged, lepetitlve test oi experllnen- tal sessmn The lmphcation here is that P3oo may be reflecting some aspect of incentive, and In this context it may be relevant to recall that Paoo is leduced when knowledge of, and payment by, results is withdrawn (Wilkinson and MoiIock, 1967)

4.2 Uncertainty

Early writers (Sut ton, Braren, Zubln and John, 1965) claimed that it was uncer- tainty about the nature or onset of the stimulus which contr ibuted to large ampli- tude P3oo waves These conclusions were based on a Stoles of guessing experiments in which the subjective uncertainty about the nature of the stimulus was manipu- lated by valylng the relative probabilit ies of two stimuli The lower the probabil i ty of a stimulus, the larger was P3oo But there are two P30o relatmnshlps in the present results which signal probabil i ty cannot explain The first is that P3oo was larger for Hits than for Misses or CRs m the signal detection paradigm, confirming the iesults of Hillyard et al (1971) and Cael, Nash and Singer (1974) This cannot have been due to a low probablhty of Hits for in both our study and that of Hillyard and his colleagues there were approximately 90% correct detections and the signal and non-signal stimuli were equally probable The second relationship is that P3oo dechned from fust to second half of the present test Prior uncertainty cannot have varied during the course of the test since the proport ion of signal and non-signal events was identical in each quarter, nor can the decline in P30o have been due to the fall in the number of Hits (which yield the largest P3ooS) during the course of the test This could account for no more that one fifth of the reduction in overall P3oo amphtude from first to second half of the test

4 3 Internal template

To account for the larger amplitude of Paoo to Hits as compared with Misses and CRs Hillyard et al (1971) suggested that P3oo was produced when an internal tem- plate (presumably of the signal) was matched by the external event This hypothesis too is difficult to reconcile with the present data, on two counts We introduced a faint visual cue as a time marker for the point at which the signal would rather appear or not, as did Squires et al (1975a) This provided a physical stimulus for the non-signal, for which there could also be an appropriate internal template Yet P3oo was still larger to Hits than to CRs Why should the signal template be the one favoured9 Second, the physical nature of the signal stimuli remained the same throughout the present test and so therefore should the matching of the subjects' Internal template, yet the amplitude of Paoo declined with time With a low signal frequency it might be ~rgued that the template itself may become debased with the passage of time, but in the present setting signals were being detected at an average of one every 20 sec

R T Wtlktnson, D M Seales / Evoked potenttals 23

4 4 Orienting response

Another hypothesis of the nature of P3oo is that it may constitute an EEG com- ponent of the orienting response (OR) (Ritter, Vaughan and Costa, 1968, Vaughan and Rltter, 1970, Roth, 1973, Courchesne, Hlllyard and Galambos, 1975) In settings where Paoo follows unpredictable, infrequent, Irrelevant stimuli and is absent to regular background ones this may be a tenable hypothesis, but not in set- tings where P~0o appears to valy with behavloural parameters, such as the present one Here a relatively high ~ate of signal presentation should have resulted in a rapid decline in any OR to the signal stlnmlus As a correlate of OR, P3oo should have done the same during the first minute or two of the test and levelled off to a steady amplitude thereafter Instead the fall in P3oo was a steady one throughout the test

4 5 Signal mtenstty and response confidence

When the present data were analyzed on a trial-by-trial basis Paoo reflected both signal intensity and response confidence as in previous studies (Hlllyard et al., 1971, Squires et al, 1973), that is, high levels of signal intensity and confidence were associated with high amplitude of P3oo However,when time on task was the param- eter these relationships failed to hold Although P3oo declined in amplitude from first to second half of the test, signal intensity remained the same In terms both of the physical stimulus (a feature of the experimental design) and the 'subjective' stimulus as indicated by d' Response confidence remained the same too in terms of the proportion of 'signal present' reports made in the high confidence category It follows that the fall in P3oo with time on task must have been due to the Influence of some factor quite different from that which caused P3oo to vary with signal Intensity and response confidence from trial to trial

4 6. Prtor state/reactive change

An Important way in which time on task effects differ from those of signal intensity, trial outcome, and response confidence is that, whereas the latter three were of necessity post-stimulus influences (since in the present experiment the order of the various trials was random as far as the sublect could tell), time on task effects could have been due to influences prevailing either before or after the pre- sentation of the stimulus

An explanation of the way In which the pre-stlmulus effect of ttme on task could influence the P30o measure is at hand in the prior state/reactive change hypothesis of Karhn (1970) and particularly the version of this hypothesis proposed by Wilkin- son and his co-workers following a series of empirical tests of the original formula- tlon (Wilkinson and Lee, 1972, Wilkinson and Spence, 1973, Wilkinson and Ashby, 1974, Wilkinson, 1976) This is that P3oo may be modulated by summation with the positive-going resolution of negativity which IS present in the EEG when the

24 R T Wtlkmson, D M Seale~ / k voaed potenttals

stunulus occurs FoJ convemence we will refer to this negahvaty as "CNV" m what follows The degree of this resolution, and the consequent exaggregatlon of P3oo amphtude, will then be a function of two factors (1) The level ofnegatlvzty (prior state) (2) The degree to which this negativity is reduced oi 'resolved" as a function of the stmiulus and the subject 's response to it (reactive change) This reactive change may vary independently of prior state negatlvity, but with the unpoItant proviso that the extent of this variation must be a function o f the level of that negativity, being in the hmlt , zero when negativity is zero

There are two features of present data which suggest that the effects of time on task in this experiment may be best explained m terms of the above two-factor for- mulat ion The first and most dnect of these concerns the CNV, a measure which is usually taken as providing some indication of the prior state level of negativity CNV decreased significantly from first to second half of the test and in doing so it correlated positively with the level of P3oo across quarrels of the test This would appear to establish a clear relationship between a prior state index and P3oo am- phtude when tlme on task is the variable The second feature concerns the corre- lations we have observed between Paoo and the criterion measure ~ as derived from the analysis of the performance data in terms of theory of signal detection When t i m e o n task was the parameter this correlation was negative When response con-

~ d e n c e was the (post-stimulus) parameter the correlation was positive On any sin- gle factor model these results appear irreconcilable, but on the two-factor one pre- sented above they can be harmonized It is not unreasonable that two indepen- dently acting variables, in this case a prior state one of time on task, and a post- stimulus one reflecting either reactive change or a 'real ' change in a P3oo component of the stimulus-evoked potential , should influence P3oo m opposite ways in relation

to the same parameter of performance Thus the present study at once shows that prolonged work on a task can reduce

the amphtude of P3oo following the test stimuli and also suggests that this effect on P3oo is due to prior state influences operating within a two-factor form of the prior state/reactive change hypothesis It follows from this that in no experiment where condit ions are such as to produce systematic changes in the pre-stlmulus negativity can researchers afford to ignore the possible contr ibut ion of these prior state influ- ences to what they measure as the amplitude of P3oo Examples of such experi- ments are those involving long, boring periods of experimental testing, or variables whach may influence the motivational state of the subject such as incentive, drugs, and sleep deprlvatlon

From the methodological point of view it will be necessary in such studies to ensure that experimental conditions are balanced for order during the course of prolonged sessions, and that pre-stlmulus negativity is reduced as far as possible by avoiding the practlce of presenting trials at regular and short (say less than 4-sec) intervals. The ideal is a syst}m which prevents the subject from predicting when a stimulus will come and what it will be, thus in the present experiment inter-trial intervals varied unpredictably from 3 to 13 sec Finally, where the conventional

R T Wdklnson, D M Seales / Evoked potenttals 25

$ 1 ~ 2 CNV can be measured this should be done to compare its response with that of P3oo to the experimental variables

From a more theoretical point of view our results suggest that It is impossible to consider the aetlology of changes in P3oo in response to psychological influences without taking mto account the role of prior level of negativity in the context of a two-factor version of the prior state/reactive change hypothesis, as outlined earlier.

Let us now consider the influence of the post-stimulus factor Itself. In the present experiment this is the factor relating to the effects of signal intensity, response confidence, and trial outcome upon the amplitude of P3oo In the context of the two-factor prior state/reactive change hypothesis outlined above we have no way of knowmg from current evidence, whether the observed effects of these variables on P3oo were due to changes in the reactive resolution of CNV (given prior CNV level) or to 'real' changes in the amplitude of Paoo as a direct index of cortical Information processmg

It is true that most authorities m this field would reject the reactive change explanation on the basis of current correlational and topographical evidence relatmg to CNV and Paoo (for example, Pooh, Thompson, Williams and Marsh, 1974, Donchin, Tuetmg, Rltter, Kutas and Heffley, 1975) But these data only reject the prior state/reactive change hypothesis in its early form as Karlin (1970) appeared to present It (his position is not wholly clear on this point), namely that P3oo is 'nothing but' CNV resolution. Once it is conceded that 'both CNV return and a discrete P3oo wave may contribute in varymg degrees to posltlvIty at 300 msec' (Wilkinson and Lee, 1972) this form of the prior state/reactive change hypo- thesis is not mvahdated by the topographical distributions of CNV and Paoo being different (Donchm et al, 1975), always provided they are wide enough to overlap, which in most locations they are.

Turning to correlational studies, once It is conceded that given the level of prior negativity its resolution may vary as a function of the transactions which follow the stimulus, (and this was demonstrated by the results of Naatanen, 1967 Donald and Goff, 1971, and Wilkinson and Ashby, 1974) a lack of correlation between CNV and P3oo is no argument for denying the influence of CNV resolution through summation upon what is measured as P3oo, nor, can it be said in this case that 'reac- tive change becomes a euphemism for cortical potential' (DonchIn, in McCallum and Knott, 1975) because, unhke the cortical potential in a narrow sense, reactive change will be modulated directly by the level of the prior negativity from which It emanates, much, to take an electronic analogy, as the signal through a valve may have its amplitude (cf resolution) modulated by the bias on the grid, (cf prior negativity) yet still vary mdependently of the grid bias in all other respects

Let us now stay with the electronic analogy to insert necessary subscript If there be no bias on the grid no signal will pass through the valve Similarly with no prior negativity there can be no reactive resolution to influence the P3oo measure. It seems unlikely, however, that lack of CNV, as conventionally measured In the S~-$2 paradigm, can be advanced as a sufficient token of this state (see Donchm

26 R T Wdkmson, D ~I Seales / k i oa.ed potemtals

et a l , 1975, p 1 for review of these ~.ontrlbutmns) Some or all of the CNV may have developed prior to S~ (Knott and Peters, 1974), paJtlcularly whele mtertrlal mteivals are short, or, there may be a tomc level of negatlVlt~ which prevails throughout the task (Wilkinson and Lee, 1972, Wilkinson, 1976) Demonst)atmg the absence of such negativity may well be &fficult, but the onus to do so is upon those who would use such a demonstratmn to deny an influence of pre-stmlulus

negativity on P3oo

4 7 C N V a n d m d w t d u a l & f f e r e n c e s

Consldenng CNV m its own sight, we have observed that its average amplitude over the four quarters of the vigilance test predicted fairly well an individual's abil- ity to maintain signal detection performance Specifically, high overall CNV amph- tude was associated with relatively constant performance, low amphtude w~th a drop m performance from the first to second half of the test Thus CNV might serve in the selection of mdlvlduals for tasks requiring sustained concentratmn m repetitive situations On the other hand, as noted above, the CNV may still be an eqmvocal measure m that it is difficult to be sure that the point from which it is measmed, usually EEG level at the onset of the warning stimulus ($1), is a true baseline Knott and Peters (1974) have already suggested that some CNVs in highly anxious or aroused subJects may appear artlflcmlly small because they are already close to then celhng before the warning stnnulus arrives Similar suggestions have been made by Wilkinson (1976), implying that a slowly varying resMent level of negativity may be present between trials of tasks similar to the present one, and that this may make the true level of negativity difficult to judge at the imperative stmmlus In the present case, raised or ' resident ' negatlvity between trials, as would occur plesumably m individuals who failed to relax their wgflance between trials, would produce low amphtude CNVs That it should be these same subjects who show perfolmance deczements ~s consistent with the collective evMence of Broadbent (1971, p 83) who reports 'It seems to be a broadly acceptable conclu- sion that a task which allow moments of relaxanon will show a constant d' during the work period, while one winch reqmres continuous and sustained observatmn will give a drop m d' during the run ' Only furthe) experiments therefore, m which between-t~lal as well as within-trial levels of EEG ale averaged, can resolve the lntm- pie ta tmn of the present ~elatmnslup between mdlvMual CNV level and the ability to sustain performance during prolonged vigilance

Acknowledgements

This lesealch was suppmted m part by Naval Medical Resealch and Development Conunand, Navy Department , Resealch Task No MROOO 01 01-6010 The views presented m this papm ale those of the authors No endorsement by the Depart-

R T Wtlkmson, D M Seales/ Evoked potentials 27

ment of the Navy has been given or should be referred The semor au thor carned

out thas work whde on de tachment from Apphed Psychology Umt , Medaeal Re-

search Councd, Cambridge, England. He would hke to acknowledge with much

grati tude the part played by Dr L C Johnson o f NHRC and Mr G H Lawrence o f

Advance ProJects Research Agency , U S in arranging and support ing thas tour o f

duty. The t ime generously devoted by Dr. Johnson and his s taff to dascussmg this

paper is also gratefully acknowledged Fanally we wash to c o m m e n d the enthusmsm

o f V Ddeo an helping with the admlmst ra t lon o f the exper iment

References

Broadbent, D E and Gregory, M (1963) Vigilance considered as a statistical declsxon British Journal of Psychology, 54,309-323

Broadbent, D E (1971) Decision and stress New York, Academic Press Cael, W W, Nash, A and Stager, J J (1974) The late positive components of the human EFG

m a signal detection task Neuropsychologla, 12,385-387. Courchesne, E , Hdlyard, S A and Galambos, R (1975). Stimulus novelty, task relevance and

the visual evoked potential m man Electroencephalography and Chnlcal Neurophyslology, 39,131-143

Deaton, M , Tobtas, J.S and Wilkinson, R T (1971) The effect of sleep deprivation on signal detection parameters Quaterly Journal of Experimental Psychology, 23,449-452

Deese, J (1955) Some problems m the theory of vlgalance Psychological Rewew, 62, 359- 368

Donald, M W and Goff, W R (1971) Attention-related increases m cortical responslvlty disso- ciated from the contingent negative variation Science, 172, 1163-1166

Donchm, E (1976) CNV-P3o0 relationships. In W C McCallum and J R Knott (Eds) The Responswe Brain Proceedings of the Third International Congress on Event-Related Slow Potentials of the Brain Bristol, England John Wright, p 230.

Donchm, E , Tuetmg, P , Rltter, W, Kutas, M and Heffley, E (1975) On the independence of the CNV and the P300 components of the human averaged evoked potential Electroen- cephalography and Chmcal Neurophyslology, 38,449-461

Freeman, P R (1973) Table of d' and ~3 Cambridge Umverslty Press Hlllyard, S A, Sqmres, K C , Bauer, J W and Llndsay, P.H (1971) Evoked potentml correlates

ot auditory signal detection Scxence, 172, 1357-1360 Karhn. L (1970) Cogmtlon, preparation, and sensory-evoked potentmls Psychological Bulle-

tin, 73,122-136. Knott, J R and Peters, J F (1974). Changes m CNV amphtude with progresswe mductxon of

stress as a function of sex Electroencephalography and Chnlcal Neurophyslology, 36, 4 7 - 51

Naatanen, R (1967 Selective attention and evoked potentials. Annales Academme Sclentxarum l"ennlcae, 151,1-226

Paul, D D and Sutton, S (1972) Evoked potentml correlates of response criterion m auditory signal detection Science, 177,362-364.

Poon, L W, Thompson, L W, Wdhams, R.B. and Marsh, G R (1974) Changes of antero-pos- tenor distribution of CNV and late positive component as a function of Information proces- sing demands Psychophyslology, 11,660-673.

Rltter, W, Vaughan. H G and Costa, L D (1968) Orienting and habituation to auditory stlm- uh A study of short term changes m average evoked responses Electroencephalography and Clanlcal Neurophyslology, 25,550-556

28 R T WllAlnson, D 41 Sealcs / Evolted l)otentmls

Roth, W T (1973) Autidoty evoked responses to unpredictable Stllnuh Ps>choph)slolog), 10, 125-138

Siegel, S (1956) Nonparametrlc Statistics, New • ork McGraw-Hill Sqmres, K C, Hlllyard, S A and Llndsa}, P tl (1973) Vertex potentials evoked during audi-

tory signal detection Relation to decision critena Perception and Ps} choph~ SlCS, 4 , 2 6 5 - 272

Sqmres, K C, Squires, N K and Hlllyard, S A (1975a) Decision-related cortical potenuals during an auditor}' signal detection task x~lth cued observation intervals Journal ot Experi- mental Psychology Human Perception and Performance, 1 ,268-279

Squires, K C, Squires, N K and Hlllyard, S A (1975b) Vertex evoked potentials in a rating- scale detection task Relation to signal probabfllt~ Behavioral Blolog}, ] 3 21 34

Sutton, S, Braren, M Zubln, J , and John, L R (1965) Evoked potential correlates oI stimulus uncertainty Science, 150, 1187-1188

Tanner, W P and Swets, J A (1954) A deClslon-lnaklng theory of visual detection Ps)chologl- cal Review, 61 ,401-409

Tuetlng, P Event-related potentials, cognitive events, and inlormatlon processing In Otto, D (Ed) Proceedings of the Fourth International Congress on Fvent Related Slow Poten- tials of the Brain, US Government, (In press)

Tuetmg, P and Sutton, S (1973) The relationship between pre-stlmulus negative shifts and post-stimulus components of the averaged evoked potential In Kornblum, S (Ed I &tten- tlon and Performance IV, New York Academic Press

Vanghan, H G and Rltter, W (1970) The sources ot auditory evoked responses recorded trom the human scalp Electroencephalography and Chmcal Neurophyslology, 28,360 367

Walter, W G, Cooper, R , Aldrldge, V J , McCallum, W C and Winter, A L (1964) Contingent negative variation an electrical sign of sensorlmotor association and expectancy in the human brain Nature (London), 203,380 384

Wilkinson, R T (1976) Relationship between CNV, its resolution and the evoked response In McCallum, W C and Knott, J R (Eds), The Responsive Brain, Bristol John W'rlght and Sons

Wilkinson, R T and Asby, S A (1974) Selective attention, contingent negative variation and the evoked potential Biological Psychology, 1, 167-179

Wilkinson, R T and Lee, M V (1972) Auditory evoked potentials and selective attention Elec- troencephalography and Chnical Neurophyslology, 33, 411 418

Wilkinson, R T and Morlock, H C 11967) Auditor} evoked response and reaction rune Iclec- troencephalography and Cllmcal Neurophyslology, 23, 50-56

Wilkinson, R T and Spence, M T (1973l Deterlnmants of the post-stimulus resolution of con- tmgent negative variation ICNVI Flectroencephalography and Cllmcal Neurophs}lology, 35,503-509