journal of experimental psychology: 1990, vol. 16, …...through sony mdr m66 headphones. a room...

Journal of Experimental Psychology:Learning, Memory, and Cognition1990, Vol. 16, No. 5, 902-915

Copyright 1990 by the American Psychological Association, Inc.O278-7393/9O/$OO,75

Consciousness of Attention and Expectancy as Reflected inEvent-Related Potentials and Reaction Times

Werner Sommer, Juliana Matt, and Hartmut LeutholdUniversitat Konstanz

Konstanz, Federal Republic of Germany

The effects of conscious expectancies and attention on event-related potentials (ERP) and choicereaction times (RT) and their modulation by stimulus sequence were studied. Subjects retrospec-tively reported their expectancy of, and attention to, the terminal tones of short series. ERPs andRTs showed the usual sequential effects that were modulated by practice. As ratings were affectedby only a few of the stimulus sequence, conscious access to sequence-based expectancy orattention appears to be fragmentary. Increased P300 amplitude with attention indicates consciousaccess to processing capacity. RTs and P300 latencies suggest stimulus processing time to decreasewith sequence-based and consciously accessible expectancy. Differential effects of stimulussequences and conscious expectancies on P300 amplitude indicate influences of two varieties ofexpectancy.

Expectancy is an important concept both in theories of theP300 component in event-related potentials (ERP) and forthe explanation of the effects of stimulus sequences on reac-tion times (RT). The P300 is an electrically positive deflectionin the ERP at 300 or more milliseconds, elicited by task-relevant stimuli. A model of the variables controlling theamplitude of this component has been presented by Johnson(1988). According to his model,

P300 Amplitude =f[Tx(\/P+M)],

where T represents the amount of information transmitted bya stimulus, P reflects its subjective probability, and M corre-sponds to the meaning or significance of a stimulus.

Together with the prior probability of a stimulus class,expectancy determines the subjective probability variable inJohnson's model and has most often been studied in thecontext of sequential effects. In the first systematic study ofsequential effects on ERPs (Squires, Wickens, Squires, &Donchin, 1976), one of two tones of different pitch, presentedin random order (Bernoulli series), had to be counted. Theamplitude of the ERP elicited by a given stimulus was foundto be inversely proportional to expectancy, which was deter-mined according to the following model:

Portions of this research were conducted by Hartmut Leuthold inpartial fulfillment of the requirements for the diploma in psychology,Universitat Konstanz, and were presented at the International Con-ference on Event-Related Potentials of the Brain, Noordwijk, TheNetherlands, May 1989, and at the meeting of the Deutsche Gesell-schaft fur Psychophysiologie und Ihre Anwendungen, Wuppertal,Federal Republic of Germany, May 1989. This research was sup-ported by the Deutsche Forschungsgemeinschaft. We thank RobertB. Freeman, Jr., Robert M. Chapman, Albrecht Inhoff, and twoanonymous reviewers for their helpful comments on earlier drafts ofthe article.

Correspondence concerning this article should be addressed toWerner Sommer, Fachgruppe Psychologie, Universitat Konstanz,Postfach 5560, 7750 Konstanz, Federal Republic of Germany.

Expectancy = M + A + P + constant,

where P stands for the prior probability of the stimulus andM represents the exponentially decaying contribution of priorstimuli at specific sequential positions to the expectancy forthe presentation of a like stimulus. In other words, the morerecently and the more often a particular class of stimuli hasbeen presented, the stronger the expectancy for this kind ofstimulus to be presented at the next trial and the smaller theERP to this kind of stimulus if it is actually presented. Thisbasic expectancy for stimulus repetitions is modulated by anexpectancy for the continuation of alternation patterns (A) inthe stimulus sequence. Sequential effects in ERPs in accordwith this model have been reported many times for countingtasks (Duncan-Johnson & Donchin, 1977; Johnson & Don-chin, 1980, 1982; Squires, 1977; Petuchowski, Wickens, &Donchin 1977), prediction tasks (Chesney & Donchin, 1979;Munson, Ruchkin, Ritter, Sutton, & Squires, 1984; Tueting,Sutton, & Zubin, 1970) and choice reaction tasks (Duncan-Johnson, Roth, & Kopell, 1984; Ford, Duncan-Johnson, Pfef-ferbaum, & Kopell, 1982).

In choice reaction times, two kinds of sequential effects arecommonly observed when different stimuli are presented inBernoulli series and when response stimulus intervals (RSIs)are longer than 500 ms. First, the RT to a stimulus is shorterwhen it is preceded by a different stimulus than when it ispreceded by a like stimulus (Hale, 1967; Kirby, 1972, 1976;Moss, Engel, & Faberman, 1967; Soetens, Boer, & Hueting,1985; Williams, 1966; for a review, see Kirby, 1980). This so-called first-order alternation effect is attributed to subjects'tendency to expect an excess of stimulus alternations overrepetitions in random stimulus series. Note the contrast tothe notion of subjects expecting stimulus repetitions in themodel of Squires and coworkers. Occasional reports of first-order repetition effects at long RSIs are attributed to the useof few trials (Schvaneveldt & Chase, 1969) or to a complexrelationship between stimulus and response (Bertelson & Ren-kin, 1966; Entus & Bindra, 1970).

902

CONSCIOUSNESS OF ATTENTION AND EXPECTANCY 903

A second kind of sequential effect is observed after contin-ued runs of stimulus repetitions or stimulus alternations,where RT decreases with the length of that run. Conversely,RT increases with run length if the run is discontinued by thecurrent stimulus (Remington, 1969; Soetens et al., 1985).These so-called higher order sequential effects are attributedto the consequences of expecting the continuation of a run ofevents. In either case, a stimulus that matches the expectancyis considered to be processed faster than a stimulus that doesnot.

There are discrepant views regarding the expectancy-gen-erating processes as automatic or as controlled and whetheror not these expectancies are conscious (i.e., subjectivelyaccessible). In reaction time research with long RSIs, on theone hand, the buildup of sequence-based expectancies isconsidered a slow, controlled process (Kirby, 1980; Soetenset al., 1985), at the end of which there should be a consciouslyaccessible expectation concerning the forthcoming stimulus(Shiffrin & Schneider, 1977). In ERP research, on the otherhand, the expectancies underlying the sequence-related mod-ulation of P300 are usually considered to be based on auto-matic processes (Johnson, 1988), and there is a tendency toview these expectancies as largely unconscious (Johnson &Donchin, 1982; Karis, Chesney, & Donchin, 1983). Even thisview, however, is not entirely unitary. In a recent controversy(Donchin & Coles, 1988; Verleger, 1988), Verleger claimedthat expected rather than unexpected events elicit large P300amplitudes. After long runs of repetitions, for example, sub-jects are supposed to "await" an alternation, and when itoccurs, a large P300 is elicited because of the closure of theperceptual epoch. This awaiting of particular stimuli on thebasis of the stimulus sequence is explicitly understood as aconscious expectation.

When consciously accessible expectancies have been di-rectly assessed by means of predictions, the subjects usuallyhave predicted the repetition of the immediately precedingstimulus (Geller & Pitz, 1970; Hale, 1967; Jarvik, 1951;Munson et al., 1984; Schvaneveldt & Chase, 1969; but for anexception see Keele, 1969, Experiment 3). After runs ofstimulus repetitions, this so-called positive recency effect turnsinto a negative recency effect or "gambler's fallacy", that is,an increasing expectancy for a stimulus change (Hale, 1967;Jarvik, 1951). Particularly the gambler's fallacy is in clearcontradiction to the model of Squires et al. (1976), whichholds that expectancy increases with the length of a repetitionrun. Donchin (1981) and Donchin and Isreal (1980) havepointed to various heuristics that may interact with perceivedprobability and therefore bias subjects' predictions. They sug-gested that P300 amplitude may serve as a more reliableindicator of subjective probability than the overt predictionbehavior of the subjects.

None of the prior studies has attempted to integrate RTs,ERPs, and subjective expectancy measures in relation to acomplete mapping of stimulus sequences. In the presentstudy, we presented short Bernoulli series of stimuli that werejust long enough to induce sequential effects. Each tonerequired choice reactions, and at the end of each series self-reports about the expectedness of the terminal stimulus andabout the amount of attention payed to this stimulus were

requested. This allowed us to compare and evaluate, for eachstimulus sequence, subjects' subjective measures of expect-ancy and attention, their reaction times, and measures ofP300 amplitude and P300 latency.

Collecting self-reports after the stimulus in question repre-sents an alternative approach to the usual prediction measuresof expectancy. Such predictions (a) have been shown tointeract with the processing of the forthcoming stimulus(Geller, 1975; Naatanen & Merisalo, 1977), (b) possibly inter-fere with the generation of the expectancies, and (c) are subjectto heuristics and strategies of the subject (Kahneman & Tver-sky, 1974). Although retrospective measures may have dis-advantages of their own, there is no alternative when onewishes to control the interstimulus interval, which has beenshown to be of crucial importance for sequential effects(Kirby, 1980). By recording self-report measures not onlyacross but also within the various stimulus sequences, we wereable to assess both sequence-generated and sequence-inde-pendent expectancy effects and their interactions.

In addition to the expectancy ratings, self-reports of atten-tion were also requested. Ignored stimuli have very rarelyelicited a P300 (Duncan-Johnson & Donchin, 1977), andincreasing perceptual load on a primary task reduces the P300to target stimuli of a concurrent secondary task (Donchin,Kramer, & Wickens, 1986). Accordingly, it is of interestwhether P300 amplitude varies also as a function of self-reported attention. In Johnson's model of P300 amplitude,attention is one of the variables (T) determining the amountof information transmitted by the stimulus; correlations be-tween self-reports of attention and ERPs would thus indicateconscious access to this variable.

The concurrent recording of behavioral and ERP variablescan be particularly informative when these variables can bedissociated by some kind of experimental manipulation (Don-chin, 1984). The considerable practice provided by the exten-sive data collection in the present study represents such amanipulation. Practice has been found not only to diminishRTs in general (Leonard, 1958; Mowbray & Rhoades, 1959;see also Teichner & Krebs, 1974), but also to reduce theirmodulation by the preceding stimulus sequences (Soetens etal., 1985; Vervaeck& Boer, 1980); however, there are virtuallyno ERP data available on the relation of practice to sequentialeffects. It would be interesting to know whether the sequentialeffects observed in RTs and ERPs, which seem so neatlyassociated at first sight, would remain associated over ex-tended practice or become dissociated. Information about therelative practice effects might bear on the question of wherein the information flow sequence-based expectancies act, andhow closely coupled the sequential effects in ERPs and RTsactually are.

Method

Subjects

Complete sets of data were obtained from six women and fourmen of mean age 25.2 and 24.8 years, respectively. Two subjects hadto be excluded after Session 1 because of excessive errors (approxi-mately 60% each) or electrooculogram (EOG) artifacts. The 10 re-

904 W. SOMMER, J. MATT, AND H. LEUTHOLD

maining subjects were strongly right-handed, with laterality quotientsgreater than +66 (Oldfield, 1971). In order to maximize practiceeffects, we included only subjects who did not have prior experiencein RT or ERP studies.

Stimuli and Apparatus

Auditory stimuli were two equiprobable sinusoidal tones of 1,000and 1,500 Hz, both of 78.0 dB(A) intensity and 65-ms duration,including 8-ms rise/fall times, presented binaurally in random orderthrough Sony MDR M66 headphones. A room ventilator provided amasking noise of 42 dB(A). The interstimulus interval was 1 s. Duringthe presentation of the tone series a white fixation spot 26 mm indiameter was constantly presented on the monitor of a CommodoreAmiga 2000 microcomputer at a distance of 1 m. Two telegraph keysmounted 20 cm apart on a horizontal response panel in front of thesubject were used for recording RTs to the tones. The index fingersof the left and right hand operated the left and right key, respectively.The subjects were instructed to press one of the keys in response tothe low tone and the other key to the high tone. The relation betweentone pitch and response key was counterbalanced across subjects.

Procedure

The tone series was frequently interrupted after a number of tonesto obtain subjects' self-reports. Interruptions occurred after no lessthan five to seven tones (randomly determined), only after at leastfive consecutive correct reactions, and only if the ERP to the terminaltone had been free of ocular artifacts. A reaction was accepted ascorrect if the proper key had been pressed and if the RT was between100 and 800 ms. ERPs were accepted as- artifact free if the EOGexcursions during the recording epoch were less than 50 jiV.

One second after a tone that met these criteria had been registered,self-reports were prompted by the simultaneous appearance of threevertical columns on the Amiga monitor. Self-reports were made witha mouse-driven cursor by selecting one of the columns. The columnswere equally spaced horizontally, and each was approximately onefifth of the screen's width; the height of the left column was one thirdthat of the screen's height, the middle column was two thirds of thescreen's height, and the height of the right column was equal to thatof the screen. The kind of self-report the subject was to make wasindicated by the color of the columns. If they were blue, their heightscorresponded to increasing degrees of attention, which the subjectshad been able to pay to the terminal tone preceding the interruption.If the columns were red, their heights corresponded to differentexpectancies concerning this terminal tone. The large column was tobe chosen when the terminal tone had been expected; the small onewhen the alternative, not presented tone had been expected; and theintermediate column when neither tone had been expected. Thecolumns remained on the screen until the rating had been completed;a minimum time of 2 s was used to discourage hasty reporting.Immediately upon completion of the first rating, the second promptwas presented. The order of attention and expectancy prompts wasaltered. After each report the cursor was homed to the upper left-hand corner of the screen, and the first tone of the next series waspresented 1 s after the second report.

Subjects were instructed in writing as to the experimental proce-dure and advised to avoid muscular and eye movements or blinksduring the experiment proper. In addition to a base pay of DM 7.50per hour, correct reactions and short RTs were rewarded with 0.50DM for each percentage point below a 10% error rate and for eachunit of 10 ms below an average RT of 320 ms.

Each subject was tested in four sessions, 2 to 3 days apart, usuallyat the same time of day. In each session 416 tone series of variable,

subject-controlled length were presented, with short breaks after eachblock of 104 series. The first 16 series of a session were consideredwarm-up trials and discarded from data analysis. The average numberof tones presented were, in order of session, 3,264 (range: 2,762-4,055), 3,159 (2,746-3,715), 3,229 (2,800-3,992), and 3,566 (2,771-7,162). The increase of the range during Session 4 was caused by atechnical problem for one subject that increased the number of tonesin the series but did not affect the data recorded to the terminal tonebefore the interruption. Without that subject the average number oftrials in Session 4 would have been 3,166 (2,771-3,952). The averagelength of the tone series preceding the self-reports was, in order ofsession, 7.9, 7.6, 7.8, and 8.6 tones (7.7 when the subject with thetechnical problem is disregarded).

Recording and Data Analysis

For the purposes of data analysis only the RTs and ERPs to theterminal tones were considered, to which the self-reports also refer.In order to be able to distinguish between the two types of self-reports,as well as between different stimulus sequences of maximal length,the data were collapsed over tone pitch. The pitch of a particular tonewas considered only inasmuch as it constituted a repetition (R) or analternation (A) of the pitch of the preceding tone. This allowed forthe differentiation of 16 stimulus sequences, defined by the terminaltone and all possible combinations of the four preceding tones.Collapsing over tone pitch seemed justified also by the fact that pitchdid not affect RT, P300 area, or P300 latency (see Table 1).

For the analysis of the effects of the tone sequences on the depend-ent variables, we consistently used the model of Soetens et al. (1985).The 16 sequences were subdivided into those for which the terminaltone before the self-reports represented either a repetition or analternation of the second-to-last tone (first-order sequence). The 8higher order sequences for each first-order sequence were orderedfrom three consecutive stimulus repetitions (RRR) to three alterna-tions (AAA; compare also Figure 1). When calculating linear regres-sions over these higher order sequences in the analyses of variance(ANOVAS) we do not wish to imply equal spacing between thesesequences; rather, we are using the linear regressions merely to checkfor the existence of monotonic trends over the higher order sequencesand to compare their direction and strength over different conditions.

The electroencephalogram (EEG) from Fz, Cz, and Pz, referencedto linked earlobes, and the vertical EOG were recorded with GrassE5SH Ag/AgCl electrodes and Beckman electrode electrolyte paste.The EEG and EOG signals were amplified with time constants of 5.5s and low-pass filters set to 40 Hz (-3 dB attenuation, 12 dB rolloff/octave). All signals were sampled at 100 Hz for 900 ms, starting 90ms prior to stimulus onset, and stored on magnetic tape.

The averaged ERPs were digitally low-pass filtered at 4.9 and 8.8Hz (-3 dB; Ruchkin & Glaser, 1978) for the rating and the practiceanalysis, respectively. Then the average activity during a 90-ms pre-stimulus baseline was subtracted from each waveshape and an areameasure of P300 (270 to 370 ms) was taken. These voltage integralsover time were divided by the number of time points involved,resulting in a measure representing the average amplitude during the

Table 1Effects of Tone Pitch

Tonepitch(Hz)

10001500

Reactiontime(ms)

270.4263.8

P300area(MV)

8.28.1

P300latency

(ms)

302300


component interval in ^V. P300 latency was measured following theprocedure of Pfefferbaum, Ford, Johnson, Wenegrat, and Kopell(1983). A negative (electrical positivity corresponds to numericalnegativity in our data) 2 Hz sinus half-wave template was alignedwith the averaged ERP waveshapes from the Pz electrode. The centerof the template was moved across the ERP from 270 ms after stimulusonset to 370 ms in steps of 10 ms, and cross-products betweentemplate and ERP were calculated at each lag. At the lag of themaximum cross-product, P300 latency was taken as the time pointin the ERP corresponding to the center of the template.

All dependent variables were subjected to repeated measuresANOVAS with the following factors: first-order stimulus sequence (R,A) and higher order sequence (RRR, ARR, RAR, AAR, RRA, ARA,RAA, AAA). The electrode sites (Fz, Cz, Pz) (for the P300 amplitudemeasures) and the sessions (1 to 4) (for the practice-related analysis)were also included as repeated measures factors in the ANOVA. If anymain effect or interaction involving the higher order sequence wassignificant, the linear trends across the levels of this factor were alsoconsidered. In order to correct for possible violations of sphericityassumptions, conservative (Huynh-Feldt) F tests were used through-out, with degrees of freedom before and p values after the adjustmentreported. For a posteriori comparisons, the levels of significance werecorrected according to the Bonferoni criterion.

Sequential Effects in ERPs and RTs and TheirModulation by Practice

Results

Reaction times. Throughout all four sessions RTs wereabout 10 ms longer after first-order stimulus repetitions thanafter first-order alternations (see Table 2), F\ 1, 9) = 5.73, MSC

= 3,129.90, p < .05. If the terminal tone discontinued runsof higher order repetitions or alternations, RT was delayedin proportion to the length of that run (Figure 1),F(7, 63) = 35.47, * = 0.50, MS, = 989.04, p < .001. This delaydiminished over the session, F(21, 189) = 4.06, t = .62,MSC = 201.77, p < .001, in a way that was not specific toonly a few isolated sequences but held true also for the lineartrends over the higher order sequences, F(linear: 1,9) = 21.43,MSC = 5,632.48, p<.0\.

Event-related potentials. Figure 2 depicts sequential effectson the waveshapes of the averaged ERPs from the Pz elec-trode. In Figure 3 first-order repetitions and alternations arehighlighted by collapsing over higher order sequences, andthe scalp topography is visualized by depicting all three leads.Like RT, the P300 latency (Figure 4) was increased when runsof repetitions or alternations were discontinued by the ter-minal stimulus, F(7, 63) = 16.9, t = .75, MS, = 8.85, p <.001, and /^linear: 1, 9) = 43.0, MS* = 154.22, p < .001.Although there was an overall shortening of P300 latencyfrom Session 1 to Session 4 (Ms = 323, 309, 305, and 302ms, respectively), F(3, 27) = 7.9, e = 1.28, MS, = 17.81, p <.001), this shortening was independent of the stimulus se-quence. P300 amplitude showed a centroparietal scalp topog-raphy, Aft (Fz to Pz) = 4.5,10.3, and 9.3 tiV, F{2, 18) = 20.6,t = .91, MSe = 29,176.25, p < .001. Similar to the RT, itincreased with the length of repetition and alternation runs,if discontinued by the terminal stimulus (Figure 5), F(7, 63)= 17.7, f = 0.66, MS, = 2,668.64, p < .001, and ^linear: 1,9)= 30.93, MSC = 28,770.15, p< .001.

Table 2Effects of First-Order Stimulus Sequence

Session

1234

Reactiontime (ms)

R

280272274264

A

273258260255

P300latency (ms)

R

324313309301

A

323305301302

P300 area(n

R

6.27.88.68.6

V)

A

8.28.08.78.1

Note. R = stimulus repetition; A = stimulus alternation.

Practice affected the differences in P300 amplitude betweenfirst-order repetitions and alternations, F(3, 27) = 5.74,€=1.13, MS, = 2,233.12, p<. 01. During Session 1 the P300to first-order repetitions was smaller than the P300 to alter-nations (Table 2), F(l, 19) = 7.23, MS, = 1,856.61, p < .05,but in the course of the remaining sessions the first-ordereffects gradually vanished, F(2, 18) = 1.04, e = .92, MS, =

Reaction Times (ms)

330-1

210-1

R A R A R A R A R A R A R A R AR R A A R R A A R R A A R R A AR R R R A A A A R R R R A A A AR R R R R R R R A A A A A A A A

Session 1 2 •— 3 A— 4 4—

Figure 1. Sequential effects on choice reactions time (RT) of thefour preceding stimuli in a random series of two equiprobable tonesof different pitch. (RTs are collapsed over pitch; only repetitions (R)and alternations (A) are plotted. RTs to terminal tones are plottedseparately for first-order repetitions [R, left side] and alternations [A,right side] of the preceding tone. The eight higher order sequenceswithin each first-order sequence are ordered from three repetitions[RRR] to three alternations [AAA]. The RTs are superimposed forSessions 1 to 4.)


Session 1 — 2 — 3 4 - - -

Figure 2. Sequential effects on the ERPs from the Pz electrode,superimposed for Sessions 1 to 4. (First-order repetitions and alter-nations are shown on the left and right, respectively. The eight higherorder sequences within each first-order sequence are ordered fromtop [RRR] to bottom [AAA]. Tick marks denote intervals of 100ms.)

1,937.93, p = .37. In contrast to the RTs, the effects of thehigher order sequences on P300 amplitude or their interactionwith the first-order sequences were stable over sessions, F(21,189) = .98,f = .64, MSe = 1,245.37, p = .48, and F(2\, 189)= 1.27, f = .92,MS;= 1,009.37, p =.21.

Discussion

We will discuss higher order and first-order sequentialeffects in turn, first for Session 1 and then for the effects ofpractice in the course of the remaining four sessions. Anoverview of the findings is presented in Table 3.

Higher order effects. The higher order effects observedduring Session 1 conform to those reported by others andhave usually been interpreted to indicate that subjects expectruns of stimulus repetitions or stimulus alternations to con-tinue. When those runs are actually discontinued, prolongedRTs (e.g., Remington, 1969; Soetens et al., 1985), increasedP300 amplitudes (Chesney & Donchin, 1979; Duncan-John-son & Donchin, 1977, 1982; Duncan-Johnson et al., 1984;

Johnson & Donchin, 1980, 1982;Munsonetal., 1984; Squireset al., 1976, 1977), and increased P300 latencies (Duncan-Johnson et al., 1984; Ford, Pfefferbaum, & Kopell, 1982)have been considered as indicative of surprise.

As to the consequences of practice for the higher ordersequential effects on RTs, the study most directly comparableto ours is Experiment 3 of Soetens et al. (1985). At RSIs of500 ms the RTs initially showed the usual higher order effectsand a first-order alternation effect. With practice, both first-order and higher order effects diminished. Although they usedvery short RSIs (160 ms), Vervaeck and Boer (1980) foundexpectancy effects after long runs of alternations, facilitatingwhen continued and inhibiting when discontinued. After1,000 trials of practice this expectancy effect disappeared.Both Soetens et al. and Vervaeck and Boer interpreted thispractice-dependent reduction of sequential effects as a con-sequence of a global decrement of expectancies.

Although practice in the present study decreased the higherorder effects for RTs, it did not affect those of P300 amplitudeor latency. These data do not conform with the hypothesis ofa global decrement of the expectancies generated by thepreceding sequence, because the costs of expectancy did notdecrease (shortening of the RTs to discontinued stimulusruns) and the benefits of expectancy in continued runs didnot diminish. These sequences continued to produce theshortest RTs throughout all four sessions. The robustness ofthe expectancies against practice is also indicated by thestability of the higher order sequential effects in P300 ampli-tude.

The locus of the practice effects on the RTs may be broughtinto better focus by considering P300 latency. P300 latency isusually considered to reflect the time it takes to evaluate andcategorize an event and has been shown to be largely inde-pendent of response selection and execution processes. Whenspeed rather than accuracy is emphasized, the response mayprecede the P300; in this case, it is argued, the response isbased on only a partial analysis of the stimulus, whereas P300does not occur before the stimulus is completely evaluated(Donchin, Karis, Bashore, Coles, & Gratton, 1986, Kutas,McCarthy, & Donchin, 1977; McCarthy & Donchin, 1981).The existence of higher order effects on P300 latency dem-onstrates that stimulus evaluation and categorization proc-esses are influenced by expectancy. The stability of theseeffects over sessions indicates that this influence does notchange with practice. In addition to the sequential effects onstimulus evaluation, there seems to be an additional effect onprocesses following stimulus evaluation (e.g., response selec-tion or execution processes) because, first, the sequentialeffects in RTs are considerably stronger than in P300 latency(cf. Figures 1 and 4) and, second, only the sequential effectsin the RTs are affected by practice. Therefore, sequence-basedexpectancies seem to act upon at least two stages of theinformation flow, and only the expectancy effects on the laterstages are diminished by practice, whereas those on earlystages are stable. On the basis of the present data it cannot bedecided, however, whether we are dealing with one expect-ancy-generating process, whose effects are selectively dimin-ished, or whether there are two such processes, one that actson stimulus evaluation and categorization and is robust


Frontal Central Parietal

Stimulus Repetition — Stimulus Alternation ---

Figure 3. ERPs superimposed for first-order stimulus repetitions (solid lines) and alternations (brokenlines) for Sessions 1 to 4 and for all three electrode sites.

against practice and one that acts on later processing stagesand is susceptible to practice.

It should be noted here that there was a marked decreaseof P300 latency over sessions independent of any stimulussequences, which was accompanied by a similar, albeit insig-nificant, decrease of RT. This indicates that the sequence-independent reaction time shortening with practice might beexplained by improvements in stimulus evaluation and cate-gorization processes.

First-order effects. The first-order effects for P300 ampli-tude and RT during Session 1 agree well with the literature,but they are dissociated from each other. The RTs show afirst-order alternation effect (i.e., shorter RTs to stimulusalternations than to stimulus repetitions). The alternationeffect is consistent with most RT studies using RSIs of morethan 500 ms. Contrary to the alternation effect in the RTsbut in accord with virtually all ERP studies (Chesney &Donchin, 1979; Duncan-Johnson & Donchin, 1977, 1982;Duncan-Johnson et al., 1984; Johnson & Donchin, 1980,1982; Munson et al., 1984; Squires et al., 1976, 1977), P300amplitudes are smaller after a first-order repetition, that is, afirst-order repetition effect.

One might seek the explanation for the conflicting first-order effects in procedural differences of the studies; thepresent study, however, shows a "first-order dissociation"between RTs and ERPs for the same data set. On the otherhand, the two previous studies that have simultaneously as-sessed sequential effects in both RTs and ERPs (Ford, Dun-can-Johnson, Pfefferbaum, & Kopell, 1982; Duncan-Johnsonet al., 1984) showed conforming higher order as well as first-order (repetition) effects for both variables.

The discrepancy between the first-order effects in RTs andERPs is further emphasized by the effect of practice, which,in the present experiment, was restricted to the ERPs. Al-though practice tended to diminish the discrepancy betweenthe first-order effects of RTs and P300 amplitude after thefirst session, it did so by increasing the P300 to repetitionsand thus abolishing the well-documented first-order repetitioneffect in P300.

As the higher order sequential effects both for RTs andP300 amplitudes are in accord with each other, both effectscan be explained by expectancy. For the conflicting first-ordereffects, however, an alternative or at least additional expla-nation apart from expectancy must be found either for theRTs or the P300 amplitudes, because subjects cannot beconsidered to expect stimuli both to repeat and alternate. Oneclue may be provided by our rating data (see below). Althoughthe relationship between the expectancy ratings and the stim-ulus sequences was generally weak, subjects (insignificantly)tended to rate stimulus alternations as more expected thanrepetitions. Therefore, we are inclined to assume a process inaddition to expectancy for the first-order effects in P300amplitude rather than for the RTs. A process is needed thatacts specifically on repetitions and causes P300 to increasewith practice. Thus, one should not assume, for example, thatstimulus changes by themselves exert a P300-enhancing andpractice-dependent influence, because it is the influence ofstimulus repetitions that was altered with practice.

We tentatively suggest that for unpracticed subjects theremay be more postresponse uncertainty as to the actual pitchof a tone if it has not been preceded by a different tone towhich it can be contrasted. Similar to the reduction of P300


amplitude in conditions of low discriminability and low de-cision confidence (Johnson, 1988), the relatively greaterequivocation about stimulus identity after first-order repeti-tions might reduce P300 relative to first-order alternations.The increase of P300 amplitude to repeated stimuli withpractice might then be related to the improvement of decisionconfidence. Similar increases of P300 with practice have beenshown in learning experiments where P300 increased withdecreasing equivocation about the correct response (Johnson,Pfefferbaum, & Kopell, 1985; Peters, Billinger, & Knott,1977).

A conclusion of importance for the data yet to be describedis that the experimental procedure used here produces se-quential effects for RTs and ERPs that are very similar intheir direction to those reported for continuous stimulus series(e.g., Soetens et al., 1985; Squires et al., 1976). Discrepanciesemerged only in the context of practice and only for the ERPswhere practice has not been studied before. Also, the rathershort RTs and P300 latencies demonstrate that we havebasically been able to retain the spontaneous nature of thetask. Thus, we consider it likely that our findings can begeneralized to other experimental situations generating se-quential effects.

P300 Latency (ms)

360-

350-

340-

330-

320-

310-

300-

290-

280-

P300 Amplitude

i—i—i—i—i—i—r


Session 1 2 - ~ 3 * - - 4 * - -

Figure 4. Sequential effects on P300 latency, superimposed forSessions 1 to 4, measured at Pz as the time of maximum crosscorre-lation of a 2 Hz sinus halfwave template between 270 and 370 msafter stimulus onset.

16-

15-

14-

13-

12-

11-

10-

9-

8-

7-

6-

5-


Session 1 2 »-- 3 * - - 4 «

Figure 5. Sequential effects on P300 amplitude, superimposed forSessions 1 to 4, measured at'Pz as the average amplitude between270 and 370 ms after stimulus onset, related to a 70-ms prestimulusbaseline.

Sequential Dependencies in Self-ReportedExpectancies and Attention

Results

In order to check for dependencies between the expectancyand attention ratings, a contingency table for the total fre-quencies of both ratings (over subjects, sessions, and se-quences) was calculated (Table 4). Although the distributionof frequencies over cells was highly unequal (x2 = 254.8, p <.001), the "P'-coefficient of .089 indicated that the associationbetween the attention and expectancy ratings was negligible.

The expectancy ratings from each stimulus sequence andsession were analyzed in two ways. First, the ratio of nonneu-tral expectancy ratings (when either the actually presentedterminal tone P or the alternative tone nP had been expected)to all ratings given (including the cases when no particularexpectancy had been held) was calculated. Second, a score forthe relative expectedness of the terminal tone was calculatedby (P - nP)l(P + nP). The self-reports of attention wereanalyzed by scoring the ratings of increasing attention from 1to 3 and averaging them for each sequence and session.


Table 3Overview of Global Practice-Dependent Changes from Sessions 1 to 4, First-Order andHigher Order Sequential Effects in Session 1, and Their Changes With Practice

Variable

Reaction timeP300 latencyP300 amplitude

Globalpracticeeffect

none

r*none

First-order effect

Session 1

R > A *noneR < A *

Practiceeffect

nonenoneR " (R = A)

Higher order effect

Session 1•a*#*• ***

Practiceeffect

less t***nonenone

Note. R = stimulus repetition; A = stimulus alternation." f, I = increase and decrease of variable or effect.*p<.05. • • p < . 0 1 . ***/>< .001.

As shown in Figure 6, the stimulus sequences similarlyaffected both the self-reports of attention and the subjects'inclination to give nonneutral expectancies. With an increas-ing number of repetitions in the sequences, more nonneutralexpectancy statements and higher attention ratings weregiven. This is reflected in main effects of the higher ordersequence for the nonneutral expectancy statements and theattention ratings, i=s(7, 63) = 9.24 and 5.05, es = .44 and .65,MS'eS = .31 and = . 12, ps < .001 and .01, respectively, and inthe linear trends over the higher order sequences: /1(linear: 1,9) = 9.95, A/51. = .47, p < .05, and f\\, 9) = 5.22, MSC = .18,p < .05, respectively. There were three sequences where theratio of nonneutral expectancies was further increased—con-tinued and interrupted runs of four repetitions (RRRR andRRRA) and continued runs of four alternations (AAAA).This was reflected in an interaction of the first-order and thehigher order sequences, F(l, 63) = 4.91, t = .51, MSC = .04,p < .01. A similar tendency for the attention ratings (Figure6) failed to reach significance, F\7, 63) = 1.51. The overallinclination of subjects to give nonneutral expectancy ratingsdecreased from M = .55 in Session 1 over .48 (Session 2) and.40 (Session 3) to .39 in Session 4, F(3, 27) = 5.80, t = .60,MSt = .15, p < .05.

Only 2 of the 16 stimulus sequences affected the relativeexpectedness of the terminal tone. When runs of three alter-nations where continued (AAAA), expectedness was M = .55for the terminal tone. When such runs where discontinued(AAAR), however, the alternative, unpresented tone was rel-atively expected (M = —.25; Figure 6): first-order and higherorder interaction, F(7, 63) = 3.32, t = .43, A/5*. = .51, p <.05, and contrast between AAAR and AAAA, F(l, 9) = 13.93,A/51. = 6.04, p < .01. There was no effect of the other higherorder sequences on the relative expectedness of the stimuli.

Table 4Contingencies for Self-Reports of Expectancy and Attention

Expected stimulus

PresentedNoneNot presented

1

4251,531

689

Attention rating

2

2,0984,4731,806

3

1,5162,643

744

Similarly, there were no reliable effects of the first-ordersequences on either of the two expectancy measures. The ratioof nonneutral expectancies was M = .49 after first-orderrepetitions and M = .42 after first-order alternations. Therelative expectedness of the terminal stimulus was only M =.02 after first-order repetitions but M = .20 after first-orderalternations. However, for both expectancy measures the first-order effects were not significant, F{\, 9) = 4.11, MSC = .19,p = .07, and F( 1, 9) = 2.86, MSC = 1.94, p = . 12, respectively.

In addition, the gambler's fallacy (Jarvik, 1951) was testedby analyzing the changes of the relative expectedness of theterminal stimulus with increasing numbers of preceding stim-ulus repetitions. The mean relative expectedness of the ter-minal stimulus, averaged over sessions, was .20 after first-order stimulus alternations (RRRA, ARRA, RARA, AARA,RRAA, ARAA, RAAA, and AAAA), .03 after one stimulusrepetition (RRAR, ARAR, RAAR, and AAAR), and .04,- .01 , and -.04 after two (RARR and AARR), three (ARRR),and four repetitions (RRRR), respectively. Numerically, thisconforms with the notion that increasing runs of repetitionslead to an increase of expectancy for an alternation, butstatistical significance failed by a large margin (F < 1).

Discussion

Systematic effects of the stimulus sequences comparable tothose for RTs and ERPs were lacking for the self-reports ofexpectancy and attention. If the sequential effects on RTs orERPs were related to or caused by systematic and consciouslyaccessible variations in the kind of stimulus expected, sequen-tial effects similar to those in RTs or ERPs should be observedin reports of the relative expectedness of the terminal tone aswell. But this is evidently not the case, as demonstrated bythe absence of any first-order effects, by the lack of a lineartrend over the higher order sequences, and by the absence ofa gambler's fallacy.

Distinct expectancies are exclusively confined to continuedand discontinued runs of three alternations (AAAA, AAAR).This conforms with the findings of Munson et al. (1984) that,after runs of alternations, another alternation was usuallypredicted. On the other hand, Munson et al. found an overalltendency for their subjects to predict the repetition of theprior stimulus, whereas for our subjects first-order alternationswere somewhat, but insignificantly, more expected than first-


Self Reports

oa:

•ua.x

zI

0.6-

x-0.2-

& •

-2.1

h2.0

-1.9

L1.8i i r


Average Attention Rating *•Ratio of Non—Neutral Expectancies •Relative Expectedness of Stimulus •

Figure 6. Sequential effects on self-reports: the amount of attentionpaid to the terminal stimulus in a tone series; the ratio of nonneutralexpectancies (expecting either the presented stimulus P or its alter-native nP) to all ratings given, including neutral expectancies; and ameasure of the relative expectedness of the presented stimulus(P - nP)/(P + nP).

order repetitions. Possibly, predictions of events and retro-spective expectancy lead to different estimates of subjectiveexpectancy.

Sequence-related increases in the self-reports of attentionand in the ratio of nonneutral expectancies were observedonly after long runs of repetitions and alternations and afterthe discontinuation of repetition runs. Possibly, regularitieswithin random series or the discontinuation of such regulari-ties have an alerting effect on the subjects. But again, theseeffects cannot be directly related to the sequential effects inRTs and ERPs. Two of the three sequences that elicited mostattention and nonneutral expectancies, namely RRRR andAAAA, are related to short RTs and P300 latencies and smallP300 amplitudes. The third sequence, AAAR, is related tolong RTs and P300 latencies and large P300 amplitudes.Additionally, two of the stimulus sequences, related to ex-treme differences in RTs and ERPs and usually considered torepresent extremes of expectedness and unexpectedness,namely RRRR and RRRA (e.g., Soetens et al., 1985; Squireset al., 1976), have no particular impact on relative expected-ness or attention.

In sum, there are stimulus sequences that do elicit height-ened attention and readiness for nonneutral expectancy rat-ings, and there are also sequences that are related to particulardirections of expectancy, but these findings cannot be directlyrelated to the systematic sequential effects in RTs and ERPs.Thus, these data agree with the notion that sequence-gener-ated expectancies do not have to be conscious (Johnson,1988).

Self-Report-Related Effects on RTs and ERPs

Results

The frequency distributions of the attention and expectancyratings were inevitably under the control of the subjects.Therefore, data frequencies were low for a number of cells,even after collapsing over sessions. For the attention-relateddata, 141 of the 480 cells (3 levels of ratings, 16 stimulussequences, and 10 subjects) contained less than 10 ERPs orRTs and 73 contained only 5 or even fewer. For the expect-ancy-related data the respective figures are 75 and 44 cells,and 4 cells were empty. The empty cells were replaced by themean RTs and ERP measures of the other subjects for thiscondition after digitally low-pass filtering the originally re-corded ERPs at 4.9 Hz to improve the signal-to-noise ratio.Because the main effects and the interaction of first-order andhigher order sequence, as well as the electrode site of theANOVAS, merely restate results reported above, only effects ofthe rating categories and their interactions with the sequentialeffects or electrode sites will be reported below.

Attention ratings. There were no effects of increasing at-tention ratings on P300 latency, but the RTs tended todecrease about 10 ms (Figure 7), F(2, 18) = 3.32, t = .83,MS, = 1,433.69, p = .07, and the P300 amplitude showed ahighly significant increase of 1.5 ^V (Figures 7, 8, and 9), F(2,18) = 10.42, f = 1.10 MSC = 3,115.01, p < .001. All theseattention-related results were independent of the stimulussequences.

Expectancy ratings. The analyses according to the threelevels of the expectancy ratings revealed RTs that were shorterby about 25 ms to expected than to unexpected stimuli (Figure10), F(2, 18) = 14.14, e = .91, MS, = 1,440.37, p < .001.Similarly, P300 latency was shorter to expected than to un-expected stimuli by about 20 ms (Figures 10, 11, and 12),F{2, 18) = 7.17, MSe = 19.57, t = .75, p < .01. Again, theseeffects were independent of the stimulus sequence.

In contrast, P300 amplitude showed an interaction betweenself-reported expectancy and the first-order stimulus sequence(Figures 10, 11, and 12), F(2, 18) = 6.18, e = .70, MSC =2,337.02, p < .05. After first-order stimulus repetitions, P300was insignificantly larger by .5 n\ when the terminal toneactually presented, rather than its alternative, had been ex-pected: F(l, 9) = .72. After stimulus alternations, however,P300 was larger by 1.7 fiV when the terminal tone rather thanits alternative had been expected, F(\, 9) = 8.19, MSC =203,937.59, p < .05.


UV

8,0-

7.5-

7,0-

6,5

6,0

ms

315-i

310

305

275

270

265 J

1 2

Attention Rating

P300 Amplitude # —

1 2 3

Attention Rating

P300 Latency • — Reaction Time •

Figure 7. Effects of self-reported attention on P300 amplitude (left panel), reaction times, and P300latency (right panel).

RAM\ ^ ^

Attent ion Rating 1 — 2 - - - 3 —

Figure 8. ERP waveshapes at Pz according to the attention ratingsand all 16 stimulus sequences (cf. Figure 2).

Discussion

The relatedness of expectancy and attention reports todifferences in ERPs confirms and extends our earlier findingsof subjective awareness of ERP differences (Sommer & Matt,in press). The dissociation of the expectancy effects on P300amplitude and RTs makes it unlikely that one effect can bereduced to the other. Thus, the expectancy effects on P300cannot be attributed to the differential contributions of keypress-related motor potentials or to subjects basing expect-ancy reports on the discrimination of their own reactionspeed. A similar argument pertains to a potential transferbetween the attention and the expectancy ratings, which isunlikely because of the differential effects of these ratings onP300 amplitude. Inequality of cell frequencies is also not alikely source of the reported effects. Although cell frequencieswere sometimes quite low at the highest order of factorcombinations, as mentioned above, there were at least 40 andgenerally more observations per cell and subject even for themost complex effects discussed here. Furthermore, the out-most rating categories, low and high attention, and the twononneutral expectancies were about equal in cell frequencydistribution.

Additionally, it cannot be argued that the subjects' discrim-ination of ERPs and RTs by attention and expectancy ratingsmay be based on the sensory discrimination of the differentstimulus sequences, for instance, by rating the presentedterminal tone or its alternative as expected after continued ordiscontinued alternation runs, respectively. First, as reportedabove, the sequential effects on the expectancy ratings wereweak. Second, even if subjects did base some of their ratingson the preceding stimulus sequences, the ERP and RT differ-


R

Attention Rating 1 — 2 — 3 —

Figure 9. ERP waveshapes at all three electrode sites and the aver-aged electrooculogram according to the attention ratings and for first-order repetitions (left panel) and alternations (right panel).

ences between the various ratings were independent of suchstrategies, because the differences are present within eachsequence category (cf. Figure 11). This holds true also for theinteraction of the first-order sequence and the expectancyratings, which cannot be reduced to the discrimination ofstimulus sequences.

The positive relationships between self-reported attentionand P300 amplitude conform with findings from dual-taskstudies (Donchin, Kramer, & Wickens, 1986), indicating thatP300 amplitude is related to the amount of processing re-sources allocated to a stimulus. According to our results, these

variations of resource allocation may actually be introspec-tively accessible.

The expectancy-related findings in RTs and for the P300latencies also conform well with the literature. The sequentialeffects in RTs for long RSIs are usually attributed to theexpectedness or unexpectedness of the stimuli, diminishingor increasing the RTs (Kirby, 1980; Laming, 1969). Thus, itappears plausible, as well, that stimuli that are explicitlydefined by subjects as unexpected are related to increasedRTs, as seen in our data. These findings also conform to thoseof Experiment 3 of Williams (1966), where subjects had topredict which of two alternative stimuli would be presentednext. After correct predictions, the RTs to the presentedstimulus were shorter than after false predictions. Like in thepresent study, this effect was independent of the status of thestimulus in question as first-order repetition or alternation.The increase of P300 latencies after unexpected as comparedwith expected stimuli may reflect increased processing timesfor unexpected events (Squires, Hillyard, & Lindsay, 1973).

Much less easily accounted for are the findings for the P300amplitude, which is somewhat (insignificantly) larger afterunexpected as compared with expected stimuli for first-orderstimulus repetitions but is markedly increased for expectedevents after stimulus alternations. This latter finding contra-dicts the common notion that unexpectedness is related tolarge P300 amplitudes. The evidence supporting this idea,reviewed, for example, by Johnson (1988), includes the se-quential effects on P300 and the increase of P300 amplitudewith the rareness of a stimulus.

Only a few studies, however, have directly related P300amplitude to self-reports descriptive of expectancies aboutfuture events. Horst, Johnson, and Donchin (1980) used apaired associate learning task with nonsense syllables where,in addition to responding with an associate to the first syllable,

MV

9,0-1

8,5-

8,0-

7,5

7,0 -1

nP None P nP None P

Expected Stimulus

P300-Amplitude • —

ms336-

326-

316

306

296

286

276

266

266

nP None P nP None P

Expected Stimulus

P300 Latency • — Reaction Time •—

Figure 10. Effects of self-reported expectancy and the first-order stimulus sequence on P30Q amplitude(left panel), reaction times, and P300 latency (right panel).


subjects had to rate their confidence about the correctness ofthe response. The amplitudes of the P300 elicited by thesubsequently presented "response" syllable varied inverselywith the subjective probability of the syllable. That is, largeP300 amplitudes appeared when subjects were right but hadexpected to be wrong and when they were wrong but hadexpected to be right, whereas the confirmation of expectancieselicited small P300s.

Chesney and Donchin (1979; reported in more detail inDonchin, 1979, and Donchin & Isreal, 1980) had subjectspredict the next stimulus in a visual Bernoulli series. TheERPs were averaged according to the first-order stimulussequence, the prediction made, and whether the predictionwas confirmed or disconfirmed (prediction outcome). Theresults for the P300 amplitude can be interpreted in two ways:(a) Subjects actually expect stimuli to repeat, and thus discon-firmations elicit larger P300 amplitudes than confirmationwhen repetitions are overtly predicted but smaller P300s whenalternations are predicted; (b) more simply, alternations elicitlarger P300s than repetitions, whatever the prediction.

In a similar prediction study in Bernoulli series of clicksand omissions, Munson et al. (1984) found a complex inter-action among the first-order stimulus sequence, the subjects'

Expected Stimulus

Presented — None — not Presented —

Figure 11. ERP waveshapes at Pz according to the expectancyratings and all 16 stimulus sequences (cf. Figure 2).

Expected Stimulus

Presented — None — not Presented —

Figure 12. ERP waveshapes at all three electrode sites and theaveraged electrooculogram according to the expectancy ratings andfor first-order repetitions (left panel) and alternations (right panel).

predictions, and prediction outcome. Conforming with thequantitative trend in our data, a stimulus repetition elicitedlarger P300s when disconfirming a prediction. For stimulusalternations, however, the prediction was not related to P300amplitude.

Whereas Munson et al.'s findings of larger P300 amplitudesto unexpected as compared with expected repetitions agreewith ours, neither their results nor those of Chesney andDonchin show the most striking feature of our findings—thelarger P300 amplitude to expected than to unexpected alter-nations. A possible explanation for the difference in findingsbetween our study and these studies may be the difference inmethods. Expectancies concerning particular events may beconsciously accessible to a different degree when they are tobe stated in advance of the event, as compared with theretrospective view taken in the present experiment.

Conclusion

The effects of conscious expectancies on P300 amplitudecontrast with the effects of sequence-based effects in ERPsand RTs. The sequential effects conform with the literatureand seem to be outcome dependent, revealing enhanced P300amplitudes, longer P300 latencies, and increased RTs whenruns of events are discontinued, that is, when subjects werepresumably surprised by the (unexpected) terminal stimulus.According to the analysis of the expectancy ratings, suchsequence-generated expectancies seem to be largely uncon-scious, as only 2 of the 16 stimulus sequences affected theratings.

The comparatively short RTs and P300 latencies foundhere, as well as the poor conscious access to sequence-basedexpectancies, conform with the view that these expectanciesare related to a fast, inflexible, automatic process rather thanto a controlled process (Johnson, 1988; Johnson & Donchin,


1982; Karis et al., 1983). This is partially corroborated by ourpractice results, which show robust higher order sequentialeffects in ERP measures. That there are also practice-depend-ent changes in sequential effects in RTs is a clue that theremight also be a controlled aspect in the workings of sequence-based expectancies. The relative inclination to conceptualizeexpectancies as products of controlled processes in RT workmight be seen as related to these possible contributions ofcontrolled processes to sequential effects in RTs but not inERPs. But even if there is a controlled aspect to the sequentialeffects in RTs, it is not reflected in self-reports of expectancies,either because it is too weak or because it is unrelated to whatis measured by the rating process.

When we asked subjects to report their expectancies, wemay have been tapping those aspects of expectancy that areconsciously accessible and that are related to P300 amplitudeaccording to rules quite different from those governing therelatively inaccessible, direct sequential effects on P300. Oneinterpretation of our different "expectancy" effects at thelevels of stimulus sequences and of self-reports may followKahneman and Tversky's (1982) suggestion that subjects maysimultaneously hold different types of expectations, whichmay, in turn, be consciously accessible to different degrees.In contrast to the largely unconscious sequence-based expec-tancies, there do exist conscious expectancies, not explainableby differential contributions of the stimulus sequences andthat would not have been predicted from traditional ERPresearch (e.g., Johnson, 1988; Squires etal., 1976). The findingthat within first-order stimulus alternations consciously ex-pected events are related not to particularly small but to largeP300 amplitudes conforms more with Verleger's view ofgreater P300 amplitudes to stimuli that have been "awaited."

However, if the (insignificant) P300 findings for first-orderrepetitions (cf. Figure 10) are to be taken seriously, an evensimpler interpretation of the effects of consciously accessibleexpectancies on P300 amplitude may be formulated. Onemight say that we found large P300 amplitudes wheneversubjects had expected stimulus alternations and small P300amplitudes when repetitions had been expected, with littlerelevance in either case as to whether the stimulus actuallyrepeated or alternated. When formulated this way, the ex-pectancy effects on P300 amplitude become completely in-dependent of the actual first-order sequence and solely de-pendent on the sequence expected. This reformulated effectmight be interpreted in at least two ways: (a) Any stimulus,be it a change or a repetition, is increased in relevance andthus elicits a comparatively large P300 (Donchin, Ritter, &McCallum, 1978) when environmental change rather thanconstancy is consciously expected; or (b) the internal repre-sentation of expectancy might be altered by the very processesthat control the P300 amplitude—one may experience theexpectation of a change of events when, by some chance, anevent has elicited a large P300, and a repetition when P300has been small.

In addition, our data indicate different relationships ofconsciously accessible expectancies and P300 amplitude, onthe one hand, and RTs and P300 latencies, on the other. Thelatter are accelerated for expected events and suggest that notonly sequence-based but also conscious expectancy can speedup processing time.

References

Bertelson, P., & Renkin, A. (1966). Reaction times to new versusrepeated signals in a serial task as a function of response-signaltime interval. Ada Psychologies 25, 132-136.

Chesney, G. L., & Donchin, E. (1979). Predictions, their confirmationand the P300 component (Abstract). Psychophysiology, 16, 174.

Donchin, E. (1979). Event-related potentials: A tool in the study ofhuman information processing. In H. Begleiter (Ed.), Evoked brainpotentials and behavior (pp. 13-88). New York: Plenum.

Donchin, E. (1981). Surprise!.. .Surprise? Psychophysiology, 18, 493-513.

Donchin, E. (1984). Dissociation between electrophysiology and be-havior—A disaster or a challenge? In E. Donchin (Ed.), Cognitivepsychophysiology: Event-related potentials and the study of cogni-tion (pp. 107-118). Hillsdale, NJ: Erlbaum.

Donchin, E., & Coles, M. G. H. (1988). Is the P300 component amanifestation of context updating? Behavioral and Brain Sciences,11, 357-374.

Donchin, E., & Isreal, J. B. (1980). Event-related brain potentials:Approaches to cognitive psychology. In R. E. Snow, P. A. Federico,& E. W. E. Montague (Eds.), Aptitude, learning, and instruction:Cognitive process analyses (Vol. 2, pp. 47-82). Hillsdale, NJ:Erlbaum.

Donchin, E., Karis, D., Bashore, T. R., Coles, M. G. H., & Gratton,G. (1986). Cognitive psychophysiology and human informationprocessing. In M. G. H. Coles, E. Donchin, & S. W. Porges (Eds.),Psychophysiology: Systems, processes, and applications (pp. 244-267). New York: The Guilford Press.

Donchin, E., Kramer, A. F., & Wickens, C. (1986). Applications ofbrain event-related brain potentials to problems in engineeringpsychology. In M. G. H. Coles, E. Donchin, & S. W. Porges (Eds.),Psychophysiology: Systems, processes, and applications (pp. 702-718). New York: The Guilford Press.

Donchin, E., Ritter, W., & McCallum, C. (1978). Cognitive psycho-physiology: The endogenous components of the ERP. In E. Calla-way, P. Tueting, & S. Koslow (Eds.), Brain event-related potentialsin man (pp. 349-441). New York: Academic Press.

Duncan-Johnson, C. C, & Donchin, E. (1977). On quantifyingsurprise: The variation of event-related potentials with subjectiveprobability. Psychophysiology, 14, 456-467.

Duncan-Johnson, C. C, & Donchin, E. (1982). The P300 componentof the event-related brain potential as an index of informationprocessing. Biological Psychology, 14, 1-52.

Duncan-Johnson, C. C, Roth, W. T., & Kopell, B. S. (1984). Effectsof stimulus sequence on P300 and reaction time in schizophrenics.Annals of the New York Academy of Sciences, 425, 570-577.

Entus, A., & Bindra, D. (1970). Common features of the "repetition"and "same-different" effects in reaction time experiments. Percep-tion and Psychophysks, 7, 143-148.

Ford, J. M., Duncan-Johnson, C. C, Pfefferbaum, A., & Kopell, B.S. (1982). Expectancy for events in old age: Stimulus sequenceeffects on P300 and reaction time. Journal of Gerontology, 37,696-704.

Ford, J. M., Pfefferbaum, A., & Kopell, B. S. (1982). Event-relatedpotentials to a change of pace in a visual sequence. Psvchophysi-ology, 19, 173-177.

Geller, E. S. (1975). Prediction outcome and choice reaction time:Inhibition versus facilitation effects. Ada Psychologica, 39, 69-82.

Geller, E. S., & Pitz, G. F. (1970). Effects of prediction, probability,and run length on choice reaction speed. Journal of ExperimentalPsychology, 84, 361-367.

Hale, D. J. (1967). Sequential effects in a two-choice serial reactiontask. Quarterly Journal of Experimental Psychology, 19, 133-141.

Horst, R. L., Johnson, R., & Donchin, E. (1980). Event-related brainpotentials and subjective probability in a learning task. Memory &


Cognition, 8, 476-488.Jarvik, M. E. (1951). Probability learning and a negative recency

effect in the serial anticipation of alternative symbols. Journal ofExperimental Psychology, 41, 291-297.

Johnson, R. (1988). The amplitude of the P300 component of theevent-related potential: Review and synthesis. In P. K. Ackles, J.R. Jennings, & M. G. H. Coles (Eds.), Advances in psychophysiol-ogy, Volume HI (pp. 69-137). Greenwich, London: JAI Press.

Johnson, R., & Donchin, E. (1980). P300 and stimulus categorization:Two plus one is not so different from one plus one. Psychophysi-ology, 17, 167-178.

Johnson, R., & Donchin, E. (1982). Sequential expectancies anddecision making in a changing environment: An electrophysiolog-ical approach. Psychophysiology, 19, 183-200.

Johnson, R., Pfefferbaum, A., & Kopell, B. S. (1985). P300 and long-term memory: Latency predicts recognition performance. Psycho-physiology, 22, 497-507.

Kahneman, D., & Tversky, A. (1974). On the psychology of predic-tion. Psychological Review, 80, 237-251.

Kahneman, D., & Tversky, A. (1982). Variants of uncertainty. Cog-nition, 11, 143-157.

Karis, D., Chesney, G. L., & Donchin, E. (1983). ".. .;'twas ten toone; And yet we ventured...": P300 and decision making. Psycho-physiology, 20, 260-268.

Keele, S. W. (1969). Repetition effect: A memory-dependent process.Journal of Experimental Psychology, 80, 243-248.

Kirby, N. H. (1972). Sequential effects in serial reaction time. Journalof Experimental Psychology, 96, 32-36.

Kirby, N. H. (1976). Sequential effects in two-choice reaction time:Automatic facilitation or subjective expectancy? Journal of Exper-imental Psychology: Human Perception and Performance, 2, 567-577.

Kirby, N. H. (1980). Sequential effects in choice reaction time. In A.T. Welford (Ed.), Reaction times (pp. 129-172). London: AcademicPress.

Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mentalchronometry: The P300 as a measure of stimulus evaluation time.Science, 197, 792-795.

Laming, D. R. J. (1969). Subjective probability in choice-reactionexperiments. Journal of Mathematical Psychology, 6, 81-120.

Leonard, A. (1958). Partial advance information in a choice reactiontask. General Psychology, 49, 89-96.

McCarthy, G., & Donchin, E. (1981). A metric for thought: Acomparison of P300 latency and reaction time. Science, 211, 77-80.

Moss, S. M., Engel, S., & Faberman, D. (1967). Alternation andrepetition reaction times under three schedules of event sequencing.Psychonomic Science, 9, 557-558.

Mowbray, G. H., & Rhoades, M. V. (1959). On the reduction ofchoice reaction times with practice. Quarterly Journal of Experi-mental Psychology, 11, 16-23.

Munson, R., Ruchkin, D. S., Ritter, S., Sutton, S., & Squires, N. K.(1984). The relation of P3b to prior events and future behavior.Biological Psychology, 19, 1-30.

Naatanen, R., & Merisalo, A. (1977). Expectancy and preparation insimple reaction time. In S. Dornic (Ed.), Attention and performanceVI (pp. 115-138). Hillsdale, NJ: Erlbaum.

Oldfield, R. C. (1971). The assessment and analysis of handedness:

The Edinburgh Inventory. Neuropsychologia, 9, 97-113.Peters, J. F., Billinger, T. W., & Knott, J. R. (1977). Event-related

potentials of brain (CNV and P300) in a paired associate learningparadigm. Psychophysiology, 14, 579-585.

Pfefferbaum, A., Ford, J., Johnson, R., Wenegrat, B., & Kopell, B. S.(1983). Manipulation of P3 latency: Speed vs. accuracy instruc-tions. Electroencephalography and Clinical Neurophysiology, 55,188-197.

Remington, R. J. (1969). Analysis of sequential effects in choicereaction times. Journal of Experimental Psychology, 82, 250-257.

Ruchkin, D., & Glaser, E. M. (1978). Simple digital filters for exam-ining CNV and P300 on a single trial basis. In D. Otto (Ed.),Multidisciplinary perspectives in event-related brain potential re-search (pp. 579-581). Washington, DC: U.S. Environmental Pro-tection Agency.

Schvaneveldt, R. W., & Chase, W. G. (1969). Sequential effects inchoice reaction time. Journal of Experimental Psychology, 80, 1-8.

Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatichuman information processing: II. Perceptual learning, automaticattending, and a general theory. Psychological Review, 84, 127—190.

Soetens, E., Boer, L. C, & Hueting, J. E. (1985). Expectancy orautomatic facilitation? Separating sequential effects in two-choicereaction time. Journal of Experimental Psychology: Human Per-ception and Performance, 11, 598-616.

Sommer, W., & Matt, J. (in press). Awareness of P300-related cog-nitive processes: A signal detection approach. Psychophysiology.

Squires, K. C, Hillyard, S. A., & Lindsay, P. H. (1973). Corticalpotentials evoked by confirming and disconfirming feedback fol-lowing an auditory discrimination. Perception and Psychophysics,13, 25-31.

Squires, K., Wickens, C, Squires, N., & Donchin, E. (1976). Theeffect of stimulus sequence on the waveform of the cortical event-related potential. Science, 193, 1142-1145.

Squires, K., Petuchowski, S., Wickens, C, & Donchin, E. (1977).The effects of stimulus sequence on event related potentials: Acomparison of visual and auditory sequences. Perception and Psy-chophysics, 22, 31-40.

Teichner, W. H., & Krebs, M. J. (1974). Laws of visual choice reactiontime. Psychological Review, 81, 75-98.

Tueting, P., Sutton, S., & Zubin, J. (1970). Quantitative evokedpotential correlates of the probability of events. Psychophysiology,7, 385-394.

Verleger, R. (1988). Event-related potentials and cognition: A critiqueof the context updating hypothesis and an alternative interpretationofP3. Behavioral and Brain Sciences, 11, 343-356.

Vervaeck, K. R., & Boer, L. C. (1980). Sequential effects in two-choice reaction time: Subjective expectancy and automatic after-effect at short response-stimulus intervals. Ada Psychologica, 44175-190.

Williams, J. A. (1966). Sequential effects in disjunctive reaction time:Implications for decision models. Journal of Experimental Psy-chology, 71, 665-672.

Received August 24, 1989Revision received December 11, 1989

Accepted January 19, 1990 •

journal of experimental psychology: 1990, vol. 16, …...through sony mdr m66 headphones. a room...

Documents