Discrete Versus Continuous Stage Models of Human

Journal of Experimental Psychology: Copyright 1982 by the American Psychological Association, Inc.Human Perception and Performance 0096-1523/82/0802-0273S00.751982, Vol. 8, No. 2, 273-296

Discrete Versus Continuous Stage Modelsof Human Information Processing:

In Search of Partial Output

Jeff MillerUniversity of California, San Diego

This article introduces a new technique designed to study the flow of informationthrough processing stages in choice reaction time tasks. The technique was de-signed to determine whether response preparation can begin before stimulusidentification is complete ("continuous" models), or if a stimulus must be fullyidentified prior to any response activation ("discrete" models). To control theinformation available at various times during stimulus identification, some rel-evant stimulus characteristics were made easy to discriminate and some weremade hard to discriminate. The experimental strategy was to look for effects ofpartial output based on information conveyed by characteristics that were easyto discriminate. The technique capitalized on the fact, demonstrated in Exper-iment 1, that,preparation of two response fingers on the same hand is moreeffective than preparation of two response fingers on different hands. The use-fulness of partial output was varied by manipulating the assignments of stimulito responses. For some mappings partial information could contribute to effectiveresponse preparation because the responses consistent with partial informationwere assigned to fingers on the same hand. For other mappings partial infor-mation could not contribute to effective response preparation because the re- •sponses consistent with partial information were assigned to fingers of differenthands. Performance differences between these mappings were considered evi-,dence that partial information about a stimulus was transmitted to responseactivation processes before the stimulus was uniquely identified, and thus wereconsidered evidence against discrete transmission of information about the stim-ulus as a, whole. A variety of stimulus sets were studied; the results suggest thatinformation is transmitted discretely with respect to stimulus codes, althoughdistinct codes activated by a single stimulus may be transmitted at differenttimes.

One goal of the study of human infor- to represent a series of operations that mustmation-processing capabilities is to describe be performed to produce an appropriate re-performance in terms of a set of component sponse when a stimulus is presented. In themental processes. These processes are taken most general terms, for example, it is thought

that a stimulus must first be perceived, thatthe perception must then be used to decide

a response and that this decision mustIn the course of this research, I have profited partic- be passed to the motor system so that the

ularly from discussions with David Bauer, Geoffrey correct response can be activated and exe-Hinton, James Johnston, James McClelland, David cuted (eg Neisser, 1967' Sanders, 1977)

^^SfS^^SSS^&^y. These processes have a contingent relation-sition by J. A. Deutsch, Raymond Gibbs, Jonathan »hlP. » that the Output of one process mustGrudin, Raymond Klein, and the anonymous reviewers serve as input to the next.were also quite helpful. Anne Suiter and Diane Fisher Recent attempts to build precise modelsworked heroically to collect most of the data reported of human performance in speeded choice^Requests for reprints should be sent to Jeff Miller, tasks have emphasized the importance of anDepartment of Psychology, C-009, University of Call- issue regarding the temporal relations amongfomia, San Diego, La Jolia, California 92093. the perceptul, decision, and response pro-

273

274 JEFF MILLER

cesses: When is information transmittedfrom one process to the next? The idea thateach process operates over a significant pe-riod of time is not new, but until recentlythe question of when information is trans-mitted from one process to the next has re-ceived scant attention. In the interest offormulating precise human informationprocessing models, however, it has becomeimportant to consider the timecourse overwhich the products of one process becomeavaiable to subsequent processes. This issueis probably best clarified by considering twoextreme views.

The simplest view is that the output of oneprocess is passed to subsequent processesonly after that process has finished. Accord-ing to this view, performance is carried outby a series of discrete processes or "stages"operating in a contingent fashion. That is,the output of one stage serves as the inputto the next stage, and stage N + 1 cannotbegin until stage N has finished (Broadbent,1958; Sperling, 1960; Sternberg, 1969a).For the purposes of temporal modeling, theimportant feature of the discrete model isthat total reaction time (RT) is the sum ofthe durations of the component mental pro-cesses (Donders, 1969; Schweikert, 1978;Sternberg, 1969a, 1969b).

Recently, many theorists have given se-rious consideration to the sorts of modelsthat would be needed to allow temporal over-lap of processes in successive stages (An-derson, 1977;, Anderson, Silverstein, Ritz,& Jones, 1977; Eriksen & Schultz, 1979;McClelland, 1979; Norman & Bobrow, 1975;Rumelhart, 1977; Taylor, 1976; Turvey,1973). Exploration of such models seems tohave been motivated largely by argumentsof physiological and neuroanatomical plau-sibility. Alternatives to the discrete approachare based on the idea of "partial" or "con-tinuously available" output from one processto the next; such models are sometimes re-ferred to as "concurrent-contingent" models(Turvey, 1973). According to these models,it is not necessary for one process to finishcompletely before transmission of informa-tion to the next process can begin. Instead,each process continuously forwards what-ever information it has to the next process,even if this information is incomplete. In the

"continuous flow conception" of Eriksen andSchultz (1979), for example, "informationabout stimuli accumulates gradually in thevisual system, and as it accumulates, re-sponses are concurrently primed or partiallyactivated" (p. 252). In this model responseactivation begins before stimulus identifi-cation is complete. In general, continuousmodels allow different processes to operatein parallel, even though the output of oneprocess serves as input to the next. Thus,total RT is not simply the sum of the du-rations of the component processes, as as-sumed by many standard procedures for bas-ing inferences on RT data (cf., Smith, 1968).

The crux of the dispute between discreteand continuous models concerns the fate ofinformation that becomes available rela-.tively early in the course, of processing doneby a particular stage. According to discretemodels, such preliminary information is sim-ply held until full stimulus information isavailable. Continuous models, however, as-sume that preliminary information is trans-mitted to subsequent processes immediately.Furthermore, continuous models asume thatthis preliminary information is used by thesubsequent processes to accomplish some"preparation" or "priming" (e.g., Eriksen& Schultz, 1979). This preparation is be-lieved to heighten readiness for a set of stim-uli suggested by the preliminary informa-tion, generally facilitating the processingthat is done when full information becomesavailable (cf., Posner, 1978, chap. 4).

There have been many studies of the prep-aration that occurs when a subject is givenpartial information about a forthcomingstimulus (Seller, 1971; Doll, 1969; Goodman& Kelso, 1980; Klapp, 1977; La Berge, VanGelder, & Yellot, 1970; Neely, 1977; Posner& Snyder, 1975; Posner, Snyder, & Dav-idson, 1980; Rosenbaum, 1980); these stud-ies support the assumption that preliminaryinformation would produce useful prepara-tion if it were transmitted from one processto the next. These studies are not directlyrelevant to the discrete versus continuousissue, however, because partial informationdid not come from the stimulus itself butrather from a separate cue given well in ad-vance of the stimulus. For example, Good-man and Kelso (1980) asked subjects to

DISCRETE VERSUS CONTINUOUS STAGE MODELS 275

make one of eight choice responses to oneof eight stimulus lights. About 4 sec beforethe stimulus light was turned on, a subsetof the eight lights was illuminated. This sub-set served as a cue to prepare for certainstimuli, since the test stimulus always camefrom the indicated subset. Not surprisingly,responses to the test stimulus got faster asthe cued subset got smaller; In general, cuingstudies demonstrate that preparation occurswhen a cue provides partial informationabout a stimulus. To get evidence relevantto the issue of discrete versus continuousmodels, however, it is necessary to find outwhether partial information provided byearly analysis of the stimulus itself also leadsto useful preparation.

Response Preparation Effect

The goal of the research reported here wasto develop an experimental method for dis-criminating between discrete and continuousmodels. The method developed in this articlespecifically examines the process of activat-ing a response. The question addressed inthese experiments is, whether informationextracted early in the processing of a par-ticular stimulus can be used to prepare re-sponses before full information about thestimulus is available. Continuous modelssuggest that such preparation should be pos-sible, whereas discrete models suggest thatit should not.

It must be emphasized that discrete andcontinuous models make different predic-tions only when the information provided bya stimulus is recognized little by little overan extended perid of time. If all the infor-mation in a stimulus were to become avail-able at precisely the same time, there wouldbe no preliminary information to be trans-mitted or used for preparation. Therefore,an experimental comparison of the two mod-els must use stimuli designed so that somepreliminary information about a stimulusbecomes available well before complete in-formation is available.

One way to ensure the presence of prelim-inary information is to manipulate pairwisestimulus discriminability (Miller & Bauer,1981). Suppose, for example, a stimulus set

consists of four letters chosen such that theycan be grouped into two pairs of visuallysimilar letters (e.g., M, N; U, V). With ap-propriate pilot studies, it is easy to show thatdiscriminations between visually similar let-ters (e.g., MN) are much slower than dis-criminations between visually dissimilar let-ters (e.g., MU). Thus, it is reasonable tosuppose that preliminary information wouldindicate to which visually similar pair (MNor UV) the stimulus belonged and that thisinformation would be available well beforestimulus identity was uniquely established.The problem, then, is to determine whetherthis preliminary information is actually usedto prepare responses.

The basic idea behind the method used inthese experiments is this: Manipulating theefficiency with which preparation can takeplace will reveal whether or not it does takeplace. Suppose, for example, we want to testthe hypothesis that students prepare for ex-ams (i.e., study), and exam performance isthe only available dependent variable. Oneway to test the hypothesis would be to giveone group of students a "good" study guideand another group a "bad" study guide. Thegood study guide would focus students' at-tention on material specifically covered inthe exam and provide clear, accurate sum-maries of the course material. The bad onewould stress material not covered on the test,and give incomprehensible, inaccurate sum-maries. If students actually do prepare forexams, then students who get the good studyguide should perform better than studentswho get the bad one. If students do not pre-pare for exams, the study guides shouldmake no difference, and the two groupsshould do equally well. In general, an ex-perimental manipulation of the efficiency ofhypothesized preparation should only pro-duce an effect if the hypothesized prepara-tion actually takes place. In this example,then, we can determine whether studentsprepare for exams by manipulating the ef-ficiency with which this hypothesized prep-aration could take place.

The test for hypothesized response prep-aration is based on the same idea. As dis-cussed above, it is possible to construct stim-ulus sets with the property that any particularstimulus can provide partial information

276 JEFF MILLER

well before full information is available (e.g.,by controlling pairwise stimulus discrimi-nability). If this preliminary information isactually used for preparation of responses,as assumed by the continuous models, thenan experimental manipulation of prepara-tion efficiency should produce a main effect.If preliminary information is not used toprepare responses, as assumed by the dis-crete models, then such an experimentalmanipulation should not produce any effect.Thus, it would be! possible to discover whetherpreliminary information is used to prepareresponses by manipulating the efficiencywith which the hypothesized preparation canoccur.

Studies of the structure of the motor sys-tem suggest a way to manipulate the effi-ciency of the hypothesized response prepa-ration. There is considerable evidence fromcuing studies that some combinations of re-sponses can be prepared together more ef-ficiently than other combinations of re-sponses. For example, Rosenbaum (1980)showed that two responses made by the samearm can be prepared together more effi-ciently than two responses made by differentarms. To control the efficiency of responsepreparation, then, it is necessary to produceone condition in which preliminary infor-mation is consistent with two responses pre-pared together relatively efficiently and an-other condition in which preliminary infor-mation is consistent with two responsesprepared together relatively inefficiently.

In the present experiments subjects wererequired to respond with one of four fingers:the middle and index fingers of the right andleft hands. Experiment 1 used a cuing par-adigm to show that when two of these re-sponses can be prepared in advance, prepa-ration of two fingers on the same hand ismore efficient than preparation of two fin-gers on different hands. That is, subjectscould respond to a stimulus more quicklywhen an advance cue indicated that the re-sponse would be made with (say) one of thetwo fingers on the left hand than when thecue indicated that the response would bemade with one of the two index fingers. Sincethis was a cuing study, it is not directly rel-evant to the issue of discrete versus contin-uous models, as discussed earlier. The expe-

ment does establish, however, that pre-liminary information allowing preparationof two fingers on the same hand leads tomore effective preparation than informationallowing preparation of two fingers on dif-ferent hands.

Given these preliminaries, the proposedtest for response preparation can be statedas a comparison between two conditions. Thestimulus set is identical in the two conditions,consisting of two pairs of visually similarstimuli. Preliminary information is availablein both conditions because it is much easierto recognize to which similar pair a stimulusbelongs than to recognize which particularstimulus it is. The response set (middle andindex fingers of the right and left hands) isalso identical in the two conditions. The dif-ference between the two conditions is in theassignment or mapping of stimuli to re-sponses. In the same^hand condition, stimuliin one visually similar pair are assigned toresponse fingers on the left hand, and stimuliin the other pair are assigned to responsefingers on the right hand. In the different-hand condition, stimuli in both visually sim-ilar pairs are assigned to response fingers ondifferent hands (e.g., one pair to the two in-dex fingers and the other pair to the twomiddle fingers).

In the proposed, experimental design,preparation of responses based on prelimi-nary information should be more efficient inthe same-hand condition than in the differ-ent-hand condition. In the same-hand con-dition, preliminary information is consistentwith two response fingers on the same hand,and the system can prepare these two fingersrelatively efficiently. In the different-handcondition, preliminary information is consis-tent with two response fingers on differenthands. The response system can also preparethese two fingers, but it does so less effi-ciently. The efficiency of preparation willonly have an effect, of course,, if preparationbased on preliminary information actuallyoccurs. Thus, if preparation does occur, per-formance should be better in the same-handcondition than in the different-hand condi-tion.

Predictions of discrete and continuousmodels for the same- and different-hand con-ditions are reasonably clear, assuming that

PISCRETE VERSUS CONTINUOUS STAGE MODELS 277

more efficient preparation leads to faster re-sponses. According to continuous models,preparation does take place. Since prepa-ration is more efficient in the same-hand con-dition than in the different-hand condition,responses should be faster in the former con-dition than in the latter. According to dis-crete models, response preparation does notbegin until the stimulus has been fully iden-tified. At that time, of course, only a singleresponse needs to be prepared. Thus, a dis-crete system would not prepare the two re-sponses consistent with preliminary infor-mation, and it would be unaffected by therelative efficiency of preparing these two re-sponses together. Discrete models, then, can-not explain any difference between the same-and different-hand conditions in terms ofresponse preparation.'

To summarize, the method is based onpresenting preliminary information that isconsistent with two efficiently prepared re-sponses in one condition (same-hand), andconsistent with two inefficiently preparedresponses in a second condition (different-hand). Continuous models assume that re-sponse preparation occurs before full stim-ulus information is available. Greater effi-ciency of response preparation should leadto faster responses in the same-hand condi-tion than in the differen,t-hand condition.Discrete models asssume that response prep-aration does not begin until full stimulus in-formation is available. Therefore, prepara-tion efficiency should have no effect, and thesame-hand and different-hand conditionsshould not differ. The difference in RT be-tween the same- and different-hand condi-tions is a measure of the extent to whichresponse preparation actually occurs. In thepresent article, the term response prepara-tion effect (RPE) will be used to refer to thedifference in overall RT (different-hand mi-nus same-hand) between these two condi-tions.

The present method may at first appearto be a roundabout approach to the problemof examining the preparation of responses.However, it overcomes the serious problemof finding a dependent variable that accu-rately reflects continuous response prepara-tion. Making a response in an information-processing task is, for all practical purposes,

a discrete process. Given the discreteness ofthe response, it is hard to determine whetherprocesses prior to the response are also dis-crete, or whether there is simply a discretecriterion at the end of a continuous system(e.g., McClelland, 1979). The method pro-posed here circumvents this problem by Ic-ok-ing at a variable that need not reach a dis-crete threshold on each trial: the efficiencyof response preparation.

Experiment 1

The first experiment examined the re-sponse set used in these experiments: the in-dex and middle fingers of the left and righthands. The purpose of the experiment wasto show that responses are faster when a cuespecifies which hand will make the responsethan when it specifies which type of finger(e.g., index vs. middle) will make the re-sponse.

The experimental design was similar tothat employed by Goodman and Kelso(1980), as discussed earlier (see also Rosen-baum, 1980). The stimulus was a plus sign(+) in one of four spatial locations; stimuliwere assigned compatibly to four horizon-tally arrayed response keys. On some trialsthe subject was given a cue indicating thatthe stimulus would appear in one of two par-ticular positions. Trials on which no cue wasgiven will be referred to as unprepared trials.Trials with a cue indicating two positionsassigned to fingers on the same hand will bereferred to as trials in the prepared: handcondition; trials with a cue indicating twopositions assigned to homologous fingers(i.e., index or middle) on different hands willbe referred to as trials in the prepared: fingercondition; and trials with a cue indicatingtwo positions assigned to different fingers ondifferent hands will be referred to as trialsin the prepared: neither condition.

To examine effects of deliberate prepa-ration strategies, one group of subjects wasexplicitly instructed to prepare for the re-

' Readers may wonder whether discrete models mightaccount for the sort of effect considered here in some,way other than response preparation. Alternative ex-planations will be considered after the effect is demon-strated in Experiment 3.

278 JEFF MILLER

sponses indicated by the cues, and anothergroup was not. The group given instructionsto prepare was also allowed more time be-tween cue and test stimulus to give delib-erate preparation strategies ample time toestablish a set. Furthermore, this group wasalso occasionally presented with a test stim-ulus that was not one of the ones indicatedby the cue. For the group not specificallyinstructed to prepare, all test stimuli camefrom the cued set.

MethodSubjects. A total of 44 right-handed undergraduates

at the University of California, San Diego (UCSD),served in a single experimental session of 20-30 minutes.Participation in the experiment partially fulfilled a re-quirement of an introductory psychology course.

Apparatus and stimuli. Stimuli were presented andresponses and response latencies recorded by a Terak8510 microcomputer. Stimuli were plus signs presentedin the standard character set of the computer. The stim-uli were presented in the center line of the display screen.The horizontal positions of the four possible stimuli cov-ered a total width of eight character positions, spanningabout 2 cm. Stimuli could appear in the first, third,sixth, or eighth position. Thus, the two leftmost and tworightmost positions were separated by about 3 mm; thetwo center positions were separated by 6.5 mm.

Procedure. A trial began with the presentation ofa warning signal consisting of plus signs in all four lo-cations. The warning signal appeared on the screen twolines above the line in which the stimulus would appearand remained on throughout the trial. After a delay of250 msec, the cue appeared on the line directly abovethe stimulus line. In the prepared conditions, the cueconsisted of plus signs in two of the four locations, asdescribed above. In the unprepared condition, the cueconsisted of plus signs in all four locations. Then therewas a variable delay, with interstimulus intervals (ISIs)ranging from 0 to 1,000 msec, followed by the onset ofa single plus sign as the, test stimulus. The entire displaythus generated remained on until a response was made,at which time accuracy feedback was given. The nexttrial began approximately 2 sec later. Responses weremade by pressing one of the four outside buttons on thebottom row of a standard typewriter keyboard (Z, X,.[the period], and / [the slant]).

For one group of 24 subjects, the test stimulus wasalways one of the two expected stimuli on preparedtrials. Subjects were informed of this fact but were notexplicitly told to prepare. For these subjects, the ISIswere 0, 125, 250, 375, and 500 msec. Each subject wastested in a single block of 430 trials, with the first 30being practice. There were 5 tests at each ISI in eachexpected position for each preparation condition.

A second group of 20 subjects was explicitly in-structed to try to prepare for the responses indicated bythe cues. These-subjects were informed that the teststimulus would appear in one of the expected positionson 80% of all trials but would appear in one of the

unexpected positions on the remaining 20%. For thesesubjects, ISIs between cue onset and stimulus onset were0,250, 500,750 and 1,000 msec. Each subject was testedin a single block of 370 trials, of which the first 30 werepractice. Subjects were tested twice on each finger ateach ISI in the unprepared condition, four times on eachfinger at each ISI in every expected condition, and onceon each finger at each ISI in every unexpected condition.

Results and Discussion

Separate analyses were performed for thetwo groups of subjects. For each group, av-erage RTs and percentages of correct re-sponse were computed for each of the prep-aration and test conditions.

Average RTs are shown in Figure 1 forthe group that was not explicitly instructedto prepare. An analysis of variance (ANOVA)confirmed the significance of type of prep-aration, F(3, 69) = 28, p < .01, MS, = 7,626,ISL, F(4, 92) = 97, p < .01, MSt = 5,111,and their interaction, F(12, 276) = 2.4, p <.025, MSe = 3,444. Planned comparisons re-vealed that responses were faster in the handpreparation condition than any of the othersand that this difference increased with ISI.An analysis of the percentages of error in-dicated only a signficant difference amongerror rates for the four preparation condi-tions, F(3, 69) = 7.9, p < .01, M5e = 63.Fewer errors were made in the hand prep-aration condition (1.6%) than in the finger(3.7%) and neither (3.5%) preparation con-ditions, with the unprepared condition yield-ing an intermediate number (2.4%).

N

M

SE HBO

C

—• UNPREPRRED—• "PREPRREDi HflND—"PREPRREO. FINGER-"•PREIWEDi NEITHER

250ISI

375 500

Figure I. Experiment 1: average reaction time as a func-tion of the delay between the cue and the stimulus (ISI)and the type of preparation indicated by the cue. (Theseresults came from subjects who were not explicitly in-structed to prepare.)


Figures 2 and 3 show average RTs for thegroup that was instructed to prepare. To theextent that a subject is prepared for one ofthe expected stimuli, a response to one ofthese stimuli should be especially fast anda response to one of the unexpected stimulishould be especially slow. Figure 2 showslatencies from trials with the test stimulusin one of the expected locations, and Figure3 shows latencies from trials with the teststimulus in one of the unexpected locations.The neutral unprepared condition is in-cluded as the solid line on both figures toexpedite comparison between the two. Ex-pected responses were about 50 msec fasterin the hand preparation condition than ineither of the other two preparation condi-tions, whereas unexpected responses wereabout 35 msec slower in the hand condition.It is particularly striking that there was ben-efit for expected stimuli only when the cueled to preparation of two fingers on the samehand.

Statistical analyses confirmed the reli-ability of the results apparent in Figures 2and 3. In an ANOVA with factors of type ofpreparation, expected versus unexpected,and ISI, the type of Preparation X Expec-tancy interaction was highly signficant, F(2,38) = 30.8, p < .01, MSe = 4,204. The three-way interaction with ISI was not significantF(8, 152) = 1.5, p > .10, MSe = 3,896, how-

600 .

5SO .

° • UNPREPRREDa-.—° PREPRREOi HflND*— 'PREPRREDt FINGER•""• PREPHREDi NEITHER

500ISI

1000

Figure 2. Experiment 1: average reaction time as a func-tion of the delay between the cue and the stimulus (ISI)and the type of preparation indicated by the cue. (Theseresults came from subjects who were explicitly in-structed to prepare for the stimuli indicated by the cue,and represent only trials on which the test stimulus wasone of the two stimuli indicated by the cue.)

700 .

(SO .

600 .

SSO .

BOD .

•* UNPREPflRED"PREPRRED: HflND•«PREPRREDi FINGER-^PREPRREOt NEITHER

0 250 SOD 750 1000ISI

Figure 3. Experiment 1: average reaction time as a func-tion of the delay between the cue and the stimulus (ISI)and the type of preparation indicated by the cue. (Theseresults came from subjects who were explicitly in-structed to prepare, and represent trials on which thetest stimulus was not one of the two stimuli indicatedby the cue.)

ever. Thus, the RT data support the ideathat preparation is more effective when tworesponses on the same hand are preparedthan when two responses on different handsare prepared. An analysis of the error dataconfirmed this hypothesis. For expectedstimuli, fewer errors were made with handpreparation (3.5%) than with finger (6.9%)or neither (6.0%) preparation. For unex-pected stimuli, more errors were made withhand preparation (6.0% vs. 4.5% and 4.7%).This preparation by expectancy interactionwas statistically reliable, F(2, 38) = 4.5, p <.05, MSe = 72.5.

Overall, the results from both groups sup-port the claim that preparation to respondwith one of two fingers on the same hand ismore effective than preparation to respondwith one of two fingers on different hands.Thus, with this response set it is possible tomanipulate the efficiency of response prep-aration. As described in the introduction,these preparation conditions will be used inan attempt to find out whether responsepreparation can begin based on preliminaryinformation from a single stimulus.

The finding that same-hand preparationis more efficient than different-hand prepa-ration is consistent with several distinct the-ories of the motor system. For example, onemight account for the result with a hierar-chical model (Kerr, 1978; Rosenbaum,

280 JEFF MILLER

1980), and argue that it is necessary to spec-ify which hand will respond before differ-ential activation of fingers can take place.The failure to find any difference betweenthe prepared: finger and prepared: neitherconditions could be interpreted as supportfor this model, as could the evidence thatcerebral control of hand and finger move-ments is almost completely localized in thecontralateral frontal lobe (Brinkman &Kuypers, 1973; Crosby, Humphrey, & Lauer,1962; Gazzaniga, 1970; Wiley, 1975). Cer-tainly, though, other explanations of the ef-fect could be based on a more dynamic viewof the motor system (e.g., Goodman &Kelso, 1980) or on the idea of differentialresponse competition among different com-ponents of the motor system (Kornblum,1965).

The important point about the alternativeexplanations of the same-hand advantage isthat for present purposes, it does not matterwhich explanation is correct. In any case,this set of responses and preparation con-ditions can be used to manipulate the effi-ciency of response preparation. In terms ofthe study guide analogy, the alternative ex-planations have to do with the reasons whyone study guide is better than the other. Itis not necessary to understand these reasonsfully to determine whether preparation takesplace.

It is important, however, to rule out onepossible nonmotoric explanation ot the prep-aration effects observed in Experiment 1. Itmight be argued that these effects werereally due to perceptual preparation ratherthan response preparation, in which casethey would not be useful in looking for aresponse preparation effect. For example,one could claim that preparation for twostimulus positions is more efficient the closertogether they are, possibly because of anadvantage in sharing attention across nearbypositions. Under this hypothesis, preparationcould be most effective in the hand prepa-ration condition because the two primed po-sitions are close together. Such an expla-nation is not particularly plausible, however,in view of the small visual angles separatingstimulus positions. Eriksen and Hoffman(1972, 1973) have shown that attention can-not be focused more precisely than 1° ofvisual angle; position separation was less

than that for all but one of the preparationconditions in this experiment (the conditionin which the two middle fingers were pre-pared). Furthermore, the observed effects donot support the explanation. The cues for thetwd index fingers were nearly as close to-gether as the cues for the two fingers on thesame hand, and they were only half as farapart as the cues for the two middle fingers.However, preparation effects were virtuallyidentical for preparing two middle or twoindex fingers.

Finally, an interesting effect produced bythe instructional manipulation should bementioned. The extent of preparation wasconstant across ISI for the group explicitlyinstructed to prepare, but increased acrossISI for the group not explicitly instructed toprepare. It is likely that subjects adopted adeliberate strategy of preparing on everytrial in the former condition and that thisstrategy was so strong that they never re-sponded to the stimulus until after they hadprepared. In the latter condition, subjectsmay have started preparing when given thecue and then aborted the preparation processif the stimulus was presented before prepa-ration was complete.

Experiment 2Experiment 1 demonstrated response

preparation based on the information in acue and showed that the efficiency of prep-aration depends on which two response fin-gers are prepared. The purpose of this ex-periment was to begin examining conditionsunder which response preparation could takeplace during a trial, based on informationthat is part of the actual stimulus. The stim-uli were the four letter pairs BE, BO, ME,and MO. Each stimulus thus had two bi-nary-valued components, and the temporalonset of the two components could be con-trolled independently. Preliminary infor-mation about the stimulus pair was conveyedby presenting one letter slightly before theother letter. Varying stimulus-response (S-R) mappings were used to determine whichletter signaled "hand" and which letter sig-naled "finger within a hand."

MethodSubjects. Subjects were 24 undergraduates recruited

from the same pool as in Experiment 1.


Apparatus and stimuli. Stimulus presentation andresponse timing and recording were controlled by anAutomated Data Systems Micro-8 computer. The fourstimuli were the letter pairs BE, BO, ME, and MO.Stimuli were presented as light figures on a dark back-ground and were viewed from a distance of abut 50 cm.Individual letters were about 9.5 mm high by 6 mmwide, and were separated by 3 mm. Responses weremade with the index and middle fingers of the left andright hands on a microswitch keyboard located directlyin front of the subject. „

Procedure, Two groups of 12 subjects were tested.For one group, the four response buttons, frpm left toright, were assigned to the stimuli in the order BE, BO,ME, and MO. Thus, for this group the consonant in-dicated which hand would be used in making the re-sponse, and the vowel indicated which finger on a hand.For the other group, the assignment was BE, ME, BO,and MO, so the vowel indicated the response hand.

Each subject was tested in two blocks of 200 trials.The first 8 trials in each block were considered practiceand were not recorded, and the remaining 192 trialswere divided evenly among 24 conditions. These con-ditions were defined by three factors: which of the fourstimulus pairs was presented, whether the vowel or theconsonant was presented first, and the ISI between theonset of the first letter and the onset of the second letter(100, 350, or 600 msec). The two letters in a pair werealways presented in the same spatial order, even thoughthe temporal order varied. Both letters remained on thescreen until a response was made, at which time ac-curacy feedback was given.

Subjects were instructed as to the nature of the stim-ulus set, the response buttons, and given the S-R map-ping appropriate for their groups. They were told thatthe two letters would not appear at the same time butwere not explicitly instructed to try to prepare responses.


Average RTs are shown in Figure 4 as afunction of response assignment, letter pre-sented first, and ISI. For the BEBOMEMOgroup, there was no advantage for the handpreparation condition at the 100 msec ISI,but there was an advantage of about 70 msecat ISIs of 350 and 600 msec. For the BE-MEBOMO group, there was no advantage forthe hand preparation condition at any ISI.

Statistical analysis confirmed the reliabil-ity of these results. The factors in the anal-ysis were group, ISI, and whether the firstletter presented on the trial indicated handor finger-within-hand information. An AN-OVA indicated that the between-subjects fac-tor of group was not significant, F(l, 22) =1.26,p > .10, MSe = 534,314, but the within-subjects factors of type of information in thefirst letter, F(l, 22) = 20, p < .01, MSe =8,722 and ISI, F(2, 44) = 158, p <.01,

BOO .

100 -

650 x.

600 .

-« BEBOHEMO. CONSONRNT FIRST-'' BEBOhEMOi VOUEL FIRST-• BEHEBOMOi CONSONflNT FIRST"• BEHEBOMOi VOUEL FIRST

100 3SOIS]

600

Figure 4. Experiment 2: average reaction time as a func-tion of delay between the onset of the two componentsof the stimulus (ISI) and the stimulus-response map-ping.

MSe= 7,628, were significant. In addition,the interactions of group with type of infor-mation, F(l, 22) = 19, p < .01, MSe = 8,-722, type of information with ISI, F(2, 44)= .3.87, p < .05, MSe = 7,505, and the three-way interaction, F(2, 44) = 4.9, p < .025,MSC = 7,505, were all significant. Error ratesranged from 3.5% to 6.5% across these con-ditions, with no significant differences in anANOVA. Conditions producing faster re-sponses also produced more accurate re-sponses, so the possibility of a confoundingspeed-accuracy tradeoff can be discounted.

The most important result of this exper-iment is to demonstrate the possibility ofmeasuring response preparation that takesplace during a trial. The finding of a signif-icant preparation effect for the BEBOMEMOgroup at ISIs of 350 and 600 msec showsthat information abut the consonant can ac-tivate responses before the vowel becomesavailable. The interaction with response as-signments, however, shows that not all pairsof responses are equally easy to activate.

One plausible explanation of the lack ofa preparation effect for the BEMEBOMOgroup is that the decision system is inflexiblewith respect to the order of processing theletters in a pair. The decision system mightbe constrained to reach the first decisionabout the consonant, since reading is nor-mally done from left to right. This systemcould produce preliminary output when pre-sented with the consonant first, but not whenpresented with the vowel first. Such a system

282 JEFF MILLER

would allow hand preparation for the BE-BOMEMO response assignment, but not forthe BEMEBOMO response assignment.

Both the absolute size^f the RPE and theISI required to obtain it are difficult to judgefrom the first two experiments. The handadvantage in the second experiment wasabout 70 msec. This figure is somewhatlarger than the hand advantage over the un-prepared conditions in Experiment 1. Ofcourse, cue processing was not obligatory inExperiment 1, since the stimulus uniquelydetermined the response. Subjects may sim-ply have ignored the cue on some trials. Inthe second experiment, however, both lettershad to be processed to determine the re-sponse, so subjects could not avoid extractingthe information on which preparation wasbased. On the basis of the first two studies,then, 70 msec is a reasonable guess as to thesize of the RPE. It appears that the ISIneeded to obtain the RPE is at most 350msec with stimuli comparable to these. It isimpossible to establish the minimum ISI re-quired to obtain an RPE without a moreprecise model of how the effect builds up.In Experiment 2, for example, we wouldneed to know whether processing of the firstletter took the same amount of time as pro-cessing the second letter and whether re-sponse preparation continued during pro-cessing of the second letter or was terminatedwhen that letter appeared. Without this in-formation, it is difficult to tell why there wasno RPE at the 100 msec ISI. It is possiblethat this ISI allowed 100 msec for responsepreparation and that this was not enough,but it is also possible that considerably lessthan 100 msec was actually available forpreparation.

Experiment 3Experiment 2 demonstrated that re-

sponses can be activated on the basis of earlystimulus information obtained within a trialin a task with two-part stimuli. Finding ev-idence of partial output with such stimuli issuggestive of the power of the method, butit is a long way from demonstrating thatpartial output is always available to the re-sponse stage. In particular, the spatial andtemporal separation of the different stimulus

components would encourage subjects totreat them as two entities rather than one.The results of Experiment 2 indicate that thewhole stimulus is not processed discretely asa single unit; but processing could be discretewith respect to spatially and temporally sep-arate components of a stimulus.

The purpose of Experiment 3 was to lookfor an RPE with single entity stimuli. Tworelated stimulus sets were used. One con-sisted of four letters: a large S, a small s,a large T, and a small T. The actual letterheights were 11 mm for the large letters and8 mm for the small letters. Pilot testing witha separate group of eight subjects indicatedthat with these four letters, the letter dis-crimination could be made about 85 msecfaster than the size discrimination (SE ofdifference = 35 msec). If partial output wereavailable, then, it should indicate which let-ter was presented before size informationwas available. Assignments of stimuli to re-sponses were varied in order to look for aresponse preparation effect. Two letters ofthe same name could be assigned to responsefingers on the same hand ("letter-same-handmapping condition"), two letters of the samesize could be assigned to response fingers onthe same hand ("size-same-hand mappingcondition"), or two different letters of dif-ferent sizes could be assigned to responsefingers on the same hand ("neither-same-hand mapping condition"). If the decisionabout letter identity were available to re-sponse processes before the decision aboutsize, then responses should be fastest in theletter-same-hand mapping condition be-cause of the RPE. Otherwise, there shouldbe no difference between the mapping con-ditions.

With the srST stimulus set, it was im-possible to manipulate on a trial-to-trial ba-sis the time interval between the onset of thetwo sources of information (i.e., letter iden-tity and size). An attempt in this directionwas made, however, by using a second stim-ulus set ("irlT") in which the letter discrim-ination was more difficult than the size dis-crimination. This set used the letters I andT, in heights of either 15 mm or 5 mm. Pilottesting of eight subjects indicated that withthis set the size discrimination could be made


approximately 72 msec faster than the letterdiscrmination (SE of difference = 14 msec).

MethodSubjects. Ninety-six subjects were recruited on the

UCSD campus, and each served in a single 30-minuteexperimental session in partial fulfillment of a require-ment for an introductory psychology course,

Apparatus. The same apparatus was used as in Ex-periment 2.

Procedure. Each subject was tested in six blocks oftrials. The first block was identical for all subjects; datafrom this block were used in an analysis of coyariance(ANCOVA) to reduce the between-subjects error term.The stimuli in this block were the letters A, B, C, andD, assigned in order to the left middle finger, left indexfinger, right index finger, and right middle finger. Theblock was 84 trials long, with trials equally dividedamong the four stimuli. The stimulus letter was dis-played until the subjects made a response, at which timeaccuracy feedback was given. Approximately 1 sec laterthe next stimulus appeared.

At the end of the first block, subjects were told thatfor the rest of the experiment the stimulus set wouldconsist of four letters. Half of the subjects were thengiven the srST stimulus set, and the other half weregiven the ITIT set. The subject was instructed as towhich response key was assigned to each of the fourstimuli, and this assignment remained constant through-out the remainder of the experiment. All possible as-signments of stimulus letters to response keys were usedequally often across subjects. Five blocks of 84 trialswere run under these instructions, with 20-sec breaksbetween blocks. The trial structure was identical to thatof the first block.


For each subject, the average RT and per-centage of error were computed for each of

RE

? 860 .

N

T 750 .

I

? 100 J

•—• LETTER SflME HflND"•—° SIZE SflME HflND'—"NEITHER

COVRRIflTE 3 4. BLOCK

E

CTI0N

TIME

IN

SEC

B7S -

825 .

775 .

725 _

675 _

625 .

A,,

^•v.

0. '*"-.

'"'•-. '^"~*.

a ''--... '"'"'~'-y "'>* fr o

• — • LETTER SRHE HRNOO....D 5|ZE sflME Kfluo

g *— "NEITHERA

COVBBIflTE 3 4BLOCK

Figure 5. Experiment 3: average reaction time as a func-tion of experimental block and stimulus-response map-ping for the STST stimulus set.

Figure 6. Experiment 3: average reaction time as a func-tion of experimental block and stimulus-response map-ping for the ITIT stimulus set.

the six blocks. Figure 5 shows the averageRTs for subjects in the three groups corre-sponding to the different stimulus-responseassignment conditions for the STST stimulusset. As is evident from the figure, the letter-same-hand mapping condition was about 85msec faster than either of the other two,which did not differ from each other. Theresults for the irlT groups are shown in Fig-ure 6. Again, the letter-same-hand mappingwas fastest, but with this stimulus set thesize-same-hand mapping was somewhatfaster than the neither-same-hand mapping.

To test the significance of the RPEs ob-tained with these stimulus sets, ANCOVASwere performed on RTs. Each subject's av-erage RT for the first (ABCD) block was thecovariate, and the dependent measures werethe average RTs for the five experimental(STST or ITIT) blocks. Separate analyseswere performed for the two different stim-ulus sets, and the independent variables weremapping, subject within mapping, and block.In the STST analysis, the main effects ofmapping, F(2, 44) = 6.1, p<.01, MS, =32,548, and block, F(4,180) = 17.5, p < .01,MSe = 2,667, were significant, as was thecovariate, F(l, 44) = 44, p< .001). The in-teraction of mapping and block was not sig-nificant, F(8, 180) = .98, p> .20. The sig-nificant main effect of mapping indicatesthat the advantage for the letter-same-handmapping in Figure 5 was statistically reli-able. In the ITIT analysis, the main effect

284 JEFF MILLER

of mapping was also significant, F(2, 44) =5.85, p < .01, MS, = 28,699, as was themain effect of block, F(4, 180) = 11.6, p <.01, MSe = 3,356 and the covariate F(l,44) = 67.1, p < .01. Paired comparisons in-dicated that the letter-same-hand and thesize-same-hand mappings did not differ sig-nificantly from each other, but both werefaster than the neither-same-hand mapping.For both stimulus sets, average error rateswere about 5-6%, with conditions that pro-duced longer RTs also generally producinghigher error rates. The differences amongerror rates were not statistically reliable.

These results are unambiguous with re-spect to the difficulty of the various S-Rmapping conditions, and they suggest fur-ther conditions under which subjects can usepartial information about a stimulus to pre-pare responses. Subjects respond relativelyquickly if the large and small versions of agiven letter are assigned to the same hand,indicating that they can use letter name in-formation to prepare responses. To a lesserextent, subjects are able to use size infor-mation to prepare responses, but only if thesize discrimination is very easy relative tothe letter discrimination. It appears that de-cisions about letter identity normally but notinvariably precede decisions about size, andthat response preparation can begin as soonas either one is complete. Again, preparationis more efficient if the two responses consis-tent with the outcome of the first decisionare assigned to fingers on the same hand thanif they are assigned to fingers on differenthands, so the final response in the formercase.

It is interesting that letter identity couldserve as a basis for priming with the ITITstimulus set, even though the pilot data in-dicated that identity discriminations weremuch slower than size discrimination. Oneinterpretation of this result is that the mostnatural system for coding the stimuli is interms of letter identity, with size being afeature that discriminates among responsesonce letter identity has been selected (cf.,Hardzinski & Pachella, 1980). Imagine, forexample,' that the decision process used ahierarchical decision tree to look up the re-sponse to a given stimulus. The most naturalchoice for the decision node at the highestlevel is the most salient stimulus character-

istic. Letter identity would tend to be moresalient than size, since identity involves well-learned categories. Using such a decisiontree, subjects would, hold perceptual infor-mation about size until letter identity hadbeen processed. Letter identity informationcould be used to prime responses during thetime needed for the second level of decisionin the decision tree (i.e., the decision usingsize information).

If the order of nodes in the decision treewere inflexible, of course, no priming basedon size would be expected. However, somesuch priming was observed. This suggeststhat subjects in the size-same-hand conditionmay have been somewhat successful in rear-ranging the order of nodds in the decisiontree so as to make the first decision basedon size. With these stimuli size and letteridentity may have been about equally sa-lient, because size was so much easier todiscriminate. If the structure of the decisiontree were based on salience, then, it is notunreasonable to suppose that the structurewould be flexible with these stimuli. Thisflexibility would help both size-same-handand letter-same-hand mappings, relative tothe neither-same-hand mapping.

The RPE obtained in this experiment isimportant if it truly reflects response prep-aration. By looking at response preparationin this way, it is possible to get constraintson the order in which various sorts of infor-mation are transmitted to the response sys-tem. In a sense, then, this effect provides asmall window enabling us to observe trans-mission of information within the traditional"black box" of the human information-pro-cessing system. The observed RPE is alsoimportant because it allows rejection of mod-els in which the stimulus is processed as adiscrete whole. The RPE indicates that in-formation about different properties of aunitary stimulus becomes available to re-sponse processes at different times, so it isimpossible to characterize processing of thewhole stimulus in terms of a discrete se-quence of perception, decision, and responseoperations.

Alternative Interpretations of the RPE

Before applying the method to other stim-ulus sets, it seems important to consider pos-


sible alternative explanations for the RPEobserved in Experiment 3. The effect consistsof an advantage for conditions with similarstimuli assigned to responses on the samehand as opposed to conditions with similarstimuli assigned to responses on differenthands. RT is commonly found to depend onthe exact assignment of stimuli to responses,and effects of this nature are often attributedto "S-R compatibility." Perhaps, then, dis-crete models could be reconciled with theresults by attributing the effect to somethingother than response preparation based ontransmission of partial stimulus information.

It is difficult to reject firmly a compati-bilty-based explanation of the results of Ex-periment 3, because there is no consensusabout the mechanism(s) responsible for S-R compatibility effects (cf., Duncan, 1977a,1977b). However, it is even more difficult togive a compelling explanation of the RPEin terms of compatibility. The most clearcutexamples of S-R compatibility involve anobvious and natural correspondence betweenstimulus and response sets defined in termsof the same sorts of attributes. For example,when both stimuli and responses are definedin terms of spatial position, the most com-patible mappings are those that require aresponse in the same position as the "pre-sented stimulus (e.g., Fitts & Seeger, 1953;Morin & Grant, 1955; Rabbitt, 1967; Si-mon, 1969; Wallace, 1971). Similarly, withlinguistic stimuli and verbal responses, themost compatible mappings are those inwhich the response is simply to read thename of the stimulus (Blackman, 1975;Hawkins & Friedin, 1972; Sternberg, 1969a,1969b). However, no such correspondencecould account for the RPE in Experiment3, because the variation among stimuli wasdifferent in character from the variationamong responses.

It is possible that some of the left-to-rightstimulus orderings used in Experiment 3 aremore natural than others because of sub-jects' previous experience with the alphabet.One could perhaps argue that the letter-same-hand orderings are easier to rememberthan the others, and that this accounts forthe faster responses to them. Such an expla-nation seems somewhat implausible in viewof the limited memory demands of the task(four stimuli assigned to four responses), as

well as the invariance of the effect acrosspractice (see Figure 5). More significantly,the explanation is ruled out by the resultsof two experiments that will be describedonly briefly here.2 One experiment used thesame stimulus orderings as Experiment 3,but the four responses were made with theindex finger of the left hand and the index,middle, and ring fingers of the right hand.No advantage was obtained for the letter-same-hand orderings in this experiment, sothe advantage does not simply result fromease of remembering, the left-to-right order-ing of S-R assignments. The other experi-ment used the same stimulus orderings andresponse fingers as Experiment 3, but variedthe assignment of response fingers to re-sponse positions.3 One condition was iden-tical to Experiment 3, whereas the other re-quired subjects to cross their index fingers.In the latter condition, the four responsebuttons were pressed, from left to right, bythe left middle finger, the right index finger,the left index finger, and the right middlefinger. Responses were faster when lettersof the same name were assigned to responsefingers on the same hand, regardless of theassignment of fingers to response positions.Thus, what is important is the match of stim-ulus set structure with motor system struc-ture, exactly as predicted by the idea of re-sponse preparation. This finding is alsoembarrassing for an S-R compatibility ex-planation of the RPE, since compatibilityeffects depend on the assignment of stmiulito response position rather than responselimb (Wallace, 1971, 1972).

What is the "Grain Size" of InformationTransmission?

As noted earlier, the RPE observed inExperiment 3 seems to rule out simple dis-crete models with successive stages in whichthe stimulus is processed as a whole. How-ever, it is possible to reconcile the resultswith a discrete model in whiph the stimulusis processed in terms of two discrete com-ponents, letter identity and size. If letteridentity and size were processed separately

2 A more detailed summary of these studies is availlip. nn rRnilftstable on request,e on request.

This experiment was suggested by Geoffrey Hinton.

286 JEFF MILLER

and discretely, response preparation couldbegin as soon as processing was complete foreither characteristic. This model would pro-duce an RPE, even though it assumes dis-crete processing for both identity and size.The model accounts for the RPE by breakingup stimulus identification into two subpro-cesses, each of which has a discrete output.By allowing response preparation to beginas soon as either subprocess finishes, themodel permits response preparation to beginbefore the stimulus as a whole has been iden-tified.

The above discrete model illustrates a log-,ical problem inherent in attempting to rejectthe discrete view of information processing.Whenever information about one stimuluscharacteristic is shown to be available forresponse preparation before informationabout another stimulus characteristic, dis-crete models can be reconciled with the re-sults simply by allowing different processesto handle the two stimulus characteristics.In the limit, models could be constructedfrom discrete processes handling arbitrarilysmall units of stimulus information, and theywould be indistinguishable from continuousmodels. There would be less and less pointin calling such models "discrete," however,as their units of information transmissionbecame smaller and smaller. In a sense,models are only discrete to the extent thatthey assume transmission of relatively largeunits of information.

It appears that the discrete versus contin-uous distinction reduces to a debate aboutthe size of the units of information trans-mission (i.e., the "grain size" of informationtransmission). Continuous models state orimply that the grain size is effectively zero,so that any available information immedi-ately begins to prime responses (cf., Mc-Clelland, 1979). Discrete models, on theother hand, must claim that information istransmitted discretely with respect to somefairly information-rich internal "codes,"perhaps of the type often studied within in-formation processing psychology (e.g., Pos-ner & Taylor, 1969). In these models, thegrain size is considerably larger than zero,though it does not necessarily encompass allof the information in the whole stimulus. Thenotion of "code" required here is closely re-lated to that of Posner (1978), who said that

"by a code I mean the format by which in-formation is represented" (p. 27). The ideais that the system may represent a stimulusin terms of several internal codes (e.g., letteridentity and size). Each code could be pro-cessed discretely in the sense that no partialinformation about the code is ever madeavailable to later processes. Yet responsepreparation could begin as soon as any codewas completely activated, without waitingfor full recognition of the other relevantcodes. This model is referred to as the asyn-chronous discrete coding (ADC) model,since each code is transmitted discretely butdifferent codes activated by a single stimulusmay be transmitted at different times.

The results of Experiment 3 do not dis-criminate between continuous models andasynchronous discrete models, because eachstimulus could have been represented interms of two rather information-rich codes:letter identity and letter size. Though theexperiment demonstrates that the informa-tion grain is something smaller than a codefor the full stimulus, it does not indicate theminimum units of information transmission.These grains of information could be as fineas the level of features processed by the vi-sual system, or they could be as large as thelevel of codes for letter identity and size. Therest of the experiments in this article weredesigned to study the RPE with a variety ofstimulus sets to get more information aboutthe grain size of information transmission.

Experiment 4

This experiment tested for partial outputconcerning the single attribute of letter iden-tity by capitalizing on visual similarity re-lations among letters. Four stimulus letterswere assigned to four response keys, as inExperiment 3. In this experiment the fourletters always consisted of two pairs of vi-sually similar letters (e.g., MN and UV).Pilot testing of 12 subjects showed that dis-criminations between two letters in differentpairs were about 112 msec faster than dis-criminations between two letters in a singlepair (SE of difference = 22 msec). Thus,preliminary information, if it were availableto the response system, should produce anadvantage when two similar letters were as-signed to fingers of the same hand, as com-


pared with a condition in which similar let-ters were assigned to fingers of differenthands.

If an RPE were found with this stimulusset, it would be difficult to defend the ADCmodel. The idea of asynchronous processingfor distinct stimulus codes is reasonable forstimulus sets like those used in Experiments2 and 3. When the stimuli are four distinctletters having different names, however, itseems unlikely that they would be coded interms of anything other than letter name(Hardzinski & Pachella, 1980). Thus, re-sponse preparation with this stimulus setwould have to be based on partial outputcaused by activation within codes corre-sponding to similar letters (say M and N),rather than discrete output from a singlecode activated by both of them (an "M orN detector").

A failure to find the RPE in this experi-ment would be evidence against continuousmodels, since these models suggest that anyinformation available from the stimulus canbe used to prime responses. In such a system,partial information about letter identityshould be available to prime responses. Thus,responses should be faster when early infor-mation indicates two possible responses madeby the same hand than when this informa-tion indicates two responses made by differ-ent hands.

Method

Subjects. Subjects were 48 students recruited fromthe same pool used in the previous experiments.

Apparatus and procedure. This experiment wasidentical to Experiment 3, with the exception of thestimulus set. Each subject saw one of the followinggroups of four stimulus letters: CGKR, EFOQ, andMNUV. These groups of letters were chosen to maxi-mize intrapair similarity while minimizing interpair sim-ilarity. Half of the subjects had similar letters assignedto fingers on the same hand, and half had similar lettersassigned to fingers on different hands. Equal numbersof each of these groups were assigned to respond to eachof the three stimulus sets. Within these constraints, 48different S-R mappings were systematically selected tocounterbalance alphabetical order.

Results

Mean RTs are shown in Figure 7. On av-erage, subjects with similar letters assignedto the same hand were 45 msec faster in theexperimental blocks than the subjects with

RERcT10N

iME

1N

MSEC

850 .

BOO .

ISO .

700 .

650 .

600 .

Q-Q *

B...O

» » _

° ~~~-°

D

" • — ° SIMILflR SRME HRNDG—-° SIMILflR DIFFERENT HRND

COVflfllRTE 2 3 BLOCK 4 5 6

Figure 7. Experiment 4: average reaction time as a func-tion of experimental block and stimulus-response map-ping.

the other assignments. However, these sub-jects also averaged 30 msec faster in the co-variate blocks. Neither the ANCOVA nor anANOVA suggested that the difference be-tween the preparation conditions might bereliable (Fs between 1 and 1.5, p > .10).

Experiment 5

It seemed important to replicate this nullresult, both because a small effect in the ex-pected direction was obtained and becauseof the theoretical interest of the conclusionit supports. This experiment used a within-subjects design for added power.

MethodSubjects. Twenty-one right-handed subjects re-

cruited at UCSD were tested.Apparatus and procedure. Apparatus and procedure

were identical to those of Experiment 3 except as fol-lows. Subjects were tested in six blocks of 100 trials. Inorder to get within-subjects information on the differentmapping conditions, each subject performed in twoblocks with each group of four letters. For one groupof four letters (i.e., one pair of blocks) the four letterswere assigned to hands so that similar letters were as-signed to response fingers on the same hand. For anothergroup of letters, the four letters were assigned to keysso that similar letters were assigned to the same finger,though on different hands, For the third group of letters,the four letters were assigned to hands so that similarletters would be on both different hands and differentfingers.4

4 In this design, subjects received less practice in eachof the three S-R mappings than did the subjects inExperiment 3 (200 trials vs. 400 trials). Since the RPEdoes not seem to increase with increases in practice level(see Figure 5), however, this difference in procedureshould not work against finding an RPE.

288 JEFF MILLER


The average RTs and percentages of errorfor the three mapping conditions were: 870msec (6.8%) in the same hand condition, 846msec (8.7%) in the same-finger condition,and 867 msec (4.7%) in the both-differentcondition. An ANOVA revealed that neitherthe differences in RT nor the differences inpercentage error were reliable: for RT, F(2,36) = 0.5?, p > .20, MSe = 47,655; forpercentage error, F(2, 36) = 2.8, p > .10,MS, = 243.

Taken together, Experiments 4 and 5 pro-vide no support for the idea that partial out-put about letter identity can be used to primeresponses. Using difference scores to correctfor group differences on the covariate block,there was a 15-msec effect5 in the expecteddirection in Experiment 4. In the within-sub-jects design of Experiment 5, a 13-msec re-versal was observed. It seems clear that thesestimuli produce much less than the 85-msecpriming observed with the srST stimulus setin Experiment 3. This smaller effect cannotbe attributed to differences in relative dis-criminability in the two stimulus sets. Thepilot data indicate that the advantage of theletter discrimination over the size discrimi-nation with the srST stimulus set was com-parable to or slightly smaller than the ad-vantage of the between-pair discrimination(e.g., MU) over the within-pair discrimi-nation (e.g., MN). If priming were relatedsimply to relative discrirflinability of the twosources of information, Experiments 4 and5 should have revealed an RPE at least aslarge as the one in Experiment 3.

Experiment 6

The ADC view of decision output predictsthat response preparation should be obtainedif and only if partial stimulus informationis sufficient to activate fully a unique code.Experiments 4 and 5 indicated that when allstimuli activate distinct codes, no prepara-tion is found. In this experiment stimuli weretwo letters and two digits. These stimuli werechosen so that each one would activate twocodes. One code was a distinct, highly over-learned name code sufficient to determine

the response. The second code, letter versusdigit category, was in a sense ancillary andirrelevant to the task. On the other hand,there is considerable evidence that categoryinformation is automatically available at thesame time as or a little before identity in-formation (Jonides & Gleitman, 1972,1976).If so, these letter and digit category cpdesmay be capable of generating partial outputthat can activate responses.

As in earlier experiments, it seemed im-portant to control the relative discrimina-bility of the items that were expected to becoded together. Since the RPE depends onthe order in which information becomesavailable, this variable should have a largeeffect on the size of the RPE. If it did not,then alternative explanations of the effectwould have to be considered. Pilot work ledto the selection of three letter-number stim-ulus sets: UV49, EF17, and IS15. The firsttwo sets provided stimuli for which the dis-crimination between categories was mucheasier than the discrimination within a cat-egory (average advantage = 75 msec, SE =13 msec). The IS 15 set was just the reverse,with the discrimination between categoriesmuch slower than the discrimination withina category (average difference = 163 msec,SE = 28 msec).

Method

Subjects. Five groups of subjects all recruited atUCSD were tested, with 16 subjects per group.

Apparatus and procedure. The apparatus and pro-cedure were .identical to that of Experiment 3 except asfollows. For one group, the stimuli provided easy dis-criminations between categories (UV49 aftd EF17); thestimuli were assigned to responses so that making thecategory discrimination allowed the activation of tworesponses on the same hand. For a second group, the.same stimuli were used but the category discriminationallowed activation only of responses on different hands.

The other three groups had the IS 15 stimulus set, forwhich the between-category discrimination is relativelydifficult. Of these, one group had two members of thesame category assigned to the same hand, one grouphad visually similar items assigned to the same hand(i.e., 1 and I vs. 5 and S), and one group had visuallydissimilar members of different categories assigned tothe same hand.

5 The analogous correction based on ANCOVA ratherthan difference scores gives an effect of 21.7 msec.

\



Figure 8 shows the average RTs for thetwo groups that saw the UV49 and EF17stimulus sets. The large difference for thetwo mappings was highly significant in theANCOVA, F(l, 29) = 7.28, p < .025, Af5e =25,541, as was the effect of block, F(4,120) = 4.98. p < .01, MSe = 1,324 and thecovariate, F(l, 29) = 8.16,p < .01. The RPEdid not change significantly over blocks F(4,120) = .83, p> .10.

Figure 9 shows average RTs for threegroups responding to the IS 15 stimulus set.Though the covariate F( 1, 44) = 24, p < .01and the effect of blocks F(4, 180) = 7.79,p < .01, MSe = 1,894, were both significant,the differences among the mapping condi-tions were actually somewhat less thanwould be expected by chance, F(2, 44) =.03, p> .95, MSe = 28,814.

This is the first experiment in which anRPE has been found with four distinctlynameable stimuli, The only difference be-tween these stimuli and the visually similarletters used in Experiments 4 and 5 is thatthere was a categorical difference here: let-ters versus digits. It seems clear that thiscategory difference was responsible for pro-ducing the RPE, so these results provide sup-port for the idea that the RPE depends ondiscrete activation of a code. Even when thiscode is ancillary to the final decision, infor-mation transmitted through the code can beused to prepare responses.

750 -

700 _

650 .

600 .

-« CRTEGORY SRME HflND••" CRTEGORY DIFFERENT HflND

COVflRIflTE 3 4BLOCK

Figure 8. Experiment 6: average reactoin time as a func-tion of experimental block and stimulus-response map-ing for the EF17 and UV49 stimulus sets.

750 .

700 .

650 .

' • PHYSICAL SflME HflND•"• ° CflTESORY SflME HflNO"—'NEITHER

COVflRIRTE .2 3 HBLOCK

Figure 9. Experiment 6: average reaction time as a func-tion of experimental block and stimulus-response map-ping for the IS 15 stimulus set.

It is important to emphasize the differencein findings between this experiment and theexperiments with similar letters, because thisdifference removes some of the difficultiesinherent in accepting the null hypothesiswith respect to the experiments with letters.The continuous model predicts RPEs in bothexperiments, since it does not acknowledgethe importance of discrete code activationin the transmission of information. That theeffect was large in one experiment and smallor nonexistent in the other is positive evi-dence against the continuous model. Thus,these results are more than simply a failureto confirm a prediction of the continuousmodel. Similarly, the finding of response ac-tivation only when visually similar items be-longed to the same category (compare Fig-ures 8 and 9) is also evidence against thecontinuous model. In total, the results pro-vide rather strong evidence for the impor-tance of code activation in the transmissionof evidence about stimulus characteristics.

Experiment 7

To this point, the question of whether re-sponse preparation begins before stimulusidentificatipn is complete has been examinedonly with alphanumeric stimulus sets. Re-sults suggest the^generalization that responsepreparation can occur only when early in-formation discretely activates an internalcode used in representing the stimulus. Ex-

290 JEFF MILLER

Figure 10. Experiment 7: stimulus set.

periments 7 and 8 examine this generaliza-tion in the context of geometric stimulussets.

The purpose of Experiment 7 was to es-tablish that an RPE can be obtained withgeometric stimuli that would activate ap-propriate internal codes. The stimuli areshown in Figure 10. Both intuition and datafrom a previous study (Miller, 1979) suggestthat these stimuli would be coded in termsof two unequally salient binary attributes:shape of central figure (more salient) andlocation of horizontal bar (less salient). Theprevious study showed that RT decreases notonly with increases in the probability of thetest stimulus itself, but also with increasesin the probability of the other stimulus withthe same central shape. This finding supportsthe intuition of binary coding, since it sug-gests two stimuli with the same central shapejointly influence the level of activation in thecode for that shape. Given the hypothesisthat code activation is responsible for re-sponse preparation, then, it seems that in-formation about the central shape of thestimulus should produce response prepara-tion.

MethodSubjects. Subjects were 32 right-handed undergrad-

uate students recruited at UCSD.Apparatus and procedure. The apparatus and pro-

cedure were identical to that of Experiment 3, with theexception of the stimulus set. Two classes of S-R map-

pings were used: same shape on the same hand, andsame bar location on the same hand. Sixteen subjectswere tested in each of the two mapping conditions.


Average RT as a function of experimentalblock and type of S-R mapping is shown inFigure 11. An ANCOVA RT indicated thatthe main effects of mapping F(l, 29) = 13,^<.01, MSe= 18,397, and block, F(4,120) = 19, p < .01, MSe = 3,279, were sig-nificant, as was the covariate, F(l, 29) =64.5, p < .01, MSe = 18,397. The 79-msecRPE obtained with these stimuli demon-strates that response preparation can occurwith codes other than highly overlearned,alphanumeric ones.

Experiment 8

Since the previous experiment demon-strated that response preparation effects canbe obtained with geometric in addition tolinguistic stimuli, it is again of interest toask whether the effect depends on completeactivation of codes used in recognizing thestimuli, or whether any kind of preliminaryinformation can be used to prepare re-sponses. The purpose of this experiment wasto look for evidence of response preparationwith a geometric stimulus set that wouldhave unique codes for each stimulus. Thestimuli are shown in Figure 12. These stimuliwere also used in the previous studies of

1000 .

900 .

• SHflPE SflME HflNO° BflR SflME HflND

3 4BLOCK

Figure 11. Experiment 7: average reaction time as afunction of experimental block and stimulus-responsemapping.


probability effects (Miller, 1979), but it wasfound that RT to one of these stimuli de-pended only on the probability of that stim-ulus and not on the probability of the similarstimulus. This suggests that these stimuli arecoded using unique codes for each stimulusrather than using two shared binary codesas were the stimuli in Figure 10. If responsepreparation depends on stimulus coding evenfor geometric stimuli, then, there should belittle or no RPE with these stimuli.

Method

Subjects. Subjects were 32 right-handed undergrad-uate students recruited at UCSD.

Apparatus and procedure. The apparatus and pro-cedure of this experiment were identical to that of theprevious experiment, with two exceptions. First, ob-viously, a different stimulus set was used. Second, sincethe stimulus set was not composed of binary attributes,it was impossible to construct mappings logically equiv-alent to the "bar-same-hand" mappings in the previousstudy. The two sets of mappings studied in this exper-iment were those with the figure (X vs box) on the sameor on different hands.


Average RT as a function of experimentalblock and S-R mapping are shown in Figure13. An ANCOVA indicated that the main ef-fect of block, F(4, 12) = 12.8, /?<.01,MS, = 2,779, and the covariate, F(l, 29) =20.6, p < .01, MSe = 38,548, were signifi-cant. However, the 3 8-msec effect of map-

1000

700 .

600 .

~° SHRPE SflME HflND•'" SHRPE DIFFERENT HflND

COVRRIflTE 3 1BLOCK

Figure 12. Experiment 8: stimulus set.

Figure 13. Experiment 8: average reaction time as afunction of experimental block and stimulus-responsemapping.

ping condition was not statistically reliable,F(l, 29) = 1.47,^ > .20, MS. = 38,548. Thefinding of a smaller, nonsignificant RPE withthese stimuli thus supports the idea that re-sponses are only (or at least primarily) pre-pared through the activation of codes usedin recognizing the stimuli. The differencebetween this experiment and the previousone, under this theory, can be accounted forby arguing that subjects code the Figure 10stimuli in terms of distinct binary codes, butuse unique codes for each of the four stimuliin Figure 12.

General Discussion

This series of experiments shows how re-sponse preparation effects can be used todetermine when different units of stimulusinformation are transmitted to response ac-tivation processes. The technique is based onthe idea that manipulating the efficiency ofresponse preparation will produce an effectonly when response preparation actuallytakes place. Preparation efficiency was ma-nipulated by allowing preparation of two fin-gers on the same hand or on different hands.

The logic of the method is most easily il-lustrated by considering stimulus sets con-structed from two binary dimensions. Theamount of time available for preparation wascontrolled by varying the relative discrimin-abilities of the two dimensions. The S-Rmappings controlled which dimension, if ei-

292 JEFF MILLER

ther, determined the hand to be used in theresponse and thus controlled the type ofpreparation that could be made given infor-mation about one particular dimension.

Within this framework, the effect of re-sponse preparation (RPE) associated witha dimension was measured as the differencein overall RT between a condition in whichthat dimension determined which hand wouldrespond and a condition in which neitherdimension alone was sufficient to determinewhich hand would respond. Significant RPEswere taken to imply that information aboutthe dimension was indeed available to pre-pare possible responses before a unique re-sponse was determined. In other words,RPEs were taken as evidence that all of theinformation about the stimulus did not reachthe response preparation mechanism at thesame time, but rather information about thepriming dimension reached the responsemechanism before the other informationneeded to specify the response.

RPEs were also logically possible withstimulus sets not easily classified in terms oftwo binary dimensions. In principle an RPEcould be observed whenever easy-to-perceiveinformation restricted the possibilities to tworesponses, with a more difficult discrimina-tion being required to decide between those.For example, the stimuli could be four dis-tinct letters with strong visual similaritieswithin two pairs (e.g., MNUV). With suchstimulus sets it was possible that the easilyperceivable information could be used toprepare responses while the more difficultdiscrimination was made. As with the binarydimension stimuli, this would result in anadvantage for a condition in which visuallysimilar stimuli were assigned to responses onthe same hand, as compared with a conditionin which visually similar stimuli were as-signed to responses on different hands.

In the present analysis of response prep-aration, the debate between discrete andcontinuous models hinges on the unit or"grain" size of information transmission. Atone extreme are fully continuous models, inwhich any information about the stimulusbecomes available to "prime" later process-ing mechanisms. Fully continuous modelsthus predict RPEs with all stimulus sets. Atthe other extreme are fully discrete models,

in which no information about a stimulushas been processed by that stage. Fully dis-crete models predict that RPEs should neverbe obtained with any stimulus set. Modelsintermediate between fully discrete and fullycontinuous vary according to the grain sizeof information transmission. A model maybe regarded as discrete to the extent that itrequires transmission of relatively largechunks of stimulus information and contin-uous to the extent that it allows transmissionof relatively small chunks of stimulus infor-mation.

Results from a variety of stimulus setsplace some interesting bounds on the grainsize of information transmission. Contraryto fully discrete models, complete informa-tion about a stimulus need not be availablebefore response preparation can begin. Con-trary to fully continuous models, all easilyperceived stimulus information is not im-mediately transmitted to response prepara-tion processes. It appears that responses canbe prepared on the basis of preliminary in-formation that completely activates an in-ternal code used in classifying a stimulus,but not on the basis of information that par-tially activates such a code. The only de-monstrable instances of response prepara-tion, in fact, were cases in which the stimuluscould easily be represented using multiplecodes, and one code was available well beforethe other. Thus, it appears that information-processing stages transmit information aboutdimensions by means of discrete codes, bothfor semantic/linguistic stimulus dimensionsand for physical stimulus dimensions (cf.,Posner, 1978, chap. 3). The RPEs werefound only when a code corresponding to anobvious stimulus dimension could be acti-vated discretely to provide preliminary in-formation about the stimulus. No prepara-tion effects were found when the informationavailable early could not be readily codedin a discrete fashion. These results thus sup-port a model, referred to here as the asyn-chronous discrete coding, or ADC, model,in which partial information abut a stimuluscan be transmitted only when the informa-tion is complete with respect to an internalcode.

The evidence in support of this general-ization is strong. The RPEs were found when


early information indicated that a stimuluswas of a particular size, that a stimulus wasa particular letter, or that a stimulus be-longed to a particular category (letter ordigit). However, no RPE was found whenearly information constrained the stimulusto be one of two distinct letters (MN or UV)or one of two distinct alphanumeric char-acters in different categories (II or S5). Thecontrast between Experiments 4 and 6 isespecially strong support for the notion thatoutput is1 discrete with respect to codablestimulus properties. Both experiments usedtwo pairs of visually similar stimuli, and ev-ery stimulus had a unique and highly over-learned name in both studies, Responsepreparation was obtained only when therewas a letter versus digit category differencebetween the two pairs of visually similarstimuli. Thus, preliminary information aboutvisual characteristics seems to be useful forresponse preparation only if it can be usedto activate a discrete code by which the in-formation can be transmitted.

Results from two studies with geometricstimuli also support the importance of cod-ing for information transmission. When geo-metric stimuli were constructed from or-thogonal dimensions, a large preparationeffect was found. When the stimuli were notconstructed from orthogonal dimensions,however, a much smaller (possibly zero) ef-fect of response preparation was obtained.These results are exactly what would be pre-dicted from the hypothesis, supported byprevious evidence (Miller, 1979), that geo-metric stimuli constructed from orthogonaldimensions tend to be coded in terms of thesedimensions whereas stimuli constructed fromnonorthogonal properties tend to be codedas independent wholes.

Even the results of Experiment 2 suggest,in retrospect, the importance of stimuluscoding for response preparation effects.Stimuli in that experiment were the four let-ter pairs BE, BO, ME, and MO, and the twoletters in a stimulus pair were presented atslightly different times. Response prepara-tion was obtained when the consonant waspresented before the vowel, but not when thevowel was presented before the consonant.This result is easy to interpret in terms ofa model in which stimulus coding is con-'

strained to process letter pairs in the normalleft-to-right reading order. It is hard to seehow to reconcile the result with a fully con-tinuous model, however. On logical grounds,information provided by the vowel shouldhave been juist as useful for response prep-aration as information provided by the con-sonant.

The idea of an information-processingstage is an old and honored one within ex-perimental psychology (Donders, 1969; Jas-trow, 1890; Wundt, 1880), and the work re-ported here suggests there is real merit inthe view of processing as a series of discreteoperations. In general, a stage may be de-fined as a process or set of processes thatreceive input only after necessary prior pro-cesses have finished completely and transmitoutput only when all processes performed bythe stage have completely finished. In otherwords, it seems reasonable to say that a stageneither receives nor sends partial output, Ofcourse, it is important to be explicit aboutthe units of information (codes) that a stagereceives as input and transmits as output.Also, in specifying the processing accom-plished by a stage, it is important to be ex-plicit about what processes cannot begin un-til the stage is finished. The discrete outputproperty requires that a stage be defined rel-ative to a subsequent process that awaits itscompletion, so a stage cannot be defined inisolation.

A definition of a stage, of course, is onlyuseful to the extent that there is evidence forthe existence of such an entity. The presentwork has shown that the processes respon-sible for preparing responses seem to receiveinformation only about complete activationof internal codes used in representing thestimulus. Thus, it appears that responsepreparation really is a stage separate fromthe perceptual and decision processes. Al-though this does not demonstrate discreteinformation transmission between otherstages classically assumed to be involved inRT tasks (e.g., transmission from perceptionto decision), it certainly raises the a priorilikelihood of discreteness at other points inthe system.

The idea being proposed on the basis ofthese experiments is simply that there existsome points of processing for which output

294 JEFF MILLER

is discrete with respect to some units of in-formation. Clearly, processing is not alwaysdiscrete for the stimulus as a whole. Nor isthere any reason to believe that processingis discrete at all points, so the results shouldnot be taken as evidence against models withcontinuous processing within a stage. Con-sider, for example, the reading model ofRumelhart (1977). In this model, the word-identification process is based on a complexinteractive system examining stimulus evi-dence in the light of an ever-more-complexset of constraints generated by other partsof the stimulus field. Partial activations offeature codes continuously activate lettercodes, which in turn continuously activateword codes. All this continuous processinghappens within the stage responsible forword identification, however. The word-identification process might transmit infor-mation to response processes discretely, eventhough the information built up as the resultof processes with continuous output. Simi-larly, Turvey (1973) proposed a concurrent-contingent continuous-output model of per-ceptual processing on the basis of maskingdata. He concluded that partial output fromthe level of crude visual features was! avail-able to letter-recognition processes before allfeatures had been extracted. Thus his con-clusions are limited to perceptual processes,and are not incompatible with the conclusionthat information about letters is transmitteddiscretely to response activation processes.

For purposes of modeling RT, the presentresults suggest a formalism in which pro-cessing is discrete for codable stimulus com-ponents or dimensions rather than for.wholestimuli. For stimulus sets represented interms of unitary internal codes (e.g., mem-ory scanning with letters), the results suggestthat the discrete formalism required by theadditive factor method (Sternberg, 1969a,1969b) is reasonable. For stimulus setscoded in terms of multiple dimensions, how-ever, the formalism that seems most appro-priate is that used in critical path analysis(Kelley, 1961; Schweikert, 1978; Wiest &Levy, 1969). This formalism is a general-ization of the additive factor method inwhich each process is contingent on somebut not necessarily all of the prior processes.

For example, a decision process using letteridentity would be contingent on perceptualprocesses responsible for recognizing iden-tity but not on perceptual processes that de-termined size. In critical path analysis, aprocess cannot begin until all processes onwhich it is contingent have completely fin-ished, but several different processes can op-erate at the same time as long as they arenot contingent (e.g., identity decision andperception of size).

Schweikert (Note 1) discussed severalpatterns of data that can be accounted forusing critical path analysis but that are in-compatible with a simple additive factormethod analysis. He showed some data fromdouble-stimulation experiments that fit thesepatterns, and concluded the assumptions ofthe additive factor method are violated whensuch stimuli are used. This is exactly whatwould be expected based on the ADC model,since more than one code would almost cer-tainly be needed to represent more than onestimulus. Similarly, Bauer (1981) has de-monstrated patterns of data favoring criticalpath analysis over the additive factor methodframework in experiments with single stim-uli having multiple dimensions.

In conclusion, it should be noted that themost important result of this work may bethe introduction of a new method for study-ing the transmission of information throughthe human perceptal-motor system. Themethod used here can easily be extended toa variety of stimulus sets, tasks, and responsedomains. Ultimately, this method may pro-vide an important tool for learning aboutboth the time course of response preparationand the codes of information transmission.

Reference Note

1. Schweikert, R. A generalization of the additive fac-tor method with an application to psychological re-fractoriness. Paper presented at the annual meetingof the Psychonomic Society, St. Louis, November1976.

References

Anderson, J. Neural models with cognitive implications.In D. LaBerge & S. Samuels (Eds.), Basic processesin reading: Perception and comprehension, Hillsdale,N.J.: Erlbaum, 1977.


Anderson, J., Silverstein, J., Ritz, S., & Jones, R. Dis-tinctive features, categorical perception, and proba-bility learning: Some applications of a neural model.Psychological Review, 1977, 84, 413-451.

Bauer, D. A critical path analysis of perceptual pro-cesses. Unpublished doctoral dissertation, Universityof California, San Diego, 1981.

Beller, H. Priming: Effects of advance information onmatching. Journal of Experimental Psychology, 1971,87, 176-182.

Blackman, A. Test of the additive-factor method ofchoice reaction time analysis. Perceptual and MotorSkills, 1975, 41, 607-613.

Brinkman, J., & Kuypers, H. Cerebral control of con-tralateral and ipsilateral arm, hand and finger move-ments in the split-brain rhesus monkey. Brain, 1973,96, 653-674.

Broadbent, D. Perception and communication. NewYork: Pergamon Press, 1958.

Crosby, E., Humphrey, T,, & Lauer, E. Correlativeanatomy of the nervous system. New York: Mac-millan, 1962.

Doll, T. Short-term retention and preparatory set ascovert rehearsal. Journal of Experimental Psychol-ogy, 1969, 82, 175-182.

Donders, F. On the speed of mental processes. InW. G. Koster (Ed. and trans.), Attention and per-formance II. Amsterdam: North-Holland, 1969.(Reprinted from Ada Psychologica, 1969, 30, 412-431.)

Duncan, J. Response selection errors in spatial choicereaction tasks: Quarterly Journal of ExperimentalPsychology, 1977, 29, 415-423. (a)

Duncan, J. Response selection rules in spatial choicereaction tasks. In S. Dornic (Ed.), Attention and Per-formance VI. Hillsdale, N.J.: Erlbaum, 1977. (b)

Eriksen, C., & Hoffman, J. Temporal and spatial char-acteristics of selective encoding from visual displays.Perception & Psychophysics, 1972, 12, 201-204.

Eriksen, C., & Hoffman, J. The extent of processing ofnoise elements during selective encoding from visualdisplays. Perception & Psychophysics, 1973,14, 155-160.

Eriksen, C., & Schultz, D. Information processing invisual search: A continuous flow conception and ex-perimental results. Perception & Psychophysics, 1979,25, 249-263.

Fitts, P., & Seeger, C. S-R compatibility: Spatial char-acteristics of stimulus and response codes. Journalof Experimental Psychology, 1953,4(5, 199-210.

Gazzaniga, M. The bisected brain. New York: Apple-ton-Century-Crofts, 1970.

Goodman, D., & Kelso, J. Are movements prepared inparts. Not under compatible (naturalized) conditions.Journal of Experimental Psychology: General, 1980,109, 475-495.

Hardzinski, M., & Pachella, R. The manipulation ofstimulus quality and the definition of stimulus encod-ing operations in memory scanning experiments. Per-ception & Psychophysics, 1980, 27, 232-240.

Hawkins, H., & Friedin, B. The relative frequency ef-fect and S-R compatibility. Psychonomic Science,1972, 28, 329-330.

Jastrow, J. The time-relations of mental phenomena.Fact & Theory Papers No. VI. New York: N. D. C.Hodges, 1890.

Jonides, J., & Gleitman, H. A conceptual category ef-fect in visual search: O as letter or as digit. Perception& Psychophysics, 1972, 12, 457-466.

Jonides, J., & Gleitman, H. The benefit of categoriza-tion in visual search: Target location without identi-fication, Perception & Psychophysics, 1976, 20, 289-298.

Kelley, J. Critical path planning and scheduling, math-ematical basis. Operations Research, 1961, 9, 296-320.

Kerr, B. Task factors that influence selection and prep-aration for voluntary movements. In G. Stelmach(Ed.), Information processing in motor control andlearning. New York: Academic Press, 1978.

Klapp, S. Response programming as assessed by reac-tion time does not establish commands for particularmuscles. Journal of Motor Behavior, 1977, 9, 301-312.

Kornblum, S. Response competition and/or inhibitionin two-choice reaction time. Psychonomic Science,1965, 2. 55-56.

LaBerge, D., Van Gelder, P., & Yellot, J. A cueingtechnique in choice reaction time. Perception & Psy-chophysics, 1970, 7, 57-62.

McClelland, J. On the time relations of mental pro-cesses: A framework for analyzing processes in cas-cade. Psychological Review, 1979, 86, 287-330.

Miller, J. Cognitive influences on perceptual processing.Journal of Experimental Psychology: Human Per-ception and Performance, 1979, 5, 546-562.

Miller, J., & Bauer, D. Visual similarity and discrim-ination demands. Journal of Experimental Psychol-ogy: General, 1981, 110, 39-55.

Morin, R., & Grant, D. Learning and performance ofa key-pressing task as a function of the degree ofspatial stimulus-response correspondence. Journal ofExperimental Psychology, 1955, 49, 39-47.

Neely, J. Semantic priming and retrieval from lexicalmemory: Roles of inhibitionless spreading activationand limited-capacity attention. Journal of Experi-mental Psychology: General, 1977, 106, 226-254.

Neisser, U. Cognitive psychology. New York: Appleton-Century-Crofts, 1967.

Norman, D., & Bobrow, D. On data-limited and re-source-limited processes. Cognitive Psychology, 1975,7, 44-64.

Posner, M. Chronometric explorations of mind. Hills-dale, N.J.: Erlbaum, 1978.

Posner, M., & Snyder, C. Facilitation and inhibition inthe processing of signals. In P. Rabbitt & S. Dornic(Eds.), Attention and performance, V. New York:Academic Press, 1975.

Posner, M., Snyder, C., & Davidson, B. Attention andthe detection of signals. Journal of ExperimentalPsychology: General, 1980, 109, 160-174.

Posner, M., & Taylor, R. Subtractive method appliedto separation of visual and name components of multi-letter arrays. Acta Psychologica, 1969, JO, 104-114.

Rabbitt, P. Signal discriminability, S-R compatibility,

296 JEFF MILLER

and choice reaction time. Psychonomic Science, 1967,7, 419-420.

Rosenbaum, D. Human movement initiation: Specifi-cation of arm, direction and extent. Journal of Ex-perimental Psychology: General, 1980,109, 444-474.

Rumelhart, D. E. Toward an interactive model of read-ing. In S. Domic (Ed.),'Attention and PerformanceVI. Hillsdale, N.J.: Erlbaum, 1977.

Sanders, A. Structural and functional aspects of thereaction process. In S. Domic (Ed.), Attention andPerformance VI. Hillsdale, N.J.: Erlbaum, 1977.

Schweickert, R. A critical path generalization of theAdditive Factor Method: Analysis of a Stroop task.Journal of Mathematical Psychology, 1978,18, 105-139.

Simon, J. Reactions toward the source of stimulation.Journal of Experimental Psychology, 1969,81, 174-176.

Smith, E. Choice reaction time: An analysis of the majortheoretical positions. Psychological Bulletin, 1968,69, 77-110.

Sperling, G. The information available in brief visualpresentations. Psychological Monogrpahs, 1960, 74,11.

Sternberg, S. The discovery of processing stages: Ex-tensions of Donders' method, Acta Fsychologica,1969, 30. 276-315. (a)

Sternberg, S. Memory scanning: Mental processes re-vealed by reaction-time experments. American Sci-entist, 1969, 57. 421-457. (b)

Taylor, D. Stage analysis of reaction time. Psycholog-ical Bulletin, 1976,55, 161-191.

Turvey, M. On peripheral and central processes in vi-sion: Inferences from an information-processing anal-ysis of masking with patterned stimuli. PsychologicalReview, 1973, 80, 1-52.

Wallace, R. S-R compatibility and the idea of a re-sponse code. Journal of Experimental Psychology,1971,55, 354-360.

Wallace, R. Spatial S-R compatibility effects involvingkinesthetic cues. Journal of Experimental Psychol-ogy, 1972, 93, 163-168.

Wiest, J. & Levy, F. A management guide to PERT/CPM. Englewood Cliffs, N.J.: Prentice-Hall, 1969.

Wiley, J. Hemispherectomy and consequent motor be-havior. In W,. Spirduso & J. King (Eds), Proceedings:Motor Control Symposium. Austin: University ofTexas Press, 1975.

Wundt, W. Grundzuge der physiologischen Psychologie(Vol. 2, 2nd ed). Leipzig: Engelmann, 1880.

Received April 13, 1981

Documents

Discrete Versus Continuous Stage Models of Human