information reduction during skill acquisition: the ... · than participants under accuracy stress....

Journal of Experimental Psychology: Applied1999, Vol. 5, No. 2,129-151

Copyright 1999 by the American Psychological Association. IP.C.i076-898X/99/.$3.00

Information Reduction During Skill Acquisition:The Influence of Task Instruction

Hilde HaiderUniversity of the Armed Forces

Peter A. FrenschHumboldt-Universitv Berlin

H. Haider and P. A. Frensch's (1996) information reduction hypothesisholds that with practice, people leam to distinguish task-relevant fromtask-redundant information and to ignore task-irrelevant information. In 2experiments, the authors examined whether degree of information reduc-tion can be directly affected by task instruction. Participants verifiedalphabetic arithmetic tasks containing task-relevant and task-irrelevantinformation. Participants were asked to optimize their accuracy or theirspeed of performance throughout the entire experiment or to optimizeaccuracy for half of the experiment and speed for the other half of theexperiment or vice versa (first speed, then accuracy). In addition, timeduration of stimulus presentation under speed instruction was systemati-cally reduced over practice in Experiment 2. Results showed that amount ofinformation reduction was affected by instructions, suggesting that theprocess of information reduction is at least partially under voluntarycontrol.

Think of two people, one an experienced userof computer software and the other an inexperi-enced user, both trying to install some newcomputer software. During the installation pro-cess, a number of messages are displayed on thescreen, some of which need to be attended to andsome of which may be safely ignored. Whereasthe inexperienced user is likely to thoroughlyread through all messages displayed on thescreen, the skilled user very likely pays attention

Hilde Haider, Department of Psychology, Univer-sity of the Armed Forces, Hamburg, Germany; PeterA. Frensch, Department of Psychology, Humboldt-University Berlin, Berlin, Germany.

This research was supported, in part, by German-American Commission on Collaborative ResearchGrant DAAD-ACLS.

Correspondence concerning this article should beaddressed to Hilde Haider, who is now at the Psycho-logisches Institut n, Universitat Hamburg, Von MellePark 5, D-20146 Hamburg, Germany, or to PeterA. Frensch, Institut fur Psychologic, Humboldt-Universitat zu Berlin, Hausvogteiplatz 5-7, D-10117Berlin, Germany. Electronic mail may be sent [email protected] or to peter.frensch®psychologie.hu-berlin.de.

to only those messages that require him or her torespond and more or less ignores all othermessages. Consequently, the skilled user needsless time to install the new software than does theunskilled user.

What this example suggests is that the transi-tion from unskilled to skilled task performancecan be partly characterized in terms of an increas-ing ability to sort out task-relevant from task-irrelevant information. In general, skilled perform-ers are likely to know which of the giventask-information is relevant and which is irrel-evant and are therefore able to focus their process-ing on task-relevant information. Unskilled per-formers, in contrast, do not possess the knowledgethat would let them distinguish between task-relevant and task-irrelevant information. Oneaspect of acquiring a skill, then, appears to be theability to limit task processing to task-relevantinformation, an ability that we refer to as informa-tion reduction (Haider & Frensch, 1996).

Surprisingly, this aspect of a skill has receivedlittle attention in most current theories of skillacquisition (e.g., Anderson, 1983, 1987, 1992;Lassaline & Logan. 1993; Logan, 1988; Logan &Etherton, 1994; but see, Anderson, Matessa. &

129

130 HAIDER AND FRENSCH

Douglass, 1995). According to most theories,practice-related improvements in the perfor-mance of a skill (e.g., increasing speed andefficiency; decreasing effort) are due to (a) quali-tative changes in the task processing (e.g., Cheng,1985; Logan, 1988), (b) an increasing efficiencyto perform individual task components (e.g.,Anderson, 1982,1987,1992), (c) increases in theefficiency to perform sequences of task compo-nents (e.g., Anderson, 1982,1987; Frensch, 1991,1994; Newell & Rosenbloom, 1981), or (d) somecombination of these factors (e.g., Anderson,1987; Frensch & Geary, 1993). In general, exist-ing theories view performance changes that occurwith increasing level of skill as being a result ofan increased efficiency of information process-ing. In addition, the learning mechanisms under-lying skill acquisition are typically seen as mecha-nistic and entirely data-driven (e.g., Anderson,1987; Logan, 1988; Newell & Rosenbloom,1981).

The neglect of information reduction in theo-ries of skill acquisition is even more surprising ifone considers mat in a great number of everydaysituations, such as driving or using modernelectronic information systems (e.g., Internet),information has to be processed efficiently. Also,the phenomenon of information reduction iscommonly acknowledged in many different areasof psychology (e.g., Christensen et al., 1981;Fisher & Tanner, 1992; Holding, 1985; Lambert,Spencer, & Mohindra, 1987; Myles-Worsley,Johnston, & Simons, 1988; Shapiro & Raymond,1989; Strayer & Kramer, 1994a, 1994b). Thephenomenon has been discussed theoretically forthe domains of visual perception (e.g., EL J.Gibson, 1963; J. J. Gibson & Gibson, 1955;Kaptein, Theeuwes, & Van der Heijden, 1995;Van der Heijden, 1992), sports and perceptualmotor skills (e.g., Abemethy, 1993; Helsen &Pauwels, 1993; Petrakis, 1993), educational psy-chology (e.g., Bransford, Sherwood, Vye, &Rieser, 1986), and expertise (e.g., Ericsson &Lehmann, 1996; Shanteau, 1992; Shapiro &Raymond, 1989; Vickers, 1988). For example,Helsen and Pauwels (1993) analyzed the eyemovements of experts and novices in tacticalsoccer situations and found that experts differedfrom nonexperts in the selectivity of information.Similarly, Shapiro and Raymond (1989) wereable to show that novices' performance could be

improved when novices were trained to attend totask-relevant information only and to ignoretask-irrelevant information. Taken together, theseobservations suggest the acquisition of an internal-ized evaluation function that helps to distinguishbetween task-relevant and task-redundant infor-mation and to base task processing on the rel-evant aspects of a task. Thus, an importantquestion in skill acquisition is how people learnto selectively use information with increasingpractice.

We have recently formulated a theoreticalaccount of how information reduction mightproceed when a cognitive skill is acquired (Haider& Frensch, 1996). Our theoretical view relates tofindings we obtained with the alphabetic arith-metic task (AAT). In the AAT, participants areasked to verify alphabetic letter strings, that is, tojudge whether a presented string follows thealphabet or not. Strings consist of an initialletter-digit-letter triplet, and a varying number ofadditional letters (e.g., D [4] I, E[4]JKL,D [4]J K, and E [4] J K L), and are either correct (i.e.,they follow the alphabet; D [4] I) or incorrect(i.e., they do not follow the alphabet; D [4] J).Because errors in incorrect strings always occurat String Position 3 (i.e., the letter to the right ofthe digit), the letters in String Positions 4 andhigher are always correct and are therefore irrel-evant for the verification task.

Because the total number of letters variesacross strings, one can infer whether or notparticipants ignore the task-irrelevant informa-tion from the magnitude of the string-lengtheffect for correct strings. Specifically, if partici-pants verify the entire alphabetic string, thenverification time should systematically increasewith string length. However, if participants havelearned to limit their task processing to therelevant letter-digit-letter triplet, then verifica-tion time should be independent of string length.The magnitude of the string-length effect forcorrect strings is thus an indirect measure of thedegree to which task-irrelevant information isignored.

For our initial set of experiments with the AAT,we reported two main results (cf. Haider &Frensch, 1996). First, the effect of string lengthon verification times for correct strings (i.e.,increased verification times for longer strings)declined significantly with task practice and

INFORMATION REDUCTION 131

disappeared altogether after approximately 500trials, indicating that the task-irrelevant informa-tion was no longer processed. Second, the magni-tude of the string-length effect was unaffected bysudden and unexpected changes in the stimulusmaterial. We interpreted these results in terms ofa two-stage information reduction process: Dur-ing the first phase, task-relevant and task-irrelevant information are distinguished, and dur-ing the second stage, task-relevant information isactively selected for processing (e.g., Allport,1987; Neumann, 1990), and task-irrelevant infor-mation is actively ignored (e.g., Neisser &Becklen, 1975). In general, the two-stage view isconsistent with arguments that have been made inother psychological contexts (e.g., J. J. Gibson &Gibson, 1955; Neisser, 1976; Trabasso & Bower,1968).

We further theorized that the ability to distin-guish between relevant and irrelevant informa-tion results from implicit (bottom-up) and/orexplicit (top-down) learning mechanisms. In con-trast, the active selection of relevant informationand the ignoring of irrelevant information is theresult of a voluntary strategic decision to nolonger attend to the redundant information (Haider& Frensch, 1999).

Purpose of the Present Experiments

The assumption that limiting task-processingto the task-relevant information is under strategiccontrol has important practical implications. Ifinformation reduction is indeed at least partiallythe result of a voluntary decision, then one shouldbe able to externally influence the degree towhich information reduction mechanisms areused. Insight in external facilitating and inhibit-ing conditions may eventually lead to the capac-ity to accelerate information reduction when thisis wanted (e.g., implementing software by anexperienced user) or decelerate it when it needsto be prevented (e.g., airline pilots' preparationsfor a flight).

The primary purpose of the present studies wasto experimentally explore whether degree ofinformation reduction can be manipulated exter-nally. One possible, and relatively straightfor-ward, way to do just that is to instruct participantsto optimize either their speed or accuracy ofperformance. We hypothesized that manipulating

task instructions in this manner may have at leastthree different consequences. First, in accordancewith a large literature on speed-accuracy trade-offs (Pachella, 1974; Pachella & Pew, 1968;Rabbitt, 1989; Sperling & Dosher, 1986; Wickel-gren, 1977), participants under speed stress shoulddemonstrate faster latencies and higher error ratesthan participants under accuracy stress. Second,and more important in the present context, weexpected that participants under speed stresswould terminate processing of task-irreievantinformation earlier during practice than partici-pants under accuracy stress. Third, because infor-mation reduction is assumed to be under aparticipant's voluntary control, instruction shouldaffect degree of information reduction, irrespec-tive of whether speed or accuracy instructions aregiven at the beginning of or during training.

In relation to the first hypothesized outcome, itis well known in the response-time literature thatparticipants can adjust their speed and accuracyof responding as a function of task instruction.The preferred explanation is that participantsadopt different response criteria for differentinstructions: a conservative one under accuracystress and a liberal one under speed stress(Pachella, 1974; Pachella & Pew, 1968; Rabbitt,1989; Sperling & Dosher, 1986; Wickelgren,1977). Frequently, random walk models are usedto capture the effects of setting response criteriaon latencies and accuracy (e.g., RatclifL 1981,1985; Treisman & Williams, 1984). In thesemodels, it is assumed that setting response crite-ria affects both the speed of processing taskinformation and the amount of information—extracted from a stimulus—that is necessary totrigger a response. If the response criterion ismoved closer to the starting point (i.e., where theaccumulation of information begins), then theamount of information that is needed to trigger aresponse is reduced. If the criterion is movedfarther away from the starting point, then theamount of information necessary to trigger aresponse is increased.

In the language of random walk models, partici-pants instructed to optimize speed of processingare likely to adopt a liberal response criterion,meaning that these participants will increase theirspeed of task processing and/or will strategicallymove their response boundaries closer to thestarting point. In contrast, participants instructed


to optimize accuracy of performance are morelikely to adopt a conservative response criterion,meaning that these participants will strategicallymove their response boundaries farther awayfrom the starting point or will process taskinformation more slowly.

With regard to the second hypothesized out-come, the setting of a response criterion needs, inprincipal, to be distinguished from the decision ofwhich stimulus information the response crite-rion is applied to. Thus, for example, a participantmay decide to use a liberal response criterion oninformation selected randomly from the stimulus.Or, alternatively, the participant may use a liberalresponse criterion on information that is extractedsystematically, for example from left to right, ifthe stimulus is a written word.

The empirical question of how the manipula-tion of task instruction affects degree of informa-tion reduction relates to the question of whichtask information a selected response criterion isapplied to. On the one hand, one possibility is toconsider the setting of a response criterion asindependent of the decision of which task infor-mation is processed. Deciding to respond quicklyis, after all, a different decision than deciding toprocess only parts of the presented task informa-tion. According to the independence view, thepossible prediction that speed stress will leadparticipants to ignore task-irrelevant informationmore than accuracy stress is not an automaticconsequence of the prediction that speed stresswill lead participants to respond more quicklythan accuracy stress. It would indeed be a distinctpossibility for participants under speed stress toprocess the entire task information very quicklybut nevertheless decide not to ignore parts of theinformation. Conversely, participants under accu-racy stress may adopt a conservative responsecriterion, that is, process task information veryslowly and yet decide to no longer processtask-irrelevant information.

On the other hand, however, a rational viewsuggests that the setting of a liberal responsecriterion may be accompanied by limiting taskprocessing to task-relevant information. Thus,under speed instruction, participants may decideon the basis of little evidence to ignore task-redundant information in order to quicken taskprocessing. In contrast, the setting of a conserva-tive response criterion makes most sense if all

available task information is processed. On thebasis of this rational view, we predicted thatparticipants who are instructed to optimize theirspeed of task processing should very quickly—onthe basis of high uncertainty—decide to processtask-relevant information only, resulting in a highrate of information reduction. Participants whoare instructed to stress accuracy should ignoretask-irrelevant information only if they are abso-lutely certain that the information is truly irrel-evant, resulting in a much smaller degree ofinformation reduction.1

It is important to point out that methodologi-cally, the effects of task instruction on the settingof a response criterion and on deciding whichinformation is processed (i.e., information reduc-tion) are difficult to separate, at least if thepreferred measure of information reduction islatency-based, as is true in the present experi-ments. Indeed, a decreasing string-length effect—which is taken as evidence for information reduc-tion in the present experiments—is an unavoidableconsequence of declining latencies. Thus, it isone of the challenges of the present article todemonstrate that the manipulation of task instruc-tion has any effect on information reductionabove and beyond its influence on overalllatencies.

Concerning the third hypothesized outcome,the ability to change a response criterion duringtraining, Strayer and Kramer (1994b) have shownthat participants are unable to dynamically adjusttheir response criteria. However, as mentionedabove, the information reduction hypothesis holdsthat participants first notice that the AAT tasks

1 Any difference in degree of information reductionbetween conditions is affected by the number ofparticipants who notice that task-irrelevant informa-tion exists because the decision to consistently andsystematically ignore parts of the presented informa-tion can only follow the detection of task-irrelevantinformation. In principle, empirically obtained differ-ences on measures of information reduction couldtherefore reflect differences in noticing that task-irrelevant information exists rather than differences ininformation reduction per se. Because participantsunder speed stress may be at a disadvantage fordetecting task-irrelevant information, differences innoticing should, at worst, reduce (and not increase) thehere predicted higher degree of information reductionin the speed condition over the accuracy condition.


contain task-redundant information and then vol-untarily decide to ignore this information. Conse-quently, if the manipulation of task instructiongenerally affects degree of information reductionand if task instruction primarily affects a partici-pant's voluntary decision and not the process ofdetecting task-redundant information, then amountof information reduction should depend on in-struction, irrespective of when speed or accuracyinstructions are given. Thus, we expected that theeffects of task instruction would be independentof whether the instructions are given at thebeginning of or during training.

Overview of Experiments

In the experiments described in this article, allparticipants performed the AAT, As alluded toabove, the task requites participants to verifyalphabetic strings by judging whether or not thepresented strings are correct or incorrect- In bothExperiments 1 and 2, participants in two pureinstruction conditions were instructed to optimizeeither their accuracy or speed of performancethroughout the entire experiment (accuracy condi-tion and speed condition, respectively). In addi-tion, one control condition and two mixed instruc-tion conditions were realized. In the controlcondition, participants were asked to simulta-neously optimize both the speed and accuracy oftheir performance. In the accuracy-speed condi-tion, participants were initially instructed to opti-mize the accuracy of their performance and werethen later given speed instructions; in the speed-accuracy condition, participants were initiallyinstructed to optimize their speed of processingand then later their accuracy. Experiment 2differed from Experiment 1 mainly in that thetime duration of stimulus presentation underspeed stress was systematically reduced withincreasing practice. This was done in order toexamine the effect of time pressure on informa-tion-reduction. Time pressure is one characteristicof those real-life situations in which humansare required to make important decisions (e.g.,flying an airplane, driving a car. or playingfast ball sports). Thus, time stress might forcethem to limit task-processing to task-relevantinformation.

Experiment 1

There were five experimental conditions inExperiment 1: speed, accuracy, speed-accuracy,accuracy-speed, and control. The control condi-tion allowed us to examine whether speed andaccuracy instructions have symmetrical yet oppo-site effects or whether the effects are highlyasymmetrical. Asking participants to optimizeboth speed and accuracy was deemed a moreappropriate control condition than leaving theoptimization of speed and/or accuracy up to theindividual participants.

In accordance with the main hypotheses out-lined above, we expected the speed and accuracyinstructions to affect overall error rats, overalllatency, and degree of information reduction,First, participants in the accuracy condition shouldset a more conservative response criterion thanparticipants in the speed condition and, conse-quently, should respond more slowly and commitfewer errors than participants in the speed condi-tion. Second, information reduction should bereduced under accuracy instructions relative tospeed instructions. Thus, the string-length effectfor correct strings—our measure of informationreduction—should decline more rapidly for par-ticipants in the pure speed condition than forparticipants in the pure accuracy condition. Inaddition, if participants can flexibly adapt tochanges in task instruction, then performance anddegree of information reduction during the sec-ond half of practice in the speed-accuracy condi-tion should differ significantly from performance(i.e., degree of information reduction) in the purespeed condition. Conversely, performance anddegree of information reduction during the sec-ond half of practice in the accuracy-speed condi-tion should differ from performance (i.e., degreeof information reduction) in the pure accuracycondition.

Method

Participants

One hundred forty-one female and 54 malestudents at the University of Missouri at Colum-bia served as participants in the experiment. Theparticipants ranged in age from 17 to 41 years(M — 19,2 years. SD = 2.7) and received coursecredit in an introductory psychology class for


their participation. Because of technical prob-lems, data from 6 participants were lost.

Materials

A total of 30 correct and 30 incorrect alpha-betic strings_was used in the experiment. Correctstrings consisted of an initial letter-digit-lettertriplet (e.g., E [4] J) that began with one of 10letters: D, E, F, G, H, I, J, K, L, or M. The 10possible correct triplets were followed by zero,two, or four additional letters such that the entirestring followed the alphabet. The digit in bracketswas always 4. Thus, in all experimental condi-tions, there were 3 X 10 = 30 correct letterstrings that varied in total length (three, five, orseven symbols). The construction of the incorrectletter strings was identical to the construction ofthe correct letter strings except that the letterimmediately following the digit 4 was the letterthat followed the correct one in the alphabet (e.g.,E [4] K instead of E [4] J).

Strings were presented at the center of a 9-in.(22.9-cm) diagonal video screen controlled by aMacintosh SE microcomputer. The letters wereapproximately 0.3 cm X 0.3 cm in size. Adjacentletters appeared approximately 0.2 cm apart onthe screen. Participants responded by pressingeither the "z" or the "3" key on the second rowfrom the bottom of an extended Macintosh key-board. Half of the participants were instructed topress the "z" to indicate that a string was correctand the "3" to indicate that the string wasincorrect; for the other half, the key assignmentwas reversed.

Procedure

Participants were randomly assigned to one ofthe five experimental conditions (speed, accu-racy, speed-accuracy, accuracy-speed, and con-trol) and were tested individually in a moderatelylit room. The experiment began with computer-ized instructions and a short practice session.Participants were told that their task was to verifyalphabetic strings. They were informed aboutwhat constituted a correct string and were shownexamples of correct and incorrect strings. Then, ashort practice session followed in which partici-pants evaluated five correct and five incorrectpractice strings. In contrast to the experimental

practice blocks, errors in the incorrect stringscould occur both within and outside the triplet.Strings with errors outside the triplet were pre-sented to ensure that participants would notignore the irrelevant letters right from the begin-ning of the experimental phase. If participantsmade more than three errors on the practicestrings, instructions and practice session wererepeated.

When all participants had successfully com-pleted the practice session, participants in theaccuracy and accuracy-speed conditions wereinstructed to optimize their accuracy of perfor-mance, whereas participants in the speed andspeed-accuracy conditions were instructed tooptimize their speed. Participants in the controlcondition were instructed to optimize both speedand accuracy of performance. All participantswere told to pay close attention to the entire stringbecause errors could occur anywhere in thestring. Thus, participants were not informedabout the strings containing task-irrelevantinformation.

After they finished the fourth practice block,participants in the two mixed instruction condi-tions received new instructions. In the accuracy-speed condition, participants were told that fromnow on they should optimize their speed ofresponding. In the speed-accuracy condition,participants were told that from now on theyshould optimize their accuracy of responding.

Each trial began with the presentation of afixation point at the center of the screen for 500ms. The disappearance of the fixation point wasimmediately followed by the presentation of astring that remained on the screen until a re-sponse was made. The screen then went blankfor 1,000 ms, after which the next fixationpoint appeared. The entire experiment lastedbetween 45 and 60 min, depending on experimen-tal condition.

The experimental phase consisted of eightblocks of trials. Each block contained the 30correct and 30 incorrect strings, such that allparticipants verified a total of 240 correct and 240incorrect alphabetic strings (eight repetitions perstring). The order of strings was randomly deter-mined for each participant and each block. At theend of each block, participants received feed-back. Under accuracy instructions, participantsfirst received feedback concerning their error


rates followed by feedback about their meanresponse times in the previous block. Underspeed instruction, the feedback sequence wasreversed.

Design

Dependent variables were participants' medianverification times (reaction times) and their meanerror rates per trial block. The only between-subjects variable was experimental condition(accuracy vs. speed vs. accuracy-speed vs. speed-accuracy vs. control). Within-subjects variableswere practice block (one to eight), string length(three, five, seven), and string type (correct vs.incorrect).

Results

A first check ensured that only participantswith reasonable error rates in the experimentalphase were entered in the data analysis. To thisend, mean error rates were computed for eachparticipant and each trial block. Participants wereexcluded from all subsequent data analyses iftheir mean error rate was higher than 15% in eachtrial block (n = 5, n = 3, n = 15, n = 8, and n =9, in the control, accuracy, speed, accuracy-speed, and speed-accuracy conditions, respec-tively). Participants with very high error rateswere excluded from the data set because higherror rates did not allow us to compute reliableindividual regression slopes for the assessment ofdegree of information reduction in each trialblock. At a group level, the results reported forthe reduced sample did not differ qualitativelyfrom those for the entire sample, however. Theelimination of participants resulted in 16 remain-ing participants in the control condition, 42remaining participants in the accuracy condition,26 participants in the speed condition, 33 partici-pants in the accuracy-speed condition, and 32participants in the speed-accuracy condition. Forall remaining participants, median response times(RTs) for correct responses were computed tocorrect and incorrect alphabetic strings separatelyfor each block and each string length.

Manipulating task instructions bears the dan-ger of generating speed-accuracy trade-offs thatcan make comparison of RTs and error ratesacross conditions difficult. In order to examine

speed-accuracy trade-offs, the correlation be-tween mean error rate and median RT wascomputed in each of the five experimental condi-tions, separately for Blocks 1-4 and Blocks 5-8.Results are shown in Table 1.

As can be seen in the table, reliable correla-tions between RT and error rate occurred onlywhen participants were instructed to optimizespeed. For Blocks 1̂ , the correlation was nega-tive and significant in the speed-accuracy condi-tion and just failed to reach significance in thespeed condition (p < .07). In Blocks 5-8, onlythe accuracy-speed condition showed a signifi-cant negative correlation; in the speed condition,the speed-accuracy trade-off disappeared withincreasing practice. In general, a speed-accuracytrade-off occurred only in the speed conditions,not in the accuracy conditions. However, thecorrelations were all very small and do notqualify the findings reported below for RTlatencies.

In all further analyses of Experiment 1 andExperiment 2, the adopted level of significancewas a = .05. For significant effects, individual pvalues are not reported.

The discussion of the main results is dividedinto two sections. First, we discuss the effects oftask instruction on overall mean error rates andRTs. Then, we turn to a comparison of the degreeof information reduction in the five experimentalconditions.

Table 1Speed-Accuracy Trade-Offsin Experiments 1 and 2

Blocks 1-4ControlAccuracySpeedAccuracy-speedSpeed-accuracy

Blocks 5-8ControlAccuracySpeedAccuracy— speedSpeed-accuracy

-.019-.053-.174-.126-.195*

-.160-.028

.156-.189*-.023

-.123-.025

.215-.018

-.101.271

-.052-.082

*p < .05.


Overall Error Rate and RT

Error rate. Figure 1 displays the mean per-centage error rate in the five experimental condi-tions as a function of practice block. Because aninitial analysis had not revealed any qualitativedifferences for correct and incorrect strings, indi-vidual error~rates were averaged across the twostring types. As can be seen in the figure, themean error rate was higher in the pure speedcondition than in the pure accuracy condition,indicating that the between-subjects instructionmanipulation affected accuracy of performance.The control condition was in between the two butwas somewhat closer to the pure accuracy condi-tion, at least during the second half of theexperimental phase.

14

13-

12-

11

10-

9

8

Control

Accuracy

•••••• Speed

Accuracy-Speed

Speed-Accuracy

•• Accuracy-Speed— Speed-Accuracy

1 2 3 4 5 6 7 8

Practice block

Figure 1. Mean percentage error rate as a function ofcondition and practice block (Experiment 1). Errorbars represent 95% between-subjects confidence inter-vals (Loftus & Masson, 1994). Italic numbers on theright indicate caps end length: control = medium caps(3), accuracy = longest caps (1), speed = shortestcaps (5), accuracy-speed = long caps (2), and speed-accuracy = short caps (4).

A comparison of error rates for the mixed andthe pure instruction conditions shows that partici-pants were able, at least to some extent, to modifytheir performance when task instructions werechanged. In the accuracy-speed condition, theerror rate was initially indistinguishable from thatin the pure accuracy condition but increased from2% in Blocks 1-4 to 6% in Blocks 5-8. Incontrast, in the speed-accuracy condition, theerror rate was initially identical to that in the purespeed condition but decreased from 6% in Blocks1-4 to 4% in Blocks 5-8.

In order to separately capture the effects of theinitial setting and later changing of task instruc-tions, we computed two Condition (control vs.accuracy vs. speed vs. accuracy-speed vs. speed-accuracy) X Practice Block mixed-design analy-ses of variances (ANOVAs) on participants' errorrates for the first (Blocks 1-4) and second(Blocks 4-8) half of practice. Block 4 wasincluded in the second analysis in order toprovide a baseline for the expected effects ofchanging task instructions.

The analysis for the first half of practicerevealed significant main effects of condition,F(4,144) = 16.34, MSE = 35.36, and of practiceblock, F(3,432) = 5.69, MSE - 17.83, as weU asa significant Condition X Practice Block interac-tion, F(12,432) = 2.60, MSE = 17.83. A plannedcontrast confirmed the impression conveyed byFigure 1 that the difference between the twocombined accuracy and the two combined speedconditions was significant, F(l, 144) = 64.99,MSE — 35.36. The speed and speed-accuracyconditions did not differ significantly from eachother, nor did the accuracy and accuracy-speedconditions (bothps > .5). In addition, the controlcondition differed significantly from the com-bined speed and speed-accuracy conditions, F(l,144) = 10.81, MSE = 35.36, whereas the com-bined accuracy and accuracy-speed conditionsjust fell short of significance, F(l, 144) = 3.14,MSE = 35.36, p<. 08.

The ANO VA for Blocks 4—8 revealed a signifi-cant main effect of condition, F(4, 144) = 9.29,MSE — 63.89, and a significant interaction be-tween condition and practice block, F(16,576) =4.79, MSE = 20.71. As can be seen in Figure 1,the pattern of results was considerably noisier forthe second half of practice, as compared with thefirst half. Whereas the error rates in the pure


instruction conditions remained at first-half level,the error rates in the mixed conditions changed inthe expected direction and eventually reached thelevel of the pure conditions.

Separate planned follow-up contrasts compar-ing the speed-accuracy condition with the speedcondition and the accuracy-speed condition withthe accuracy condition confirmed these impres-sions. The former analysis revealed a significantdifference between the speed-accuracy conditionand the speed condition, F(l, 144) = 7.99,MSB = 63.89, that was qualified by a significantinteraction with practice block, F(4,576) = 4.64,MSE = 20.71. The latter analysis showed asignificant difference between the speed-accu-racy condition and the speed condition, F(l,144) = 13.38, MSE = 63.89, that was qualifiedby a significant interaction with practice block,F(4, 576) = 6.49, MSE = 20.71. In addition, theaccuracy-speed condition differed significantlyfrom the control condition, F(l, 144) = 3.46,MSE = 63.89 (F < 2, for the difference betweenthe speed-accuracy and control conditions).

RT. Figure 2 displays the mean RTs in thefive experimental conditions as a function ofpractice block. Because we did not obtain qualita-tive differences for correct and incorrect strings,individual median RTs were averaged across thetwo string types. The general pattern of resultsdepicted in Figure 2 is very similar to the one forerror rates shown in Figure 1. First, throughoutpractice, participants in the pure accuracy condi-tion verified strings more slowly than participantsin the pure speed condition. The control conditionwas almost exactly in between the two pureinstruction conditions. Second, the mean RTs inthe accuracy-speed and speed-accuracy condi-tions changed systematically when task instruc-tions were modified, although the changes weremuch less dramatic than they were for error rates.

Two separate Condition X Practice Blockmixed-design ANOVAs were again computed,one for Blocks 1-4 and a second one for Blocks4-8. The Blocks 1-4 ANOVA yielded significantmain effects of condition, F(4, 144).= 14.31,MSE = 4,292,653.0, and of practice block, F(3,432) = 235.82, MSE = 553,800.6. The maineffect of condition was, as is evident in Figure 2,due primarily to the difference between thecombined speed and speed-accuracy conditions,on the one hand, and between the combined

TOOO-i

6000-

5000-

4000-

3000-

2000-

1000-

Control

Accuracy

•••— Speed

Acouacy-Speed—#•-• Accuracy-Speed

Sp££d-Accuracy A Speed- Accuracy

1 2 3 4 5 6 7 8

Practice block

Figure 2. Mean reaction times (RTs; in millisec-onds) for correct and incorrect strings (Experiment 1).Error bars represent 95% between-subjects confidenceintervals; control = medium caps, accuracy = longestcaps, speed = shortest caps, accuracy-speed = longcaps, and speed-accuracy = short caps.

accuracy and accuracy-speed conditions, on theother hand, F(l, 144) = 55.13, MSE =4,292,653.0. The difference between these twopairs of conditions did not interact with practiceblock, p > .5. Although the speed and speed-accuracy conditions just failed to differ signifi-cantly from the control condition, F(l, 144) =2.82, MSE = 4,292,653.0, p < .1, the differencebetween the two accuracy conditions versus thecontrol condition was significant, F(l, 144) =9.10, MSE = 4,292,653.0.

The Blocks 4-8 ANOVA yielded significantmain effects of condition, F(4, 144) = 10.55,MSE = 3,731,489.0, and of practice Wock, F(4,576) = 103.09, MSE = 188,755.6, as well as asignificant interaction between condition andpractice block, F(16, 576) = 3.48, MSE =188,755.6. Separate planned follow-up contrastscomparing the accuracy-speed condition with theaccuracy condition and the speed-accuracy con-


dition with the speed condition did not revealsignificant interactions, all ps > A. The smalljumps in the latencies of the accuracy-speed(downward) and the speed-accuracy (upward)conditions that can be seen in Figure 2 were thusnot significant.

Taken together, the analyses of the overallerror rates and RTs suggest that the manipulationof task instruction was effective. Participantsunder speed stress performed more quickly andcommitted more errors than participants underaccuracy stress. Moreover, changing task instruc-tions affected participants' error rates and RTs aswell, although the effect on the latter measurewas small relative to the effect on the former.

Information Reduction

Correct strings. As discussed earlier, thestring-length effect for RTs on correct strings canbe considered an indirect measure of informationreduction (Haider & Frensch, 1996, 1999). Thestring-length effect is expected to differ systemati-cally across experimental conditions for correctstrings but not for incorrect strings. For the latter,string length should not matter because the errorin incorrect strings was fixed at the third stringposition. We computed the best fitting linearregression line across the three different stringlengths separately for each participant in eachpractice block. Figure 3 contains the mean slopesfor correct alphabetic strings in the five experi-mental conditions as a function of practice block;Figure 4 contains the corresponding mean slopesfor incorrect strings.

Inspection of Figure 3 shows a pattern qualita-tively similar to that obtained for overall errorrates and latencies. First, with practice, all slopesdeclined, F(7, 287) = 9.55, MSE = 20,977.3;F(7, 175) = 17.14, MSE = 9,647.6; F(l, 224) -20.53, MSE = 26,274.2; F(7, 217) - 11.99,MSE = 10,747.1; and F(7, 105) = 7.91, MSE =1,4091.1, for the accuracy, speed, accuracy-speed, speed-accuracy, and control conditions,respectively.

Second, and more importantly, the five experi-mental conditions differed in degree of slopedecline. In the two pure instruction conditions,the slope for the speed condition was smaller thanthe slope for the accuracy condition in the firsttrial block already and also declined much faster

500

450-

400

350-

300

250-

35 200-

150-

100-

50-

0-

-50

Control

Accuracy

-•••-• Speed

Accuracy-Speed

Speed-Accuracy

• Accuracy-Speed

Speed-Acmiacj

1 2 3 4 5 6 7 8

Practice block

Figure 3. Means of the best fitting regression slopesfor correct strings (Experiment 1). Error bars represent95% between-subjects confidence intervals; con-trol = medium caps, accuracy = longest caps,speed = shortest caps, accuracy-speed = long caps,and speed-accuracy = short caps.

over practice than in the accuracy condition. Thechange of instruction after the fourth practiceblock in the accuracy-speed condition led to animmediate dip in the slope, resulting in a slightdifference between this condition and the accu-racy condition in Blocks 5-8. In the speed-accuracy condition, the slope remained relativelyconstant after the instruction was changed, yield-ing a slight difference between this condition andthe speed condition in Blocks 5-8.

Although these results essentially mirror thosereported for overall error rates and latencies, asurprising finding that is conveyed by Figure 3concerns the control condition. In essence, partici-pants in the control condition seemed to behavevery similarly to the participants in the speed-accuracy condition. In fact, although the overalldifference between these two conditions was


350

300

250-

200

150

100-

50-

0-

-50

ControlAccuracy

••*••• SpeedAccuracy-Speed

Speed-Accuracy

-• Accuracy-Sjjeed

— Spged-Aggnracy

T 4-

4 5

Practice Mock

Figure 4. Means of the best fitting regression slopesfor incorrect strings (Experiment 1). Error bars repre-sent 95% between-subjects confidence intervals; con-trol = medium caps, accuracy = longest caps, speed =shortest caps, accuracy-speed = long caps, and speed-accuracy = short caps.

significant, F(l, 40) - 4.19, MSE = 47,426.7, itdid not interact with practice block, p > .2.

In order to examine the effects of instructionon the slopes representing the string-length ef-fect, we again computed two Condition X Prac-tice Block ANOVAs separately for Practice Blocks1-4 and Practice Blocks 4-8. The two ANOVAsyielded significant main effects of condition, F(4,144) = 12.08, MSE = 70,868.2, and F(4, 144) =6.88, MSE = 51,753.6, for Blocks 1-4 and 4-8,respectively, and of practice block, F(3, 432) =27.35, MSE = 22,556.5, and F(4, 576) = 17.78,MSE = 9,765.7, for Blocks 1-4 and 4-8, respec-tively. In addition, the Condition X PracticeBlock interaction was significant in both analy-ses, F(12, 432) = 2.87, MSE = 22,556.5, andF(16, 576) = 2.51, MSE = 9,765.7, for Blocks1-4 and 4-8, respectively.

Planned comparisons following up the Blocks

1-4 analysis confirmed the impression con-veyed by Figure 3 that the two combined speed-stress conditions differed significantly from thetwo combined accuracy-stress conditions, F(l,432) - 3.89, MSE = 22,556.5. The two speed-stress conditions and the two accuracy-stressconditions did not differ significantly from eachother, both/JS > .1.

Planned comparisons following up the Blocks4-8 analysis revealed a significant Condition XPractice Block interaction when only the accu-racy-speed and the accuracy conditions werecompared, F(4, 576) = 3.17, MSE = 9,765.7.The corresponding interaction for the comparisonbetween the speed-accuracy and speed condi-tions was not significant, p > .2.

Taken together, the analyses on the slopes forcorrect alphabetic strings show that the manipula-tion of task instruction yielded results that werequalitatively consistent with our expectations.However, the within-subjects manipulation oftask instruction was less effective than the be-tween-subjects manipulation. In addition, theeffects of speed stress and accuracy stress werenot symmetrical relative to the control condition.There are two different possible explanations forthis finding that we cannot distinguish on thebasis of our data. The first explanation takes thenature of the control condition seriously byarguing that accuracy instructions slow the pro-cess of information reduction, whereas speedinstructions have no effect. The second possibleexplanation holds that participants who are askedto maximize both speed and accuracy are morelikely to orient themselves toward maximizingspeed of performance rather than toward maximiz-ing accuracy. The second explanation, thus, ques-tions the validity of the control condition. Inretrospect, it seems clear that without additionalinformation on the behavior of individual partici-pants, the usefulness of the control condition islimited. Consequently, we did not use a controlcondition in Experiment 2.

Incorrect strings. Figure 4 displays the slopesfor incorrect alphabetic strings as a function ofpractice in the five experimental conditions. Incontrast to the slopes for correct strings, theslopes for incorrect strings did not vary withinstruction and, furthermore, declined only fromthe first to the second practice block. One-way(practice) ANOVAs, computed separately for


each condition, indicated a significant decline ofmean slope for all but the speed-accuracy condi-tion, F(7, 105) = 3.70, MSE = 15,368.2; F(7,287) = 6.89, MSE = 25,135.7; F(7, 175) = 4.30,MSE = 15,564.4; and F(7, 224) = 6.28, MSE =37,980.5, for the control, accuracy, speed, andaccuracy-speed conditions, respectively (in thespeed-accuracy condition, F < l.50,p > .2).

Two separate Condition X Practice BlockANOVAs (for Blocks 1-4 and Blocks 4-8)yielded only significant main effects of practiceblock, F(3, 432) = 12.63, MSE = 33,039.4, forBlocks 1-4, and, F(4, 576) = 4.33, MSE =12,708.4, for Blocks 4-8. All other effects werenot significant (all Fs < 1.50), indicating that theslopes for incorrect strings did not vary systemati-cally as a function of task instruction.

Discussion

Experiment 1 produced four main results:First, the initial between-subjects manipulation oftask instruction affected overall performance interms of both error rate and latency. Participantsstressed for speed performed less accurately andmore quickly than participants stressed for accu-racy. Second, the initial manipulation of taskinstruction affected amount of information reduc-tion as well. Participants stressed for speedshowed a larger reduction in the string-lengtheffect for correct strings than did participantsstressed for accuracy. Third, changing task instruc-tions after Block 4 affected overall error rates andalso had effects, albeit relatively small, on laten-cies and on the magnitude of the string-lengtheffect. Thus, participants were able to modifytheir behavior in accordance with new instruc-tions—at least to some extent. Fourth, the effectsof speed stress and accuracy instructions werenearly symmetrical relative to the control condi-tion on measures of overall error rate and latencybut not on the measure of information reduction.

In general, the results of Experiment 1 areconsistent with the interpretation that task instruc-tion affects both the setting of a response criterionand the selection of which information is pro-cessed. It appears that, under speed stress, partici-pants adopt a liberal response criterion and, inaddition, quickly cease processing of task infor-mation that they deem irrelevant. Under accuracy

instructions, participants adopt a conservativeresponse criterion and also process all or most ofthe given task information. These findings are inaccordance with our rational analysis providedabove, in which we suggested that in order toavoid systematic errors, the setting of a liberalresponse criterion needs to be accompanied byneglect of task-irrelevant information. In con-trast, the setting of a conservative responsecriterion makes most sense if all available taskinformation is processed. At a somewhat highertheoretical level, the obtained findings are alsoconsistent with the information-reduction viewthat we proposed earlier (Haider & Frensch,1996), according to which the decision to limittask processing to task-relevant information isbased, at least partly, on strategic considerations,and therefore can be influenced by manipulatingtask instructions.

One potential problem with the interpretationthat type of instruction affects degree of informa-tion reduction as referred to above is that method-ologically, the effects of instruction on overalllatency and on information reduction are difficultto separate. The reason for this is that a decreas-ing string-length effect, which we take as evi-dence for information reduction, is also an un-avoidable consequence of declining latencies.That is, as latencies decrease, the difference in RTfor shorter and longer strings decreases as well.

One possibility to separate task instructioninfluences on the string-length effect from taskinstruction influences on overall latency is tostatistically equate all experimental participantsin terms of their overall latencies and thencompute the individual string-length effects again.To do this, we first converted individual RTsseparately for each block into z scores and thencomputed string-length effects. The results of thisanalysis are shown in Figure 5 (correct strings)and Figure 6 (incorrect strings). When comparedwith corresponding Figures 3 and 4, it is clearthat the pattern of results for the string-lengtheffect is virtually the same. Thus,.we can bereasonably confident that the effect of task instruc-tion on the string-length effect is not a statisticalartifact of the effect of task instruction on overalllatency.

The main purpose of Experiment 2 was toreplicate the results of Experiment 1 with a


different sample. A second purpose was to furtherexplore one surprising finding in the presentexperiment: Specifically, although participants'behavior in the mixed-instruction conditions wasclearly affected by the change of instruction, theaccompanying changes in performance and, moreimportantly, in information reduction were rela-tively small (see Figures 3 and 5). For example,one might have expected that changing to a speedinstruction in the accuracy-speed condition wouldhave immediately resulted in a string-lengtheffect that was indistinguishable from that in thepure speed condition.

One possible explanation for the pronounceddifference hi information reduction between thespeed and accuracy-speed conditions during thesecond half of practice is that relatively fewparticipants in the accuracy-speed condition mighthave realized during Blocks 1-4 that some of the

0.35-

0.30

0.25-

0.15-

0.10-

0.05-

Contzol

Accuracy

Speed

Speed-Accuracy Speed-

1 2 3 4 5 6 7 8

Practice block

Figure 5. Means of the best fitting regression slopeson the z-score transformed reaction times for correctstrings (Experiment 1). Error bars represent 95%between-subjects confidence intervals; control = me-dium caps, accuracy = longest caps, speed = shortestcaps, accuracy-speed = long caps, and speed-accuracy = short caps.

0.40-1

035

0.30

0.25-

0.20

0.15

0.10-

0.05-

0.00

Control

Accuracy

Speed

>— Accuracy-Speedk— Speed-Atmmcy

0 1 2 3 4 5

Practice block

Figure 6. Means of the best fitting regression slopeson the z-score transformed reaction times for incorrectstrings (Experiment 1). Error bars represent 95%between-subjects confidence intervals; control — me-dium caps, accuracy = longest caps, speed = shortestcaps, accuracy-speed = long caps, and speed-accuracy = short caps.

letters were irrelevant. Therefore, they were notable to limit their task processing to the relevantinformation when this would have helped them toimprove their speed of processing. The assump-tion that noticing task-irrelevant information oc-curs to a larger extent under speed than underaccuracy instruction is, however, not very appeal-ing and clearly lacks face validity. Alternatively,one might therefore assume that manipulatingonly task instruction might not have been suffi-cient to force participants in the accuracy-speedcondition to ignore the task-irrelevant letters evenif they were relatively certain that the letters wereirrelevant.

In Experiment 2, we concentrated on theswitch from an initial accuracy to a later speedinstruction and tried to design conditions thatwould equate the string-length effect in theaccuracy-speed and speed conditions during the


second half of practice.2 For this purpose, weexamined whether additional time pressure underspeed-instruction would eliminate the differencesin string-length effect between the accuracy-speed and speed conditions during the secondhalf of practice. We assumed that additional timepressure would increase the likelihood that partici-pants would ignore task-irrelevant information—provided they had discovered the irrelevant infor-mation. To this end, we systematically decreasedthe time that strings were visible on the screenunder speed instructions. In me pure speed instruc-tion condition, presentation time was decreasedfrom 2,000 ms in the first practice block by 100ms in each subsequent trial block. In the accuracy-speed condition, presentation time was initially6,000 ms for the first half of practice and was thenidentical to that of the pure speed condition in thesecond half of practice. Under accuracy instruc-tions, presentation time was held constant through-out the entire experiment (fixed at 6,000 ms).Thus, in the speed-accuracy condition, presenta-tion time was identical to that for the pure speedcondition during the first half of practice and wasfixed at 6,000 ms during the second half ofpractice.

If the difference in string-length effect betweenthe speed condition and the accuracy-speed con-dition during the second half of practice was dueto participants' difficulty in the accuracy-speedcondition to ignore task-irrelevant informationeven though they deem this information as taskirrelevant, then adding time pressure should leadto equal performance or at least to a reduceddifference in performance relative to Experiment1. If, on the other hand, the difference betweenthe accuracy-speed condition and the speed con-dition was due to a difference in detectingtask-irrelevant information during the first half ofpractice, then the difference between the twoconditions during the second half of practiceshould be just as pronounced as it was inExperiment 1.

Experiment 2

Experiment 2 was identical to Experiment 1,with the exception described above. In addi-tion, we did not use a control condition inExperiment 2.

Method

Participants

Sixty female and 26 male undergraduate stu-dents at the University of Missouri at Columbiaserved as participants. All students received coursecredit in an introductory psychology class. Partici-pants ranged in age from 18 to 34 years (M =19.5 years, SD = 2.82). Because of technicalproblems, data from 3 participants were lost.

Materials

Stimuli and apparatus were identical to thoseused in Experiment 1.

Procedure

The procedure and presentation of stringsfollowed the format described for Experiment 1,with the exception noted above.

Design

Dependent variables were mean error rate andmedian RT per practice block. Independent vari-ables were experimental condition (accuracy,speed, accuracy-speed, speed-accuracy; be-tween subjects), practice block (one to eight;within subjects), string type (correct vs. incorrect;

2 We concentrated on the change from accuracy tospeed instructions because we assumed that it wouldbe easier to convince participants who knew about theirrelevant letters to no longer process these letters thanit would be to convince participants who had ignored—with no obvious drawback—the irrelevant letters forsome time to now not ignore this information anylonger. Our results concerning the switch from speedto accuracy instruction in Experiment 1 seemed toshow that it may well be that participants who hadsuccessfully ignored the irrelevant information forfour trial blocks simply did not understand why theyshould now process information they knew was irrel-evant. The finding that the string-length effect re-mained at the Block 4 level when the task instructionwas changed is consistent with the explanation thatparticipants in the speed-accuracy condition who hadcome to ignore the irrelevant information continued todo so.


within subjects), and string length (three, five, orseven; within subjects).

Results

As in Experiment 1, overall error rates werefirst computed for each participant and eachpractice block. Participants were excluded fromall subsequent data analyses if their mean errorrate was higher than 15% in all eight practiceblocks (n = 1, n = 13, n = 5, and n = 8, in theaccuracy, speed, accuracy-speed, and speed-accuracy conditions, respectively). This resultedin 19 participants remaining in the accuracycondition, 8 participants remaining in the speedcondition, 16 participants remaining in the accu-racy-speed condition, and 13 participants remain-ing in the speed-accuracy condition. For allremaining participants, median RTs were com-puted for correct responses to correct and incor-rect alphabetic strings, separately for each prac-tice block and each string length.

As argued above, manipulating task instruc-tions bears the danger of generating speed-accuracy trade-offs that can make comparison ofRTs and error rates across conditions meaning-less. In order to examine speed-accuracy trade-offs, the correlation between mean error rate andmean latency was computed separately for eachcondition and separately for Blocks 1-4 versus5-8. The results are shown in Table 1. As can beseen, none of the computed correlations wassignificant. Thus, despite the higher speed stressthat was realized in Experiment 2, no significantspeed-accuracy trade-off occurred.

The discussion of the main results follows thesame pattern as in Experiment 1. First, we discussthe effects of manipulating type of task instruc-tion on overall mean error rates and RTs. Then,we compare the string-length effects in the fourexperimental conditions.

Overall Error Rate and RT

Error rate. Figure 7 displays the mean per-centage error rate in the four experimental condi-tions as a function of practice block. Again,individual error rates were averaged across thetwo string types because an initial analysis hadnot revealed any qualitative differences for cor-rect and incorrect strings.

— D — Accuracy

•-•— Speed

Accrnacy-Speed

Spfifidr Accuracy

•• Acraracy-Spefid

— Speed-Accnragy

4 5

Practice block

Figure 7. Mean percentage error rates as a functionof condition and practice block (Experiment 2). Errorbars represent 95% between-subjects confidence inter-vals; control = medium caps, accuracy = longestcaps, speed = shortest caps, accuracy-speed = longcaps, and speed-accuracy = short caps.

Figure 7 shows an error pattern that is practi-cally identical to that displayed in Figure 1. Ascan be seen in the figure, the mean error rate washigher in the speed condition than in the accuracycondition, indicating that the between-subjectsmanipulation of instructions was effective. Acomparison of the error rates for the mixed- andthe pure-instruction conditions shows that partici-pants were able to modify their performance inaccordance with the new task instructions. In theaccuracy-speed condition, the error rate was,during the first four blocks, indistinguishablefrom that in the pure accuracy condition andincreased almost to the level of the pure speedcondition when task instruction was changed.Similarly, in the speed-accuracy condition, theerror rate was initially identical to that in the pure


speed condition and decreased to the level of thepure accuracy condition in Blocks 5-8.

In order to separately assess the effects of theinitial task instruction manipulation and the sub-sequent change of task instruction, we againcomputed separate Condition X Practice BlockANOVAs for. Blocks 1-4 and Blocks 4-8. TheBlocks 1-4 analysis revealed only a significantmain effect of condition, F(3, 52) = 34.12,MSE — 28.57. A planned comparison confirmedthat the difference between the two combinedaccuracy and the two combined speed conditionswas significant, F(l, 52) = 82.86, MSE = 28.57.The speed and speed-accuracy conditions did notdiffer significantly from each other nor did theaccuracy and accuracy-speed conditions, bothps > .2.

The Blocks 4-8 ANOVA revealed a significantmain effect of condition, F(3,52) = 4.27, MSE =71.48, and a significant interaction between con-dition and practice block, F(12, 208) = 3.56,MSE = 22.30.

Separate planned follow-up contrasts compar-ing the speed-accuracy condition with the speedcondition and the accuracy-speed condition withthe accuracy condition confirmed the impressionthat changing task instruction had an immediateeffect on participants' error rates. The formeranalysis revealed a significant difference betweenthe speed-accuracy and speed conditions, F(l,52) = 4.30, MSE = 71.48, that was qualified by asignificant interaction with practice block, F(4,208) = 3.22, MSE = 22.3(XThe latter analysis,the difference between the accuracy-speed andaccuracy conditions, just failed significance, F(l,52) = 3.43, MSE = 71.48, p < .1, but yielded asignificant Condition X Practice Block interac-tion, F(4, 208) = 2.65, MSE = 22.30.

On the whole, the results of the analyses onmean error rate replicate Experiment 1 and showthat (a) the initial between-subjects manipulationof task instruction affected accuracy of perfor-mance and (b) participants were able to modifytheir behavior when task instructions werechanged.

RT. Figure 8 displays the mean RTs in thefour experimental conditions as a function ofpractice block. Again, individual RTs were aver-aged across me two string types because an initialanalysis had not revealed any qualitative differ-ences for correct and incorrect strings. The gen-

5000-

3000-

3H

2000'

1000-

—O— Accuracy

--••• Speed

AccnracySpegd

Speed-Accoracy

Accuracy-Speed

Spged-Accm-acy

I T

0 1 2 3 4 5 6 7 8

Practice block

Figure 8. Mean reaction times (RTs; in millisec-onds) for correct and incorrect strings (Experiment 2).Error bars represent 95% between-subjects confidenceintervals; control = medium caps, accuracy = longestcaps, speed = shortest caps, accuracy-speed = longcaps, and speed-accuracy = short caps.

era! pattern of results depicted in Figure 8 is verysimilar to the one for Experiment 1 shown inFigure 2. First, throughout practice, participantsin the accuracy condition verified strings moreslowly than participants in the speed condition.Second, the latencies in the accuracy-speed andspeed-accuracy conditions changed systemati-cally and qualitatively more dramatically than inExperiment 1 when task instructions werechanged.

Two separate Condition X Practice Blockmixed-design ANOVAs were again computed,one for Blocks 1—4 and a second one for Blocks4-8. The Blocks 1-4 ANOVA yielded significantmain effects of condition, F(3, 52) = 5.47,MSE = 3,055,970.0, and of practice block, F(3,156) = 85.30, MSE = 451,168.3. The main effectof condition was due to the difference betweenthe combined speed and speed-accuracy condi-


tions, on the one hand, and the combined accu-racy and accuracy-speed conditions, on the otherhand, F(l, 52) = 9.15, MSE = 451,168.3. Thedifference between these two pairs of conditionsdid not interact with practice block, p > . 1.

The Blocks 4-8 ANOVA yielded significantmain effects of^condition, F(3, 52) = 6.90,MSE = 1,890,915.0, and of practice block, F(4,208) = 39.53, MSE = 173,312.0. In addition, theCondition X Practice Block interaction was sig-nificant as well, F(12, 208) = 4.74, MSE =173,312.0. As Figure 8 shows, the interactionbetween condition and practice block was causedprimarily by an immediate increase in the meanRT of the speed-accuracy condition and decreasein the mean RT of the accuracy-speed conditionwhen task instructions were changed.

Separate planned follow-up contrasts on theCondition X Practice Block interaction revealeda significant interaction between practice blockand the difference between the speed-accuracyand speed conditions, F(4, 208) = 8.36, MSE =173,312.0. The interaction between practice blockand the difference between the accuracy-speedand accuracy conditions just failed to reachsignificance, F(4,208) = 2.30, MSE = 173,312.0,p < .06.

In summary, the results of the overall RTsreplicated Experiment 1. Under speed stress,participants performed more quickly and withmore errors than participants under accuracystress. Furthermore, as was expected, changingtask instructions had a larger effect on partici-pants' RTs than it had in Experiment 1. However,this was true for both the accuracy-speed andspeed-accuracy conditions, although our manipu-lation had aimed primarily at the accuracy-speedcondition.

Information Reduction

Correct strings. The magnitude of the string-length effect was again used as an indirectmeasure of information reduction. As a reminder,the string-length effect was expected to differsystematically across experimental conditions forcorrect strings but not for incorrect strings. Inorder to separate type of instruction influences onoverall latency from type of instruction influ-ences on the string-length effect, we first con-verted individual RTs separately for each practice

block into z scores and then computed string-length effects (see Discussion to Experiment 1).The results of this analysis are shown in Figure 9for correct strings and Figure 10 for incorrectstrings.

Inspection of Figure 9 shows a pattern that isessentially what we had expected to find. First,and replicating Experiment 1, the slope in thepure speed condition declined faster over practicethan the slope in the pure accuracy condition.Second, the change of instruction after the fourthpractice block in the accuracy-speed conditionled to an immediate dip in the slope, resulting inonly a very slight difference between this condi-tion and the pure speed condition for Blocks 5-8.Surprisingly, although our presentation time ma-nipulation was aimed at the accuracy-speedcondition and not the speed-accuracy condition,the slope in the speed-accuracy condition also

—D— Accuracy

--••-- Speed

Accuracy-Speed -•+•

Speed-Accuracy A Speed- Accuracy

0.006 7 8

Figure 9. Means of the best fitting regression slopeson the z-score transformed reaction times for correctstrings (Experiment 2). Error bars represent 95%between-subjects confidence intervals; control = me-dium caps, accuracy = longest caps, speed = shortestcaps, accuracy-speed = long caps, and speed-accuracy = short caps.


O.«h

035-

0.30-

0.20-

0,15-

0.10-

0.05-

0.00

Figure 10. Means of the best fitting regressionslopes on the z-score transformed reaction times forincorrect strings (Experiment 2). Error bars represent95% between-subjects confidence intervals; con-trol = medium caps, accuracy = longest caps,speed = shortest caps, accuracy-speed = long caps,and speed-accuracy = short caps.

changed dramatically after Block 4 and was notdifferent from the slope in the pure accuracycondition for Blocks 5-8.

In order to examine the effects of instructionon the slopes, we again computed two Condi-tion X Practice Block ANOVAs separately forBlocks 1-4 and Blocks 4-8. The Blocks 1̂ANOVA yielded significant main effects of condi-tion, F(3, 52) = 9.95, MSB = 0.0262, and ofpractice block, F(3,156) = 3.43, MSB = 0.0107,as well as a significant Condition x PracticeBlock interaction, F(9, 156) = 2.14, MSE =0.0107. The Blocks 4-8 ANOVA just failed toreach significance for the main effect of condi-tion, F(3, 52) = 2.28, MSE = 0.0347, p < .09.Only the Condition X Practice Block interactionwas significant, F(12, 208) = 4.22, MSE =0.0123.

Planned comparisons following up the Blocks

1-4 analysis confirmed the impression conveyedby Figure 9 that the two speed-stress conditionsdiffered significantly from the two accuracy-stress conditions, F(l, 52) = 26.20, MSE =0.0262. The two speed-stress conditions and thetwo accuracy-stress conditions did not differsignificantly from each other, bothps > .1.

Planned comparisons on the Condition XPractice Block interaction following up the Blocks4-8 analysis revealed a significant interactionbetween practice block and the difference be-tween the accuracy-speed and accuracy condi-tions, F(4, 208) = 4.41, MSE = 0.0123. Theinteraction between practice block and the differ-ence between the speed-accuracy and speedconditions just fell short of significance, F(4,208) = 2.15, MSE = 0.0123, p < .08.

On the whole, the analyses on the slopes forcorrect alphabetic strings replicate the results ofExperiment 1 by showing that the initial between-subjects manipulation of task instruction waseffective. In contrast to Experiment 1, the changeof instruction, accompanied by additional timepressure, pushed the mean slope in the accuracy-speed condition down to the level of the purespeed condition. Surprisingly and unexpectedly,the slope in the speed-accuracy condition in-creased to the level of the pure accuracy condi-tion as well.

Incorrect strings. Figure 10 displays theslopes for incorrect alphabetic strings as a func-tion of practice in the four experimental condi-tions. As we had done for correct strings, wecomputed two separate Condition X PracticeBlock ANOVAs for Blocks 1-4 and for Blocks4-8. The only significant effect we found was forpractice block in the first analysis, F(3, 156) =7.49, MSE = 0.012. No other effects weresignificant (all Fs < 2.00; ps > .1). Thus, as inExperiment 1, type of instruction did not affectthe slopes for incorrect alphabetic strings.

Discussion

The primary purpose of Experiment 2 was toreplicate the results of Experiment 1. A secondarypurpose was to explore the effect of time pressureon information reduction and to examine whethera lack of force in Experiment 1 had caused theremaining differences in degree of information


reduction between the accuracy-speed and purespeed conditions found for Blocks 5-8.

First, the results of Experiment 2 clearly repli-cated those of Experiment 1. Again, participantsstressed for speed performed less accurately andfaster than participants stressed for accuracy,indicating that participants were able to establishdifferent response criteria for different instruc-tions. Furthermore, type of instruction affecteddegree of information reduction. For correctstrings, slopes were higher under accuracy stressthan under speed stress. Second, as in Experiment1, the within-subjects change of instruction hadan immediate effect on the slopes in the accuracy-speed and speed-accuracy conditions. The slopeof the accuracy-speed condition reached the levelof the speed condition during the second half ofthe practice, probably because of the higherspeed stress in the current experiment.

In addition, and quite unexpectedly, the slopefor the speed-accuracy condition did not differfrom that for the accuracy condition when taskinstruction was changed. The latter finding wassurprising because one might expect that it isdifficult to imagine any experimental manipula-tion that might force participants to processinformation that they assume is irrelevant. Onepossible reason for why the speed-accuracycondition did not differ from the accuracy condi-tion for Blocks 5-8 in Experiment 2 as it did inExperiment 1 may be that participants in Experi-ment 2 were uncertain whether errors would nowoccur in the task-redundant letters and withouttime pressure in Blocks 5-8, used the additionaltime to evaluate these letters.

General Discussion

Taken together, Experiments 1 and 2 providedtwo main results: First, the manipulation of taskinstruction affected overall task performance interms of RT and accuracy. Second, and moreimportant in the present context, task instructioninfluenced the string-length effect for correctstrings as well. These results were found not onlywhen task instruction was manipulated betweensubjects but also when it was manipulated withinsubjects. The latter effects were particularly im-pressive in combination with additional timepressure under speed stress.

At a first level of explanation, the present

findings are consistent with the assumption thattask instruction affects both the setting of aresponse criterion (e.g., Pachella, 1974; Rabbitt,1989; Sperling & Dosher, 1986) and the samplingof information from a presented stimulus. In thelanguage of random walk models (Ratcliff, 1981,1985; Treisman & Williams, 1984), participantswho were asked to optimize their speed of taskprocessing moved their response boundary closerto the starting point, meaning that their selectionof a response was based on relatively littleinformation about the stimulus. Participants whowere instructed to optimize accuracy of perfor-mance, in contrast, were more likely to adopt aconservative response criterion, that is, to movetheir response boundary farther apart from thestarting point and to base their response selectionon more information extracted from the taskstimulus. Thus, the results suggest that the settingof response criteria not only affects the speed ofprocessing but also the amount of informationprocessed to generate a response, as is predictedby random walk models.

In the language of information reduction, par-ticipants who are instructed to optimize speed ofprocessing try to reduce the amount of informa-tion they process. After having identified task-irrelevant information, they try to ignore it on thebasis of relatively little prior experience. If theyhave not identified the irrelevant information yet,they appear to sample randomly from the stimu-lus information and become extremely inaccu-rate, as was the case in the speed condition ofExperiment 2. Participants who are instructed tooptimize accuracy of processing, in contrast, areunlikely to reduce the amount of stimulus infor-mation they process.

A rational analysis suggests, and the presentfindings support this view, that the setting of aresponse criterion and the reduction of processedinformation should not be independent. Indeed, aliberal response criterion appears to be accompa-nied by neglect of task-irrelevant information. Incontrast, the setting of a conservative responsecriterion seems to imply that all available taskinformation is processed. These findings extendthe view expressed by Strayer and Kramer (1994a,1994b) that more liberal settings of responsecriteria are one characteristic of advanced skill.

At a second level of explanation, the presentfindings are consistent with our information reduc-


tion hypothesis (Haider & Frensch, 1996). Thisview holds that the decision to limit task process-ing to task-relevant information is based onstrategic considerations and therefore can bemanipulated by task instructions. The currentfindings, particularly the decrease of the meanslope in theaccuracy-speed condition and theincrease of the mean slope in the speed-accuracycondition in Experiment 2, clearly demonstratethat information reduction is at least partly undervoluntary control.

Finally, the present results suggest that cogni-tive skill acquisition may, in general, be based notonly on data-driven learning mechanisms (e.g.,Logan, 1988), as has been argued sometimes, butalso on strategic considerations. The latter possi-bility has recently been discussed by Strayer andKramer (1994a, 1994b) and Doane, Aldeiton,Sohn, and Pellegrino (1996). Doane et al. (1996),for instance, demonstrated that an initially adoptedlearning strategy influences which information alearner attends to when the task consists ofdiscriminating targets from distractors (see alsoFisher & Tanner, 1992).

Practical Implications

The findings presented here are importantbecause of their potential practical implicationsfor many areas of human performance, such asthe domain of expertise or instructional psychol-ogy (e.g., Bransford et al., 1986; Larkin, 1983;Myles-Worsley et al., 1988; Shanteau, 1992;Shapiro & Raymond, 1989). The results stronglysuggest that with practice, humans learn to ignoreinformation that has no diagnostic value (e.g.,redundant task information) with respect to agiven goal. Importantly, the amount of evidencethat is needed to judge information as task-redundant and to ignore this information dependson task instruction. In contrast to schema theory(Minsky, 1975; Rumelhart, 1975; Schmidt, 1975,1988), which focuses primarily on knowledgerepresentation, that is, on the result of learning,the purpose of the present experiments was toshow that information reduction is a general andnonspecific-domain learning process that takesplace during practice. The information reductionhypothesis holds that learning to limit task pro-cessing to task-relevant information is based onthe detection that task information contains irrel-

evant information. This process is assumed to bea data-driven process (bottom-up) that is fol-lowed by a second strategical process, limitingtask processing to the task-relevant information.The latter assumption is strikingly confirmed bythe results reported here. Thus, our findings oninformation reduction extend schema theory andare important for domains in which routineactivities play a major role.

For example, for the domain of instructionalpsychology, Bransford et al. (1986) criticized thatmany teaching activities fail to educate studentsto attend to the relevant information. For ex-ample, students learning to solve word problemsin arithmetic have difficulties differentiating be-tween task-relevant and task-irrelevant numericfacts. They are taught abstract schemata or rulesof how to find the relevant facts. However,despite these schemata, the students are poor indifferentiating between relevant and irrelevantfacts in concrete tasks. As the current findings oninformation reduction suggest, an alternative wayto teach students to identify the relevant facts inarithmetic word problems might be to trainstudents with a repeated number of problemscontaining the same structure. In this case, stu-dents learn to identify the relevant facts in abottom-up manner. In contrast to schema theory,learning in this situation is equivalent to adifferentiation of perception (J. J. Gibson &Gibson, 1955), such that the learner becomesincreasingly sensitive to relevant environmentalcues.

Moreover, as we discussed earlier, informationreduction is consistent with several findings inthe domain of expertise (e.g., Ericsson &Lehmann, 1996; Ericsson & Smith, 1991; Patel &Groen, 1991). A robust difference between ex-perts and novices is that experts focus on relevanttask information and ignore task-irrelevant infor-mation. For example, De Maio, Parkinson, Lesho-witz, Crosby, and Thorpe (1976) compared thevisual scanning processes of student and instruc-tor pilots. They found that, particularly undertime pressure, the scanning strategies of instruc-tor pilots did not follow the rules they had beentaught. Instead, they fixated only task-relevantinformation while neglecting the less importantinstruments. This allowed the instructor pilots tomanage situations with high time pressure,whereas student pilots who scanned all instru-


meats failed to manage these situations. How-ever, it is unclear exactly how experts learn toefficiently use information or how to teach some-one to process relevant information only (e.g.,E. J. Gibson, 1963; J. J. Gibson & Gibson, 1955;Shapiro & Raymond, 1989). It is our hope thatinsights in the relation between external facilitat-ing and inhibiting conditions may eventuallyallow us to control the degree to which factualand seemingly task-irrelevant information is ig-nored. The present experiments on informationreduction are a first step in this direction.

References

Abernethy, B. (1993). Searching for the minimalessential information for skilled perception andaction: Motor control and sensory-motor integra-tion. Psychological Research, 55,131-138.

Allport, A. (1987). Selection for action: Some behav-ioral and neurophysiological considerations of atten-tion and action. In H. Heuer & A. F. Sanders (Eds.),Perspectives on perception and action (pp. 395-419). Hfflsdale, NJ: Erlbaum.

Anderson, J. R. (1982). Acquisition of cognitive skill.Psychological Review, 89,369-406.

Anderson, J. R. (1983). The architecture of cognition.Cambridge, MA: Harvard University Press.

Anderson, J. R. (1987). Skill acquisition: Compilationof weak-method problem solutions. PsychologicalReview, 94,192-219.

Anderson, J. R. (1992). Automaticity and the ACT*theory. American Journal of Psychology, 105, 165-180.

Anderson, J. R., Matessa, M., & Douglass, S. (1995).The ACT-R theory and visual attention. In J. D.Moore & J. F. Lehman (Eds.), Proceedings of theXVI. Annual Conference of the Cognitive Psychol-ogy Society (pp. 506-511). Hillsdale, NJ: Erlbaum.

Bransford, J. D., Sherwood, R. D., Vye, N. J., &Rieser, J. (1986). Teaching thinking and problemsolving. American Psychologist, 41, 1078-1089.

Cheng, P. W. (1985). Restructuring versus automatic-ity: Alternative accounts of skill acquisition. Psycho-logical Review, 92,414-423.

Christensen, E. E., Murry, R. C, Holland, K., Rey-nolds, J., Landay, M. J., & Moore, J. G. (1981). Theeffect of search time on perception. Radiology, 138,361-355.

De Maio, J., Parkinson, S., Leshowitz, B., Crosby, J.,& Thorpe, J. A. (1976). Visual scanning: Com-

parisons between student and instructor pilot(AFHRL-TR-76-10, AD-A023 634). Williams AirForce Base, AZ: Air Force Human ResourcesLaboratory, Flying Training Division.

Doane, S. M., Alderton, D. L., Sohn, Y. W., &Pellegrino, J. W. (1996). Acquisition and transfer ofskilled performance: Are visual discrimination skillsstimulus specific? Journal of Experimental Psychol-ogy: Human Perception and Performance, 22,1218-1248.

Ericsson, K. A., & Lehmann, A. C. (1996). Expert andexceptional performance: Evidence of maximaladaptation to task constraints. Annual Review ofPsychology, 47,273-305.

Ericsson, K. A., & Smith, J. (1991). Toward a generaltheory of expertise. New York: Cambridge Univer-sity Press.

Fisher, D. L., & Tanner, N. (1992). Optimal symbol setselection: An automated procedure. Human Fac-tors, 34,79-92.

Frensch, P. A. (1991). Transfer of composed knowl-edge in a multistep serial task. Journal of Experimen-tal Psychology: Learning, Memory, and Cognition,17,997-1016.

Frensch, P. A. (1994). Composition during seriallearning: A serial position effect. Journal of Experi-mental Psychology: Learning, Memory, and Cogni-tion, 20,423-442.

Frensch, P. A., & Geary, D. C. (1993). The effects ofpractice on component processes in complex mentaladdition. Journal of Experimental Psychology:Learning, Memory, and Cognition, 19,433-456.

Gibson, E. J. (1963). Perceptual learning. AnnualReview of Psychology, 14, 29-56.

Gibson, J. J., & Gibson, E. J. (1955). Perceptuallearning: Differentiation or enrichment? Psychologi-cal Review, 62, 32-41.

Haider, H., & Frensch, P. A. (1996). The role ofinformation reduction in skill acquisition. CognitivePsychology, 30, 304-337.

Haider, H., & Frensch, P. A. (1999). Eye movementduring skill acquisition: More evidence for theinformation reduction hypothesis. Journal of Experi-mental Psychology: Learning, Memory, and Cogni-tion, 25, 172-190.

Helsen, W., & Pauwels, J. (1993). A cognitive ap-proach to skilled performance and perception insport. In G. d'Ydewalle & J. Van Rensbergen(Eds.), Perception and cognition (pp. 127-139).Amsterdam: Elsevier Science.


Holding, D. H. (1985). The psychology of chess skill.Hillsdale, NJ: Erlbaum.

Kaptein, N. A,, Theeuwes, J., & Van der Heijden,A. H. C. (1995). Search for a conjunctively definedtarget can be selectively limited to a color-definedsubset of elements. Journal of Experimental Psychol-ogy: Human Perception and Performance, 21,1053-1069.

Lambert, A. J., Spencer, M. B., & Mohindra, N.(1987). Automaticity and the capture of attention bya peripheral display change. Current PsychologicalResearch and Reviews, 6,136-147.

Larkin, J. H. (1983). The role of problem representa-tion in physics. In D. Gentner & A. L. Stevens(Eds.), Mental models (pp. 75-98). Hillsdale, NJ :Erlbaum.

Lassaline, W., & Logan, G. D. (1993). Memory-basedautomaticity in the discrimination of visual numer-osity. Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 19, 561-581.

Loftus, G. R., & Masson, M. E. (1994). Usingconfidence intervals in within-participant designs.Psychonomic Bulletin & Review, 1, 476—490.

Logan, G. D. (1988). Toward an instance theory ofautomatization. Psychological Review, 95, 492-527.

Logan, G. D., & Etherton, J. L. (1994). What islearned during automatization? The role of attentionin constructing an instance. Journal of Experimen-tal Psychology: Learning, Memory, and Cognition,20,1022-1050.

Minsky, M. (1975). A framework for representingknowledge. In P. H. Winston (Ed.), The psychologyof computer vision (pp. 211-277). New York:McGraw-Hill.

Myles-Worsley, M., Johnston, W. A., & Simons, M. A.(1988). The influence of expertise on x-ray imageprocessing. Journal of Experimental Psychology:Learning, Memory, and Cognition, 14, 553—557.

Neisser, U. (1976). Cognition and reality. San Fran-cisco: Freeman.

Neisser, U., & Becklen, R. (1975). Selective looking:Attending to visually-specified events. CognitivePsychology, 7,480-494.

Neumann, O. (1990). Visual attention and action. In O.Neumann & W. Prinz (Eds.), Relationships betweenperception and action (pp. 227-268). Berlin, Ger-many: Springer.

Newell, A., & Rosenbloom, P. S. (1981). Mechanismsof skill acquisition and the law of practice. In J. R.

Anderson (Ed.), Cognitive skills and their acquisi-tion (pp. 1-56). Hillsdale, NJ: Erlbaum.

Pachella, R. G. (1974). The interpretation of reactiontime in information-processing research. In B. H.Kantowitz (Ed.), Information processing: Tutorialsin performance and cognition (pp. 41-82). Hillsdale,NJ: Erlbaum.

Pachella, R. G., & Pew, R. W. (1968). Speed-accuracytradeoff in reaction time: Effect of discrete criteriontimes. Journal of Experimental Psychology, 92,378-384.

Patel, V. L., & Groen, G. J. (1991). The general andspecific nature of medical expertise: A critical look.In K. A. Ericsson & J. Smith (Eds.), Toward ageneral theory of expertise (pp. 93-125). NewYork: Cambridge University Press.

Petrakis, E. (1993). Analysis of visual search patternsof tennis teachers. In G. d'Ydewalle & J. VanRensbergen (Eds.), Perception and cognition (pp.159-168). Amsterdam: Elsevier Science.

Rabbitt, P. M. A. (1989). Sequential reaction. In D. H.Holding (Ed.), Human skills (pp. 153-175). NewYork: Wiley.

Ratcliff, R. (1981). A theory of ordered relations inperceptual matching. Psychological Review, 88,552-572.

Ratcliff, R. (1985). Theoretical interpretations of thespeed and accuracy of positive and negative re-sponses. Psychological Review, 92,212-225.

Rumelhart, D. E. (1975). Notes on a schema forstories. In D. G. Bobrow & A. Collins (Eds.),Representation and understanding (pp. 211-236).New York: Academic Press.

Schmidt, R. A. (1975). A schema theory of discretemotor skill learning. Psychological Review, 82,225-260.

Schmidt, R. A. (1988). Motor control and learning. Abehavioral emphasis. Champaign, IL: HumanKinetics.

Shanteau, J. (1992). How much information does anexpert use? Is it relevant? Acta Psychologica, 81,75-86.

Shapiro, K. L., & Raymond, J. E. (1989). Practice ofefficient oculomotor strategies enhances skill acqui-sition. Acta Psychologica, 71, 217-242.

Sperling, G., & Dosher, B. (1986). Strategy andoptimization in human information processing. InK. R. Boff, L. Kaufman, & J. P. Thomas (Eds.),Handbook of perception and human performance(Vol. 1, pp. 2-1-2-65). New York: Wiley.

Strayer, D. L., & Kramer, A. F. (1994a). Strategies and


automaticity: I. Basic findings and conceptual frame-work. Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 20,318-341.

Strayer, D. L., & Kramer, A. F. (1994b). Strategies andautomaticity: H Dynamic aspects of strategy adjust-ment. Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 20, 342-365.

Trabasso, T. R., & Bower, G. D. (1968). Attention andlearning. New York: Wiley.

Treisman, M., & Williams, T. C. (1984). A theory ofcriteria setting with an application to sequentialdependencies. Psychological Review, 91, 68-111.

Van der Heijden, A. H. C. (1992). Selective attentionin vision. London: Routledge.

Vickers, J. N. (1988). Knowledge structures of expert-novice gymnasts. Journal of Motor Behavior, 19,401-415.

Wickelgren, W. A. (1977). Speed-accuracy tradeoffand information processing dynamics. Acta Psycho-logica, 41,67-85.

Received April 7,1997Revision received September 22,1998

Accepted September 24,1998 •

New Editors Appointed, 2000-2005

The Publications and Communications Board of the American Psychological Associa-tion announces the appointment of three new editors for 6-year terms beginning in 2000.

As of January 1,1999, manuscripts should be directed as follows:

• For Experimental and Clinical Psychopharmacology, submit manuscripts toWarren K. Bickel, PhD, Department of Psychiatry, University of Vermont, 38Fletcher Place, Burlington, VT 05401-1419.

• For the Journal of Counseling Psychology, submit manuscripts to Jo-Ida C.Hansen, PhD, Department of Psychology, University of Minnesota, 75 EastRiver Road, Minneapolis, MN 55455-0344.

• For the Journal of Experimental Psychology: Human Perception and Perfor-mance, submit manuscripts to David A. Rosenbaum, PhD, Department of Psy-chology, Pennsylvania State University, 642 Moore Building, University Park,PA 16802-3104.

Manuscript submission patterns make the precise date of completion of the 1999 volumesuncertain. Current editors, Charles R. Schuster, PhD; Clara E. Hill, PhD; and Thomas H.Carr, PhD, respectively, will receive and consider manuscripts through December 31,1998. Should 1999 volumes be completed before that date, manuscripts will be redirectedto the new editors for consideration in 2000 volumes.

information reduction during skill acquisition: the ... · than participants under accuracy stress....

Documents