Analysis of Part-Task Training Using the Backward-Transfer ...myweb.fsu.edu/vshute/pdf/shute...

Journal of Experimental Psychology: Applied1996, Vol. 2, No. 3, 227-249

In the public domain

Analysis of Part-Task Training Usingthe Backward-Transfer Technique

Barry P. Goettl and Valerie J. ShuteArmstrong Laboratory, U.S. Air Force

Researchers conducted 2 experiments that used backward transfer to improvethe efficiency of part-task training for a desktop flight simulator. In Experi-ment 1, a part-task group showed positive transfer but did not perform as wellas a whole-task group. Backward-transfer analysis indicated that only a subsetof the component tasks was critical to the criterion task. In Experiment 2, apart-task training regime that used the critical component tasks was comparedwith a whole-task regime and a part-task regime composed of noncritical compo-nent tasks. Results indicated that the critical part-task regime was as effective as thewhole-task regime, validating the utility of the backward-transfer technique.

The research discussed in this article examinedcomplex skill acquisition involving a dynamic spa-tial task, coupled with an analysis of the effects of abackward-transfer approach to training. A discus-sion of part-task training and the factors thatinfluence the effectiveness of part-task training isfollowed by an overview of the transfer of trainingparadigm used to test part-task training effective-ness. We conclude our literature review with anin-depth discussion of the specific training tech-nique used in our two experimental studies; namely,the backward-transfer approach.

Part-Task Training

Part-task training has long been a popular ap-proach to training complex manual control and

Barry P. Goettl and Valerie J. Shute, ArmstrongLaboratory, Human Resource Directorate, TechnicalTraining Research Division, Intelligent Training Branch(AL/HRTI), Brooks Air Force Base (AFB), U.S. Air Force.

Funds for this research were provided by the U.S. AirForce Office of Scientific Research and the TrainingResearch for Automated Instruction (TRAIN) Projectof AL/HRTI at Lackland AFB under the direction of J.Wesley Regian. The opinions expressed in this articleare those of the authors and do not necessarily reflectthose of the U.S. Air Force.

Correspondence concerning this article should be ad-dressed to Barry P. Goettl, AL/HRTI, 1800 Carswell Ave-nue, Lackland AFB, Texas 78236-5507. Electronic mail maybe sent via Internet to [email protected]. af.mil.

tracking tasks. The basic tenet underlying part-task training is that drill on individual componentsof a complex skill will improve performance on thecomplex skill. Carlson, Khoo, and Elliot (1990)refer to this hypothesis as the component fluencyhypothesis (p. 267). Carlson, Sullivan, and Schneider(1989) point out that this hypothesis rests on threeassumptions generally agreed on by theories ofcognitive skill: (a) complex skills consist of ahierarchy of basic component skills and organizingstrategies; (b) there are capacity limits placed oncognitive processing; and (c) fluency on the compo-nent skills is critical to skilled performance on thecomplex task. Thus, part-task training is thought topermit individuals to acquire critical componentskills that transfer to the whole task withoutimposing the cognitive demands of the whole task.

In spite of the intuitive appeal and theoreticalfoundations of the component fluency hypothesis,part-task training has garnered, at best, only mod-est empirical support and has frequently been lesseffective than whole-task training (for a review, seeWightman & Lintern, 1985). We discuss fourchallenges that a part-task training regime mustovercome to be effective. First, part-task trainingeffectiveness depends on the identification of validcritical component tasks. Second, the skills identi-fied as most critical early in training may not becritical later in training. Third, interactions amongthe component tasks play an important role in thewhole task. Finally, individual differences in abilityand style of learning play a large role in skill

227

228 GOETTL AND SHUTE

acquisition. We examine each of these challengesbelow.

Identification of Critical Skills

One of the problems faced by part-task trainingis the difficulty in identifying component skills thatare critical for proficiency on the whole-task.Wightman and Lintern (1985) identified threegeneral approaches to part-task training and foundthat the effectiveness of training depends on theapproach used. One approach they examined iscalled simplification, in which a difficult task ismade easier by adjusting certain characteristics ofthe task. For example, Briggs and Naylor (1962)trained male undergraduates on tracking taskshaving relatively simple control dynamics and trans-ferred them to tracking tasks having more complexcontrol dynamics. They found that training on thesimplified tasks produced positive transfer, buttransfer was less than 100%. Although some re-searchers have shown that simplification can pro-duce transfer (Briggs & Naylor, 1962; Briggs &Waters, 1958), Lintern and Gopher (1980) con-cluded in their review of research on simplificationthat the technique is not as effective as whole-tasktraining.

Another technique for part-task training is calledsegmentation. In segmentation, tasks are parti-tioned on the basis of temporal or spatial dimen-sions. For example, Bailey, Hughes, and Jones(1980) segmented an air-to-ground attack task intocomponents of downwind leg, base leg, roll-in, andfinal approach and used a backward-chaining pro-cedure. In backward chaining, the final segment inthe sequence (e.g., final approach) is trained tocriterion first; then the preceding segment is added(e.g., roll-in and final approach) and so on until thefull task is "chained" together. Bailey et al. foundthat the backward chaining approach was moreeffective than whole-task training. Wightman andSistrunk (1987) used segmentation to train a car-rier landing task. Participants performed segmentsof the task starting at 2,000 feet (609.6 m) fromtouchdown, then 4,000 feet (1,219.2 m) from touch-down, and then 6,000 feet (1,828.8 m) from touch-down. Those receiving segmentation training per-formed better on the criterion task than thosereceiving equivalent training on the criterion task.Ash and Holding (1990) used segmentation todivide a keyboard task into three segments of eight

sequential notes. They also found that segmenta-tion was superior to whole-task training.

The third approach reviewed by Wightman andLintern (1985) is called fractionation, wherein thetask is divided into components that are performedsimultaneously. For example, Mane, Adams, andDonchin (1989) used a task analysis of SpaceFortress (Mane, Coles, Wickens, & Donchin, 1983)to develop a part-task training regime based onfractionation. Their part-task regime was moreeffective than whole-task training. This approachwas also successfully used by Frederiksen andWhite (1989) and Fabiani et al. (1989). However,other studies have failed to show that part-tasktraining based on fractionation is more effectivethan whole-task training (Adams, 1960; Stammers,1980). Wightman and Lintern concluded that frac-tionation is more effective than whole-task train-ing, only when a systematic procedure is used todecompose the task.

Critical Skills and Skill Acquisition

The challenge of identifying the critical compo-nents of a complex task is exacerbated by the factthat the critical skills may change as expertisedevelops. Ackerman's (1992) theory of skill acqui-sition proposes that during the early stages oflearning, knowledge is declarative and is basedlargely on working memory and general intelli-gence. However, later in learning, consistent com-ponents become automated, and the importanceof working memory and general intelligence be-comes attenuated and the importance of percep-tual speed increases. Kanfer and Ackerman (1989,1996) have presented skill acquisition data on anair traffic control task that were consistent withAckerman's predictions. The point here is thatidentification of the critical components based onempirical data may depend on the skill level of theindividuals being studied. Moreover, changes inthe criticality of skills during acquisition suggestthat the order in which component tasks arepresented maybe important.

Interactions Among Component Tasks

Part-task training effectiveness also depends onthe interaction among task components and thestrategies for organizing the components into thewhole. Gopher, Weil, and Siegel (1989) argued

ANALYSIS OF PART-TASK TRAINING 229

that a complex task is best thought of as "anorganized set of response schemas" (p. 148) ex-ecuted and coordinated by high-level schemas orstrategies. They developed a training regime thatrequired individuals to shift emphases between thevarious components (i.e., ship control, mine man-agement, and bonus management) of a complextask (i.e., Space Fortress). The advantage of theirapproach is that individuals can avoid the cognitivelimits of the whole task by focusing on the compo-nents as they occur in the context of the wholetask.

A production system approach also emphasizesthe importance of interactions among the compo-nents and their relationship to the whole task(Anderson, 1983). This approach holds that trans-fer depends on whether identical procedures canbe used in the acquisition and transfer contexts.Singley and Anderson (1989) argued that "knowl-edge comes in declarative form, is used by weakmethods to generate problem solutions, and as abyproduct, new productions are formed. The keystep is the knowledge compilation process, whichproduces the domain-specific skill" (p. 50). Declara-tive knowledge serves as input for the knowledgecompilation phase, thus the quality and structureof this information influences the compilationprocess. According to this view, exposure to thecriterion task (i.e., the whole task) would be mostinfluential early in training when it can provide anappropriate template for the development of do-main-specific productions during the knowledgecompilation phase. Without early exposure to thecriterion task, productions are developed outsidethe context of the criterion, and therefore may notmap onto the relevant domain.

In contrast, Schneider and Detweiler (1988)suggested that exposure to the whole task will bemost helpful during the middle phase of trainingafter the component task procedures have beencompiled to the point that they are stable. Theyargued that integration of component skills isaccomplished through compensatory activities andthat these compensatory activities must be basedon stable component task procedures in order tobe effective. The data of Carlson, Khoo, and Elliot(1990) supported this hypothesis. They concludedthat trainees should not be introduced to the targetdomain until after they have received sufficientpractice on the component tasks. The question ofwhen to introduce the target task is beyond the

scope of the present article; however, we providedall participants with a brief, early exposure to thetarget task (i.e., whole task) in the form of apretest.

Individual Differences

Differences in spatial ability may play a role inthe effectiveness of part-task training of complexspatial tasks such as that used in the present study.Spatial skill represents a major individual differ-ences factor distinct from other abilities like ver-bal, quantitative, or reasoning skill (see Lohman,1979, for an excellent review of this factor). Effec-tive use of spatial information is one aspect ofhuman cognition and shows up in situations asdiverse as navigating through a novel environmentto determining the trajectories of approachingobjects. One interesting aspect of spatial ability isthat there have been countless studies reported inthe literature that have revealed significant genderdifferences (e.g., Linn & Petersen, 1985; Maccoby& Jacklin, 1974; McGee, 1979; Voyer, Voyer, &Bryden, 1995). Invariably, these studies have shownthat men perform better on a wide array ofsmall-scale spatial tasks compared with women.

For instance, Linn and Petersen (1985) con-ducted a meta-analysis on studies reported in anearlier review on this topic (i.e., Maccoby &Jacklin, 1974). They found gender differences intwo categories of spatial tests: spatial perception(effect size M = 0.44,p < .05) and mental rotation(effect size M = 0.73, p < .05). Their third cat-egory of spatial tests (i.e., spatial visualization) wasnot significantly different between women andmen (effect size M = 0.13, p > .05). More re-cently, Voyer, Voyer, and Bryden (1995) per-formed a very large meta-analysis of studies citedin the Maccoby and Jacklin (1974) review, studiescomprising the Linn and Petersen (1985) meta-analysis, as well as studies culled from an extensiveliterature search through 1993 on the PsycLitdatabase of the American Psychological Associa-tion. Specific tasks were included in this meta-analysis only if at least five studies had alreadybeen conducted with the task. This was done toallow meaningful tests of homogeneity and reason-able estimates of effect size for each test. Cohen'sd was used in the final analysis of 286 studies. Theauthors found a mean weighted d = 0.37 (z = 2.61,


p < .01), which indicated highly significant genderdifferences in spatial abilities that favor men.

Spatial aptitude may also interact with trainingcondition to influence skill acquisition. For in-stance, inefficiencies in a particular training re-gime may be most evident for low-aptitude learn-ers. There is a large body of research documentingAptitude x Treatment interactions (ATI) thatshows that the effectiveness of training interven-tions (treatments) depends on the aptitudes of thelearners. For example, Wightman and Sistrunk(1987) showed that high-aptitude learners, but notlow-aptitude learners, were able to overcome theinitial disadvantage of whole-task training in anaircraft simulator landing task. Similarly, Shute(1993) showed that in the context of a flight-engineering tutor, learners with low incomingknowledge but high working-memory capacity ben-efited from an extended practice environment.Learners with high incoming knowledge and lowworking-memory capacities benefited from theabbreviated practice environment. Other learners(i.e., high working memory-high knowledge andlow working memory-low knowledge) did not showdifferential learning in either environment. Wewill pay particular attention to the potential ofATIs in the current study because of the strongdifferences in spatial aptitudes that have beenobserved in the literature.

Transfer and Backward Transfer

Of the challenges faced by part-task trainingdescribed above, we focus on the problem ofidentifying component tasks that are critical to thewhole task. Wightman and Lin tern (1985) sug-gested that to implement part-task training effec-tively, critical skills must be identified with acomponential analysis of the target task. Subse-quently, tasks designed to enhance the criticalskills must be developed and validated. One valida-tion technique recommended by Wightman andLintern (1985) is referred to as backward transfer.To illustrate the backward-transfer paradigm, wecompare it with a more familiar paradigm calledtransfer of training (TOT).

When the TOT paradigm is used to test part-task training effectiveness, the experimental de-sign should include at least one control group thatreceives training on the whole or criterion taskthroughout training and at least one experimental

group that receives an equivalent amount of train-ing on one or more part tasks. In addition, bothgroups should also be given an appropriate test onthe criterion task. The test phase is often called atransfer phase, even though the control group doesnot transfer to a different task. This experimentaldesign permits calculation of transfer as well asdifferential transfer. In estimating transfer, thetransfer performance of the experimental group(i.e., part-task group) is compared with the perfor-mance of the control group (i.e., the whole-taskgroup) during initial training. Differential transferinvolves the comparison of the control and experi-mental groups relative to equal amounts of experi-ence. In other words, the transfer performance ofthe experimental group is compared with thetransfer performance of the control group.

There are many different formulas for measur-ing transfer and differential transfer, but in gen-eral, transfer can be negative, positive, or evenexceed 100%. Negative differential transfer indi-cates that the training regime disrupted or im-peded skill development. Positive differential trans-fer that is less than 100% indicates that some skillswere acquired, but part-task training was lessefficient than whole-task training. Differentialtransfer that exceeds 100% indicates that part-tasktraining is more efficient than whole-task training.Although a training regime that produces positive,differential transfer exceeding 100% is ideal, onethat produces positive differential transfer below100% is of value in applied settings where part-task training is sufficiently safer or less expensivethan whole-task training (e.g., aircraft simulatortraining).

When the backward-transfer paradigm is ap-plied to part-task training, the general experimen-tal design is similar to a TOT design, but the goal isto measure transfer from the component tasks tothe criterion task. At least one group is trained onthe criterion task (i.e., whole-task group), and atleast one group is trained on the component tasks(i.e., part-task group) during the training phase.During the transfer phase, both groups are testedon the component tasks rather than the criteriontask. Backward transfer to the component taskscan be estimated by comparing the transfer perfor-mance of the whole-task group with the initialtraining performance of the part-task group. Com-ponent tasks that show positive backward transferinvolve skills presumably acquired by the whole-


task group during training and thus are critical tothe whole task.

Salthouse and Prill (1983) used a modifiedbackward-transfer technique to identify the compo-nent skills critical to a trajectory intersection task.They trained individuals on the trajectory-intersec-tion task and divided them into high- and low-ability groups that were based on performancelevel. Then they compared the two groups onseveral component tasks derived from a prelimi-nary model of the trajectory-intersection task. Themodel assumed that component tasks were ex-ecuted sequentially. Although their study is a goodillustration of the backward-transfer approach,Salthouse and Prill found that component effective-ness was not strongly related to skilled perfor-mance. In addition, improvements in trajectory-intersection performance were not accompaniedby improvements on the component tasks. Theyconcluded that skilled performance was associatedwith more effective strategies for executing compo-nents rather than skill on the individual compo-nents. That is, high- and low-skill performersshowed similar skill levels on the component tasks,but high-skill performers showed evidence of ex-ecuting the component sequence repetitively, andlow-ability individuals executed the entire se-quence only once.

There were two major goals of the present study:(a) to determine the utility of the backward-transfer technique for developing effective part-task training regimes and (b) to examine the roleof spatial ability differences in the acquisition of acomplex spatial task. In Experiment 1, we com-bined the backward-transfer methodology with aTOT design to identify several component taskscritical to a complex flight task. In Experiment 2, apart-task training regime composed of componenttasks deemed critical in Experiment 1 was com-pared with (a) a whole-task training regime and(b) a part-task training regime composed of non-critical component tasks.

The component tasks were developed on thebasis of observations of individuals in several pilotstudies and were intended to represent a hierarchyof skills. Some of the tasks focused on basic flightskills related to pitching and rolling1 the plane. Inaddition, several tasks required integration of thebasic skills into more complex maneuvers. Thesetasks required individuals to change heading and/oraltitude. At the next level in the hierarchy were

several tasks requiring participants to fly through asingle gate. The tasks at the highest level of thehierarchy required participants to locate a gate ona navigational display and then find it and flythrough it. These tasks require three-dimensionalspatial orientation to translate the "god's eye"view into the perspective "out-of-the-cockpit" view.These tasks were intended to integrate skills usedin all the other tasks. In this hierarchy, the gate-aiming tasks represent the part-task method ofsegmentation, and all other tasks represent frac-tionation.

Experiment 1

There were three primary goals underlying Ex-periment 1. First, we were interested in determin-ing the overall effectiveness of a part-task trainingapproach compared with a whole-task approachusing a desktop flight simulator as the criteriontask. The second goal was to isolate the componenttasks and determine their relevance to the crite-rion task. Finally, we wanted to explore the relation-ship between spatial aptitude and training ap-proach. The experimental design is a combinationof the TOT and backward-transfer paradigms. Thewhole-task group received in order a trainingphase on the criterion task, a transfer phase on thecriterion task, and then a backward-transfer phaseon the component tasks. The part-task traininggroup received a training phase on the componenttasks, a transfer phase on the criterion task, and abackward-transfer phase on the component tasks.This design allows us to estimate: (a) transfer fromthe component tasks to the criterion task bycomparing the transfer performance of the part-task group with the initial training performance ofthe whole-task group; (b) differential transfer bycomparing the transfer performance of the part-task group with the transfer performance of thewhole-task group; and (c) backward transfer fromthe whole task to the component tasks by compar-ing the backward-transfer performance of thewhole-task group to the initial training perfor-mance of the part-task group. Differential back-

1 To pitch and roll the plane means to rotate the planearound axes of the plane so that the nose moves up ordown relative to the tail (i.e., pitching), or the right wingmoves up or down relative to the left wing (i.e., rolling).


ward transfer cannot be estimated because thepart-task group experiences the component tasksas well as the criterion task prior to the backward-transfer phase.

Assessment of backward transfer is crucial foridentifying the relevant and irrelevant componenttasks. Elimination of component tasks irrelevant tothe criterion task would presumably render thepart-task training regime more efficient. Finally,we wanted to examine performance data on thistask in relation to differences in spatial ability.Specifically, we hypothesized a main effect ofspatial ability and an interaction between trainingcondition and spatial ability where low-aptitudelearners would show larger benefits of part-tasktraining than high-aptitude learners. This hypoth-esis was motivated by findings of Wightman andSistrunk (1987) that high-aptitude learners, butnot low-aptitude learners, were able to overcomethe negative effects of whole-task training. Ourhypothesis is also based in part on consideration ofpotential ceiling effects.

Method

Participants

We recruited 42 men and 38 women throughlocal temporary employment agencies in San Anto-nio, Texas, to take part in the study. Participantswere paid about $5.00 an hour. Participants rangedin age from 18 to 30 and reported spending lessthan 20 hr per week playing computer and videogames (Mdn — 2.0 h). All participants had a highschool diploma or general equivalency diploma(GED), but none had completed a 2- or 4-yearcollege degree. Also, none of the participants hadever flown the flight simulator used in the presentstudy. The participants were assigned randomly toone of two groups (i.e., the whole-task and part-task training). There were 21 men and 19 womenin each group; however 2 women in the whole-taskgroup did not complete the study, and their datawere excluded from all analyses.

Equipment

The study was conducted at the ArmstrongTraining Research for Automated Instruction(TRAIN) Laboratory (Lackland Air Force Base,Texas), which contains 30 laboratory stations, each

equipped with a Deskpro 486/33L (Compaq),Multisync 4DS monitor (NEC), and a flight stick(CH Products). A desktop flight simulator calledPhoenix (Galaxy Scientific, San Antonio, TX) wasused for training and data collection. This desktopsimulator utilizes simplified dynamics in that thethree axes of aircraft rotation are independent.For example, changes in roll do not effect pitch oraltitude. The Phoenix display (see Figure 1) showsflight relevant information and a simulated environ-ment as depicted from inside the cockpit lookingout of the windscreen (perspective view). Thebottom half of the display represents the cockpitpanel with indicators for thrust, missile range,target distance, a dynamic navigational display,and various status indicators, such as afterburnersand landing gear. In addition, the display containsa dialog box for on-line instructions and feedbackthat appears on the cockpit panel. The top half ofthe display shows a simulated world and a head-updisplay (HUD) containing flight information. Inthe simulated world, the horizon is depicted as ablue line, and the ground is displayed as a red gridagainst a black screen simulating night flying. TheHUD includes indicators for airspeed (left side),altitude (right side), and heading (bottom) andshows a climb/dive ladder that indicates pitch androll. Figure 1 also shows how a typical slalomcourse was depicted.

Tasks

Slalom task. The criterion flight task consistedof an airborne slalom course where participants"flew" the simulator through "gates" in the sky. Agate was represented as four octahedrons ar-ranged in a square suspended in the simulatedenvironment. The gates were positioned so thatparticipants had to turn left or right and climb ordive in order to fly from one gate to the next. Fourdifferent courses were created that varied in diffi-culty. The two easy courses consisted of gateswhose centers were 1,000 simulated feet (304.8 m)apart in altitude and averaged 4,000 simulated feet(1,219.2 m) between gates. Given these param-eters, participants could generally see the next gatefrom the gate they were currently flying through.In the two difficult courses, the gates were 2,000simulated feet (609.6 m) apart in altitude withabout 3,200 simulated feet (975.4 m) betweengates. These courses required sharper turns and


470-

480- * *

- 4.0

- 4.5

490-

0 V » 8

-T~ —-1

» » - 5.0

- 5.5

Figure 1. Phoenix computer display.

steeper climbs and dives than the easy courses, andparticipants rarely saw the next gate from the gatethey were flying through. However, for both easyand difficult courses, the radar map showed thehorizontal position of the gates at all times. Whenparticipants successfully flew the simulator througha gate, the following message was presented in thedialog box at the bottom of the screen: "Good!You made it through gate n. Go on to gate n + 1."When participants missed a gate, the followingmessage was displayed: "Sorry! You missed gate n.Go on to gate n + 1." Participants receivedinstructions to proceed to the next gate when theymissed one and were not given credit for returningto a missed gate. Only seven gates were displayedat a time. When participants completed the lastdisplayed gate, another set of seven gates waspresented and so on until the end of the trial. Eachtrial lasted 3 mins. Participants received instruc-tions to fly through as many gates as possible whileminimizing misses. After each trial, participantsreceived information on how many gates theysuccessfully flew through.

Component tasks. There were a total of 19different component tasks derived from a system-

atic analysis of skills involved in the slalom task.These tasks represented a hierarchy of skills rang-ing from elementary "stick and rudder" skills (e.g.,controlling pitch and roll) to more complex ones(e.g., spatial orientation). The following descrip-tion of component tasks represents the order inwhich participants performed them.

One of the most basic skills in flying involvesmaneuvering the plane back to straight and levelflight. This skill is required to recover from un-usual attitudes (i.e., nonzero pitch and/or roll).Thus, the first three tasks (unpitch, unroll, andunpitch-roll) were intended to train participantson recovering from nonzero pitch and roll. Trialswere started with the simulator pitched up or down(unpitch), rolled left or right (unroll), or pitchedand rolled (i.e., unpitch-roll). Participants weregiven 30 s to bring the pitch, roll, or both aspectsback to level. Trials were presented in blocks of 10trials.

The next three tasks represented additionalbasic skills required for maneuvering the plane(i.e., pitching and rolling skills). In all three tasks,trials started with the simulator flying straight andlevel. Participants were instructed to pitch and/or


roll the plane to designated values. In the pitchtask, participants pitched the plane up or down; inthe roll task, participants rolled the plane left orright. In the pitch-roll task, participants pitchedand rolled the plane to specified degrees. As in theprevious tasks, participants had 30 s for each trial,and trials were presented in blocks of 10 trials.

In the next three tasks, participants adjustedtheir altitude, heading, or both. These tasks areslightly more complex than the previous onesbecause they require a combination of the afore-mentioned basic skills. For example, changingaltitude requires individuals to pitch the plane up(or down), watch the altimeter while ascending (ordescending), and level out at the proper altitude bybringing the pitch back to zero. Thus, skill inchanging altitude and heading depended on thebasic skills emphasized in the first six tasks. In thecurrent group of component tasks (i.e., altitude,heading, and altitude-heading), the simulator wasstarted in a straight and level orientation, andparticipants achieved a new altitude, heading, orboth. Once attained, the new altitude and/orheading had to be maintained for 5 s. Participantshad 2 min for each trial, and trials were presentedin 10-trial blocks.

The next six tasks emphasized gate-aiming skills,requiring participants to fly through gates in thesky. Starting from a position in front of a singlegate, participants flew the simulator through thegate as quickly as possible. These six tasks repre-sent difficulty manipulations on three dimensions:sharpness of turn, airspeed, and gate size. In the"easy gate" task, the starting positions were almostdirectly in front of the gate, and participantsneeded to make only small turns. In the "hardgate" task, participants had to make sharper turnsto fly through the gates. In the "slow gate" task,the speed of the simulator was fixed at half thrust;in the more difficult version, "fast gate," thesimulator was fixed at full thrust. The "big gate"task consisted of gates that were 50% larger thanin the slalom task; in the "tiny gate" task, gateswere 50% smaller. These tasks were designed toprovide practice on approaching gates from vari-ous angles, at various speeds, and with varyingdegrees of precision. Participants had 30 s to flythrough the gates comprising these componenttasks.

The final four tasks focused on spatial orienta-tion skills. In these four tasks, participants located

a gate on the radar map and flew through it.Successful performance on this task required inte-gration of information from the top-down view ofthe navigational display with the perspective viewof the cockpit windscreen. In these tasks, one gatewas located in each quadrant of the airspace.Trials started at random positions in the environ-ment, and participants located their position andthe position of the specified gate, and then flewthrough the gate. In two of the four tasks (Orient-Plan I and Orient-Plan II), participants had unlim-ited time to plan their course. That is, they werefirst shown their position and told which gate to flythrough. Participants then initiated each trial them-selves after they had planned their course to thegate. In the other two tasks (Orient I and OrientII), participants did not have time to plan theircourse. Rather, they were told which gate to flythrough, but were not shown their position untilafter the trial had started. Thus, once the trialbegan, participants had to orient themselves inspace quickly and find the correct gate. The romannumerals following the task names represent diffi-culty level, with I indicating easy versions and IIindicating difficult versions of the tasks. Difficultyon these trials was defined in terms of the positionof the plane relative to the target gate, how easy itwas to find the correct gate, and how sharp a turnwas needed. In all four tasks, participants had 30 sto fly through the correct gate.

Cognitive ability tests. Three tests were used toassess spatial aptitude from an on-line battery ofcognitive ability measures (CAM-4; Kyllonen etal., 1990). These spatial tests assessed participants'working-memory capacity, information-processingspeed, and inductive reasoning skill, using spatial(as opposed to verbal or quantitative) stimuli. Alltests were presented on the same computers usedfor the Phoenix task.

The working memory test was a four-term order-ing task with blocks as the stimuli. In this test,participants were required to relate what wasdescribed in three pictorial statements to thesequence of four block figures. Figures consisted ofblocks divided by a diagonal line and colored pinkwith black or blue with black (e.g., one side ofdiagonal is pink, and the other is black). Thedirection of the diagonal could change positions,allowing for different combinations (e.g., a diago-nal going from the top left to the bottom right maycause pink to be on the top and black to be on the


bottom; a diagonal going from the top right to thebottom left may cause the black to be on top andthe pink to be on the bottom). For each statement,two blocks of the same color (i.e., either pink/black or blue/black) appeared with an arrow. Thearrow described the sequence in which the twoblocks should appear (e.g., one on top of theother). The arrow could have had a slash throughit, which was interpreted as meaning not (e.g.,Block 1 will not appear below Block 2). The thirdstatement merely displayed solid pink and solidblue blocks, describing the sequence of the pinkand blue blocks (e.g., pink will not appear beforeblue).

The pictorial statements appeared one at a timeat the top of the screen. Participants determinedthe sequence of blocks as the statements appeared.After the final statement, eight numbered alterna-tives appeared on the screen with a timer. Thesealternatives expressed the possible combinationsusing the presented statements, and the partici-pants were required to type the number corre-sponding to the correct sequence. Correct re-sponses were followed by music, incorrect responsescaused a buzzer to be sounded, and the threestatements were then displayed to show how theitem was incorrect. Next, three asterisks appearedto warn participants that the next item was aboutto be presented. This test contained 24 items. Thealternative responses were 1-8 using the numberkeys at the top of the keyboard.

The information-processing speed test used simi-lar stimuli as the working memory test above, butdiffered in that it consisted of a two-term orderingtest. Participants decided as quickly as possiblewhether the presented figure combinationsmatched the simple sequence formula specified byfigure statements initially provided on the screen.Shortly after each figure statement was presented,a set of two blocks was shown in the middle of thescreen. If the figure and the initial statementmatched, L was the correct response; if they didnot match, D was the correct response. Upon acorrect response, music was sounded. Incorrectresponses were followed by a buzzer. The nextitem, preceded by three warning asterisks, wasthen presented. At the end of the test, the percent-age correct and average response time were dis-played. This test contained 12 items.

In the inductive reasoning test, participantswere shown a 3 x 3 matrix in which a figure was

contained in all but one of the cells. Participantslooked at the figures and applied horizontal and/orvertical rules to determine what figure belonged inthe empty cell. The matrix was shown on thescreen concurrent with the eight alternative re-sponses. Participants typed in the alternative theybelieved corresponded to the matrix. Some of thethemes used were gradual shading of figures,alternating positions of figures (e.g., a square,circle, and triangle; a circle, triangle, and square; atriangle, square, and circle), successive movementsof shapes to form new figures, successive additionsor deletions to figures, and rotation of figures. Allthemes were used in combination to form bothhorizontal and vertical rules for participants toinduce. Participants determined the missing figureas quickly as possible without sacrificing accuracy.After entering a response, participants were in-formed whether it was correct. If no response wasentered, the item was counted wrong. Participantshad 10 min to solve all nine problems contained inthis test. There were eight alternative responsesusing the 1-8 keys at the top of the keyboard.

Procedure

Participants completed training and testing ingroups of 18 to 22. Approximately half of eachgroup was assigned randomly to each of the twoexperimental groups. The study took 3 consecutivedays to complete. All training was completed onDay 1. Participants signed a consent form, com-pleted the CAM tests, and then received Phoenixinstructions and a brief computer-based introduc-tion to the Phoenix simulator. This introductionfamiliarized participants with the Phoenix displaysand controls. All participants then received apretest consisting of 4 trials on the slalom task (1trial for each course). This pretest not only servedto assess initial ability but also provided partici-pants the criterion task context for subsequenttraining (Carlson et al., 1990; Singley & Anderson,1989). Following the pretest, the whole-task groupcontinued to practice on the slalom task for fiveblocks of 16 trials each. Each block of trials wasseparated by either a 15-min rest period or a 1-hlunch period.

For the part-task group, the component skillswere presented in the order in which they weredescribed above, and all were self-paced. How-ever, total time on training was equated for all


participants in the part-task group by requiringfaster participants to repeat trials when they fin-ished early. Equating part-task participants ontotal time, rather than equating total number oftrials (or some other objective criterion), poses apotential problem. Skilled participants who com-pleted the task faster would receive more practicetrials compared with the less skilled participants.However, this same potential problem existed forthe whole-task group in that highly skilled partici-pants would be able to fly faster and completemore gates per trial compared with the lesserskilled participants. Hence, the two groups weretreated similarly with regard to practice trials.Moreover, our procedure equated all participantson total time on the simulator and allowed us tosynchronize rest and lunch periods between thetwo groups.

On Day 2, the transfer phase, participants re-ceived three blocks of 16 trials on the slalom task.As during training, blocks lasted approximately 50min and were separated by 15-min breaks. Thistransfer phase was the first time since the pretestthat the part-task group had performed the slalomtask. On Day 3, the backward-transfer phase, allparticipants completed the component skills tasks.Similar to Day 1, the component tasks were pre-sented in five 1-h blocks separated by either a15-min break or a 1-h lunch period.

Results

Spatial Aptitude

To explore the role of spatial aptitude on train-ing effectiveness, we conducted a factor analysis onthe accuracy scores on the three spatial aptitudetasks (i.e., working memory, processing speed, andinductive reasoning). Because of difficulties in thedata collection process, only 65 participants hadvalid data on the spatial aptitude tasks (32 partici-pants in the whole-task group and 33 in thepart-task group). With a principal axis factoranalysis, one factor was extracted, and the solutionaccounted for 67% of the variance. Factor scoreswere saved, and a median split procedure was usedto identify high- and low-aptitude individuals.

Slalom Task Performance

The two criterion measures of performance onthe slalom task were speed (i.e., the total number

of gates flown through) and accuracy (i.e., totalgates successfully completed divided by the totalnumber of gates possible). Of the 16 trials perblock, 8 were from easy courses (Trials 1, 2, 5, 6, 9,10, 13, and 14), and 8 were from difficult courses(Trials 3, 4, 7, 8, 11, 12, 15, and 16). For thepurpose of reducing within-subjects variability, weaveraged performance across blocks. The data inFigures 2 and 3 represent block averages on easy(see Figures 2a and 3a) and difficult courses (seeFigures 2b and 3b). Thus, each point in thesefigures represents average performance on eighttrials in each transfer block. An alpha level of .05was used for all significance tests.

Speed data. Figure 2 shows average training(whole-task group only) and transfer scores oneasy (see Figure 2a) and difficult (see Figure 2b)courses for high- and low-aptitude groups. Sepa-rate 2 (training condition) x 2 (aptitude) x 3 (block)multivariate analyses of variance (MANOVAs),2 withblock as a repeated measures variable, were per-formed on easy and difficult trials. For the easycourses, as shown in Figure 2a, whole-task partici-pants performed better (M - 16.41) than part-taskparticipants (M = 12.83), and high-aptitude indi-viduals (M = 15.75) performed better than low-aptitude individuals (M = 13.47). In addition,scores improved from Block 1 (M = 12.07) toBlock 2 (M = 15.34) and Block 3 (M = 16.31). Thethree-way MANOVA supported all these observa-tions by showing significant main effects of trainingcondition, F(\, 61) = 6.70, p = .012, spatialaptitude, F(l, 61) = 4.08, p = .048, and block,Wilks's exact F(2, 60) = 42.52,;? = .001. None ofthe interactions reached statistical significance.

For difficult courses, the analysis revealed asignificant main effect of block, Wilks's exact F(2,60) = 18.07, p = .001, and a significant TrainingCondition x Spatial Aptitude interaction, F(l,61) = 4.27, p = .043. The main effect of blockindicates participants improved across transferblocks (M = 8.72, 10.69, and 10.56 for Blocks 1, 2,and 3, respectively). Further analysis of the Train-ing Condition x Spatial Aptitude interaction re-vealed a main effect of training condition forlow-aptitude individuals, F(2, 31) = 5.54,p = .025,but not for high-aptitude participants, F(2, 30) <

2 The MANOVA procedure was used for repeatedmeasures analysis as recommended by O'Brien andKaiser (1985).

ANALYSIS OF PART-TASK TRAINING

(a) Easy Courses (b) Difficult Courses

237

20

15

O 10

20

15

10

1

High-Whole High-Part«• •«•

Low-Whole Low-Part•B- •«•

1 2 _3_ 4 5 1 2 3TralringBlock Transfer Block

1 2 3 4 5 1 2 3

Training Block Transfer Block

Figure 2. Total number of gates flown through by high- and low-aptitude participantsin the whole-task training and part-task training groups on the slalom task.

1.0. These findings and the general pattern inFigure 2b suggest that low-aptitude participantsdid not perform well following part-task training.

Accuracy data. Figure 3 shows the averageaccuracy on easy (see Figure 3a) and difficult (seeFigure 3b) courses during the transfer phase. Inthis figure, the abscissa differs from that in Figure 2because the training data are not shown. Thesedata were submitted to the same 2 x 2 x 3MANOVAs that were used to analyze the speed

data. Data from the easy courses showed thathigh-aptitude individuals performed more accu-rately (M = 73.4%) than low-aptitude individuals(M = 56.8%). This difference was supported by asignificant main effect of spatial aptitude, F(\,61) = 13.78, p = .001. The analysis also revealed amain effect of block, Wilks's exact F(2, 60) = 5.74,p = .005, and a Spatial Aptitude x Block interac-tion, Wilks's exact F(2, 60) = 3.28, p = .044. Theinteraction was further examined by testing the

(a) Easy Courses (b) Difficult Courses

90

80

70

b« 602| 50

40

30

20

4-

QEl

€>

90

80

70

60

50

40

30

High-Whole High-Part

Low-Whole Low-Part

J^ «^v

Transfer Block Transfer Block

Figure 3. Accuracy performance (i.e., percentage of gates made) of high- andlow-aptitude participants in the whole-task training and part-task training groups on theslalom task.


simple main effect of spatial aptitude for eachtransfer block separately. These analyses revealedthat the main effect of spatial aptitude was verystrong for Block 1, F(l, 61) = 14.70,p = .001, andBlock 2, F(l, 61) = 15.27, p = .001, relative toBlock 3, F(l, 61) = 7.11,p = .011. These analysesreveal that differences in spatial aptitude wereattenuated across transfer blocks but were noteliminated.

For difficult courses, the 2 x 2 x 3 MANOVArevealed significant main effects of spatial apti-tude, F(l, 61) = 7.92, p = .007, and block, Wilks'sexact F(2, 60) = 5.86, p = .005. In addition,training condition interacted with spatial aptitude,F(l, 61) = 4.68, p = .034, and block, Wilks's exactF(2,60) = 4.15,;? = .020. To examine the TrainingCondition x Spatial Aptitude interaction, themain effect of training condition was examined forlow- and high-aptitude individuals separately. Low-aptitude individuals showed better performancefollowing whole-task training (M = 49.9%) thanpart-task training (M = 36.9%), F(l, 31) = 4.38,p = .045. High-aptitude individuals performedslightly better following part-task training (M =59.3%) than whole-task training (M = 52.9%),although this difference was not statistically signifi-cant, F(l, 30) < 1.0.

The Training Condition x Block interaction wasexamined by testing the simple main effect of blockfor each group separately. Individuals receivingpart-task training showed significant improvementacross the three test blocks, Wilks's exact F(2,30) = 11.03, p = .001. Whole-task individualsshowed a decline in performance across test blocksthat approached statistical significance, Wilks'sexact F(2, 30) = 3.20,p = .055. The improvementin scores by the part-task group may be due in partto the lower overall performance on the initialposttest block. The slight decline in performanceby the whole-task group may be due in part tofatigue.

Differential transfer. To determine the amountof transfer, we used the formula of Katona (1940):percentage transfer = 100 x (E, — Cj)/(C, - C,),where E, is the average performance of the part-task group on the criterion task during transfer, C,is the average performance of the whole-taskgroup during initial training, and C, is averageperformance of the whole-task group during trans-fer. For easy courses, E, = 10.13 gates, C, = 4.71gates, C, = 14.07, and the percentage transfer is

57.9%. For difficult courses, E, = 7.78 gates, C, =5.48 gates, C, = 9.68, and the percentage transfer is54.9%.

Component Task Performance

Next, we wished to examine performance on thecomponent tasks. Positive transfer from the wholetask to the component tasks indicates which of thecomponent tasks are important to the whole task.That is, if the whole-task group performs better ona component task during transfer than the part-task group does during initial training, then wemay infer that the component task involves skillsacquired during whole-task training. These skillsare presumed to be important for skilled perfor-mance on the whole task.

Analysis of component factors. Rather than con-duct separate significance tests for each of the 19component tasks, we wished to reduce family-wiseerror rate by performing analyses on a small set ofcomposite scores created by combining similarcomponent tasks. One method of generating com-posite scores would be to combine componenttasks that were based on our task hierarchy.Instead, we performed a global factor analysis oncomponent task scores across all participants.Even though a global factor analysis could becriticized for masking important differences be-tween the groups, we used the global analysis fortwo reasons. First, we wanted to combine compo-nent tasks that were based on empirical datarather than theory. Second, factor scores areattractive in that they are standardized scores(M = 0, SD =1). Running separate factor analy-ses for each group makes it impossible to comparegroups because the means and standard deviationswould be identical across groups. In summary, ourgoal was not to create definitive factor solutions foreach group but to reduce the number of signifi-cance tests, the family-wise error rate, and thecomplexity of the interpretation.

Using the principal-axis factoring procedurewith varimax rotation, we extracted five factorsthat accounted for 80.0% of the variance. Therotated factor matrix is shown in Table 1. As shownin this table, Factor 1 loads strongly on the six gatestasks. Recall that these tasks required individualsto fly through a single gate of variable size fromdifferent angles and speeds. Hence, Factor 1 isGate Aiming. The second factor loads most heavily


Table 1Task Loadings on the Five Rotated Factors Extracted From the Analysisof the Backward-Transfer Group

Task

GateHardEasySlowBigTinyFast

OrientPlan IIIIPlan II

UnpitchUnrollUnpitch-rollPitchHeadingAltitude-HeadingAltitudeRollPitch-roll

Gateaiming

.87

.86

.85

.80

.76

.71

.17

.38

.34

.24

.11

.08

.14

.24

.04

.21

.15

.20

.41

Spatialorientation

.15

.26

.27

.25

.13

.32

.82

.79

.73

.72

.17-.01

.18

.20

.17

.29

.38

.13

.14

Factor

Recovery

.06

.13

.16

.26

.22

.08

.15

.22

.21

.04

.78

.76

.74

.59

.15

.29

.38

.17

.16

Altitude-heading

.21

.18

.12

.06-.14

.08

.16

.22

.22

.10

.23

.05

.28

.01

.84

.72

.41

.03

.23

Roll

.08

.09

.17

.15

.16

.27

.19

.02

.17

.03

.04

.08

.18

.09

.06

.11

.26

.85

.73

on the four spatial orientation tasks. The distin-guishing feature of these tasks is that they allrequire the individual to use the radar map inorder to navigate to a specific gate. Thus Factor 2is Spatial Orientation. Factor 3 loads most stronglyon the three tasks requiring participants to "re-cover" from nonzero pitch and roll to return tostraight and level flight (i.e., unpitch, unroll, andunpitch-roll). In addition, this factor also loads onthe pitch task. Because of the nature of the firstthree tasks, this factor is called Recovery. Factor 4loads most strongly on the two tasks requiringindividuals to adjust their heading and altitude(i.e., heading and altitude-heading). For this rea-son, Factor 4 is Altitude-Heading. The fifth factorloads almost exclusively on the two tasks thatrequire individuals to roll to a specific angle. Inthese tasks, individuals judge their roll by observ-ing the slant of the horizon line in the simulatedenvironment. There is not a digital indicator forroll, so individuals must be given on-line feedbackabout their roll. This factor is Roll.

It is interesting to note that the groupingsderived from the factor analysis are not very

different from what we would have derived fromthe hierarchy. Had we grouped tasks by the hierar-chy, the groupings would have been recovery tasks(unpitch, unroll, and unpitch-roll), pitch-roll tasks(pitch, roll, and pitch-roll), altitude-heading tasks(altitude, heading, and altitude-heading), gatestasks (all six gates tasks), and spatial orientationtasks (all four). Notice that the only differencebetween this grouping and the one derived fromthe factor analysis is that the pitch task is groupedwith the recovery tasks in the factor analysissolution but is included with the pitch-roll tasks inthe hierarchy-derived groupings.

Average factor scores, separated by trainingcondition and spatial aptitude, are shown in Figure4. Because factor scores are standardized, the datain Figure 4 can be viewed as points above andbelow the mean in standardized units. The fivefactors were submitted to a 2 (TrainingCondition) x 2 (Spatial Aptitude) MANOVA.This analysis revealed significant effects of trainingcondition, Wilks's exact F(5, 57) = 5.06, p = .001,spatial aptitude, Wilks's exact F(5,57) = 2.79,^ =.025, and the interaction, Wilks's exact F(5, 57) =


(a) Gate Aiming (b) Spatial (c) Recovery (d) Altitude- (e)RollOrientation Heading

1.0

U.J

1*£ o.o

-0.5

-1.0

^;;;

//

/4

1 0

.5

00

-0.5

-1.0

.--•..*

* '"

1 0

.5

no

-0.5

-1.0

High Low

\**C""

* \i

1 0

0.5

o n

-0.5

-1.0

•̂ .4.

1 0

U.5

0 0

-0.5

-1.0

/?^/

Part Whole Part Whole Part Whole Part Whole Part Whole

Figure 4, Factor scores of high- and low-aptitude participants in part- and whole-taskgroups on the five factors extracted from the component tasks.

2.S5,p = .023. Univariate tests for the main effectof training condition were significant only for theGate Aiming factor, F(l, 61) = l8A3,p = .001, andRoll, F(l, 61) = 4.17,p = .046. As shown in Figure4, the whole-task group performed better than thepart-task group on gate-aiming and roll tasks. Theunivariate tests for the main effect of spatialaptitude were only significant for gate aiming, F(l,61) = 5.21,p = .026, and spatial orientation, F(l,61) = 5.44, p = .023. Low-aptitude individualsperformed worse than high-aptitude individuals onboth components. For the interaction betweentraining condition and spatial aptitude, the onlyunivariate test that reached significance was forgate aiming, F(\, 61) = 11.84,/? = .001. The data inFigure 4a show the nature of the interaction.Low-spatial ability participants in the part-taskgroup performed much worse on the gate-aimingtasks than any other group. This observation issupported by a significant main effect of trainingcondition for low-aptitude participants, F(l, 31) =25.36, p = .001, but not for high-aptitude partici-pants, F(l, 30) < 1.0. The extremely poor perfor-mance of the low-aptitude participants in thepart-task group may account for the significantmain effects of training condition and spatialaptitude observed for gate aiming.

Backward transfer. Although factor scores areattractive for the reasons cited above, they are notfeasible for the calculation of transfer. Thus, toestimate the amount of transfer from the wholetask to the component task factors, we generated

composite scores by averaging across the compo-nent scores in each factor derived from the factoranalysis. Using these composite scores, we com-puted with Gagne, Forster, and Crowley's (1948)modification of Katona's (1940) equation: percent-age transfer = 100 x (Pbt - />,)/( T - Pt), where Pbl

is the average performance of the whole-taskgroup on the component tasks during backwardtransfer, P, is the average performance of thepart-task group during initial training, and T is thetotal possible score on the component tasks. Wemodified the notation so that in backward transferanalysis, the part-task group is the control orcomparison group instead of the whole-task group,as in the preceding transfer analysis. In addition,we used the total possible score on the componenttasks (i.e., 100%) instead of part-task group scoreson component tasks during backward transferbecause subjects in the part-task group had lengthyexperience on the component tasks during trainingand were thus more likely to approach the maximalpossible score on each component.

Table 2 shows the means and percentage trans-fer scores for the five composite scores. Notice thatGate Aiming and Roll show the largest amount oftransfer. This is consistent with the findings thatthe whole-task group performed significantly bet-ter on the gate-aiming and roll tasks during trans-fer than the part-task group did during training. Itis also worth noting that two of the factors (i.e.,Recovery and Heading) produced negative trans-fer. That is, the part-task group performed better


Table 2Composite Scores and Estimates of Transfer From the Whole Taskto the Component Tasks

Score

Factor

Gate AimingSpatial OrientationRecoveryHeadingRoll

Part task

68.5735.1283.8877.3141.67

Whole task

88.4140.7983.1076.1361.86

Transfer (%)

63.128.74

-4.84-5.2034.61

Note. The total possible score for each factor was 100%.

on these factors in training than the whole-taskgroup did during backward transfer.

Discussion

The primary results of this experiment can besummarized as follows: (a) whole-task training wassuperior to part-task training on speed measuresfor easy courses, (b) low-aptitude individuals in thepart-task training group performed more poorlythan low-aptitude individuals in the whole-taskgroup on difficult courses, (c) the 19 differentcomponent tasks reduced to five unique factors,and (d) only two of the five factors (Gate Aimingand Roll) appeared to be related to proficiency onthe slalom task.

From these results, we have concluded that thepart-task training procedure was only moderatelyeffective. That is, the part-task training groupshowed some transfer from the component tasks tothe slalom task, but the amount of transfer wasrather modest (55% to 58%). The modest benefitsof part-task training observed in the present studyare rather typical of much part-task training ap-proaches in the literature (Wightman & Lintern,1985).

As discussed in the introduction, one explana-tion for why transfer of the part-task traininggroup was less than expected is that the compo-nent tasks did not adequately represent the criticalskills underlying the criterion task. Of the fivefactors we extracted, only two (Gate Aiming andRoll) showed substantial backward transfer fromthe whole task to the part tasks. This findingsuggests that only the Gate Aiming and Rollfactors involve skills that are directly related to thecriterion task. One factor (Spatial Orientation)showed small, nonsignificant backward transfer,

suggesting that the part-task group may haveaccrued small benefits from these component tasks.Two of the five factors (Recovery and Heading)showed small, though nonsignificant, negative back-ward transfer, suggesting that presentation of thesetasks may have a slightly disruptive effect onlearning. Overall, the results suggest that a substan-tial portion of the component tasks did not pro-duce learning benefits. Elimination of the nonben-eficial component tasks may produce a moreefficient part-task training regime.

Results from the gate-aiming analysis (see Fig-ure 4) are particularly interesting with regard tothe Training Condition x Spatial Aptitude interac-tion. Low-aptitude individuals in the part-taskgroup performed dramatically worse than low-aptitude individuals in the whole-task group. High-aptitude individuals showed no effect of trainingcondition. This pattern was also obtained on diffi-cult courses in the criterion task. These findingscontradict our prediction that low-aptitude indi-viduals would perform better in part- than whole-task training because they would be overwhelmedby cognitive demands of whole-task training.Rather, these findings suggest that the low-aptitude individuals were particularly sensitive tothe inefficiencies in the part-task training condi-tion as revealed by the backward-transfer analysis.Stated another way, high-aptitude individuals wereable to overcome the inefficiency of part-tasktraining, but low-aptitude individuals were not.Thus, improving the efficiency of part-task trainingshould have a disproportionately greater benefitfor low-aptitude individuals than high-aptitudeindividuals. In other words, we can specificallyposit an Aptitude x Treatment interaction (Cron-bach & Snow, 1977; Shute, 1992).

Another interesting finding that emerged was


the failure of spatial orientation to show a strongeffect of training condition. Because high-aptitudeindividuals performed better than low-aptitudeindividuals on both the criterion task and thespatial orientation tasks, it is tempting to concludethat spatial orientation is important for the crite-rion task. However, there was no effect of trainingcondition on the spatial orientation tasks. Thissuggests that practice on the criterion task did notimprove performance on the spatial orientationtasks. One possible interpretation of these findingsis that spatial orientation skill is not critical to theslalom task. Note that this conclusion, however, isonly correlational. An alternative interpretation isthat spatial orientation is critical, but not mal-leable in the context of this study (i.e., it representsa fixed ability rather than one that can be manipu-lated through instructional environment).

Experiment 2

In general, the results from the component taskanalysis from Experiment 1 suggest that a signifi-cant portion of the part-task training regime wasdevoted to tasks that were not critical to thecriterion task. Of the five factors identified in thefactor analysis, only Gate Aiming and Roll werestrongly related to proficiency on the slalom task.Moreover, two factors (Recovery and Altitude-Heading) appeared to show negative backwardtransfer, suggesting that part-task training on themmay have disrupted learning on the criterion task.These findings led us to hypothesize that by elimi-nating the "deadwood" tasks, the effectiveness ofthe part-task training regime would be improved.The goal of Experiment 2 was to provide a directtest of this hypothesis.

Specifically, Experiment 2 focused on two hy-potheses suggested by Experiment 1. The firsthypothesis is that a part-task training program thatconcentrates on gate-aiming skills will be moreeffective than one that does not. Second, individu-als showing lower proficiency on the gate-aimingskill should show larger benefits from concentratedpractice compared with participants with higherproficiency. To test these hypotheses, we com-pared a part-task training program, focusing on thegate-aiming component tasks, with one that in-cluded altitude-heading and spatial orientationtasks. Both of these part-task conditions werecompared with a whole-task training condition.

Method

Participants and Equipment

Participants consisted of 66 men and 66 womenrecruited by local temporary employment agenciesin San Antonio, Texas. Participants were paidabout $5.00 an hour. They ranged in age from 18 to30 years of age and reported spending less than20 hr per week playing video games (Mdn = 1.25 hr).Of the 132 participants, 13 did not complete thestudy. Experiment 2 used the same hardware andsoftware as Experiment 1.

Tasks

The slalom task was identical to that used inExperiment 1. In addition, 12 of the 19 componentskills tasks used in Experiment 1 were selected (onthe basis of the factor analysis) and used in thepart-task training conditions. One part-task train-ing group received practice on the six gate-aimingtasks. In these tasks, participants flew through asingle gate from various angles and speeds. Be-cause the gate-aiming tasks could be completedmore quickly than the altitude-heading and spatialorientation tasks, a seventh task (i.e., the pitch taskfrom Experiment 1) was included in the gate-aiming regime to equate the total training time.The second part-task training group received prac-tice on the three tasks comprising the altitude-heading factor (i.e., Altitude, Heading, and Alti-tude-Heading) and two of the four tasks from theSpatial Orientation factor (i.e., locate a specificgate on the radar map and then fly through thegate).

Procedure

Participants, randomly assigned to one of threegroups, (a) whole-task training, (b) part-task train-ing on the gate-aiming tasks, and (c) part-tasktraining on altitude-heading tasks, took part in a2-day study. Individual attrition resulted in slightlyunequal groups. The group receiving whole-tasktraining consisted of 20 men and 20 women, thegroup receiving gate-aiming part-task training had18 men and 19 women, and the group receivingaltitude-heading part-task training had 21 menand 21 women.

On Day 1, participants signed a consent form


and then proceeded to take the spatial abilitypretests. After completing these tests, participantsreceived general instructions on the Phoenix simu-lator, a pretest on general Phoenix skills, and apretest on the slalom task. The slalom pretestprovides important contextual information for sub-sequent training (Anderson, 1983; Carlson et al.,1990). The general flight skill pretest consisted ofone trial in which the participants changed thethrust of the aircraft, two trials in which theychanged pitch, and two in which they changed roll.The pretest on the slalom task consisted of twotrials on the easy courses and two trials on the hardcourses. Following the pretests, participants re-ceived three 1-hr blocks of training in their respec-tive regimes, each separated by a 15-min break. OnDay 2, participants completed the transfer phaseon the slalom task. Transfer consisted of four30-min blocks of trials on the slalom task with easyand difficult courses.

Results

Spatial Aptitude

To identify and group participants into high-and low-spatial ability groups, we conducted afactor analysis on the accuracy scores on the threespatial aptitude tasks (i.e., inductive reasoning,processing speed, and working memory). Becauseof difficulties in the data collection process, only 88participants had valid data on the spatial aptitudetasks. With a principal axis factor analysis, onefactor was extracted, and the solution accountedfor 58.2% of the variance. Factor scores weresaved, and a median split procedure was used toidentify high- and low-aptitude individuals. Thisprocedure resulted in having unequal numbers ofparticipants in each group. The whole-task groupconsisted of 13 low-aptitude individuals and 18high-aptitude individuals. The gate-aiming grouphad 18 low- and 16 high-aptitude participants. Thealtitude-heading group had 13 low- and 10 high-aptitude individuals.

Pretests

Nine pretest measures were assessed. Five of thepretest measures came from the general Phoenixskills pretest: one thrust task latency, two pitchlatencies, and two roll latencies. The other four

pretest measures came from the slalom pretest:two slalom speed scores (one mean score for easycourses and one mean score for hard courses) andtwo slalom accuracy scores (mean scores for theeasy and difficult courses, respectively). To reducethe number of pretest measures, we performed aprincipal axis factor analysis on the nine pretestmeasures. Three factors were extracted and ro-tated with the varimax rotation procedure. Thesolution accounted for 75.6% of the variance. Thefirst factor, Slalom Pretest, consisted of the fourslalom task measures, which were weighted posi-tively and strongly (>.85). The second factor,Pitch Pretest, consisted of the two pitch pretestsand the thrust pretest, with most of the weight onthe down-pitch pretest (.85), followed by the up-pitch pretest (.53), and the thrust pretest (.46).The third factor, Roll Pretest, consisted of the tworoll pretests with the right roll pretest showing alarger weight (.74) than left roll pretest (.49).Factor scores were saved for use as covariates insubsequent analyses.

Although multivariate and univariate testsshowed that groups did not differ on the pretestfactor scores, we tested whether the pretest scorescould be used as covariates. We first calculated thecorrelations between the pretest factor scores andthe measures of performance on the criteriontasks. The first two factors (i.e., Slalom Pretest andPitch Pretest) showed significant correlations withall the performance measures, but the third factor(i.e., involving roll) was not significantly correlatedwith any of the performance measures. Next, wetested the assumption of equivalent slopes andconcluded that the first two factors did not violatethis assumption. Therefore, in subsequent analysesof slalom task performance, we included SlalomPretest and Pitch Pretest scores as covariates.

Component Task Performance

Performance during part-task training was aboutequal for the two part-task training groups. Thegate-aiming group showed slightly higher accuracyduring training (M = 81.4%, SD = 17.1) than thealtitude-heading group (M = 75.2%, SD = 18.1).However, because the two groups performed differ-ent tasks, direct statistical comparisons betweenthe two groups cannot be made.


16.0

14.0

12.0

10.0

6.02 3Transfer Block

Figure 5. Speed scores of high- and low-aptitude partici-pants across four transfer blocks (adjusted by covari-ates). Alt-Hdg = altitude-heading.

Slalom Task Performance

Figure 5 shows speed measures (i.e., meannumber of gates made) on the easy courses on thefour transfer blocks in the posttest, adjusted forthe covariates. These data were analyzed in a 3(Training Condition) x 2 (Spatial Aptitude) x 4(Block) mixed factors MANOVA, with both pre-test factor scores as covariates. Block was the onlyrepeated measure. This analysis revealed only amain effect of block, Wilks's exact F(3,80) = 24.16,p = .001. Planned polynomial contrasts showedsignificant linear, F(l, 80) = 48.52, p = .001, andquadratic effects, F(l, 80) = 15.85, p = .001. Fordifficult courses, the same pattern was observed,with only a main effect of block reaching statisticalsignificance, Wilks's exact F(3, 80) = 6.24, p =.001. Planned polynomial contrasts indicated thatonly the linear effect was significant, F(l, 80) =15.88,p = .001.

Slalom task accuracy scores for the easy courseswere adjusted by the pretest factor scores and areshown in Figure 6. Figure 6a displays data forhigh-aptitude participants; Figure 6b shows datafor low-aptitude participants. The 3 x 2 x 4 mixedfactors MANOVA revealed main effects of train-ing condition, F(2, 80) = 3.45, p = .037, spatialaptitude, F(l, 80) = 8.67, p = .004, and block,Wilks's exact F(3, 80) = 10.83, p = .001. Inaddition, the Training Condition x Block interac-tion was statistically significant, Wilks's exact F(6,160) = 3.29, p = .004. The main effect of spatialaptitude indicates that high-aptitude individuals

performed better (M = 69.6%) than low-aptitudeindividuals (M = 57.8%). The planned compari-sons of the main effect of training conditionindicate that the whole-task group did not differfrom the average of the two part-task groups, F(l,80) = 1.08, p = .303, but the gate-aiming groupperformed better than the altitude-heading group,F(l, 80) = 6.5l,p = .013.

The main effect of training condition must beinterpreted in the context of the training conditionby block interaction. This interaction was exam-ined by testing the simple main effect of trainingcondition for each block separately. For transferBlock 1, the main effect of training condition wassignificant, F(2, 80) = 5.14,p = .008, and plannedcontrasts indicated that the whole-task group per-formed better than the two part-task groups, F(l,80) = 8.54, p = .005, but the two part-task groupsdid not differ from each other, F(l, 80) = 3.20,p =.077. For Block 2, the main effect of trainingcondition was significant, F(2,80) = 4.15,;? = .019,and planned contrasts revealed that the whole-taskgroup did not differ from the average of the twopart-task groups, F(l, 80) < 1.0, but that thegate-aiming group performed significantly betterthan the altitude-heading group, F(l, 80) = 8.29,p = .005. The main effect of training conditionapproached significance for block 3, F(2, 80) =2.95, p = .058, and was not significant for Block 4,F(2, 80) = 1.98, p = .144. Despite the apparentdifferences between the high- (see Figure 6a) andlow-aptitude (see Figure 6b) learners, the interac-tion between spatial aptitude and training condi-tion was not statistically significant, F(2,80) < 1.0.

For difficult courses, accuracy scores only re-vealed a main effect of spatial aptitude, F(l, 80) =5.44, p = .022. High-aptitude participants per-formed better (M = 59.2%) than the low-aptitudeparticipants (M = 50.7%). No other main effectsor interactions were significant for the difficultcourses.

Transfer

To estimate percentage transfer, we again usedthe formula of Katona (1940): percentagetransfer = 100 x (E, - C,)/(C, - C,). Scores andcomputed transfer values for the gate-aiming andaltitude-heading groups on easy courses are shownin Table 3. Separate transfer scores were calcu-lated for transfer Blocks 1 and 2. In addition,


(a) High Aptitude (b) Low Aptitude90

80

70

60I

1 50

40

30

Whole Alt-Hdg Gates

1 2 3Transfer Block

90

80

70

60

50

40

30

,-*

1 2 3Transfer Block

Figure 6. Accuracy scores of high- and low-aptitude participants on four transferblocks (adjusted by covariates). Alt-Hdg = altitude-heading.

transfer was calculated separately for speed andaccuracy data. As indicated in this table, transferwas higher for the gate-aiming group than for thealtitude-heading group.

Discussion

The results of this experiment confirm the hy-pothesis that concentrated practice on the gatestasks leads to better performance on the slalom

task compared with practice that focuses on alti-tude-heading tasks. Overall, the gate-aiming groupperformed more accurately on the slalom taskcompared with the altitude-heading group. Addi-tionally, the gate-aiming group scored as well asthe whole-task group on the slalom task. Spatialaptitude showed only a main effect on slalom taskperformance and did not interact with trainingcondition.

It is important to note the conditions under

Table 3Adjusted Criterion Task Scores and Estimates of Transfer on Each HalfofPosttest Block 1 for Speed and Accuracy Data on Easy Courses

Task

Block 1Gate AimingAltitude-Heading

Block 2Gate AimingAltitude-Heading

Whole-tasktraining

5.315.31

5.315.31

Transfer

Part taskSpeed

9.578.20

12.3911.17

Whole task

10.6910.69

11.6511.65

Transfer(%)

79.1853.72

111.6792.43

Accuracy

Block 1Gate Aiming 38.66 58.36 65.68 72.91Altitude-Heading 38.66 49.76 65.68 41.08

Block 2Gate Aiming 38.66 72.87 66.75 121.79Altitude-Heading 38.66 57.04 66.75 65.43


which we observed the benefits of the gate-aimingtraining. First, the benefits were primarily cen-tered on accuracy. Speed showed a similar patternbut did not reach statistical significance. Second,the benefits of gate aiming were not evident oninitial posttest scores. The whole-task group per-formed better than the part-task groups on the firsttransfer block. The benefits of gate aiming peakedon the second transfer block. This suggests thatgate-aiming training prepared learners to acquirethe criterion task more readily than did altitude-heading training. Third, the relative benefits ofgate-aiming training waned as the altitude-headinggroup continued to improve. This is to be expectedbecause transfer trials serve as additional practicetrials. Finally, the data in Figure 6 suggest thatgate-aiming training was particularly beneficial forlow-aptitude learners. They showed nearly thesame level of transfer performance as their high-aptitude counterparts, and low-aptitude learnersin the other two conditions showed lower perfor-mance than their high-aptitude counterparts. How-ever, because the interaction was not significant,we must conclude that the benefits of gate-aimingtraining did not differ for low- and high-aptitudelearners. This contradiction to our prediction willbe discussed in conjunction with Experiment 1findings below.

Perhaps the most impressive finding is the trans-fer effect. The gate-aiming condition resulted intransfer that exceeded 100% for both speed andaccuracy in the second transfer block. The altitude-heading condition produced moderate transfer forspeed but not accuracy. Even though direct com-parisons of transfer in the two experiments shouldbe made with caution, we observed that the trans-fer of the gate-aiming group in Experiment 2(79.1% and 111.6% for speed in transfer Blocks 1and 2) was greater than that of the part-task groupin Experiment 1 (57.9% for easy courses in Block1). Note that the transfer blocks in Experiment 1were 60 min in duration and those in Experiment 2were only 30 min in duration.

Summary and Conclusions

Even though there was an initial advantage forthe whole-task group in Experiment 2, it may beconcluded that removing the irrelevant componenttasks identified in Experiment 1 resulted in a moreefficient part-task training regime. Furthermore,

the initial advantage of whole-task training shownin transfer Block 1 (Experiment 2) was eliminatedrelatively quickly. Although statistical comparisonsshowed there was no difference between the whole-task group and the mean of the two part-taskgroups, gate-aiming learners showed higher meanscores than whole-task learners on speed andaccuracy. This is reflected in the differential trans-fer scores that exceed 100% in Experiment 2. InExperiment 1, part-task training scores approachedbut did not exceed, whole-task training scores.

The effectiveness of the gate-aiming regime inExperiment 2 testifies to the value of the backward-transfer procedure in identifying the componenttasks that are most critical to the whole task. Byshowing that only a portion of the tasks in theoriginal part-task regime produced positive back-ward transfer, we were able to eliminate irrelevanttasks from the part-task training regime. By focus-ing on the critical tasks, we created a more efficientpart-task training regime. We should also point outthat we did not select all the component tasks thatshowed positive backward transfer. The two com-ponent tasks comprising the Roll factor in Experi-ment 1 were not included in gate-aiming training.Also, the four spatial orientation tasks that showedpositive, albeit nonsignificant backward transfer,were included in the altitude-heading regime in-stead of the gate-aiming one. It is interesting tospeculate whether these tasks would further im-prove the efficiency of the gate-aiming condition,but we leave that for future investigation.

Wightman and Lintern (1985) observed thatpart-task training based on segmentation is moreeffective than part-task training based on fraction-ation. Because they were unable to ascertainwhether they were observing a general benefit ofsegmentation or a unique effect of backward chain-ing, it would be interesting to compare directly theeffectiveness of the gate-aiming tasks with that ofthe Roll factor tasks. Such a comparison would testthe question of whether segmentation (i.e., gate-aiming) is more effective than fractionation (i.e.,roll). However, the results may be biased in favorof the gate-aiming tasks because they show greaterbackward transfer than the roll tasks in Experi-ment 1. To test the question of fractionation versussegmentation, relevant tasks that are equivalent interms of backward transfer should be selected andthen compared directly.

We predicted that manipulating the efficiency of


the part-task training condition would have agreater impact on individuals with low- rather thanhigh-spatial aptitude. This prediction was based onthe findings of Wightman and Sistrunk (1987).They found that high-aptitude learners are af-fected less adversely by whole-task training thanare low-aptitude learners. Furthermore, this repre-sents a common finding in the ATI literature. Thatis, ATIs often take the form of large treatmenteffects for low-aptitude learners and small (ornonsignificant) treatment effects for high-aptitudelearners, primarily because of ceiling effects withinthe high-aptitude population (e.g., Shute, 1992,1995; Snow, 1994; Swanson, 1990; Tobias, 1994).

Data from the present experiments were par-tially consistent with these findings. Although wedid not find an interaction between training condi-tion and spatial-aptitude as predicted in Experi-ment 2, we did eliminate the Aptitude x Treat-ment interaction obtained in Experiment 1.Specifically, in Experiment 1, low-aptitude, but nothigh-aptitude participants showed a disadvantagefor part-task training. We concluded that low-aptitude participants were more sensitive to theinefficiencies of part-task training revealed in thebackward-transfer analysis. In Experiment 2, theATI was not statistically significant, and the gate-aiming group overall did not differ in performancecompared with the whole-task group. Thus, al-though low-aptitude participants did not showsignificantly greater benefits from gate-aiming train-ing compared with high-aptitude participants, theywere able to overcome the disadvantage shown inExperiment 1. In other words, these findings indi-cate that irrelevant part tasks in a part-task regimecan have a negative effect on low-aptitude learn-ers. Thus, if the reason for the introduction ofpart-task training in the first place was to helplow-aptitude trainees cope with the high demandsof the whole task, then the present outcomeindicates that unless these tasks provide a relevantcontext for the performance of the whole task, parttraining may worsen their situation. Such a findinghighlights the importance of identifying and remov-ing irrelevant part tasks from part-task training,and the important role that backward transfer canplay in creating a training environment in which alllearners may excel.

It should be noted that although the altitude-heading condition was worse than the gate-aimingcondition, participants in the former group did

acquire skills relevant to the criterion task. More-over, transfer data suggests that the altitude-heading condition produced as much or moretransfer (53.6% and 92.4%, for the first and secondtransfer blocks) as the part-task training regime inExperiment 1 (57.9%). This can be explained bypointing out that the spatial orientation tasksincluded in the altitude-heading condition showedpositive backward transfer to the criterion task inExperiment 1. In other words, the altitude-headingcondition included tasks that showed positive trans-fer as well as negative transfer. There may havebeen enough relevant tasks to produce as muchtransfer as in Experiment 1. An alternative expla-nation is that participants may acquire generalabilities about display and control dynamics when-ever they "fly" the simulator. These skills may notbe directly relevant to the criterion task, but theystill may provide the general skills that transfer tocriterion performance. However, these argumentsdo not detract from the conclusion that removingmany of the less relevant tasks improves efficiencyof part-task training.

One question that can be raised is whetherbackward transfer is the only way to identify thecritical component skills or if some general heuris-tic can be derived from the present data. Consider,for instance, spatial aptitudes. In both experi-ments, we found that gate-aiming skill was bestpredicted by spatial processing speed. In contrast,altitude-heading skill was predicted best by spatialinductive reasoning. Although these findings sug-gest that spatial processing speed may serve as apredictor of critical skills, it should be noted thatthe results may not generalize beyond the criteriontask of the current study. There are likely to beother flight tasks that are best predicted by spatialinductive reasoning or spatial working memory.

In conclusion, we suggest that the superiority ofwhole-task training in Experiment 1 may havebeen the result of including many noncriticalcomponent tasks in the part-task training regimeof Experiment 1. Experiment 2 supported thisconclusion, whereby concentrated practice on criti-cal skills resulted in performance that was equiva-lent to the whole-task group and produced transferthat was greater than 100%. Moreover, the gate-aiming condition effectively eliminated an ATIobtained in Experiment 1. This suggests that low-aptitude participants benefited more from therevised part-task training regime than did high-


aptitude participants. Finally, the present dataconfirm the utility of using the backward transfertechnique to identify critical components in apart-task training regime. We do not intend tosuggest that backward transfer alone is the onlyway to identify critical component skills. We arguethat backward transfer should be used in conjunc-tion with other techniques such as cognitive taskanalysis. Although the cognitive task analysis mayprovide theoretically driven predictions about thecritical tasks, the backward transfer can provideempirical evidence for supporting or refining pre-dictions. Moreover, we contend that any task couldbenefit from part-task training provided: (a) anappropriate task decomposition is developed andvalidated, (b) the ordering of critical skills for thepart-task regime is established based on findings inthe literature, (c) individual characteristics of learn-ers are taken into consideration in developing thetraining regime, and (d) there is sufficient andeffective reintegration of the part tasks into aunified whole task.

References

Ackerman, P. L. (1992). Predicting individual differ-ences in complex skill acquisition: Dynamics of abilitydeterminants. Journal of Applied Psychology, 77, 598-614.

Adams, J. A. (1960). Part trainers. In G. Finch (Ed.),Educational and training media: A symposium of theNational Research Council (pp. 129-149; PublicationNo. 789). Washington, DC: National Academy ofSciences.

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

Ash, D. W., & Holding, D. H. (1990). Backward versusforward chaining in the acquisition of a keyboard skill.Human Factors, 32, 139-146.

Bailey, J. S., Hughes, R. G., & Jones, W. E. (1980).Application of backward chaining to air-to-surface weap-ons delivery training (AFHRL-TR-79-63). WilliamsAFB, AZ: Operations Training Division, HumanResources Laboratory.

Briggs, G. E., & Naylor, J. C. (1962). The relativeefficiency of several training methods as a function oftransfer task complexity. Journal of Experimental Psy-chology, 64, 505-512.

Briggs, G. E., & Waters, L. K. (1958). Training andtransfer as a function of component interaction.Journal of Experimental Psychology, 56, 492-500.

Carlson, R. A., Khoo, H., & Elliot, R. G., Jr. (1990).Component practice and exposure to a problem-solving context. Human Factors, 32, 267-286.

Carlson, R. A., Sullivan, M. A., & Schneider, W. (1989).Component fluency in a problem-solving context.Human Factors, 31, 489-502.

Cronbach, L. J., & Snow, R. E. (1977). Aptitudes andinstructional methods. New York: Irvington.

Fabiani, M., Buckley, J., Gratton, G., Coles, M. G. H.,Donchin, E., & Logic, R. (1989). The training ofcomplex tasks. Acta Psychologica, 71, 259-299.

Frederiksen, J. R., & White, B. Y. (1989). An approachto training based upon principled task decomposition.Acta Psychologica, 71, 89-146.

Gagne, R. M., Forster, H., & Crowley, M. E. (1948). Themeasurement of transfer of training. PsychologicalBulletin, 45, 97-130.

Gopher, D., Weil, M., & Siegel, D. (1989). Practiceunder changing priorities: An approach to the train-ing of complex skills. Acta Psychologica, 71, 147-177.

Kanfer, R., & Ackerman, P. L. (1989). Motivation andcognitive abilities: An integrative/aptitude-treatmentinteraction approach to skill acquisition [Mono-graph]. Journal of Applied Psychology, 74, 657-690.

Kanfer, R., & Ackerman, P. L. (1996). A self-regulatoryskills perspective to reducing cognitive interference.In I. G. Sarason, B. R. Sarason, & G. R. Pierce (Eds.),Cognitive interference: Theories, methods, and findings(pp. 153-171). Hillsdale, NJ: Erlbaum.

Katona, G. (1940). Organizing and memorizing. NewYork: Columbia University Press.

Kyllonen, P. C, Woltz, D. J., Christal, R. E., Tirre,W. C, Shute, V. J., & Chaiken, S. (1990). CAM-4:Computerized battery of cognitive ability tests [Unpub-lished computer program]. Brooks Air Force Base,TX.

Linn, M. C., & Petersen, A. C. (1985). Emergence andcharacterization of sex-differences in spatial ability: Ameta-analysis. Chitd Development, 56, 1479-1498.

Lintern, G., & Gopher, D. (1980). Adaptive training ofperceptual-motor skills: Issues, results, and futuredirections. International Journal of Man-Machine stud-ies, 10, 521-551.

Lohman, D. F. (1979). Spatial ability: A review andreanalysis of the correlational literature (Tech. Rep. No.8). Stanford, CA: Aptitude Research Project, Schoolof Education, Stanford University.

Maccoby, E. E., & Jacklin, C. N. (1974). The psychologyof sex differences. Stanford, CA: Stanford UniversityPress.

Mane, A. M., Adams, J. A., & Donchin, E. (1989).


Adaptive and part-whole training in the acquisition ofa complex perceptual-motor skill. Acta Psychologies71, 179-196.

Mane, A. M., Coles, M. G. H., Wickens, C. D., &Donchin, E. (1983). The use of additive factorsmethodology in the analysis of a complex skill. Proceed-ings of the 27th Annual Meeting of the Human FactorsSociety (pp. 407-411). Santa Monica, CA: HumanFactors Society.

McGee, M. G. (1979). Human spatial abilities: Psycho-metric studies and environmental, generic, hormonal,and neurological influences. Psychological Bulletin, 86,889-918.

O'Brien, R. G., & Kaiser, M. K. (1985). MANOVAmethod for analyzing repeated measures designs: Anextensive primer. Psychological Bulletin, 97, 316-333.

Salthouse, T. A., & Prill, K. (1983). Analysis of aperceptual skill. Journal of Experimental Psychology:Human Perception and Performance, 9, 607-621.

Schneider, W., & Detweiler, M. (1988). The role ofpractice in dual-task performance: Toward workloadmodeling in a connectionist/control architecture. Hu-man Factors, 30, 539-566.

Shute, V. J. (1992). Aptitude-treatment interactions andcognitive skill diagnosis. In J. W. Regian & V. J. Shute(Eds.), Cognitive approaches to automated instruction(pp. 15-47). Hillsdale, NJ: Erlbaum.

Shute, V. J. (1993). A macroadaptive approach totutoring. Journal of Artificial Intelligence in Education,4(1), 61-93.

Shute, V. J. (in press). SMART: Student ModelingApproach for Responsive Tutoring. User Modelingand User-Adapted Interaction.

Singley, M. K., & Anderson, J. R. (1989). The transfer ofcognitive skill. Cambridge, MA: Harvard UniversityPress.

Snow, R. E. (1994). Abilities in academic tasks. In R. J.Sternberg & R. K. Wagner (Eds.), Mind in context:Interactional perspectives on human intelligence (pp.3-37). New York: Cambridge University Press.

Stammers, R. B. (1980). Part and whole practice for atracking task: Effects of task variables and amount ofpractice. Perceptual and Motor Skills, 50, 203-210.

Swanson, J. H. (1990, April). The effectiveness of tutorialstrategies: An experimental evaluation. Paper presentedthe American Educational Research Association, Bos-ton, MA.

Tobias, S. (1994). Interest, prior knowledge, and learn-ing. Review of Educational Research, 64(1), 37-54.

Voyer, D., Voyer, S. & Bryden, M. P. (1995). Magnitudeof sex differences in spatial abilities: A meta-analysisand consideration of critical variables. PsychologicalBulletin, 117, 250-270.

Wightman, D. C., & Lintern, G. (1985). Part-tasktraining for tracking and manual control. HumanFactors, 27, 267-283.

Wightman, D. C., & Sistrunk, F. (1987). Part-tasktraining strategies in simulated carrier landing final-approach training. Human Factors, 29, 245-254.

Received March 29,1995Revision received January 25,19%

Accepted February 13,1996

Mentoring Program Available for International Scholars

APA's Committee on International Relations in Psychology is encouraging publication of inter-national scholars' manuscripts in U.S. journals. To accomplish this initiative, the Committee islooking for authors whose native language is not English to work with U.S. mentors. U.S. men-tors will help authors bring manuscripts into conformity with English-language and U.S. publi-cation standards.

The Committee also continues to update its mentor list and is looking for U.S. mentors, espe-cially those with translating and APA journal experience. Interested individuals should contactMarian Wood in the APA International Affairs Office, 750 First Street, NE, Washington, DC20002-4242. Electronic mail may be sent via Internet to [email protected]; Telephone:(202) 336-6025; Fax: (202) 336-5919.

Analysis of Part-Task Training Using the Backward-Transfer ...myweb.fsu.edu/vshute/pdf/shute...

Documents

Transcript of Analysis of Part-Task Training Using the Backward-Transfer ...myweb.fsu.edu/vshute/pdf/shute...