Models of Performance in Learning Multisegment Movement Tasks:Consequences for Acquisition, Retention, and Judgments of LearningDominic A. SimonMcMaster UniversityRobert A. BjorkUniversity of California, Los AngelesParticipants learned different keystroke patterns, each requiring that a key sequence be struck in aprescribed time. Trials of a given pattern were either blocked or interleaved randomly with trials on theother patterns and before each trial modeled timing information was presented that either matched ormismatched the movement to be executed next. In acquisition, blocked practice and matching modelssupported better performance than did random practice and mismatching models. In retention, however,random practice and mismatching models were associated with superior learning. Judgments of learningmade during practice were more in line with acquisition than with retention performance, providingfurther evidence that a learner’s current ease of access to a motor skill is a poor indicator of learningbenefit.In the last several decades, a number of manipulations of prac-tice and study conditions has been shown to have effects onacquisition performance that differ from the effects of those ma-nipulations on learning—as assessed by delayed tests of retentionand transfer (see reviews by Chamberlin & Lee, 1993; Christina &Bjork, 1991; Ghodsian, Bjork, & Benjamin, 1997; Schmidt &Bjork, 1992; Schmidt & Lee, 1999). Practice conditions that sup-port higher levels of immediate performance do not always fostersuperior learning, and conditions that pose difficulties for thelearner can enhance long-term retention and transfer (e.g.,Schmidt, Young, Swinnen, & Shapiro, 1989; Shea & Morgan,1979). Schmidt and Bjork (1992) reviewed a variety of evidenceindicating that such phenomena are to be found in both the verbaland motor learning domains.In the present study, we introduced a novel-practice manipula-tion: providing a matching or mismatching model of perfect per-formance before each trial on a given to-be-learned responsesequence. In the matching case, the model illustrated perfectperformances on the about-to-be-practiced response sequence,whereas in the mismatching case the model illustrated perfectperformance on one of the other to-be-learned response sequences.For reasons we outline below, we anticipated that themismatching-model condition would disrupt immediate perfor-mance but promote superior levels of learning, whereas the con-verse would be the case for the matching condition.A central goal of the present research, however, beyond simplyassessing the short-term and long-term consequences of providingsuch matching and mismatching models, was to investigate theattendant impressions that the participants developed about theirown learning. Would learners’ assessments of their learning bedriven by their immediate performance or would they have insightinto the benefits of the differing practice conditions as indicated bya delayed-performance test? Would the matching and mismatchingmodels be seen as artificially enhancing or degrading performance,respectively, and thus be taken into account when predictingperformance on a delayed test?Models of Performance as an Aid andPossible Hindrance to LearningProviding a human model for learners is a common way toenhance their understanding and appreciation of the task to belearned (see McCullagh, 1993, for a review). Generally such amodel, whether live or shown on a video or computer display,either demonstrates proper performance of the task or is a learnerhimself or herself, that is, someone who is also attempting to learnthe desired movement pattern or skill in question. Modeled infor-mation can also be provided in various nonhuman forms. A simpleexample is the metronome used by musicians to provide them withthe appropriate meter for a piece of music. Despite the establishedbenefits of using them for learning, research by Lee, Wishart,Cunningham, and Carnahan (1997) suggests that the effects ofmodels are not always positive. Indeed, they can act to nullify theeffectiveness of other manipulations known to enhance long-termretention and transfer, more specifically, the benefits of random–interleaved conditions of practice, as in the so-called contextual-interference effect first suggested by Battig (1979, p. 24).Random Versus Blocked Practice: Findings and TheoriesBroadly, the contextual interference effect, based on ideas fromverbal learning (Battig, 1972, 1979), refers to the fact that arrang-ing to-be-learned material in ways that create interference duringlearning can sometimes—unintuitively—enhance posttraining re-Dominic A. Simon, Department of Kinesiology, McMaster University,Hamilton, Ontario, Canada; Robert A. Bjork, Department of Psychology,University of California, Los Angeles.We thank Derek McAuley for helping to gather the data and AaronBenjamin for comments on an earlier version of this article.Correspondence concerning this article should be addressed to DominicA. Simon, who is now at the Department of Psychology, Mail Stop Code3452, New Mexico State University, P.O. Box 30001, Las Cruces, NewMexico 88003-8001. E-mail: [email protected] of Experimental Psychology: Applied Copyright 2002 by the American Psychological Association, Inc.2002, Vol. 8, No. 4, 222–232 1076-898X/02/$5.00 DOI: 10.1037/1076-898X.8.4.222222tention and transfer. By presenting a to-be-studied item in a con-text consisting of other to-be-learned materials, more processinginterference occurs than if the item is presented surrounded, forexample, by other presentations of that particular item. Followingon the first demonstration of contextual interference in motor skillsby Shea and Morgan (1979), studies of blocked versus randompractice have typically involved a comparison of two trainingscenarios, each of which requires that multiple responses are to belearned. In blocked practice, all trials of a particular to-be-learnedresponse are completed before moving on to begin practice onanother response. In random practice, the to-be-learned responsesare interleaved in a semirandom fashion. That is, the ordering oftrials is not completely random, but each of the to-be-learned itemsappears a set number of times in random order within a subset ofso many trials. Usually, there is also a prohibition against havingmore than two consecutive trials of the same response.The standard finding (see, Chamberlin & Lee, 1993; Magill &Hall, 1990; Schmidt & Lee,