Learning a partial-weight-bearing skill: effectiveness of two forms of feedback.

[Winstein CJ, Pohl PS, Cardinale C, et al. Learning a partial-weight-bearing skill: effectiveness of two forms of feedback. Phys Ther. 1996;76:985-993.]

Key Words: Feedback; Learning; Posture, tests and measurements; Psychomotor performance; Sensorimotor.

Partial weight bearing (PWB) is a skill that physical therapists often teach. Despite the frequency with which physical therapists teach PWB, little research has focused on what conditions of practice are the most effective for the learning of this sensorimotor skill, or even, in general, how well this skill can be learned. Attempts to train individuals in PWB have been relatively unsuccessful when using forms of concurrent feedback during practice.[1-3] For example, retraining of symmetrical weight bearing was ineffective in a group of individuals with hemiparesis using an augmented sensory feedback device that provides an auditory tone proportional to the magnitude of weight bearing during walking.[3] Despite these findings, concurrent feedback is still commonly used in clinical practice for the training of PWB. In contrast to results with concurrent feedback, a recent study[4] used a form of postresponse augmented feedback (de, summary knowledge of results [KR]) during practice and showed good learning of a PWB skill, with an accuracy at retention of within 7.2% of body weight.

Augmented feedback is one of the most powerful variables influencing motor learning.[5] Two types of augmented feedback are relevant to this research: KR and concurrent feedback. Knowledge of results has been defined as verbal (or verbalizable), postresponse feedback about the outcome of the movement related to the goal.[6] In contrast, concurrent feedback is a form of verbal (or verbalizable) augmented feedback about performance relative to the goal that is provided during the movement. Thus, the fundamental difference between these two forms of feedback is timing; KR is presented after the performance of the task, and concurrent feedback is presented during the performance of the task.

Across a large repertoire of motor skills, recent work has focused on the differential performance and learning effects of practice under conditions of several kinds of augmented feedback, including KR relative frequency (proportion of all practice trials for which KR is provided), summary KR (KR is given only after a set of trials; these trials are termed the "summary length"), KR delay (KR is given after a time delay), concurrent feedback, and presentation trials (movement is made to a physical block at the target).[7-11] The findings from these studies suggest that practice in conditions with less immediate, less frequent, and less guiding augmented feedback is more detrimental for immediate performance but more beneficial for the long-lasting learning of motor skills than practice in conditions where augmented feedback is provided with little delay, more often, or with physical guidance.[7-11]

Common to all these augmented feedback strategies that promote motor skill learning are opportunities for learners to engage in problem solving. This problem solving is not necessarily "cognitive" in nature but emerges as a function of the interaction between the motor task "problem" and the conditions of practice. Practice conditions that are less guiding (eg, have less frequent feedback) allow for the development of critical sensorimotor transformations, or the generation of solutions to motor problems, and are strongly associated with learning.[12-15]

The literature suggests that concurrent feedback is a strong performance variable but a weak learning variable. Few studies have shown the long-term benefits of practice with concurrent feedback, yet this type of feedback seems to be used often by physical therapists for the training of sensorimotor skills (eg, balance retraining). Thus, there appears to be a discrepancy between what the motor learning literature suggests about the optimal timing of feedback and what is currently used in practice. The purpose of this study was to compare the effects of practice with either postresponse or concurrent feedback on the learning of PWB. Based on theoretical predictions, our hypothesis was that practice under conditions of postresponse feedback would be more detrimental for immediate performance but more beneficial for longer-lasting learning than practice under a condition of concurrent feedback.

Method

Subjects

Sixty-one volunteers without known impairment of the neuromusculoskeletal system, between 20 and 40 years of age, were quasi-randomly assigned to one of three feedback practice groups, such that the male-female ratio was 1:2 for each group. Subjects were assigned sequentially to each of the three groups until a 1:2 male-female ratio was obtained; thereafter, group assignment was made until each group had at least 18 subjects. The mean age for all subjects was 26.2 years (SD = 3.8), the mean body weight was 66.2 kg (SD = 12.6), and the mean target weight (30% of body weight) was 19.9 kg (SD = 3.8). Exclusion criteria were (1) use of axillary crutches within 2 years before the study, (2) a physical or sensory impairment limiting the subject's ability to use crutches, or (3) body weight of less than 49.5 kg or greater than 592 kg. This weight restriction was due to limitations in the adjustablility of the scale used in the study. Most of the subjects were physical therapy students who had some knowledge of crutch use. Each subject read the institutionally approved guidelines regarding the rights of human subjects in medical experiments and signed an informed consent statement.

Task and Instrumentation

Subjects practiced supporting 30% of their body weight on their preferred limb on a floor scale with the goal of being accurate and consistent. The floor scale, with a readable accuracy of [+ or -]0.45 kg (the smallest unit of the scale dial to which the needle could be resolved), was modified to control the feedback that subjects could derive from the scale. The increments on the round scale dial were replicated on paper; however, the num bers indicating the pounds on the scale were changed, such that each 4.5-kg (10-lb) increment corresponded to 1 unit (Fig. 1A). The marking at 8 units (36 kg) was designated the PWB goal for all subjects, with the number "8" highlighted in green along with a thickened line at the corresponding hash mark on the paper dial. This paper dial was placed over the scale dial face but did not interfere with needle movement. For postresponse feedback conditions, a metal half cylinder with an opaque cover over the top end was secured upright to the scale such that it encircled half of the scale dial (Fig. 1B). This modification obstructed the view of the dial from the subject but not from the experimenter.

Wooden platforms were aligned on either side of the scale such that when subjects stepped onto the scale with the preferred limb, the contralateral limb and crutches were level with the top of the scale. The crutches were adjustable, tubular, aluminum axillary crutches with hand and axillary cushions. A portable tape player was used to provide prerecorded verbal cues, as appropriate, for each trial. All subjects wore rubber-soled shoes to provide a standard heel height and to prevent slippage on the scale.

Procedure

Subjects participated in a practice session, followed 2 days later by a retention test. Subjects were informed at the beginning of the study that a no-feedback retention test of 20 trials would follow 2 days later. During the practice session, subjects performed 80 trials (as determined from previous research4), with a 2-minute break after the first 40 trials. Prior to each day of data collection, the scale was calibrated with cast iron weights, which had been weighed against a known standard. The scale was read to the nearest 0.45 kg (1 lb).

Acquisition phase. At the start of the first session, subjects were asked to hop on one leg three times. The chosen (preferred) leg was designated the PWB limb. Each subject was weighed, and the subject's target weight was then calculated. Because 8 units was the PWB goal for all subjects, the starting point for the scale needle was adjusted for each subject (Fig. 1A). The starting point was calculated by subtracting 30% of the subject's body weight from the PWB goal of 36 kg (8 units). For example, a 90-kg subject would have an actual target weight of 27 kg (6 units). Thus, the starting position for this subject was set at 9 kg (2 units). When this subject applied 30% PWB, the scale dial would move from the starting position of 2 units to the target of 8 units.

The height of the crutches was adjusted to allow a two-finger-width space below the axilla as the subject stood erect. Each subject was instructed in a two-point step-to gait patternl6 and permitted to practice this gait pattern for a few steps. Thus, the subject simultaneously advanced both crutches and the PWB limb, and then advanced the contralateral limb to the line of the crutches and PWB limb. Subjects were allowed a few practice trials on the scale prior to data collection. An experimenter (the same person for all subjects) then demonstrated three trials of the task with the prerecorded commands and explained the procedures for feedback, depending on the group assignment. Subjects were told that the goal was to place a proportion of their body weight through their lower extremity to reach 8 units on the scale. The goal of 30% of body weight was not mentioned to control for individual differences in weight-bearing perception.[17]

There were three practice groups: concurrent feedback (CF), summarv-1-KR (KR-1), and summary-5-KR (KR-5). During practice, subjects in the KR-1 group received KR after every trial about the previous trial, those in the KR-5 group received KR after every fifth trial regarding the previous five trials, and those in the CF group received concurrent feedback during every trial.

Each trial was 7 seconds in duration and began with the subject standing in front of the scale at a starting position that was indicated by floor tape placed 15 cm from the scale. After the prerecorded "Go" signal, subjects stepped onto the scale using a two-point step-to gait pattern. One set of prerecorded verbal cues was used to instruct the subjects and to control the timing of the trial, the feedback, and the intertrial intervals across the three practice conditions (Fig. 2).

After stepping onto the scale, subjects in the CF group were told to view the scale dial and to adjust their weight so that the red scale needle was aligned with the "8" on the modified dial. Subjects in the two postresponse feedback groups (KR-1, KR-5) were told to look down at the opaque cover over the top end of the metal cylinder covering the scale dial during each trial. A "Hold" command was given 7 seconds after the "Go" command. During this 4-second interval, subjects were told to maintain the assumed weight bearing and the scale setting was recorded. Following this interval, a "Relax" command was given and feedback was provided (postresponse feedback groups only). The experimenter placed the replicated paper dial indicating the subjects' response on the top end of the metal cylinder. The subjects' performance (de, load) on the trial(s) was represented by a red line(s) referenced to the green PWB goal. This was designed to inform the subjects about the level of weight bearing they achieved. Thus, subjects in the KR-1 group saw one red line on the replicated scale dial after each trial. Subjects in the KR-5 group saw fiye red lines after every fifth trial, each labeled with a corresponding trial number (1-5), on the replicated scale dial. After the 5-second "Relax" interval, a "Return" command cued subjects to return to the starting position (Fig. 2).

The intertrial interval was 14 seconds, except after every fifth trial, when it was 19 seconds. This longer intertrial interval provided subjects in the KR-5 group with a 10-second KR interval to view feedback regarding their five preceding trials. The subjects in the KR-1 group were allowed to view the feedback from the preceding trial during these longer intertrial intervals, and subjects in the CF group remained idle. A trial was invalid if the subject misstepped by not stepping squarely on the scale or stepped off the scale before a reading could be made by the investigator. Invalid trials were repeated immediately. The duration of the acquisition phase was approximately 30 minutes per participant.

Retention phase. Subjects returned 2 days later for a 20-trial, no-feedback retention test. The procedure was reviewed to reorient the subjects to the task, and subjects were allowed to practice stepping up and down off the scale three times with the scale dial covered by the metal cylinder and opaque cover. Subjects were asked to accurately and consistently reproduce the PWB goal they had practiced 2 days earlier. All subjects were required to look down at the scale during each trial; however, the scale dial remained covered (Fig. 1B).

Data Analysis

Performances were grouped into blocks of 10 trials. Normalization of dependent measures was necessary for comparison between subjects because each subject had a different target (30% of body weight). Performance during the acquisition phase (blocks 1-8) and the retention phase (blocks 9 and 10) was analyzed separately, with normalized absolute error (NAE) and normalized variable error (NVE) as dependent measures. In addition, performance at the end of the acquisition phase (block 8) and performance during the retention phase (blocks 9 and 10) were included in a single analysis. Normalized absolute error, a measure of accuracy, is the absolute difference between the applied weight and the target weight, expressed as a percentage of the target weight. Thus, NAE represents the magni tude of error without regard to the direction of error (de, undershooting or overshooting the target). It was calculated using equation 1, where R is the scale reading, B is the initial scale setting (36 kg-30% of body weight), W is the weight applied, and T is 30% of body weight:

(1) R-B = W

|W-T| = AE

AE/T x 100 = NAE%

Normalized variable error (NVE), a measure of consistency, represents the within-subject variability about the subject's mean for a block of 10 trials. It was calculated using equation 2, where n is the number of trials in a block (10) and [bar]NAE is the normalized absolute error block mean score:

(2) [[SQRT NAE-[bar]NAE).sub.2]/n] x 100 = NVE%

Separate two-factor mixed-model analyses of variance, with group as the between-subject factor and trial block as the within-subject factor (repeated measures), were performed for NAE and NVE using BMDP statistical software.[18](*) A Greenhouse-Geisser adjusted F was used for the repeated-measures factor. Post hoc linear contrasts using a Bonferroni correction were performed to determine the locus of any significant effects. Significance was set at P<.05.

Results

Sixty-one volunteers participated in the acquisition phase. One subject, however, was unable to participate in the retention phase due to a lower-extremity injury that was incurred the day after the practice session. Thus, the results are reported on a total of 60 subjects (Table). Forty-two subjects used their right lower extremity and 18 subjects used their left lower extremity as the PWB limb. Only 11 invalid trials (less than 1%) were repeated out of the total of 6,000 trials recorded during the acquisition and retention phases.

[TABULAR DATA OMITTED]

Acquisition Phase

Accuracy (NAE). Figure 3 illustrates the mean NAE by block for each group. Across all groups, error scores decreased from 9.8% in block 1 to 4.4% in block 8. This change in performance resulted in a significant improvement across blocks (F = 31.92; df = 7,399; P<.0001).

Over blocks, there were differences in performance between groups (F = 69.25; df = 2,57; P<.0001). Subjects in the CF group had a mean NAE of 1.4%. Subjects in the postresponse feedback groups performed less accurately, with means of 6.4% and 8.2% for the KR-1 and KR-5 groups, respectively. In addition, the performance of the three groups was different across blocks for NAE (F = 9.68; df = 14,399; P<.0001). Post hoc analyses revealed that subjects in the CF and KR-1 groups showed no further improvement in performance accuracy after block 1. In contrast, subjects in the KR-5 group continued to show improvement after the first block (F = 3.99; df = 6,114; P<.006).

Consistency (NVE). With practice, subjects in all groups became more consistent in their performance. Across groups, the NVE decreased from 7.3% in block 1 to 3.2% in block 8 (Fig. 4). This improvement in consistency over practice was significant (F = 32.83; df = 7,399; P<.0001). In addition, there were differences in variability across practice blocks between groups. Subjects in the KR-1 and KR-5 groups had an average NVE of 5.0% and 5.3%, respectively. The average NVE for subjects in the CF group was 1.2%. This difference in variability between groups was significant (F = 83.18; df = 2,57; P<.0001).

The consistent performance of the subjects in the CF group over practice contrasted with the marked improvements in consistency in the postresponse feedback groups (F<4.61; df = 14,399; P<.001). Post hoc analyses within each group across blocks did not reveal a locus for this interaction.

Retention Phase

Overall, subjects performed less accurately over the two blocks of retention (Fig. 3). This decreased performance accuracy across blocks for NAE was significant (F = 5.85; df = 1,57; P<.02). Subjects in the CF group had the largest NAE, with a mean across the two blocks of 11.3%, as compared with means of 7.9% and 7.5% for subjects in the KR-1 and KR-5 groups, respectively. This difference between groups was significant (F = 3.42; df = 2,57; P<.04). Post hoc group mean contrasts for NAE, collapsed across the two retention blocks, between the CF group mean and the mean from the two KR groups, which were not different from each other,. revealed that the CF group performed with the largest error (F = 6.88; df = 1,58; P<.02).

For mean NAE, there were differences between groups across blocks (F = 3.74; df = 2,57; P<.03). Performance accuracy for subjects in the KR-1 group remained relatively stable over the two retention blocks, however, performance accuracy for subjects in the other groups declined slightly. Post hoc contrasts within each group across blocks did not reveal a locus for this interaction.

In the retention phase, all groups demonstrated a similar level of response consistency (NVE), with an average of 4.6% (right side, Fig. 4). Analysis revealed no differences.

Last Acquisition-Retention

A comparison across groups from the end of the acquisition phase, where augmented feedback was available (block 8), through the retention phase, where augmented feedback was not available, showed an overall deterioration in performance. This deterioration across blocks was statistically significant (F=36.40; df=2,114; p<.0001).

Although all groups performed less accurately in the retention phase compared with the acquisition phase, the magnitude of deterioration depended on the practice condition (Fig. 3). The accuracy of performance of subjects in the postresponse feedback groups declined approximately 1.8% from the last block of acquisition to retention. In contrast, the accuracy of performance of subjects in the CF group declined nearly 11.3%. This difference between groups across blocks was statistically significant for NAE (F=12.84; df=4,114; P<.0001). Indeed, within-group post hoc contrasts revealed a difference across blocks for only the CF group (F=50.13; df=1,18; P<.0001).

Overall, across blocks from the end of acquisition through retention, the subjects' performance became more variable as evidenced by NVE (F=17.89; df=2,114; P<.0001). More importantly, there was a difference between groups across blocks (F=10.47; df=4,114; P<.0001). Similar to the results for accuracy, there was a marked decrease in response consistency by subjects in the CF group (F=124.42; df=1,18; P<.0001). In contrast, subjects in the two postresponse KR groups performed with a similar level of response consistency at the end of acquisition and retention (Fig. 4).

Discussion

This study provided a direct comparison between the effectiveness of concurrent feedback and postresponse feedback on the learning of PWB in young adults without known neuromusculoskeletal impairment. The results demonstrated that for learning this skill, postresponse feedback was more effective than concurrent feedback. Although the subjects in the CF group performed more accurately and consistently during practice compared with the subjects in the postresponse feedback groups, subjects in the CF group performed less accurately than the subjects in the other two groups when they had to reproduce the PWB task 2 days later in the no-feedback retention test. There were no group differences, however, in performance consistency in retention. Thus, the results of this study demonstrate that concurrent feedback is a powerful performance variable in that it promotes both accuracy and consistency of performance when it is available but that this type of feedback is a poor learning variable in that performance accuracy does not persist when it is not available. In contrast, the results suggest that postresponse feedback is a more effective learning variable for the training of this PWB skill. The acquisition-retention design commonly used in motor learning research provides an inherent test of reliability. In addition, repeated measures over the acquisition phase provide a further test of reliability. Because performance measures plateaued for all groups over practice, we believe that test-retest reliability can be assumed.

The detrimental learning effects of practice with concurrent feedback are consistent with the findings of previous studies of PWB tasks and other motor skills.[1,2,9,19] Warren and Lehmann[2] used concurrent feedback training methods over 100 practice trials of a PWB task and reported errors with magnitudes greater than 50% of the target loads during retention tests. These findings and those of Bohannon and colleagues[17] suggest that PWB skills cannot be learned with any degree of accuracy. Subjects in the postresponse feedback groups in our study, however, demonstrated a minimum error in the retention test of 17% of the 30% of body weight goal, or about 5% of body weight. Thus, PWB can be learned with a reasonable degree of accuracy when practice conditions that promote learning are used.

Conditions of practice promote certain information-processing operations, including storage, retrieval, and recall. For motor learning, these operations may not be conscious or declarative in nature, but instead are implicit and procedural in nature.[20] Recent theories in motor learning and memory in general suggest that the "...act of retrieval is itself a potent learning event."[12(p37)] Thus, conditions of practice that promote the active retrieval of an action plan should be more beneficial for learning than conditions of practice where the prescription for the action is provided. For example, in conditions where feedback is immediate (eg, concurrent), the feedback can be used as a direct guide to the solution to the motor problem. In contrast, in conditions where feedback is not immediate (eg, postresponse), the feedback does not provide a direct solution to the motor problem and thus the learner must engage in active retrieval to solve the problem.

One subject in the CF group was able to overcome the detrimental effects on learning of concurrent feedback. During practice, the accuracy scores of this individual were similar to the CF group mean. In the retention phase, however, there was little change in the accuracy of this subject's response. Postexperiment debriefing revealed that this participant utilized a unique strategy during the last 40 trials of practice. During the "Relax" interval, this individual concentrated on the feeling that was associated with the target weight. Thus, self-directed strategies could be used to facilitate motor learning ever when concurrent feedback is used during practice.

Other research[3,21] suggests that practice with concurrent feedback juxtaposed with a period of no feedback may be effective for motor learning where a simple recalibration is needed. Students in a physical therapy program practiced a vertebral joint mobilization skill without feedback for three 30-second bouts. After the first practice session, concurrent feedback was provided, using an oscilloscope that displayed force production relative to a target force for one 45-second bout. Subjects showed improved accuracy in producing the target force on a 1-week retention test.[21] Similarly, Wannstedt and Herman[3] observed symmetrical weight bearing 1 month after a short training period in a group of patients with hemiparesis using the augmented sensory feedback device described earlier. Interestingly, like the students in the joint mobilization study,[21] all successful subjects in Wannstedt and Herman's study[3] corrected limb loading almost immediately in the first session using the augmented sensory feedback device. Thus, practice conditions that provide some guidance (eg, concurrent feedback), juxtaposed with opportunities for active problem solving (eg, no-feedback bouts), can be beneficial for motor learning when a simple sensorimotor calibration is required.[10]

Caution must be taken in generalizing our findings. The motor task was essentially static and limited to one target goal. Extending these findings to dynamic PWB tasks or to target goals other than 30% of body weight must be made with certain considerations. We would expect the accuracy to be lower in dynamic PWB tasks than in our static PWB task. Indeed, slightly lower accuracy scores than those reported here were found in a similar study that examined the learning of PWB during gait.4 Further, for rapid force production tasks, variability increases linearly as the target force increases up to approximately 65% of maximum.[22,23] Thus, we might predict that accuracy and consistency of PWB under conditions with target goals greater than 30% of body weight would follow a similar rule.

The participants in this study were all young adults without known neuromusculoskeletal impairment. Although there is no evidence to suggest that the processes underlying the learning of these types of motor skills are different in older individuals or in individuals with disability, tests of such claims are needed. Previous work examining the effects of augmented feedback for motor learning in older adults[24,25] and individuals with stroke[26] has revealed a similar pattern of improvement with practice when compared

with younger adults and age-matched control subjects, respectively. Differences were revealed in the absolute magnitude of best performance but not in the rate of improvement. Thus, we would predict that if this study were conducted with older individuals or individuals with impairment, the same pattern of results would emerge but the magnitude of error would be greater.

Conclusion

The results of this study showed that practice conditions with postresponse feedback were more effective for learning a PWB skill, as measured by a 2-day retention test, than was a practice condition with concurrent feedback. Further, provided with conditions of practice that facilitated active problem solving, young adults without known neuromusculoskeletal impairment were able to learn this static PWB skill quite well (de, accuracy to within 5% of body weight). These results are consistent with a more general hypothesis that conditions of practice (eg, concurrent feedback) that provide a direct prescription for action may be beneficial for immediate performance but detrimental for long-term learning. In contrast, conditions of practice (eg, postresponse feedback) that promote active problem solving such as the development of specific sensorimotor transformations may be relatively detrimental for immediate performance but beneficial for longer-lasting motor learning.

[Figures 1 to 4 ILLUSTRATION OMITTED]

(*) BMDP Statistical Software Inc, 1440 Sepulveda Bivd, Suite 316, Los Angeles, CA 90025.

References

[1] Baxter ML, Allington RO, Koepke GH. Weight-distribution variable in the use of crutches and canes. Phys Ther.: 1969;49:360-365.

[2] Warren CG, Lehmann JF. Training procedures and biofeedback methods to achieve controlled partial weight hearing: an assessment. Arch Phys Med Rehabil. 1975;56:449-455.

[3] Wamlstedt GT, Herman RM. Use of augmented sensory feedback to achieve symmetrical standing. Phys Ther. 1978;58:553-559.

[4] Winstein CJ, Christensen S, Fitch N. Effects of summary knowledge of results on the acquisition and retention of partial weight bearing during gait. Physical Therapy Practice. 1993;2:40-51.

[5] Salmoni AW, Schmidt RA, Walter CB. Knowledge of results and motor learning: a review and critical reappraisal. Psychol Bull. 1984;95: 355-386.

[6] Winstein CJ. Knowledge of results and motor learning: implications for physical therapy. Phys Ther. 1991;71:140-148.

[7] Schmidt RA, Young DE, Swinnen S, Shapiro DC. Summary knowledge of results for skill acquisition: support for the guidance hypothesis. J Exp Psychol Learn Mem Cogn. 1989;15:352-359.

[8] Swinnen SP, Schmidt RA, Nicholson DE, Shapiro DC. Information feedback for skill acquisition: instantaneous knowledge of results degrades learning. J Exp Psychol Learn Mem Cogn. 1990;16:706-716.

[9] Vander Linden DW, Cauraugh JH, Greene TA. The effect of frequency of kinetic feedback on learning an isometric force production task in nondisabled subjects. Phys Ther. 1993;73:70-87.

[10] Winstein CJ, Pohl PS, Lewthwaite R. Effects of physical guidance and knowledge of results on motor learning: support for the guidance hypothesis. Res Q Exerc Sport. 1994;65:316-323.

[11] Winstein CJ, Schmidt RA. Reduced frequency of knowledge of results enhances motor skill learning. J Exp Psychol Learn Mem Cogn. 1990;16:677-691.

[12] Bjork RA, Bjork EL. A new theory of disuse and an old theory of stimulus fluctuation. In: Healy AF, Kosslyn SM, Shiffrin RM, eds. From Learning Processes to Cognitive Processes: Essays in Honor of William K Estes, Volume 2. Hillsdale, NJ: Lawrence Erlbaum Associates Inc; 1992:35-67.

[13] Schmidt RA. Frequent augmented feedback can degrade learning: evidence and interpretations. In: Stelmach CE, Requin J, eds. Tutorials in Motor Neuroscience. Dordrecht, the Netherlands: Kluwer Group; 1991 :59-75.

[14] Schmidt RA, Bjork RA. New conceptualizations of practice: common principles in three paradigms suggest new concepts of training. Psychological science. 1992;3:207-217.

[15] Singer RN, Pease D. Effect of guided versus discovery learning strategies on learning, retention, and transfer of a serial motor task. Res Q. 1976;47:788-796. [16] Minor MA, Minor SD. Patient Care Skills. Reston, Va: Reston Publishing; 1984.

[17] Bohannon RW, Waters C, Cooper J. Perception of unilateral lower extremity weightbearing during bilateral upright stance. Percept Mot Skills. 1989;69:875-880. [18] BMDP Statistical Software Manual, Release 7. Los Angeles, Calif: University of California Press; 1992.

[19] Annett J. Learning a pressure under conditions of immediate and delayed knowledge of results. Q J Exp Psychol. 1959;11:3-15.

[20] Squire LR. Memory and Brain. New York, NY: Oxford University Press Inc; 1987: chap 11.

[21] Lee M, Moseley A, Refshauge K Effect of feedback on learning a vertebral joint mobilization skill. Phys Ther. 1990;70:97-104.

[22] Schmidt RA, Sherwood DE. An inverted-U relation between spatial error and force requirements in rapid limb movements: further evidence for the impulse-variability model. J Exp Psychol Hum Percept Perform. 1982;8:158-170.

[23] Schmidt RA, Zelaznik HN, Hawkins B, et al. Motor-output variability: a theory for the accuracy of rapid motor acts. Psychol Rev. 1979;86: 415-451.

[24] Swanson LR, Lee TD. Effects of aging and schedules of knowledge of results on motor learning. J Gerontol. 1992;47:P406-P411.

[25] Behrman AL, Vander Linden DW, Cauraugh JH. Relative frequency knowledge of results: older adults learning a force-time modulation task. Journal of Human Movement Studies. 1992;23:233-250.

[26] Merians A, Winstein CJ, Sullivan K, Pohl PS. Effects of feedback for motor skill learning in older healthy subjects and individuals post stroke. Neurology Report. 1995;19:23-25.

CJ Winstein, PhD, PT, is Assistant Professor, Department of Biokinesiology and Physical Therapy, University of Southern California, 1540 E Alcazar St, CHP 155, Los Angeles, CA 90033 (USA) (winsteinC!hsc.usc.edu). Address all correspondence to Dr Winstein.

PS Pohl, PhD, PT, is Assistant Scientist, Center on Aging, and Assistant Professor, Department of Physical Therapy Education, The University of Kansas Medical Center, Kansas City, KS 66160. He was a doctoral candidate and Research Assistant, Department of Biokinesiology and Physical Therapy, University of Southern California, at the time the study was conducted.

C Cardinale, PT, is Physical Therapist 11, Seton Medical Center, Dayley City, CA 94015.

A Green, PT, is Lead Physical Therapist, Oakridge Care Center, Oakland, CA 94602.

L Scholtz, PT, is Staff Physical Therapist, Fortanasce and Associates, Arcadia, CA 91007.

CS Waters, PT, is Senior Physical Therapist, Imperial Valley Therapy Centers, El Centro, CA 92243.

Ms Cardinale, Ms Creen, Ms Scholtz, and Ms Waters were students in the entry-level Master of Physical Therapy degree program, Department of Biokinesiology and Physical Therapy, University of Southern California, at the time the study was conducted.

This study was approved by the Research Review Committee for Professional Staff Association, University of Southern California School of Medicine.

Preliminary results of this study were presented at the Combined Sections Meeting of the American Physical Therapy Association; February 8-12, 1995; Reno, Nev.

This article was submitted July 18, 1995, and was accepted April 15, 1996.