Acta Psychologica 111 (2002) 83100 www.elsevier.

com/locate/actpsy

Learning and performance eects of accurate and erroneous knowledge of results on time perception
Lawrence J. Ryan *, Troy B. Robey
1
Department of Psychology, 204 Moreland Hall, Oregon State University, Corvallis, OR 97330-5303, USA Received 28 May 2001; received in revised form 4 December 2001; accepted 2 January 2002

Abstract Performance feedback (also known as knowledge of results or KR) has both performance and learning eects on many tasks. Earlier studies have demonstrated performance but not learning eects on time perception tasks. In this experiment, we dissociate and identify these two phenomena on two dierent time perception tasks. Participants were presented with either accurate (100%) or erroneous (80% or 120% of actual performance) KR on either a reproduction or a numerical estimation time perception task. Accurate (100%) KR reduced the group variability and increased the accuracy of response magnitude but left individual variability unchanged. Erroneous (80% or 120%) KR also reduced the group variability while leaving individual variation unchanged, but decreased the true accuracy of the response, with response magnitudes increasing for the 80% KR group and decreasing for the 120% group. Thus external KR that is in conict with internal time cues overrides these internal cues and dictates response magnitude on these two tasks. Thus KR provides guidance for these behaviors. KR did not reduce the variability (dispersion) of participants responses, but centered each participants responses closer to the targeted performance. This decreased group response variability reected a performance enhancing eect of KR because group response variability increased after KR was withdrawn. In contrast, response magnitudes remained changed for the duration of the post-KR period, indicating that KR also induced a learned response. Thus individual response variability, group response variability and response magnitude represent dissociable features of performance on these time perception tasks. 2002 Elsevier Science B.V. All rights reserved.

Corresponding author. Tel.: +1-541-737-1371; fax: +1-541-737-3547. E-mail address: lryan@orst.edu (L.J. Ryan). 1 Present Address: Department of Educational Leadership and Counseling Psychology, Washington State University, Pullman, WA 99163, USA. 0001-6918/02/$ - see front matter 2002 Elsevier Science B.V. All rights reserved. PII: S 0 0 0 1 - 6 9 1 8 ( 0 2 ) 0 0 0 4 4 - 6

84

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

PsycINFO classication: 2343; 2340 Keywords: Learning; Knowledge of results; Feedback; Time perception

1. Introduction External performance feedback (also known as knowledge of results or KR) has been shown to improve performance on skilled motor tasks (Salmoni, Schmidt, & Walter, 1984). KR promotes better performance both while it is present and after it is withdrawn. While KR is present, it provides motivation and guidance and increases a participants interest and persistence. These performance enhancing eects dissipate rapidly once KR is withdrawn. However, KR also promotes learning and improves performance even after KR is withdrawn. Indeed, learning has been dened and dierentiated from direct performance enhancing eects by the transfer of improvement from the KR to the post-KR condition (Salmoni et al., 1984). When learning motor tasks, information as to performance outcome is valuable because it provides a means by which the nervous system can calibrate and modify the strength and timing of muscle responses in order to achieve the desired goal. It provides information that may be otherwise unavailable to the motor system. On many motor tasks, information as to performance outcome is provided by the exteroceptors, that is, participants can see or hear or touch the outcome to determine how well they just performed. Performance feedback, whether originating from the exteroceptors or as KR from an external source, provides task-specic meaning to internal physiological feedback that is itself task-independent (such as from muscle spindles or Golgi tendon organs) and that may be used for regulating vastly dierent motor behaviors (a leg muscle and its associated sensory receptors, for instance, may be used for running, kicking, stiening the body while throwing, swimming, riding a horse, etc.). The performance feedback may also directly modify motor programming, especially for ballistic movements that are preprogrammed to operate without correction by internal or exteroceptive feedback. In many cases KR provided by a coach or computer or other outside source may be redundant with the exteroceptor feedback, but, in other cases, it may provide additional information, clearer information or corroborative information to that provided by the exteroceptors. Indeed, KR is often used in preference to other sources of feedback, as is clearly shown by misguided performance following erroneous KR (Buekers, Magill, & Hall, 1992, 1994; McNevin, Magill, & Buekers, 1994). Thus, on many motor tasks, KR complements information provided by the exteroceptors. On some cognitive tasks, such as judgment of line lengths, KR also likely complements the information provided by exteroceptors. However, on other cognitive tasks the sources of either internal, task-independent feedback analogous to proprioceptive feedback, or of external, task-dependent feedback analogous to that provided by exteroceptors on a motor task, are unknown. It is unclear on these cognitive tasks what feedback is normally available to provide information for accu-

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

85

rate task performance and, therefore, how this feedback compares to KR. On a time estimation task, for instance, in which the duration of an unlled interval is observed and then reported with a numerical estimate, the source of internal feedback that would be analogous to the proprioceptive feedback is unknown. The source of outcome feedback that would be analogous to exteroceptor feedback and that would provide information for the measurement and correction of the accuracy of the time judgment is also unknown. Some models of time perception propose an internal clock whose ticks are accumulated and counted (Gibbon & Church, 1990), other models propose that mental events and changes are counted (Block & Zakay, 1997) and still others propose that the decay of memory traces of recent events is measured (Staddon & Higa, 1999). These three type of models imply very dierent sources and representations of internal time information (Grondin, 2001). Furthermore, estimation tasks, reproduction tasks and production tasks involve quite different cognitive processes to report the outcome of the time perception process (Fraisse, 1984; Montare, 1985, 1988; Zakay, 1990). Without knowing what information sources are normally used to guide timing performance it is dicult to predict how KR might act to alter and improve performance on time perception tasks. It could aect basic timing processes, calibration or interpretation of these processes, memory that either holds the current duration perception or that holds recalled intervals for comparison, or it could aect task specic factors that dier among the estimation, reproduction and production tasks. In this paper we begin an analysis of how KR aects time perception processes. KR has been shown to increase the accuracy and to decrease the variability of responses on time perception tasks using short (seconds) and ultrashort (less than 1 s) intervals (Aiken, 1965; Buckolz & Guay, 1975; Crowder & Hohle, 1970; Hicks & Miller, 1976; Hoyer, Jones, White, & Maconachy, 1970; Kladopoulos, Brown, Hemmes, & Cabeza de Vaca, 1998; Montare, 1985, 1988). KR also improves temporal discriminability (Allen & Clark, 1979). Unlike KR on motor learning tasks, these eects on time perception tasks have typically been shown to decay rapidly after KR is withdrawn (Aiken, 1965; Buckolz & Guay, 1975; Hicks & Miller, 1976; Kladopoulos et al., 1998). This is surprising as it may indicate that KR does not promote learning on time perception tasks, contrary to some claims (e.g., Montare, 1985, 1988). Of course, we may not learn about time perception because this is already an acquired and overlearned skill. Most participants have had years of experience judging short intervals of known length (TV commercials, microwave settings, etc.). Rather than inducing the learning which recalibrates our sense of time, the learning that may result from KR could be learning about the specic requirements of the task and may therefore not be obvious because most tasks are simple and easily acquired. We have three goals for this study. First, we test the hypothesis that KR does lead to learning on time perception tasks. Unlike earlier experiments, we attempt to distinguish between performance and learning enhancing eects of KR on time perception tasks by using a transfer design that allows us to examine performance during and following the presentation of KR (Salmoni et al., 1984). We also use a paradigm derived from the motor learning literature (Buekers et al., 1992, 1994; McNevin et al.,

86

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

1994) that allows us to dissociate the eects of KR on the magnitude and the variability of performance. We provide some participants with erroneous KR that either under-represents or over-represents their actual performance. Based on these motor learning studies, we predict that KR that under-represents performance will lead to an increase in response magnitude, whereas KR that over-represents performance will lead to a decrease in response magnitude. We further predict that this eect is learned and will persist following the removal of the KR. The second goal is to test the hypothesis that KR provides guidance for performance on our tasks as it is known for motor learning tasks. On such tasks, KR is believed to have both motivational eects, such as increased attention, reinforcement, persistence, interest, etc., and guidance eects, such as highlighting the correct goal, which improves performance when KR is present (Salmoni et al., 1984). These known motivational and guidance actions may well account for the observed changes in performance on time perception tasks when KR is provided. The use of erroneous KR provides a direct test of the guidance eect of KR on these participants by specifying dierent performance targets and thus dierent degrees and directions of performance errors relative to initial performance. If the guidance hypothesis is correct, the magnitude of response change should reect both initial performance and the target specied by the erroneous KR whereas other performance enhancing eects, such as increasing attention or interest, should act independently of initial performance. The third goal is to provide evidence as to whether the KR-induced learning is task specic or whether it aects more fundamental time perception processes. It has been proposed that dierent methods of measuring time perception may tap dierent cognitive processes and may therefore respond dierently to KR (Fraisse, 1984, Montare, 1985, 1988; reviewed by Zakay, 1990). Thus, we conduct two parallel experiments that use the same temporal source stimulus but use two dierent methods of measuring perceptual accuracy: a reproduction task and a numerical estimation task.

2. Method 2.1. Participants Sixty-one male and female under-graduate students enrolled in a psychology course volunteered to participate in these experiments and received extra class credit for their participation. Conditions of participation were approved by the OSU Human Participants Committee and met all current ethical guidelines for the use of human participants. 2.2. Materials and procedure The experiments were conducted using 2 DOS-based 486 computers. Each participant completed a single session comprising three phases with the entire session typically lasting 25 min and always lasting less than 35 min. Participants received

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

87

instructions and an opportunity to practice during the introductory screens of the computer program. They then initiated a series of three self-paced experimental phases. In the reproduction task (n 31), participants initiated a computer display that consisted of a red 33 37 mm rectangle centered on the screen against a black background. A white command line occupied the top 10 mm of the screen. This RED display remained on the screen for one of three xed durations: 3.7, 5.6 or 8.5 s. Participants were asked, both at the initial verbal brieng and again on the written computerized instructions, to judge the duration that this sample display remained on the screen without mentally counting. No other attempt was made to prevent counting but participants were asked at debrieng about their performance of the task. No data were excluded because of self-reports of counting. Furthermore, no participants in the erroneous KR groups reported being aware of the systematic error in the KR, which should have been obvious to them if participants had been counting. After judging the duration of the sample interval, they were to reproduce this interval by initiating a new display, identical to the sample interval display except that it was brown, and to terminate the display with a second keystroke when they estimated the BROWN display had been on the screen for as long as the previous RED sample display. Performance data were not recorded for the rst three trials (one of each duration). Eight randomly presented samples of each duration, 24 trials in total, were then presented. During phase 2, participants initiated the same RED samples and BROWN reproduction displays but after each trial received graphical feedback (KR) about the accuracy of their performance. The KR consisted of a red (84 5 mm) rectangle whose length represented the duration of the RED sample interval. Below it was a brown 5 mm in height rectangle whose length represented the length of the reproduced interval relative to the sample interval. The three groups of participants diered in the accuracy of this KR display. For one group (100% KR, n 10) the brown rectangle accurately represented their performance. For another group (80% KR, n 10), the brown rectangle represented only 80% of their reproduced interval and for the third group (120% KR, n 11) the brown rectangle represented 120% of the duration of their reproduced interval. Phase 3 resembled phase 1 in that participants completed the same task but did not receive KR. The numerical estimation task (n 30, 10 participants per KR group) was identical to that in the reproduction task with two exceptions. First, the participants were asked to make a numerical estimation of the duration of the RED sample interval rather than reproduce the interval. Participants were asked to make estimates to the nearest whole second in order to control for the strong whole number response bias that has been identied in verbal estimation paradigms (Zakay, 1990). Second, the sample durations were slightly changed from the reproduction task, being centered between whole numbers: 3.5, 5.5 and 8.5 s. 2.3. Statistical analysis Both experiments used a 3 3 3 design with two measures, sample duration (three levels) and experimental phase (three levels: pre-, during and after KR) being repeated. The third variable represented independent groups (three KR groups: 80%,

88

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

100% or 120% feedback). In both experiments, the dependent variable was the mean of the ratio of the performance (either reproduced duration or the numerical estimate) to the sample duration on each trial, eight trials per cell. Post-hoc analysis of dierences between cells were conducted using the NewmanKeuls test with difference criteria specied at a 0:05. Two measures of response variability were measured. Group variability was measured by calculating the standard error of the mean for each cell (27) in the 3 3 3 design. Statistical signicance of the change in group variability across adjacent phases (2 vs 1 and 3 vs 2) was calculated using the Wilcoxin signed ranks test. Individual variability was calculated as a coecient of variation dened as the ratio of standard deviation of each participants performance ratio (the dependent variable for response magnitude) on the 8 trials comprising each cell of the design divided by the participants mean performance ratio for that cell. Mean coecients of variations were then calculated across participants for each cell. A 3 3 3 repeated measures ANOVA was conducted on the coecients of variation. Post-hoc analysis of dierences between cells were conducted using the NewmanKeuls test with difference criteria specied at a 0:05. To test the guidance hypothesis, the perceived performance error for each participant was calculated by determining the dierence between each participants average performance during the pre-KR phase and the target performance specied by the KR (participants receiving 80% KR would have to perform at 125% of the sample duration to achieve an apparent performance of 100%, those receiving 120% KR would need 83.3% of the sample to appear to be performing at 100%). Thus a participant in the 80% KR group who initially under-estimated by 5% would have a larger apparent target (a 30% change) during KR than a participant who initially over-estimated by 5% (a 20% change). In comparison, over- and under-estimators would have similar magnitudes of change but dierent directions of change indicated by 100% KR. We examined the correlation between the change in participant performance induced by KR and their perceived error of performance.

3. Results 3.1. Pre-KR Prior to KR, participants in each group of the reproduction task reliably reproduced intervals that were slightly shorter than the sample intervals (Fig. 1A). In contrast, the numerical estimates made during the pre-KR phase in the numerical estimation task varied around the sample durations (Fig. 1B). 3.2. Eects of KR on magnitude of response The eect of KR on the magnitude of the response was pronounced and demonstrated in two ways. First, the eect of KR is demonstrated as a change in the magnitude of response from the pre-KR to KR phases (KR group Phase Interaction,

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

89

Fig. 1. KR changes the magnitude of participants reproduced or estimated durations. For the reproduction task (A) and the numerical estimation task (B) the mean response magnitude is represented as the mean of the ratio of the participants reproduced or estimated duration and the actual duration of the sample for each of the three sample durations tested. The mean ratios are indicated for the three phases of the experiment (prior to KR, receiving KR and after KR is withdrawn) and for the three experimental groups, which received either accurate KR (100% KR) or erroneous KR (80% or 120% of the participants actual performance). Eighty percent KR increased participants reproduced or estimated durations whereas 120% KR decreased participants durations. These eects persisted after KR was withdrawn, indicating the occurrence of learning. *p < 0:05 (NewmanKeuls test): the 80% or 120% KR group diered from the 100% group during the phase, p < 0:05: the 80% and 120% KR groups diered from each other.

90

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

Fig. 2. The eects of KR on response magnitude across KR phases are more clearly seen by examining the change in the ratios of reproduced or estimated duration and sample durations between the KR and preKR phases and between the post-KR and KR phases. Eighty percent KR clearly increased the magnitude of both reproduced (A) and estimated (B) durations, whereas 120% KR reduced the magnitude of participant responses. After KR is withdrawn, participants magnitude estimates increased regardless of the type of KR they had received. *p < 0:05: the matching response magnitudes diered from one phase to the next (NewmanKeuls test).

Reproduction: F 4; 56 7:517, p < 0:001; Estimation: F 4; 54 4:229, p < 0:005, and NewmanKeuls dierences as shown in Fig. 2). Second, response magnitudes for the three KR groups within phase 2 (the KR phase) were compared (NewmanKeuls dierences as shown in Fig. 1). Participants receiving 100% KR slightly

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

91

increased the duration of their reproductions and slightly decreased their numerical estimations (Fig. 1 and left half of Fig. 2). In contrast, participants receiving erroneous KR greatly altered their performance, more strongly with 80% than 120% feedback on the reproduction task, and equally on the estimation task. Participants receiving 80% feedback markedly increased the duration of their responses on both tasks. Participants receiving 120% feedback on the estimation task markedly decreased their response magnitudes, whereas, on the reproduction task, the duration of their responses was little changed. These KR eects were present at all three durations (Figs. 1 and 2), but were smaller at the longer durations (ANOVA; Reproduction: KR group DurationF 4; 56 4:164, p < 0:01; Estimation: Main eect of DurationF 2; 54 6:057, p < 0:01; KR group DurationF 4; 54 1:053, p 0:369). 3.3. Eects of KR on group and individual variability of response In both tasks, KR, regardless of whether it was accurate or erroneous, reduced the variability of the responses of each group of participants (Fig. 3A and B). For both the reproduction and estimation tasks, the SEM was smaller in all possible cases, i.e., in 9 of 9 cases for each task (3 feedback groups 3 interval durations; Wilcoxin signed ranks test: N 9, T 0, p < 0:01). In contrast, each participants variability, as indicated by the coecient of variation was essentially unchanged across experimental phase for the reproduction task (see Fig. 3C) (ANOVA: main eect of Phase: F 2; 56 0:256, p 0:775; KR group Phase: F 4; 56 1:314, p 0:276; KR group Phase Duration: F 8; 112 0:813, p 0:593; no signicant NewmanKeuls comparisons). On the estimation task, similar results were found ((see Fig. 3D) ANOVA: main eect of Phase: F 2; 56 6:685, p < 0:01; KR group Phase: F 4; 56 0:042, p 0:792; KR group Phase Duration: F 8; 112 0:998, p 0:442). Although a signicant main eect of Phase was observed on the estimation task, no signicant changes for matching cells between phases 2 and 1, phases 3 and 2, nor phases 3 and 1 were found with NewmanKeuls analysis. Signicant dierences, which may reect the signicant main eect, were only found for non-matching cells (e.g. Phase 2, 100% KR, 3.5 s duration vs. Phase 3, 80% KR, 5.5 s duration). For both the reproduction and estimation tasks, there was a small eect of duration (main eect of Duration: ReproductionF 2; 56 8:705, p < 0:001; EstimationF 2; 56 5:152, p < 0:01) with slightly smaller coecients of variation at longer sample intervals (Fig. 3C and D). 3.4. Post-KRresponse magnitude The eect of KR on response magnitude persists into the post-KR phase as indicated by dierences in the response magnitudes of the KR groups during the postKR phase. The 80% and 120% KR groups continued to dier from each other on both tasks and for all durations (signicant NewmanKeuls dierences shown in

92

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

Fig. 3. Group response variability (A and B), as indicated by the SEM for the ratio data shown in Fig. 1, was reduced by KR and increased after KR was withdrawn. The decline in SEM reects the increased targeting accuracy of each group. This increase in the post-KR phase indicates that the reduced group variability in the presence of KR is caused by a performance-enhancing eect of KR rather than by learning. In contrast, individual participants response variability (C and D), as indicated by the coecient of variation, did not change across KR conditions. Thus the dispersion of a participants response was unchanged, but this dispersion was centered more accurately around the target performance.

Fig. 1). The 80% and 120% KR groups also diered from the 100% KR group on both tasks for some durations (signicant NewmanKeuls dierences shown in Fig. 1). The absolute magnitude of responses did change slightly between the post-KR and KR phases. On both the reproduction and estimation tasks, participants showed some signicantly lengthened response magnitudes post-KR relative to their esti-

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

93

Fig. 3 (continued)

mates during the KR phase (right half of Fig. 2). This eect occurred in all three KR groups and on both tasks suggesting it is unrelated to the KR or task.

3.5. Post-KRresponse variability After the termination of KR the group variability of both reproduced durations and numerical estimates increased (Fig. 3A and B). The SEM increased for 9 of 9 conditions for the reproduction task (Wilcoxin signed ranks test: N 9, T 0, p < 0:01) and for 7 of 9 conditions in the estimation task (Wilcoxin signed ranks test:

94

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

N 9, T 4, p < 0:05). The coecient of variation did not show any consistent change between phases 2 and 3 (Fig. 3C and D). 3.6. Over- and under-estimators On both the reproduction and estimation tasks, most participants under-estimated (by more than 4%) the sample durations in the pre-KR period (Reproduction, mean all durations: NUNDER 19, NOVER 2, NWITHIN 4% 10, 1-sample sign test: z 3:71, p < 0:01; Estimation: NUNDER 18, NOVER 3, NWITHIN 4% 9, 1-sample sign test: z 3:28, p < 0:01). On both tasks, initial under-estimators increased the magnitude of their duration responses when given accurate feedback whereas initial over-estimators decreased the magnitude of their duration responses (Fig. 4). Under-estimators showed larger eects than over-estimators as a result of 80% feedback and showed smaller eects as a result of 120% feedback (Fig. 4). The magnitude of the change in performance caused by KR was a function of the dierence between the participants initial performance prior to KR and the target magnitude indicated by the KR. Results of correlation analysis of the magnitude of the change indicated by the KR (the difference between KR specied target and initial performance) and the magnitude of the actual change in performance induced by KR (change in response magnitudes between KR phase and initial performance in the pre-KR phase) are shown in Table 1. Positive correlations, ranging from 0.516 to 0.958, were seen for all groups.

Fig. 4. Participants that initially under-estimated or that initially over-estimated the sample durations on both the reproduction (left) and numerical estimation (right) tasks, shifted their performance towards the target durations indicated by the 80%, 100% or 120% KR. Thus the change in performance was guided by the indicated targets and was not based on fundamental participant dierences that accounted for their initial over- or under-estimations.

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

95

Table 1 Correlations of the magnitude of the change indicated by the KR (the dierence between KR specied target and initial performance) and the actual change in performance induced by KR (change in response magnitudes between KR phase and initial performance in the pre-KR phase) Reproduction Short 80% KR 100% KR 120% KR 0.759 0.868 0.516 Middle 0.621 0.664 0.758 Long 0.819 0.827 0.560 Numerical Estimate Short 0.564 0.958 0.940 Middle 0.580 0.937 0.867 Long 0.820 0.900 0.909

Note: For all r > 0:55, t9 > 1:83, p < 0:05.

4. Discussion These results conrm our hypothesis that KR can produce both learning and performance enhancing eects on time perception tasks. On both a time reproduction task and a time estimation task, we measured three eects in the presence of KR: (1) a shift in the magnitude of the duration responses, (2) a decrease in the variability of the group magnitude responses, and (3) no systematic change in the variability of each participants individual response. Thus, KR enhances performance by more accurately centering each participants performance around a target without altering the dispersion of that participants response. The accuracy of the centering of performance around the target values specied by the KR declined in the absence of KR, suggesting that this action of KR is a performance-enhancing action. In contrast, the relative change in response magnitude persists following the withdrawal of KR, indicating that learning has occurred. These eects are made particularly clear by the use of erroneous KR. The KR that was intentionally shorter than the persons actual performance reliably increased that persons time estimates and KR that was intentionally longer tended to decrease that persons time estimates. Thus KR appears to provide guidance information about the target performance. These eects also were true regardless of whether the participants initially over-estimated or under-estimated the sample durations suggesting that fundamental factors that might lead a person to over- or under-estimate time intervals do not aect how KR is used. 4.1. Performance eects of KR In the presence of KR, group response variability declined. Because this decline largely disappeared after KR was terminated, it appears to be associated with the known performance enhancing eects of KR (Salmoni et al., 1984). There are several ways that KR might enhance performance on these tasks. It is likely that KR increases the interest of participants in the task. During debrieng many participants reported being more interested and attentive during the feedback phase than in the non-feedback phases. This eect seems unlikely to be a major eect, though, because increased interest should cause a decline in response variability, whereas individual participants response variability did not change. Another possibility is

96

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

that KR could focus or direct attention onto the events from which the participants derived their perception of the passage of time. This seems unlikely, however, because in the erroneous KR groups, there is a conict between the internal representation of time passage and the KR. Focusing attention on these internal timing processes should magnify the conict and, presumably, increase the variability of performance. KR clearly overrides the internal representation and perceptions. This strong eect of erroneous KR has previously been reported in the motor learning literature (Buekers et al., 1992, 1994; McNevin et al., 1994). The attention eects may, in fact, direct attention towards the KR and away from the internal cues and may assist in raising the salience and apparently reliability of the KR (Buekers et al., 1994). Another possibility is that the mental representation of a short interval is changed by KR. Although some have claimed that it is implausible that KR can alter time perception per se (Kladopoulos et al., 1998), this claim is made without empirical evidence. Such an eect seems plausible if, for instance, KR interacts in working memory with a representation of the immediately preceding interval. However, this memory explanation also cannot account for the reduced variability in the erroneous feedback groups. The current experiments, though, do not provide any direct evidence for or against an alteration of mental representations of time underlying the eects of KR. Another alternative is that KR alters neither the mental representation, nor focuses attention on the judgment of interval duration, but rather KR alters the use of time information in the performance of the task itself. Hicks and Miller (1976) suggested, on the basis of KR on other, non-time related perceptual tasks, that KR alters what they called response-translation processes. Task variables which might be affected could include response criteria, motor preparation responses on the reproduction task, proprioceptive feedback on the reproduction task, or numerical labeling on the estimation task, to name a few response processes. These processes depend on the method used to measure time perception (Zakay, 1990). That is, KR may enhance performance by focusing attention on meeting the task requirements. The controlled variable may be the outcome and internal variables may be translated to t the current task requirements. This translation, could, for instance, be as simple as an 80% or 120% linear transform of internal perceptions or representations. This idea is consistent with the hypothesis that KR provides guidance (Salmoni et al., 1984; Schmidt, Young, Swinnen, & Shapiro, 1989). The idea of guidance may seem strange outside of motor tasks where physical guidance can provide proper movement information, but KR does provide guidance in the sense of specifying the target of the performance and specifying the error between the target and the performance. The specication of error could reduce response variability. It is clear from our correlational analysis that the magnitude of the change in a participants performance is largely accounted for by the magnitude of the dierence between the target specied by the KR and the participants initial performance. Prior to KR, the error specication must rely solely on other information sources, that is, the untranslated internal variables. Following KR, error specication might rely on these internal variables and the translation factor(s).

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

97

Some models of the cognitive processes underlying time perception (e.g. Gibbon & Church, 1990; Kladopoulos et al., 1998) are based on analysis of participants response variability. Our nding of no change in the coecient of variation is consonant with these (and similar) models, in which participants response variabilities derives largely from variability in the clock timing and counting mechanisms and not from the decision criteria. The increased accuracy of the centering of the group responses reected in the decreased group variability is also consistent with these models. If this were the case, KR would act by altering the decision criteria. Our data across KR groups do not test directly these models, but are not inconsistent with them. However, the small reduction in the coecient of variation seen with longer sample intervals on both tasks is not entirely consistent with scalar timing models. 4.2. Learning eects of KR In the presence of KR, participants modied the magnitude of their duration responses. These changes persisted after KR was withdrawn and therefore meet the criteria of learned responses (Salmoni et al., 1984). Changes in group variability did not persist after KR was withdrawn and changes in a participants response variability did not occur with KR. Thus changes in group response variability, participant response variability and magnitude estimations are dissociable phenomena. The central question is, What do participants learn from the KR that mediates the persistence of the changes in response magnitude? Because these changes occurred in both reproduction and numerical estimation tasks, consideration should rst be given to those elements shared by the two tasks. Both tasks share the presentation and perception of the sample intervals but dier in the method of reporting the time duration perception. Thus, it is possible that, because of the mismatch between the duration information provided by KR and the internal feedback, KR might alter the perception of the passage of time itself or the mental representation or memory of the sample interval. Such a recalibration of our sense of time would be unprecedented and seems discordant with evidence that time perception changes as we age (e.g., Craik & Hay, 1999) since the recalibrating events, such as timing with a watch, watching 15 s television commercials, etc., are widely available and should prevent age-related changes. Also, performance of the reproduction task requires matching the sample and reproduced intervals. It is hard to imagine how recalibration would aect the perception of one but not both intervals or lead to strikingly dierent perceptions of two equal intervals. The response-translation hypothesis presented above suggests that rather than our time perception being recalibrated, the unchanged perceptions may be used differently during and after KR than before. If KR provides guidance and a target for performance, then the goal of performance is minimizing the dierence between the performance and the target, not minimizing the perceived dierences between the sample duration and reproduced duration. In the estimation task, the goal becomes minimizing the dierence in verbal label of the sample duration and the KR, not of accurately representing the duration of the sample interval. A similar conclusion, that mental representations are not altered, but other decision processes are, has

98

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

been reached studying temporal generalization and bisection tasks (Ferrara, Lejeune, & Wearden, 1997). 4.3. Non-specic changes in time responses There appears to be a lengthening of both reproduced intervals and time estimates after KR has been withdrawn. These eects appear to be non-specic because they are unrelated to the KR group, all three groups showed similar lengthenings. Because we did not include a group which did not receive KR but which performed the same number of trials in the same number of phases we cannot determine if this is an eect of the KR or a result of repeated performance of the task. 4.4. Initial over- versus under-estimators In both our tasks, most participants initially under-estimated the short time intervals, though a few over-estimated them. Montare (1985, 1988), who observed a similar tendency on similar tasks, hypothesized that this dierence reects dierent internal processes and that over- and under-estimators are fundamentally dierent in some way. He attributed the reduction in the magnitude of duration estimates by KR to initial over-estimators to an unspecied increase in excitatory processes and the increase in duration estimates in initial under-estimators to a decrease in excitatory processes (italics in the original, Montare, 1988, p. 585). He inferred from this that the reduced magnitude responses during KR of initial over-estimators reected a reduction of excitatory processes and that the increased magnitude of responses of under-estimators reected increased excitatory processes. Our data suggest otherwise. Erroneous feedback demonstrates that the direction of change in response magnitudes is a function of the target duration, not of a fundamental difference in the participants themselves. The most parsimonious explanation is that participants act to minimize the dierence between the target performance and their own performance. This may also explain why 80% feedback produced larger eects than did 120% feedback. As most participants were initial under-estimators, the 120% target reects a smaller error than does the 80% target. Thus the magnitude of response shift that we observe reects the size of the error between performance and target performance. Several studies, though, have shown that indicating the direction of the response error is as eective as providing information as to the magnitude of the error (Buckolz & Guay, 1975; Salmoni et al., 1984). This does not necessarily mean that magnitude information is lacking as directional correction over the course of successive trials may provide the magnitude information. Indeed, we observed that the magnitude of the dierence between initial performance and the KR signaled target was a powerful predictor of the actual change in performance induced by KR. 4.5. Summary KR can produce both learning and performance enhancing eects on time perception tasks. On both a time reproduction task and an time estimation task, KR in-

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

99

duced three eects while it was presented: (1) a shift in the magnitude of the duration responses, (2) a decrease in the variability of the group magnitude responses, and (3) no change in the variability of each participants individual responses. Thus, KR more accurately centers each participants performance around a target without altering the dispersion of that participants responses. This eect was made particularly clear by using erroneous KR. The size of the change in response magnitude was a function of the dierence between a participants initial pre-KR performance and the target specied by the KR. Thus KR provides targeting and guidance information. When KR is removed, the accuracy of the centering of performance around the target values specied by the KR declined, suggesting that this action of KR is a performance-enhancing action. In contrast, the change in response magnitude persisted following the withdrawal of KR, indicating that learning occurred. Since the task-specic demands of the reproduction and estimation tasks are dierent, but the performance and learning eects of KR appear to be very similar, KR might affect some shared, fundamental cognitive processes as well as task-specic processes.

Acknowledgements This work was supported by the Department of Psychology at Oregon State University. We specially thank the reviewers (anonymous reviewer A and Dr. M.J. Beukers) and the journal section editor (Dr. J. Wagemans) for helpful comments in distinguishing participant and group variability.

References
Aiken, L. R. (1965). Learning and retention in the estimation of short time intervals a circuit and study. Perceptual and Motor Skills, 20, 509517. Allen, D. A., & Clark, C. A. (1979). Eect of information feedback on temporal discrimination in concurrent and post-feedback measures. Perceptual and Motor Skills, 49, 935939. Block, R. A., & Zakay, D. (1997). Prospective and retrospective duration judgments: a meta-analytic review. Psychonomic Bulletin and Review, 4, 184197. Buckolz, E., & Guay, M. (1975). Eects of feedback and procedures on human time estimation. Journal of Human Movement Studies, 1, 132142. Buekers, M. J., Magill, R. A., & Hall, K. G. (1992). The eect of erroneous knowledge of results on skill acquisition when augmented information is redundant. Quarterly Journal of Experimental Psychology. A, Human Experimental Psychology, 44, 105117. Buekers, M. J., Magill, R. A., & Sneyers, K. M. (1994). Resolving a conict between sensory feedback and knowledge of results, while learning a motor skill. Journal of Motor Behavior, 26, 2735. Craik, F. I. M., & Hay, J. F. (1999). Aging and judgments of duration: eects of task complexity and method of estimation. Perception and Psychophysics, 61, 549560. Crowder, A. M. H., & Hohle, R. H. (1970). Time estimation by young children with and without informational feedback. Journal of Experimental Child Psychology, 10, 295307. Ferrara, A., Lejeune, H., & Wearden, J. H. (1997). Changing sensitivity to duration in human scalar timing: an experiment, a review, and some possible explanations. Quarterly Journal of Experimental Psychology. B, Comparative & Physiological Psychology, 50B, 217237. Fraisse, P. (1984). Perception and estimation of time. Annual Review of Psychology, 35, 136.

100

L.J. Ryan, T.B. Robey / Acta Psychologica 111 (2002) 83100

Gibbon, J., & Church, R. M. (1990). Representation of time. Cognition, 37, 2354. Grondin, S. (2001). From physical time to the rst and second moments of psychological time. Psychological Bulletin, 127, 2244. Hicks, R. E., & Miller, G. W. (1976). Transfer of time judgments as a function of feedback. American Journal of Psychology, 89, 303310. Hoyer, W. J., Jones, E. C., White, R. S., & Maconachy, W. V. (1970). Eects of knowledge of the ITI on time estimates. Perceptual and Motor Skills, 31, 392394. Kladopoulos, C. N., Brown, B. L., Hemmes, N. S., & Cabeza de Vaca, S. (1998). The start-stop procedure: estimation of temporal intervals by human participants. Perception and Psychophysics, 60, 438450. McNevin, N., Magill, R. A., & Buekers, M. J. (1994). The eects of erroneous knowledge or results on transfer of anticipation timing. Research Quarterly in Exercise and Sport, 65, 324329. Montare, A. (1985). Learning eects of knowledge of results upon time estimation. Perceptual and Motor Skills, 60, 871877. Montare, A. (1988). Further learning eects of knowledge of results upon time estimation. Perceptual and Motor Skills, 66, 579588. Salmoni, A. W., Schmidt, R. A., & Walter, C. B. (1984). Knowledge of results and motor learning: a review and critical reappraisal. Psychological Bulletin, 95, 355386. Schmidt, R. A., Young, D., Swinnen, S., & Shapiro, D. C. (1989). Summary knowledge of results for skill acquisition: support for the guidance hypothesis. Journal of Experimental Psychology: Learning, Memory and Cognition, 15, 352359. Staddon, J. E. R., & Higa, J. J. (1999). Time and memory: towards a pacemaker-free theory of interval timing. Journal of the Experimental Analysis of Behavior, 71, 215251. Zakay, D. (1990). The evasive art of subjective time measurement: some methodological dilemmas. In R. A. Block (Ed.), Cognitive Models of Psychological Time (pp. 5984). Hillsdale, NJ: Lawrence Erlbaum Associates.