Ling-Ling Tsai, Department of Psychology, National Chung Cheng University, 168 University Road, Min-Hsiung, Chia-yi 62102, Taiwan, China. Tel.: +886 5 272 0411 extn 32201; fax: +886 5 272 0857; e-mail: email@example.com
Previous behavioral and electrophysiologic evidence indicates that one night of total sleep deprivation (TSD) impairs error monitoring, including error detection, error correction, and posterror adjustments (PEAs). This study examined the hypothesis that error correction, manifesting as an overtly expressed self-generated performance feedback to errors, can effectively prevent TSD-induced impairment in the PEAs. Sixteen healthy right-handed adults (seven women and nine men) aged 19–23 years were instructed to respond to a target arrow flanked by four distracted arrows and to correct their errors immediately after committing errors. Task performance and electroencephalogram (EEG) data were collected after normal sleep (NS) and after one night of TSD in a counterbalanced repeated-measures design. With the demand of error correction, the participants maintained the same level of PEAs in reducing the error rate for trial N + 1 after TSD as after NS. Corrective behavior further affected the PEAs for trial N + 1 in the omission rate and response speed, which decreased and speeded up following corrected errors, particularly after TSD. These results show that error correction effectively maintains posterror reduction in both committed and omitted errors after TSD. A cerebral mechanism might be involved in the effect of error correction as EEG beta (17–24 Hz) activity was increased after erroneous responses compared to after correct responses. The practical application of error correction to increasing work safety, which can be jeopardized by repeated errors, is suggested for workers who are involved in monotonous but attention-demanding monitoring tasks.
Sleepiness due to sleep loss or poor sleep quality is known to be related to decrements in some, if not all, psychomotor and cognitive performances (reviewed in Bonnet, 2005; Dinges et al., 2005). Slower, more variable responses and decreases in accuracy concomitant with increases in both the errors of commission and omission comprise the typical performance patterns after extended wakefulness (Doran et al., 2001). Recently, based on the empirical evidence from different research groups (Hsieh et al., 2007; Murphy et al., 2006; Scheffers et al., 1999; Tsai et al., 2005), error monitoring has been added to the list of the deleterious effects of extended wakefulness and sleep deprivation on cognitive function. Error monitoring is an important domain of performance monitoring and includes error detection and posterror remedy (for trial N), and posterror adjustments (PEAs; for trial N + 1). Immediate error correction (Fiehler et al., 2004) may be considered as a representative posterror remedial action. PEAs have been demonstrated in next stimulus–response trials in terms of the response speed and the probability of error. The posterror response speed may be slower (Rabbitt, 1966) or faster (Hester et al., 2005), which appears to be dependent on error awareness (Hester et al., 2005). The posterror probability of error is reduced, i.e., the probability of an error following a previous error (repeated errors) is lower than the probability of an error following a correct response (Laming, 1979).
Scheffers et al. (1999) observed that extended wakefulness reduced the amplitude of error-related negativity (ERN), an event-related potential (ERP) component of brain activity that is evoked by the committed errors. Our research team found that one night of total sleep deprivation (TSD) resulted in a reduction in both the amplitude of ERN and error positivity (Pe), another ERP component related to the committed errors, and an impairment in PEAs in reducing repeated errors (Tsai et al., 2005). Murphy et al. (2006) showed that another effect of PEAs, i.e., posterror slowing, was also reduced after 20 h of wakefulness. Recently, our research team found that one night of TSD impaired immediate error correction although the impairment was confined to specific stimulus types (Hsieh et al., 2007).
All previous studies (Hsieh et al., 2007; Murphy et al., 2006; Scheffers et al., 1999; Tsai et al., 2005) have provided different lines of evidence supporting the idea that extended wakefulness/sleep deprivation impairs the entire error monitoring function, including error detection (reduction in ERN and Pe amplitudes), posterror remedial action, i.e., error correction, and PEAs in the error rate and the respond speed in next stimulus–response trials. Nonetheless, we were puzzled by the other findings of our last study (Hsieh et al., 2007), which showed that the participants maintained comparable levels of global performance after TSD and normal sleep (NS) conditions: the overall mean response speed, response accuracy, error rate, and omission rate, and even the overall error correction rate were equivalent between the TSD and NS conditions. It appears that task difficulty-related cerebral/cognitive compensatory mechanisms (Drummond et al., 2005) might be involved in the maintenance of the abovementioned global performance and overall error correction rate after TSD. The letter version of the Flanker task used in our previous study (Hsieh et al., 2007) has a higher difficulty level, in terms of mean error rate, than the arrow version used in our first study (Tsai et al., 2005) and another letter version of the task used by Murphy et al. (2006). On the other hand, the hypothetical compensatory mechanisms in maintaining global performance and overall error correction rate after TSD may be related to the additional task demand of error correction in our previous study (Hsieh et al., 2007). Several studies have shown that immediate performance feedback provided externally, such as the knowledge of results (Jaśkowski and Włodarczyk, 1997; Steyvers and Gaillard, 1993; Wilkinson, 1961) and reward (Horne and Pettitt, 1985; Steyvers and Gaillard, 1993) can improve performance during sleep deprivation. Thus, we posited that error correction processes, based on internal information and explicitly inducing self-generated performance feedback, might be involved in a similar compensatory mechanism as externally provided feedback information.
This study thus focused on examining two hypothetical compensatory mechanisms in global performance and overall error correction rate after TSD. First, the compensatory mechanism may be dependent on task difficulty. This would manifest as impairment in global performance and overall error correction rate after TSD as task difficulty decreases. Second, the compensatory mechanism may be specific to the demand of error correction. This would manifest as comparable levels of global performance and overall error correction rate between TSD and NS conditions no matter the difficulty level of the task is. In addition, this study investigated whether intentional error correction and corrective behavior, manifesting as internally provided performance feedback to errors, would improve PEAs, particularly after TSD. We applied electroencephalogram (EEG) spectral analysis to examine the involvement of the brain in the hypothetical compensatory mechanism.
Seventy-four students from different universities in Taiwan responded to a recruitment advertisement for the study that had been posted on the campus bulletin board system of the National Chung Cheng University or by word-of-mouth and underwent questionnaire interviews. Thirty-nine of them fulfilled all the inclusion criteria, which were similar to those used in our previous studies (Hsieh et al., 2007; Tsai et al., 2005). In brief, the participants did not suffer from medical and sleep-related disorders, drug use, irregular sleep patterns, and excessive daytime sleepiness; besides, they showed anxiety and depression levels within the normal range and an intermediate type of preference for a particular time of the day. Thirteen of the 39 students did not participate in the study due to lack of time or interest or simply due to loss of touch. To maintain a comparable performance error rate between the participants, we set a criterion of an error rate that was not higher than 40% or lower than 6% for the first task test, and thereby, four participants were excluded thereafter. An additional five participants did not complete the experiment because two were very nervous, resulting in excessive muscle artifacts in the EEGs, one had visual blurring after TSD, one was excluded by mistake, and one quit. Therefore, 17 participants completed the two task tests. However, we further excluded the data of one participant and reported the data collected from the remaining 16 participants (Table 1). Unlike the others, this participant appeared to employ an extremely careless response strategy under NS and probably under TSD conditions as well, which was manifested as a very fast mean reaction time (RT) in the corrected errors (171 ms), a very slow mean error correction time (282 ms), and a very high error rate (50%; compare the data shown in Table 2). All participants were right handed and had normal or corrected-to-normal vision and had provided an informed written consent after they had been familiarized with the experimental procedure. They were provided remuneration for their participation in the experiment.
Table 1. Participant characteristics, sleep data, and subjective measures
NS, normal sleep; TSD, total sleep deprivation.
aOne participant’s sleep data under TSD were lost. Thus, the mean value was calculated from the data of 15 participants.
Certainty of the percentage of estimated correct responses
Effort for task performance
(Estimated – real) percentage of correct responses
Table 2. Effects of sleep condition and congruency on task performance
Data are depicted as the mean (SD). n = 16 except for correction time due to the absence of data from the uncorrected trials in one participant. Two-factor (S × C) anovas were applied to each behavioral variable.
NS, normal sleep; TSD, total sleep deprivation; RT, reaction time; S, sleep condition; C, congruency.
*P <0.05; **P <0.01.
Correct RT, ms
S** C** S × C*
Error RT, ms
S** S × C*
Correction time, ms (n = 15)
Error rate, %
Omission rate, %
Correction rate, %
This study was performed according to a previously described procedure (Hsieh et al., 2007; Tsai et al., 2005). Briefly, the participants received an arrow version of the Flanker task (Eriksen and Eriksen, 1974), while simultaneously being subjected to multiple-channel EEG recordings in the morning after NS and after one night of TSD, with at least a 1-week interval between the two sleep conditions in a counterbalanced sequence. They had to maintain their NS schedules for at least 1 week before the day of the task test or TSD and were supposed to complete daily sleep logs during the week. Simultaneously, they also wore an activity monitor around the dominant arm (AW64; Mini-Mitter, Bend, OR, USA) to record the activity during sleep. They were instructed to avoid alcohol and caffeine 1 day before the task test and not to nap during the daytime under TSD conditions. On the night of TSD, the participants arrived at the laboratory at approximately 22:00 hours and were then kept awake throughout the night under continuous monitoring and were required to complete the Stanford Sleepiness Scale (Hoddes et al., 1973) every hour. For the NS condition, the experimenter made telephone calls to the participants at approximately 22:00 hours on the night before the task test and reminded them to go to bed at their regular bedtime. The Flanker task started at approximately 10:00 hours under both sleep conditions. The participants performed the task alone in a sound-attenuated room with the lights switched off.
During the task test, the EEG was recorded from 32 scalp electrode sites using Ag/AgCl electrodes mounted in an elastic cap (NeuroScan, Charlotte, NC, USA). The EEG activity was referenced to the linked mastoids. Vertical and horizontal electrooculograms (EOGs) were recorded from another four Ag/AgCl electrodes mounted in the cap and placed around the outer canthi of both eyes. All electrode impedances were maintained below 10 kΩ. The EEG and EOG signals were amplified using SYNAMPS amplifiers (Neuroscan) and sampled at 500 Hz with a band-pass frequency of 0.05–50 Hz.
The participants were instructed to evaluate their present sleepiness using the Stanford Sleepiness Scale immediately before and after the task test. At the end of each task test, they also answered three questions on the subjective estimates of performance, including the percentage of correct responses, the certainty of the estimated correct response percentage (from 1 = extremely uncertain to 10 = extremely certain), and the effort required for task performance (from 1 = no effort to 10 = maximum effort).
A modified Flanker task with arrow stimuli was used, and it has been described previously (Tsai et al., 2005). Briefly, the task stimuli were presented on a computer screen (15’’) using an E-Prime-based program (Psychology Software Tools, Pittsburgh, PA, USA). The participants sat at approximately 90 cm from the screen. Each stimulus (luminance: 120 cd cm−2) comprised a visual array of five arrows subtended 2.48° horizontally and 0.38° vertically, presented in Courier New 18-point font and drawn in either dark gray (R: 156, B: 158, R: 156), light gray (R: 196, B: 194, B: 196), or white (R: 252, G: 254, B: 252) on a black background (luminance: 0 cd cm−2; R: 0, G: 0, B: 0). As shown in Fig. 1, the target arrow was in the center of the visual array and was flanked on each side by two arrows pointing in the same direction as the target (congruent) or in the opposite direction (incongruent). Congruent and incongruent trials were presented with equal probability.
Each trial started with a fixation white cross ‘+’ for 1000 ms. The visual array of arrows appeared 200 ms after the fixation disappeared, and it lasted for 50 ms. The next trial started approximately 3 s after the participant’s correct response or 2 s after the error-corrective response. A 1-s limit was set for both the maximum RT and the maximum correction time. The participants were instructed to press the designated key on the keyboard in response to the target arrow with maximal speed and accuracy and to correct their errors immediately by pressing the correct key after an erroneous response. The participants were provided 32 practice trials and at the end of each trial, they received a performance feedback informing them about the correctness of the response and the instruction ‘Speed up, please’ if the RT was >350 ms. We employed performance feedback to control the comparability of the accuracy between participants and between the sleep conditions. Subsequently, the participants completed 16 blocks of 64 trials. Only the feedback instruction ‘Speed up, please’ was then presented at the end of the last trial of each block if the mean RT of the block was >350 ms. Between the blocks, the participants were permitted to close their eyes and relax for 1 min or longer.
The daily bedtime, time of rising, and time in bed of the participants recorded in the sleep log were averaged over the entire week before the task test and were cross-examined with the activity data recorded in the wrist activity monitor. The daily activity data were automatically scored over 1-min epochs using Actiware-Sleep v. 3.3 (Mini-Mitter).
Reaction time was defined as the time between the onset of the arrow array and the first key press, and the correction time was defined as the time between the erroneous response and the subsequent corrective response. The accuracy of the response, error rate, and omission rate were calculated as the percentage of the participant’s first attempt of correct, erroneous, and omitted responses, respectively. The rate of committing errors was calculated as the percentage of committed errors over the total overt responses. The error correction rate was calculated as the percentage of corrected errors. PEAs were defined as the difference between the task performance in the trial following errors and the task performance of the trial following correct responses.
The left central (C3) EEG data segments spanning 512 ms (256 samples), from 700 to 190 ms before the onset of the arrow array, were subjected to power spectral analysis using Fast Fourier transforms without windowing. The power density values were summed at 1–24 Hz (total), 1–4 Hz (delta), 5–8 Hz (theta), 9–12 Hz (alpha), 13–16 Hz (sigma), and 17–24 Hz (beta). EEG epochs containing absolute EEG amplitudes higher than 50 μV (rejection rate: 0.91% under NS versus 1.65% under TSD; paired t-test P =0.312) or with a total power value higher than a maximum limit or lower than a minimum limit (rejection rate: 6.46% under NS versus 7.38% under TSD; paired t-test P =0.209) were excluded from statistical analysis. The maximum and minimum limits of the EEG total power were set individually and were defined as the median EEG total power value of all trials on a task test day plus and minus twice the interquartile range (obtained by subtracting the 25th percentile from the 75th percentile), respectively.
The effect of TSD on the sleep data and the subjective measures were examined using paired t-tests. A two-factor repeated-measures analysis of variance with either a 2 × 2 or a 2 × 3 design was used to evaluate how the sleep conditions interacted with the trial congruence and response type. Post hoc comparisons were performed using the LSD test or Tukey’s test if more than two comparisons were performed. All statistical analyses were performed using SYSTAT® 7.0 for Windows® (Systat Software, Chicago, IL, USA). The family-wise significance level was set at 0.05.
The participant characteristics and the sleep data in this study are shown in Table 1. Gender differences were not found in any collected participant characteristics or sleep variables. The participants felt progressively sleepier over time on the night of TSD and both before and after the task test under the TSD condition than under the NS condition (Fig. 2). After the task test, the participants subjectively estimated a lower correct rate with less confidence under the TSD condition than under the NS condition (Table 1), even though they claimed that the effort level was similar, and their underestimation of the correct rate did not differ significantly between the two sleep conditions.
Global performance and congruency effect
After one night of TSD, the participants responded more slowly in correct trials [M (SD) = 415 (51) ms versus 368 (30) ms, t(15) = 4.95, P <0.001), less accurately (79.6 (8.3) % versus 86.1 (6.3) %, t(15) = 3.06, P =0.008], and with more omissions [6.4 (7.3) % versus 0.5 (0.8) %, t(15) = 3.46, P =0.003], but maintained an equivalent number of errors [13.9 (4.7) % versus 13.4 (5.9) %, P =0.78]. The trial congruency (Table 2) and postconflict adjustment effects in the RT, error rate and accuracy were still maintained after TSD (data of the postconflict effect are not shown).
Error correction and PEAs
Immediate error-corrective behavior showed that one night of TSD induced an increase in the correction time [242 versus 178 ms; t(15) = 3.42, P =0.004] and a decrease in the correction rate [77.3% versus 95.2%; t(15) = 4.01, P =0.001]. The sleep conditions did not interact with the trial congruency, nor did trial congruency have any significant main effects on the correction time and rate (Table 2).
Next, a two-factor analysis of variance was performed to examine the effect of interaction between the sleep conditions and PEAs. The sleep conditions did not interact with the previous response type in terms of accuracy (P =0.76), error rate (P =0.44), or the rate of committing errors (P =0.53), but the main effect of the previous response type was significant with regard to all three variables [F(1,15) = 8.84, P =0.009; F(1,15) = 7.82, P =0.014; F(1,15) = 8.46, P =0.01]. Thereby, this study found that the PEAs in reducing repeated errors were still maintained at the same level after TSD as after NS. In contrast, neither the previous response type nor its interaction with the sleep conditions significantly affected the correct RT and omission rate, i.e., PEAs were not significant with regard to the RT and the omissions even under the NS condition (see below for the effect of corrective behavior).
We further examined whether the maintenance of PEAs after TSD was dependent on the presence of immediate corrective behavior. We sorted all the data from the trial following errors into two groups: one, from the trial following corrected errors and the other, from the trial following uncorrected errors. We further subtracted the data derived from the trial following the correct responses from each of the two datasets (n = 14 because two participants corrected all the committed errors under the NS condition). We then performed two-factor analyses of variance on these data and found that the effect of interaction between the corrective behavior and sleep conditions was significant with regard to the accuracy [F(1,13) = 6.96, P =0.02] and omission rate [F(1,13) = 12.02, P =0.004] and tended to be significant with regard to the correct RT [F(1,13) = 3.74, P =0.075] and non-significant with regard to the error rate (P =0.67) and the rate of committed errors (P =0.77). The main effect of corrective behavior on the omission rate [F(1,13) = 11.45, P =0.005] and correct RT [F(1,13) = 12.36, P =0.004] was also significant. Post hoc comparisons showed that the trial following corrected errors had a higher accuracy, lower omission rate, and faster correct RT than the trial following uncorrected errors, and these effects were particularly prominent under the TSD condition (Fig. 3). It was noted that the participants responded less accurately in the overall trials (79.6% versus 86.1%) and in the trial following the correct responses [80.2% versus 85.9%; t(15) = 2.91, P =0.01] under the TSD condition than under the NS condition; however, they maintained the accuracy in the trial following corrected errors at a level higher than that in the trial following uncorrected errors under the TSD condition (see Fig. 3) and at a level comparable to that of the corresponding trials under the NS conditions (84.1% versus 88.1%; P =0.16).
EEG power spectral activity
Electroencephalogram power activity was significantly affected by the sleep conditions, response type (correct response versus error or postcorrect response versus posterror response), and the interaction between the sleep conditions and response type at different frequency bands. The total [F(1,15) = 6.59, P =0.021], delta [F(1,15) = 7.68, P =0.014], and theta [F(1,15) = 18.65, P <0.001] power were higher after TSD than after NS. The delta [F(1,15) = 7.46, P =0.015] and beta power [F(1,15) = 6.28, P =0.022] varied with the response types. Delta power was higher, but beta power was lower at the beginning of the trial with errors than at the beginning of the trial with correct responses. The beta power associated with the correct responses was even higher under the TSD condition than under the NS condition (the effect of interaction between the response type and sleep conditions: [F(1,15) = 4.92, P =0.042]. The previous response type was shown to have a significant effect on the beta power, which was higher at the beginning of the trial following errors than at the beginning of the trial following correct responses [F(1,15) = 4.84, P =0.044]. Owing to the considerably small amount of artifact-free EEG data obtained from uncorrected errors under the NS condition, the interaction effect between error corrective behavior and sleep conditions was not evaluated.
In our previous study (Hsieh et al., 2007), we found that the global performance and overall correction rate were maintained under the TSD condition at a level equivalent to that under the NS condition. In contrast, this study showed a reduction in both the global performance, i.e., with slower response speed and less accurate but more omitted responses, and the overall correction rate after sleep deprivation. Task difficulty might be related to this contradiction. The letter version of the Flanker task used in the study of Hsieh et al. (2007) was more difficult than the arrow version used in the study of Tsai et al. (2005) and this study. Even though the participants were trained to maintain a comparable response speed, they committed more errors when the letter version task was used [∼25% in the study Hsieh et al. (2007) versus ∼13% in this study]. Increased task difficulty has been shown to augment cerebral compensatory recruitments after TSD in several studies (Chee and Choo, 2004; Drummond et al., 2004, 2005). However, even if task difficulty-related compensation occurs after TSD, this cannot entirely counteract the effect of TSD on immediate error correction when a specific stimulus type of the Flanker task is considered (Hsieh et al., 2007). Further systematic experiments that would involve direct manipulation of the level of task difficulty need to be carried out after TSD in order to understand the extent to which the induced compensatory mechanism can recuperate corrective behavior.
Although this study failed to show the involvement of error correction in the compensatory mechanism in the global performance and overall correction rate after sleep deprivation, it indicated that error correction might evoke a compensatory mechanism which counteracted the TSD-induced impairment in PEAs. When instructed to correct their errors immediately after committing errors, the participants maintained PEAs in reducing the repeated errors even after one night of TSD. However, it is not the corrective behavior but an explicit instruction for error correction that modulates the effect of PEAs on the errors committed after TSD. The exact manner in which the provoked intention of error correction counteracted the effect of TSD on PEAs in the error rate cannot be entirely explained by this study. Nonetheless, the EEG data suggest the involvement of the brain. EEG beta activity predicts the correctness of the following response, being lowered preceding the erroneous response as shown in our previous study (Tsai et al., 2005) and this study. In contrast, EEG beta activity did not vary with the previous response type in the absence of correction instructions (Tsai et al., 2005) but increased after erroneous responses when correction instructions were given as shown in this study.
On the other hand, PEAs in the omission rate and response speed were dependent on corrective behavior, particularly after one night of TSD. Under the TSD condition, the uncorrected errors were followed by more omissions and slower speed in the correct responses than the corrected errors. Given that the errors of omission and commission are involved in the same PEA mechanism, it appears odd that the uncorrected errors were with positive PEAs in committed error rate, i.e., a reduction in the percentage of posterror erroneous responses; but with negative PEAs in omission rate, i.e., increase in the percentage of posterror missed responses. At least two hypothesized issues might be involved in this discrepancy. First, the PEA mechanism is initiated only when overt responses, i.e., correct responses and committed errors, occur, and second, the posterror changes in the errors of omission are specific to sleep deprivation. Although PEAs in reducing the error rate in next stimulus–response trials have been reported in two studies (Laming, 1979; Tsai et al., 2005), the omission rate was differentially calculated only in one study and did not show any posterror changes under both NS and TSD conditions (Tsai et al., 2005). This study also failed to find the posterror changes in the omission rate when the errors of commission were not differentiated by corrective behavior. Thus, it is likely that posterror changes in the omission rate are independent of PEAs per se, but are specific to sleep deprivation because corrective behavior-dependent posterror changes in the omissions are present only under the TSD condition, as shown in this study.
On the other hand, posterror slowing has been reported by Rabbitt (1966) and in many other studies (e.g. Fiehler et al., 2004; Garavan et al., 2002; Hester et al., 2005; Kerns et al., 2004; Laming, 1979; Murphy et al., 2006), and it is more profound in an alert than a sleepy state (Murphy et al., 2006). However, other studies have reported no changes in (Scheffers et al., 1999; Tsai et al., 2005) or a faster (Hester et al., 2005) posterror response speed. Error awareness appears to be related to the posterror changes in the response speed, but the results are still variable. One study using a motor go/no-go response inhibition task has shown improvement in the posterror response speed is associated with explicitly recognized errors and a cautionary slowing of the posterror responses with errors that the participant is unaware of (Hester et al., 2005). In this go/no-go task, the errors had been defined as overt responses that were committed when response inhibition was required, i.e., response inhibition commission errors. Thus, error-corrective behavior was impossible to be exercised even after an error that the participant was aware of. An additional two studies used an antisaccade task and showed opposite results in the posterror changes in the response speed, with the presence of slowing only with perceived errors (Klein et al., 2007; Nieuwenhuis et al., 2001). However, as many unaware saccadic errors are corrected immediately more often than the errors that the participant is aware of, it is possible that most ‘unaware errors’ are ‘corrected errors’ and the ‘aware errors’ are ‘uncorrected errors’. Thus, it appears that our finding that the uncorrected errors were followed by a slower response speed as compared to the corrected errors does not contradict the findings of any of the above three studies.
Taken together, this study found that the demand of error correction did not induce a compensatory mechanism in the global performance and overall error correction rate after TSD but induced a counteracting effect on TSD-induced impairment of PEAs. PEAs in reducing repeated errors were maintained completely after TSD irrespective of whether error-corrective behavior was executed. In addition, corrective behavior imposed other PEAs in reducing the lapses (omissions) and enhancing the response speed, particularly after TSD. A cerebral mechanism might be involved in the effect of error correction on PEAs because EEG beta activity was increased after erroneous responses compared to after correct responses. Practically, the current findings provide useful information for increasing work safety, which can be jeopardized by repeated errors, particularly in case of monotonous but attention-demanding monitoring tasks. By explicitly instructing the workers to correct their errors immediately when the errors are committed might help maintain PEA functions. Last but not the least, further studies are required to determine the precise effect of compensatory mechanisms that are induced by intentional error correction to counteract TSD on PEAs and the mechanism behind this effect. Error correction processes are involved in internal information and explicitly inducing self-generated performance feedback. Whether error correction is mediated by the same compensatory mechanism as externally provided feedback information, such as the knowledge of results (Jaśkowski and Włodarczyk, 1997; Steyvers and Gaillard, 1993; Wilkinson, 1961) and reward (Horne and Pettitt, 1985; Steyvers and Gaillard, 1993), is being investigated in our laboratory.
The authors would like to thank the National Science Council of the Republic of China, Taiwan for financially supporting this research under Contract No. NSC-94-2413-H-194-018. The authors would like to thank I.-C. Cheng for her assistance on programming.
Conflict of interests
Shulan Hsieh declares no conflict of interest. Cheng–Yin Tsai declares no conflict of interest. Ling-Ling Tsai declares no conflict of interest.