How does error correction differ from error signaling? An event-related potential study

Abstract

It has been a question of debate whether immediate error corrections in speeded forced-choice reaction time tasks require an error detection signal from the performance monitoring system or whether they reflect delayed correct responses that are executed after the premature error like in a horserace. In contrast, signaling the error by pressing a response button that is unrelated to the primary task is based on error detection. The present study investigates the similarities and differences between immediate error corrections and signaling responses by means of behavioral and event-related potential data. In a within-subject design, participants performed two sessions of the flanker task. In one session, errors had to be corrected by immediately pressing the correct response, in the other session, errors had to be signaled by pressing an error signaling button. Compared to the signaling session, in the correction session, more errors and error corrections were made, reaction times were shorter, and the amplitude of the error-related negativity (ERN) was reduced. Whereas the error significance did not seem to differ across session, participants have most likely reduced the motor threshold in the correction session to enable efficient immediate corrections. This interpretation is supported by the lateralized readiness potentials and is consistent with the response conflict monitoring hypothesis of the ERN. The present study demonstrates that differences in error corrections may be attributable to differences in motor threshold. We conclude that the error signaling procedure is a more direct and reliable way to behaviorally test the functional integrity of the performance monitoring system than the instruction to correct errors. The consequences for studies in patients and with pharmacological challenges are discussed.

Introduction

“A man who has committed a mistake and does not correct it is committing another mistake.” (Confucius, 551–479 B.C., Chapter 15, Verse 29).

The ability to monitor for errors and to implement compensatory actions is a prerequisite of goal-directed and flexible behavior. Making use of event-related potentials (ERPs) and neuroimaging, performance monitoring research has made substantial progress over the last two decades. The error-related negativity (ERN; also called error negativity, Ne) is a negative deflection observed after errors resulting from premature responses in forced-choice reaction time tasks (Falkenstein et al., 1990, Falkenstein et al., 2000, Gehring et al., 1993). It peaks within 50 to 100 ms after the erroneous response and has a frontocentral distribution over the scalp. Source localization and functional magnetic resonance imaging (fMRI) studies have suggested the rostral cingulate zone (RCZ) to be its main generator (Ridderinkhof et al., 2004, Ullsperger and von Cramon, 2001). A recent study using concurrent EEG and fMRI recordings not only provided evidence for a trial-by-trial coupling of the ERN and the fMRI signal in the RCZ but also showed that the dynamic fluctuations in ERN amplitude predict compensatory post-error slowing on trials subsequent to errors (Debener et al., 2005). Thus, the ERN seems to reflect a signal from the performance monitoring system indicating the need for adjustments (Ridderinkhof et al., 2004, Ullsperger et al., 2004). In forced-choice reaction time tasks, this signal may be operationalized as the amount of post-response conflict. The response conflict monitoring theory (Botvinick et al., 2004, Yeung et al., 2004) suggests that the performance monitoring system monitors for the conflict between simultaneously activated response tendencies. According to the theory, the amount of conflict determines subsequent modulations in cognitive control (Ullsperger et al., 2005). Simulations in connectionist models suggested that the ERN reflects the amount of post-response conflict, i.e., the conflict between the executed erroneous response and the still-evolving correct response tendency (Yeung et al., 2004).

The relationship between ERN amplitude and corrective behavior is less clear, however. A number of ERP studies revealed a modulation of the ERN by error correction (Falkenstein et al., 1994, Falkenstein et al., 1996, Gehring et al., 1993, Rodriguez-Fornells et al., 2002). Larger ERN amplitudes were found for corrected compared to uncorrected errors. In contrast, a recent study investigating incidental (spontaneous) and intentional (instructed) error corrections failed to replicate this amplitude modulation (Fiehler et al., 2005). Interestingly, this study found larger ERN amplitudes in a group of participants that was not instructed to correct errors as compared to a group that was instructed to correct each encountered error by immediately pressing the correct response button, while no amplitude difference was found between corrected and uncorrected errors within the groups. Behavioral findings showing fewer errors, more late responses as well as a reaction time slowing subsequent to an error only in the group not instructed to correct errors suggested a more cautious response behavior in this group. One possible interpretation of these findings was the following: “It seems that when participants are explicitly told to correct their errors they view errors as expected and more acceptable than participants in the non-instructed group, who presumably believe that errors are unacceptable” (p. 8, Fiehler et al., 2005). This view was supported by concurrent recordings of phasic cardiac responses showing stronger error-related heart rate decelerations in the group unaware of the opportunity to correct errors (Fiehler et al., 2004).

It has been a matter of debate whether immediate error corrections in forced choice reaction time tasks can be interpreted as compensatory actions that result from error processing. Based on the observation that error correction time, i.e., the latency of the corrective response relative to the error, can be very short, it has been suggested that at least incidental error corrections may rather be a delayed correct response (Rabbitt, 2002). In other words, like in a horserace, the evolving incorrect and correct response tendencies could be executed sequentially without the necessity to detect the error. Fiehler et al. (2005) showed that the instruction to correct errors yields a gain in slow error corrections and suggested that these slow, intentional error corrections require the detection of the preceding error. However, they could not rule out the alternative explanation that the intention to correct errors leads to a general reduction of the motor threshold. The motor threshold concept, implemented in the response conflict monitoring models, suggests that activation of a response channel needs to exceed a certain threshold to result in an overt response. A general reduction of the motor threshold would enable even the execution of weak correct response tendencies after the error, thereby increasing the number of error corrections. As weak response tendencies need longer to reach the threshold, a selective increase of slow corrections would be expected. Thus, the motor threshold account would suggest that even instructed error corrections do not necessarily reflect activity of the performance monitoring system. This is an important issue, as many patient studies showed pathological changes of the ERN but often failed to show clear impairments of error corrections (Gehring and Knight, 2000, Ullsperger, 2006, Ullsperger and von Cramon, in press). This discrepancy has been difficult to reconcile with the notion that the ERN is a correlate of performance monitoring. The major question that needs to be answered is whether error corrections are a reliable measure of the functional integrity of the performance monitoring system.

A way to test error detection is to instruct participants to press a signaling button that is unrelated to the primary task whenever they encounter an error (Rabbitt, 2002). The horse race model is unlikely to explain the signaling response, as it is not induced by the stimuli of the primary task. When an error has been detected by the performance monitoring system, the signaling response is initiated.

The present study investigates whether intentional error corrections differ from error signaling. ERPs and lateralized readiness potentials (LRP) are used with the aim to test the motor threshold account for intentional error corrections. To this end, participants performed two sessions of a modified flanker task in a within-subjects design. In one session, they were instructed to immediately correct errors; in the other session, they were asked to signal the error by pressing a specific response button. If an increase in error corrections is reached by a general decrease of the motor threshold, a number of predictions could be made. First, reaction times should be shortened as the response tendencies exceed the motor threshold earlier after stimulus presentation. Second, error rates are expected to be increased, because even weak incorrect response tendencies would be executed. Third, whereas with high motor threshold these weak tendencies are not executed and thus do not contribute to post-response conflict, the assumed reduction in motor threshold should result in lower post-response conflict caused by a contribution of strong and weak erroneous response tendencies. It follows that a reduced ERN should be associated with intentional error correction as compared to error signaling. Fourth, a reduced motor threshold is expected to result in reduced lateralizations of the readiness potential, indicating the differential engagement of the left and right (pre)motor cortices in the preparation and initiation of unimanual motor responses (Gratton et al., 1988, Kutas and Donchin, 1980). This motor-related asymmetrical brain activity can be isolated by a double subtraction-averaging procedure resulting in the lateralized readiness potential (LRP) (De Jong et al., 1988, Rugg and Coles, 1995). Concerning the present study, the following prediction can be made: with a reduced motor threshold, only very weak erroneous response tendencies are not executed. Therefore, the lateralization to the incorrect side, typically observed in incompatible correct trials (Gratton et al., 1988), should be reduced in the correction session.

The aforementioned predictions are consistent with the results of a simulation based on a connectionist computational model of response conflict monitoring in a flanker task (Yeung et al., 2004). In Simulation 3 (p. 938), speed–accuracy shifts were modeled by varying the motor threshold and an additional parameter. While predictions for the LRP were not made explicit, the simulation predicts (1) shorter reaction times, (2) more errors, and (3) a smaller ERN for the speed condition, modeled by lowering the motor threshold.