Abstract
The error negativity or error-related negativity (Ne/ERN), a correlate of errors in choice tasks, is related to posterror adjustments indicating that it signals the need for behavioral adjustments following errors. However, little is known about how the error monitoring system selects appropriate posterror adjustments for a given error to ensure that future errors are effectively prevented. This could be achieved by monitoring error precursors indicating potential error sources and then scaling the Ne/ERN according to the strength of the error precursor upon error occurrence. We isolated such an error precursor in alpha oscillations and tested whether it predicts the size of the Ne/ERN. A total of 28 participants (23 female, 5 male) had to classify a target in one hemifield but ignore a distractor in the opposite hemifield. Because responding to the distractor always led to an error, misallocating spatial attention to the distractor as reflected in posterior alpha was a viable error precursor in this paradigm. We found that an alpha asymmetry reversal indicated a shift of spatial attention to the distractor on error trials and predicted the Ne/ERN on a single-trial level. The Ne/ERN in turn predicted alpha asymmetry on the next trial indicating a shift of spatial attention away from the distractor. This is consistent with the idea that the error monitoring system scales the Ne/ERN according to the strength of error precursors to select appropriate posterror adjustments of behavior.
Significance Statement
This study reports evidence that the error monitoring system uses misallocation of spatial attention to distracting information as an error precursor to scale error signals in the brain. This ensures that error signals convey information about the type and strength of behavioral posterror adjustments that are necessary for a given error. The idea of monitoring error precursors that reflect specific error sources significantly extends existing theories of error monitoring mechanisms in the brain.
Introduction
Even in well-learned forced-choice tasks, humans frequently commit behavioral errors. To optimize performance, the human brain detects such errors fast and reliably. A neural correlate of this process is the error negativity or error-related negativity (Ne/ERN), a frontocentral negative deflection in the event-related potential peaking within 100 ms after erroneous responses (Falkenstein et al., 1990; Gehring et al., 1993). The Ne/ERN is generated in the medial frontal cortex (MFC; Dehaene et al., 1994; Iannaccone et al., 2015), and its amplitude correlates with posterror adjustments such as posterror slowing (Gehring et al., 1993; Debener et al., 2005; Marco-Pallarés et al., 2008) and posterror reduction of interference from goal-incongruent distractors (Maier et al., 2011; Steinhauser and Andersen, 2019). Therefore, the Ne/ERN is believed to reflect an MFC signal indicating the need for posterror adjustments (Ridderinkhof et al., 2004; Ullsperger et al., 2014).
To initiate posterror adjustments, it is not only necessary to detect an error but also to select an adaptive adjustment that effectively prevents future errors (Steinhauser et al., 2017). There is evidence that posterror adjustments affect the very processes that caused the errors. For instance, errors due to response speeding entail posterror slowing (Rabbitt, 1966; Laming, 1979) achieved by a response-criterion shift (Jentzsch and Leuthold, 2006; Dutilh et al., 2012). Moreover, errors due to failures of selective attention lead to enhanced selective attention in the subsequent trial (Ridderinkhof, 2002; King et al., 2010; Danielmeier et al., 2011, 2015; Maier et al., 2011; Steinhauser and Andersen, 2019). These adjustments have been shown to correlate with the Ne/ERN amplitude (Gehring et al., 1993; Debener et al., 2005; Maier et al., 2011) and occur already at the time of the Ne/ERN (Steinhauser and Andersen, 2019). This raises the question of how the error monitoring system already knows about the nature and strength of adjustments at this early time point.
A central role in selecting an adaptive posterror adjustment could be played by monitoring error precursors, that is, neural and behavioral markers that indicate nonoptimal processing and predict the occurrence of errors. For instance, detecting failures of selective attention already before an erroneous response suggests that posterror adjustments of selective attention could effectively prevent future errors. Moreover, the strength of this error precursor signals the strength of the required adjustment and hence could be used to scale the Ne/ERN accordingly (Maier et al., 2011; Maier and Steinhauser, 2013). Previous studies have provided evidence for error precursors like deactivation of frontoparietal control networks on trials before errors (Weissman et al., 2006; Li et al., 2007; Eichele et al., 2008) or altered selective attention during stimulus processing on trials preceding errors (Steinhauser et al., 2012) and error trials (Mazaheri et al., 2009; Steinhauser and Andersen, 2019). However, none of these studies have addressed the question of whether these error precursors are related to subsequent error-related brain activity.
Here, we investigated the relationship between attentional error precursors and error-related brain activity by leveraging the phenomenon of posterior alpha asymmetry, which refers to a greater suppression in the alpha band (8–12 Hz) contralateral to attended locations (Klimesch et al., 1998; Worden et al., 2000; Sauseng et al., 2005). Alpha suppression has been shown to correlate with metacognitive judgments about attentional focus (Whitmarsh et al., 2014; see also, MacDonald et al., 2011) indicating that related attentional states are accessible to metacognitive processes. In the present paradigm, we simultaneously presented a target on one and a distractor on the other side of fixation. On error trials, we expected a reversed alpha asymmetry indicating that spatial attention is directed toward the distractor. Crucially, we hypothesized that this error-related alpha asymmetry predicts the Ne/ERN on a single-trial level. This would show that the error monitoring system uses deviations in spatial attention as an error precursor to scale the error signal according to the source of the error. Furthermore, the Ne/ERN should in turn predict the alpha asymmetry on subsequent trials, which would show that the signaled need for adjustment is translated into actual adjustments.
Materials and Methods
Participants
Twenty-eight participants (23 female, 5 male, 26 right-handed, 2 left-handed) between 18 and 28 years of age (mean = 21.9, SE = 0.445) with normal or corrected-to-normal vision participated in the study. They were recruited at the Catholic University of Eichstätt-Ingolstadt and received course credit or 8 Euro per hour for participation. The study was approved by the Ethics Committee of the Catholic University of Eichstätt-Ingolstadt (Nr. 161-2023), and all participants gave informed consent.
Stimuli and procedure
A PC running Presentation software (Neurobehavioral Systems, Albany, CA) controlled stimulus presentation and response registration. Stimuli were presented on a 21-inch color monitor at a viewing distance of 70 cm. The task is schematically presented in Figure 1. Stimuli were composed of the letters B, K, P, R, M, V, W, or X in a white Arial font in front of a black background. On each trial, one letter was presented to the left of the screen center, and one letter was simultaneously presented to the right of the screen center. One of the letters was smaller (subtending a visual angle of 0.737 × 0.900° width × height) and presented at an eccentricity of 1.15° of visual angle. The other letter was larger (subtending a visual angle of 1.06 × 1.23°) and presented at an eccentricity of 1.39° of visual angle. The positions of the small and the large letters (left vs right) varied randomly from trial to trial.
Task information. If the two letters in a stimulus were different (go trials), participants were instructed to classify the smaller of the two letters (the target; M in the example) by pressing one of four responses while keeping the gaze at the central fixation cross and ignoring the larger letter (the distractor; P in the example). In these trials, errors could consist in the response assigned to the distractor (distractor error; the left index finger in the example) or in a response not associated with the stimulus (nondistractor errors; the left or right middle fingers in the example). Please note that it depends on the stimulus which response finger corresponds to which response type.
There were two types of trials that differed with respect to whether the two letters were different or the same. If the two letters were different (“go trials”), participants were instructed to classify the small letter (the target; M in the example in Fig. 1) and ignore the large letter (the distractor; P in the example in Fig. 1). The letters were assigned to one of four responses in pairs. Participants were instructed to press the W-key with the left middle finger if the target letter was B or K, the S-key with the left index finger if the target was P or R, the L-key with the right index finger if the target was M or V, and the P-key with the right middle finger if the target was W or X. The distractor was always assigned to a different response than the target (e.g., if the target was a B or a K, possible distractors were P, R, M, V, W, or X). This resulted in 48 possible stimuli for go trials. If the two letters were the same (“catch trials”), participants were instructed to press the space bar with both thumbs. These trials served to prevent participants from fully ignoring the distractors. There were eight possible stimuli for catch trials.
While performing the task, participants were asked to keep their gaze on a white fixation cross (0.491 × 0.491°) which remained at the screen center throughout an experimental block. Each trial started with a screen containing only this fixation cross (Fig. 1). After 500 ms, the target and distractor were presented simultaneously for 150 ms. Then, the black screen with the fixation cross was again presented until a response was given. The response was followed by another 600 ms of black screen and fixation cross until the next trial started. This interval was restarted whenever further responses (e.g., spontaneous error corrections) occurred before it was terminated.
The experiment consisted of a practice session and a test session carried out on consecutive days. In each session, participants performed blocks of 112 randomized trials (two trials for each possible stimulus, resulting in 96 go trials and 16 catch trials) plus three randomly drawn practice trials at the beginning of each block. In the practice session, participants initially performed two blocks without speed pressure to learn the stimulus–response mapping. Then, they were instructed that the remaining practice blocks served to train their response speed for the test session. They were told that in the test session on the next day, they would start with 10,000 points some of which they would lose depending on their performance (minus 20 points for each slow response and each response other than the space bar on a catch trial, i.e., a missed catch trial) and that the remaining points would be converted into money (with 1,000 points corresponding to 1 €). Then they worked through 12 practice blocks, during which the error rate was adjusted by an adaptive deadline procedure. At the end of each block, participants were instructed about the number of slow responses (i.e., responses exceeding the deadline) and the number of missed catch trials. The deadline was initially set to 1,000 ms following stimulus onset. After each practice block, the deadline was decreased by 50 ms if the error rate in the block was below 15% or increased by 50 ms if the error rate in the block was above 25%. The deadline from the last practice block was used for all blocks of the test session.
At the beginning of the test session, participants were instructed that they would now lose points for errors on go trials (in addition to slow responses and missed catch trials). If they pressed the key associated with the distractor (distractor error), they would lose 10 points. If they pressed one of the keys neither associated with the target nor the distractor (nondistractor error), they would lose 1 point. This was done to encourage attentional adjustments as it ensured that responding to the distractor was particularly costly. After 2 further practice blocks, 12 test blocks were administered. After each block, participants were informed about the number of distractor and nondistractor errors, the number of slow responses, the number of missed catch trials, and the remaining points. Moreover, whenever the average error rate in the previous block fell below 15%, participants were orally instructed to respond more quickly. The EEG was recorded only during the test session, and only the test trials from the 12 test blocks were analyzed (total: 1,344 trials per participant). The test session lasted for approximately 1 h, and participants could take short breaks between blocks.
Psychophysiological recording
The electroencephalogram (EEG) was recorded during the test session using a BioSemi Active-Two system (BioSemi, Amsterdam, The Netherlands) with 64 Ag-AgCl electrodes (Fp1, AF7, AF3, F1, F3, F5, F7, FT7, FC5, FC3, FC1, C1, C3, C5, T7, TP7, CP5, CP3, CP1, P1, P3, P5, P7, P9, PO7, PO3, O1, Iz, Oz, POz, Pz, CPz, Fpz, Fp2, AF8, AF4, AFz, Fz, F2, F4, F6, F8, FT8, FC6, FC4, FC2, FCz, Cz, C2, C4, C6, T8, TP8, CP6, CP4, CP2, P2, P4, P6, P8, P10, PO8, PO4, O2, left and right mastoid). The common mode sense and driven right leg electrodes were used as reference and ground electrodes. The vertical and horizontal electrooculogram (EOG) was recorded from electrodes above and below the right eye and on the outer canthi of both eyes. EEG and EOG data were continuously recorded at a sampling rate of 1,024 Hz. EEG data were off-line rereferenced to averaged mastoids, resampled to 512 Hz, and filtered with a 0.5–30 Hz bandpass filter.
Data analyses
Behavioral data
Errors on go trials were classified as distractor errors if the error response corresponded to the response assigned to the distractor. Otherwise, errors were classified as nondistractor errors. A missed catch trial was registered if any other key than the space bar was pressed on a catch trial. RT was defined as the time interval between the onset of the stimulus and the subsequent key press. Trials on which RT was more than four standard deviations above or below the condition mean were excluded from the RT analyses to control for outliers (0.409% of trials, SE = 0.088%). Frequency data were arcsine-transformed before statistical testing (Winer et al., 1991). ANOVAs with repeated measurement and planned comparisons by t tests for dependent samples were used for statistical analyses.
Event-related potentials
ERP data were analyzed using custom routines in MATLAB R2022a (The Mathworks) as well as the EEGLAB v2022.1 (Delorme and Makeig, 2004) toolbox. Epochs of 1,500 ms before and 1,500 ms after the response were extracted from continuous EEG traces. The average voltage in an interval from 150 to 50 ms before the key press served as the baseline. This was done because error-related brain activity usually starts to build up already before the response. A channel was interpolated using spherical spline interpolation whenever it met the joint probability criterion (threshold 5) or the kurtosis criterion (threshold 10) in EEGLAB's pop_rejchan function. An average of 0.143 (SE = 0.085) electrodes per participant were interpolated in this way. Epochs were excluded if (1) the voltage on a channel deviated >300 μV from the baseline at any electrode except channels Fpz, Fp1, Fp2, AF3, AF4, AF7, and AF8 (as eye blinks were corrected in a later stage), (2) joint probability deviated >5 standard deviations from the epoch mean (pop_jointprob function), or (3) kurtosis deviated >5 standard deviations from the epoch activity (pop_rejkurt function). These methods led to the exclusion of 9.56% (SE = 0.534%) of all trials. Furthermore, epochs were excluded if a second response occurred (e.g., a spontaneous error correction). This was done because preparation of a second response can distort ERP waveforms in the postresponse period. This resulted in the exclusion of 8.87% (SE = 2.32%) of distractor error trials, 10.0% (SE = 2.18%) of nondistractor error trials, and <1% of correct trials. A mean of 1.39% (SE = 0.546%) of trials was excluded in this way. The mean trial numbers entering ERP analyses in each condition are provided in Table 1. ERP data were then subjected to a temporal independent component analysis (Makeig et al., 1996) using the infomax algorithm (Bell and Sejnowski, 1995). The resulting component matrix was screened for independent components (ICs) representing stereotyped artifact activity, such as blinks, eye movements, and muscle artifacts. This was done using a multistep correlational template-matching process using the CORRMAP plugin v1.02 for EEGLAB (Viola et al., 2009). Topographies of ICs labeled as artifacts by the CORRMAP procedure were visually inspected and then removed from the data using inverse matrix multiplication. Epochs were then averaged separately for each participant and each condition. The Ne/ERN was quantified as a peak-centered mean difference amplitude at a frontocentral cluster of electrodes (Fz, FCz, Cz, FC1, FC2). To this end, for each participant, (1) the latency of the maximal difference between each error type and correct trials at the electrode cluster in a time window of −20 to 100 ms relative to the response was identified and (2) the mean difference between each error type and correct trials was calculated at the electrode cluster in a time window of 60 ms centered around this peak. As for the behavioral data, ANOVAs with repeated measurement and planned comparisons by t tests for dependent samples were used for statistical analyses.
Trial numbers for response-locked ERP data and for stimulus-locked time–frequency data
Time–frequency analysis
Time–frequency analysis of total power was conducted using the FieldTrip open-source toolbox (Oostenveld et al., 2011). Epochs of 1,500 ms before and 1,500 ms after stimulus onset were extracted from continuous EEG traces. The same preprocessing steps as for the ERP data were applied, except that the average voltage in an interval from 400 to 100 ms before stimulus onset was used as baseline. Trial exclusion criteria led to the exclusion of 10.0% (SE = 0.625%) of all trials. Additionally, to exclude effects from horizontal eye movements on lateralized alpha power, stimulus-locked epochs were excluded if the voltage on a horizontal EOG channel deviated >150 μV from the condition mean. This resulted in an additional 1.39% (0.546%) of excluded stimulus-locked epochs. The mean number of trials entering time frequency analyses is provided in Table 1.
The stimulus-locked EEG data were subjected to a Morlet wavelet analysis (width = 6 cycles, gwidth = 3), baseline-corrected with a dB-baseline ranging from 400 to 100 ms before stimulus onset, and averaged separately for each participant and each condition. To investigate whether directing spatial attention to the distractor was an error source, (1) asymmetry in brain oscillations was calculated by subtracting ipsilateral electrodes from contralateral electrodes in each condition, and (2) the resulting asymmetry on correct trials was subtracted from the asymmetry on error trials. Positive values of this index reflect an error-related alpha asymmetry reversal indicating that spatial attention is directed more to the side of the distractor than to the side of the target on errors than on correct trials. To investigate whether this was indeed the case immediately after stimulus presentation, a cluster-based permutation test with 20,000 iterations and a cluster-inclusion threshold of p = 0.010 (Maris and Oostenveld, 2007) on this index was performed in the alpha frequency range from 8 to 12 Hz in a time interval of 400 ms following stimulus onset at a parieto-occipital electrode cluster (P7, P9, PO7, O1, P8, P10, PO8, O2), where alpha activity is typically maximal (Foxe et al., 1998; Worden et al., 2000; Sauseng et al., 2005). To investigate whether the alpha asymmetry reversal in the time–frequency window revealed by the cluster-based permutation test varied as a function of error type and target side, we computed an ANOVA with repeated measurement and planned comparisons by t tests for dependent samples.
Single-trial analyses
To examine the relation between alpha asymmetry and the Ne/ERN at the single-trial level, we used single-trial within-subjects regression analysis. Single-trial alpha asymmetry was quantified as the mean power at the posterior electrode cluster contralateral minus ipsilateral to the target on each trial in the time–frequency window revealed by the permutation test. To obtain stable single-trial Ne/ERN amplitudes, response-locked EEG data were additionally filtered with a 15 Hz low-pass filter before extraction (Debener et al., 2005). Then, the same peak-centered mean difference amplitude measure as for the response-locked averages was calculated for each single error trial at the frontocentral electrode cluster. Separately for target left trials and target right trials, the mean of the waveform of correct trials at the cluster was used as a baseline and subtracted from the single-trial error waveform at the cluster for distractor errors and nondistractor errors. To clean single-trial data from outliers due to noisy single-trial EEG data, trials were excluded if the Mahalanobis distance of Ne/ERN amplitudes or alpha asymmetry scores were >2.5 standard deviations above the condition median. For the precursor analysis, a mean of 14.8% (SE = 0.456%) and for the adaptation analysis, a mean of 12.5% (SE = 0.388%) of trials were excluded in this way. Mean trial numbers entering the regression analysis are provided in Table 2.
Trial numbers for the single-trial analysis
In a “precursor analysis,” we investigated whether the extent to which spatial attention was directed to the distractor on an error trial was used as an error precursor to scale error monitoring activity on this trial. To this end, we tested whether the single-trial alpha asymmetry in the time–frequency window revealed by the permutation test predicted the single-trial Ne/ERN amplitude for errors by fitting the following linear regression model to the data from each participant:
In an “adaptation analysis,” we investigated whether error monitoring activity on an error trial triggered attentional adjustments on the next trial, and we tested whether the single-trial Ne/ERN amplitude predicted the alpha asymmetry on the next trial by fitting the following linear regression model to the data of each participant:
For both the precursor and the adaptation analysis, all continuous variables were z-scored within subjects (i.e., using subject-specific across-trial mean and standard deviations) before fitting the respective model. Single-trial Ne/ERN amplitudes were multiplied by −1 such that larger values represent larger Ne/ERN amplitudes. In the precursor analysis, a positive beta coefficient for the variable Alpha Asymmetry represents larger Ne/ERN amplitudes for larger alpha asymmetry on the same trial. In the adaptation analysis, a negative beta coefficient for the variable Ne/ERN represents smaller alpha asymmetry for larger Ne/ERN amplitudes on the preceding trial. To test the significance of all effects in the model, within-subjects regression coefficients were tested against zero at the sample level.
To visualize the relation between single-trial alpha asymmetry and single-trial Ne/ERN amplitudes in the precursor analyses, we sorted trials according to alpha asymmetry separately for each participant, grouped them into 20 equal-sized bins, and calculated mean alpha asymmetry and mean Ne/ERN amplitude for each bin (Murphy et al., 2016). Then, we plotted single-trial Ne/ERN amplitudes against alpha asymmetry for each bin. The same was done for the adaptation analysis, using previous Ne/ERN amplitudes instead of alpha asymmetry and alpha asymmetry on correct trials instead of Ne/ERN amplitudes. Note that the dependent variables on the y-axes were residualized by partialling out all effects but those that involved the independent variable on the x-axes from the respective model.
Results
Behavioral data
Mean RTs and mean error rates per condition are provided in Table 3. For catch trials, the mean RT was 602 ms (SE = 9.68 ms), and the mean error rate was 5.69% (SE = 0.615%). This demonstrates that participants successfully executed catch trial responses, and therefore, that participants processed both the targets and distractors.
Behavioral data response times
RTs from go trials were subjected to a repeated-measurement ANOVA with the variables Target Side (left, right) and Trial Type (correct, distractor error, nondistractor error). It revealed a main effect of Target Side, F(1,27) = 6.34, p = 0.018,
Error rates were not significantly different for target left and target right trials, t(27) = 0.176, p = 0.862,
The Ne/ERN is sensitive for error type
To investigate the Ne/ERN, we examined response-locked ERP waveforms averaged over the frontocentral electrode cluster (Fig. 2). The Ne/ERN is visible as a negative deflection starting shortly before the response and lasting until ∼150 ms after the response. Ne/ERN amplitudes were subjected to a repeated measures ANOVA with the variables Target Side (left, right) and Error Type (distractor error, nondistractor error). The ANOVA revealed a main effect of Error Type, F(1,27) = 7.92, p = 0.009,
Error-related brain activity: A, Grand-average response-locked ERP waveforms for correct trials, distractor errors, and nondistractor errors. B, Error minus correct difference waveforms and topographies of the peak-centered mean error minus correct difference. All waveforms are shown averaged over a frontocentral cluster of electrodes (Fz, FCz, Cz, FC1, and FC2). Shaded areas represent within-subjects standard errors (Cousineau, 2005). DE, distractor errors; NDE nondistractor errors; R, time point of the response; ms, milliseconds; μV, microvolt.
Error precursors are reflected in alpha asymmetry
Because responding to the distractor always leads to an error, misallocating spatial attention to the distractor could be a viable error precursor in our paradigm. To test this idea, we first investigated whether posterior alpha asymmetry as an index of attentional allocation predicts errors. To this end, we analyzed stimulus-locked alpha asymmetry calculated as contralateral minus ipsilateral activity in the alpha band at posterior electrodes. Because alpha power is suppressed in the hemisphere contralateral to attended locations (Worden et al., 2000; Bacigalupo and Luck, 2019), positive values of this index reflect that spatial attention is directed to the distractor, i.e., an alpha asymmetry reversal. Figure 3A reveals an alpha asymmetry reversal for error trials in the alpha frequency range (8–12 Hz) in the first 400 ms after stimulus onset, but not for correct trials. Note that on correct trials, successful allocation of spatial attention to the target should lead to alpha suppression contralateral to the target. Such an effect is not visible in our data (Fig. 3A). This could be because the distractor was larger than the target increasing the saliency of the distractor relative to the target. This could have led to an overall decrease of alpha suppression contralateral to the target as has previously been found (Wöstmann et al., 2019; Gutteling et al., 2022). A cluster-based permutation test on the error minus the correct difference in alpha asymmetry (Fig. 3B) in a time window of 0–400 ms and a frequency range of 8–12 hz revealed a positive cluster, which was located at 44–216 ms following stimulus onset, and spanned the frequencies from 8 to 10 hz (p = 0.032; see the highlighted region in Fig. 3B, left panel). The topography in Figure 3B shows the spatial distribution of this cluster separately for target left trials (left side) and target right trials (right side). A clear lateral posterior distribution of the cluster is visible only for target right trials. We therefore included the variable Target Side in all further analysis.
Alpha asymmetry in oscillatory brain activity: A, Stimulus-locked oscillatory brain activity for contralateral electrodes minus ipsilateral electrodes for error and correct trials at a posterior electrode cluster (P7, P9, PO7, O1, P8, P10, PO8, O2). B, Error minus correct difference and topography of error minus correct difference. The black contour indicates the time–frequency range of the significant cluster revealed by the permutation test (see text for details). Topography shows oscillatory brain activity contralateral minus ipsilateral to the side of the target within the time–frequency range of the significant cluster for target left on the left side and for target right on the right side. Highlighted electrodes were used for statistical analyses (see text for details). S, time point of stimulus onset; Hz, Hertz; ms, milliseconds; dB, decibel.
Because a misallocation of spatial attention should be an error precursor particularly for distractor errors, we next subjected the error-related alpha asymmetry to a repeated-measurement ANOVA with the variables Target Side (left, right) and Error Type (distractor error, nondistractor error). The ANOVA revealed a trend toward an interaction of Target Side and Error Type, F(1,27) = 3.55, p = 0.071,
Alpha asymmetry as a function of error type and target side: A, Error minus correct grand-average stimulus-locked oscillatory brain activity for contralateral electrodes minus ipsilateral electrodes for distractor errors and nondistractor errors with target left and target right at a posterior electrode cluster (P7, P9, PO7, O1, P8, P10, PO8, O2). B, Mean alpha asymmetry in the time–frequency window revealed by the cluster-based permutation test. S, time point of the stimulus; Hz, Hertz; ms, milliseconds; dB, decibel, error bars represent standard errors of the means.
Single-trial alpha asymmetry predicts the Ne/ERN amplitude on the same trial
Next, we investigated whether the alpha asymmetry immediately after stimulus presentation on an error trial predicted the Ne/ERN that occurred after the response on the same trial using multiple within-subjects regression of single-trial Ne/ERN amplitudes on single-trial alpha asymmetry in the cluster revealed by the permutation test (Table 4 for regression coefficients and test statistics). Ne/ERN amplitudes were multiplied by −1 such that larger values represent larger Ne/ERN amplitudes and positive
β values represent a positive relationship of the respective predictor variable to the Ne/ERN. The linear regression model (Eq. 1) revealed an effect of alpha asymmetry,
Relation between alpha asymmetry and Ne/ERN amplitude on the same error trial. Trials were sorted according to the magnitude of alpha asymmetry within subjects and binned into 20 equal-sized bins. Then, alpha asymmetry and corresponding residualized Ne/ERN amplitudes were averaged for each bin. The data were then averaged across subjects within each bin and alpha asymmetry bins were plotted against corresponding Ne/ERN bins. Alpha asymmetry and Ne/ERN amplitudes were z-scored before binning. More positive values of the Ne/ERN represent larger Ne/ERN amplitudes on a trial, because single-trial Ne/ERN amplitudes were multiplied by −1. See text for details. Hz, Hertz; z, z-scores.
Beta coefficients resulting from within-subjects multiple linear regression single-trial Ne/ERN amplitude multiplied by −1 as a dependent variable
It is reasonable to assume that accidental allocation of spatial attention to the distractor occurs also on some trials with correct responses. To investigate if the presence of an alpha asymmetry reversal correlated with an Ne/ERN-like signal on correct trials, we conducted the same single-trial within-subjects regression analysis as above but with Ne/ERN-like amplitudes on correct trials as the outcome variable. Single-trial Ne/ERN-like amplitudes on correct trials were quantified using the same method as on error trials, i.e., by subtracting the mean waveform of all correct trials from the single-trial waveform on each correct trial. The model revealed an interaction of Alpha Asymmetry and Target Side,
Single-trial Ne/ERN amplitude predicts alpha asymmetry on the next trial
Finally, we examined whether the Ne/ERN amplitude on an error trial predicted the alpha asymmetry in the cluster revealed by the permutation test on the next trial (Table 5 for regression coefficients and test statistics). Ne/ERN amplitudes were multiplied by −1 such that larger values represent larger Ne/ERN amplitudes and negative
β values represent a negative relationship of the respective predictor variable to the Ne/ERN. The linear regression model (Eq. 2) revealed a main effect of RT,
Relation between Ne/ERN amplitudes on the previous trial and alpha asymmetry. Trials were sorted according to the magnitude of Ne/ERN amplitudes within-subjects and binned into 20 equal-sized bins. Then, Ne/ERN amplitudes and corresponding residualized alpha asymmetry were averaged for each bin. The data were then averaged across subjects within each bin and Ne/ERN amplitude bins were plotted against corresponding alpha asymmetry bins. Alpha asymmetry and Ne/ERN amplitudes were z-scored before binning. More positive values of the Ne/ERN represent larger Ne/ERN amplitudes on a trial, because single-trial Ne/ERN amplitudes were multiplied by −1. See text for details. Hz, Hertz; z, z-scores.
Beta coefficients resulting from within-subjects multiple linear regression single-trial Ne/ERN amplitude multiplied by −1 as a dependent variable
Discussion
In the present study, we show that error precursors related to the source of errors are reflected in early error-related brain activity. In our paradigm, participants had to respond to a target on one side while ignoring a distractor on the other side of central fixation. Because the distractor is always associated with an incorrect response, a reversed posterior alpha asymmetry indicating the allocation of spatial attention to the distractor is predictive of errors and can therefore serve as an error precursor. Accordingly, errors were accompanied by a reversed posterior alpha asymmetry shortly following stimulus onset. Crucially, on error trials, the strength of this error precursor predicted the size of the Ne/ERN. This single-trial relationship between an error precursor and an early error signal points to a possible mechanism that selects the type and strength of adaptive posterror adjustment. As discussed previously (Maier et al., 2011; Maier and Steinhauser, 2013), we propose that the error monitoring system registers error precursors that reflect specific error sources and, in case of an error, selects a corresponding adjustment that counteracts the error source. In this process, the Ne/ERN is scaled according to the strength of the error precursor and therefore signals the need for behavioral adjustment.
This interpretation receives support from the finding that the Ne/ERN in turn predicted the strength of subsequent adjustments to attention. The larger the Ne/ERN was on an error trial, the smaller the reversed alpha asymmetry on the subsequent trial indicating that the nonoptimal attentional state leading to the error was reduced on the next trial. Interestingly, this effect occurred only for distractor errors with the target on the right side (Fig. 6). This was the only condition in our experiment, where alpha asymmetry was significantly reversed as compared with correct trials (Fig. 4) indicating a shift of spatial attention to the distractor on the left. This could result from a general leftward bias in spatial attention (Benwell et al., 2014), which is often observed in studies on visual search (Foulsham et al., 2013; Nuthmann and Clark, 2023). Thus, Ne/ERN-driven attentional adjustments strongly depended on the specific attentional failure that caused the error.
The present results demonstrate that internal states such as attention can serve as readout variables for evaluative processes. This adds to corresponding findings in the literature on metacognition. In covert spatial attention paradigms, cue-induced alpha suppression correlated with metacognitive judgments on decision confidence (Trajkovic et al., 2023, 2024) and attentional focus (Whitmarsh et al., 2014; see also MacDonald et al., 2011) indicating that attentional states are accessible to metacognitive awareness. Moreover, beta power related to internal representations of self-produced time intervals predicted mediocentral alpha oscillations associated with subjective accuracy judgments, suggesting an active readout mechanism where higher-order metacognitive processes access lower-order task-related processes (Kononowicz and Van Wassenhove, 2019). This concept was formalized in computational models that conceptualize metacognition as a second-order inference about one's own performance (Fleming and Daw, 2017). The present finding that alpha asymmetry reversal can serve as an error precursor predicting the Ne/ERN suggests that this readout architecture may apply not only to later, deliberate metacognition but also to early error monitoring mechanisms. Previous studies showed that the Ne/ERN fails to distinguish between distractor errors and nondistractor errors if task processing must be executed under a high working memory load (Maier and Steinhauser, 2017). This suggests that scaling the Ne/ERN according to error precursors requires processing resources and therefore supports the idea of an active readout mechanism for linking error sources to the early error signal.
Prominent theories have linked the Ne/ERN to posterror response conflict between the executed erroneous response and an upcoming tendency to correct errors (Botvinick et al., 2001; Yeung et al., 2004) or to prediction errors arising from the unexpectedness of errors (Holroyd and Coles, 2002; Alexander and Brown, 2011). Contrary to our findings, both frameworks would predict smaller Ne/ERN amplitudes for errors with stronger alpha asymmetry reversal. According to the conflict monitoring hypothesis, allocation of attention to the distractor should lead to a weaker corrective tendency and thus to lower posterror response conflict and smaller Ne/ERN. According to expectancy-based accounts, a stronger error precursor before the response should lead to higher expectedness of the error and therefore smaller Ne/ERN amplitudes for errors with stronger alpha asymmetry reversal. As we observed the opposite, it is clear that neither postresponse conflict nor error expectancy alone determined the size of the Ne/ERN. Rather, the Ne/ERN reflected the amount of attentional failure before the response and therefore an error signal that is scaled according to the strength of the error precursor.
The idea that the Ne/ERN is not only sensitive to postresponse conflict or error expectancy but instead reflects error precursors is supported by several existing studies. For instance, despite showing smaller Ne/ERN amplitudes, nondistractor errors are corrected more often and are better consciously detectable than distractor errors (Maier et al., 2008), which is viewed as indicative of higher postresponse conflict (Yeung et al., 2004). Experimentally increasing the expectancy of distractor errors further increases the Ne/ERN for distractor errors (Maier et al., 2012; Maier and Steinhauser, 2016) contrary to the prediction by expectancy-based accounts of smaller Ne/ERN amplitudes with higher error expectancy. Instead, distractor errors were followed by stronger posterror reduction of interference than nondistractor errors, and this effect was related to the Ne/ERN amplitude on a single-trial level (Maier et al., 2011). Finally, distractor errors triggered stronger pupil dilation than nondistractor errors suggesting error-source–specific mobilization of resources for adaptive posterror adjustments (Maier et al., 2019). In sum, these findings support the idea that the Ne/ERN is increased and signals a high need for posterror adjustments under conditions where internal states indicate nonoptimal processing. In contrast, postresponse conflict and expectancy-based accounts would paradoxically predict smaller Ne/ERN, and thus a lower need for behavioral adjustments when posterror conflict is low or error expectancy is high and thus under conditions where processing is nonoptimal.
It is worth noting that alpha asymmetry predicted the size of the Ne/ERN exclusively on error trials, but not the size of an Ne/ERN-like negativity on correct trials. As the allocation of spatial attention to the distractor and hence alpha asymmetry is likely to fluctuate also on correct trials, this implies that the process of scaling the MFC signal according to maladaptive internal states is triggered only if an error is actually detected. As previously noted (Maier et al., 2011; Maier and Steinhauser, 2017), such fast error detection upon response execution could be achieved by postresponse conflict monitoring (Yeung et al., 2004), a prediction error (Holroyd and Coles, 2002; Alexander and Brown, 2011), or any other process capable of instantly distinguishing an error from a correct response.
The finding that error precursors in alpha asymmetry predict the size of the Ne/ERN expands our understanding of the role of posterior alpha oscillations in the brain. Alpha suppression contralateral to attended stimuli is well-documented as an index of attention shifts in spatial attention tasks using both endogenous (Klimesch et al., 1998; Worden et al., 2000; Sauseng et al., 2005) and exogenous methods (Feng et al., 2017; Keefe and Störmer, 2021; Arana et al., 2022). Although there is a debate about whether this alpha suppression represents an active mechanism of attentional allocation (Thut et al., 2006; Bacigalupo and Luck, 2019) or rather a passive consequence of spatial attention shifts (Antonov et al., 2020), our results show that alpha desynchronization in posterior perceptual areas serves as a proxy for the frontomedial error monitoring system to initiate adaptive behavioral posterror adjustments. This underscores the pervasive role of alpha oscillations in the modular structure of the brain.
In summary, we have demonstrated that stimulus-related alpha oscillations indicating attentional failures can predict the Ne/ERN thus informing early error monitoring mechanisms to initiate attentional adjustments which reduce future attentional failures. This indicates that error monitoring utilizes error precursors like the state of the attentional system to prioritize those errors that are crucial for successful goal-oriented behavior and initiate appropriate adjustments to counteract the error source. In this way, our findings extend the role of internal state read-outs for later metacognitive judgments to the domain of early error monitoring processes. This supports the idea of a modular architecture of behavior monitoring and evaluation, where multiple processes with similar underlying principles contribute to behavioral control. This is in line with a predictive coding framework, wherein behavior results from minimizing competing prediction errors that arise by comparing sensory states to internal models of the world (Friston and Kiebel, 2009; Friston, 2010, 2018).
Footnotes
The authors declare no competing financial interests.
- Correspondence should be addressed to Martin E. Maier at martin.maier{at}ku.de.
