Motor expertise reshapes the representational structure of graded error monitoring in midfrontal theta

Motor expertise reshapes the representational structure of graded error monitoring in midfrontal theta | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL This is a preprint and has not been peer reviewed. Data may be preliminary. 12 December 2025 V2 Latest version Share on Motor expertise reshapes the representational structure of graded error monitoring in midfrontal theta Authors : Yiheng Chen 0009-0006-7467-1677 , Guanghui Zhang 0000-0003-0134-3614 , Yinyue Wang , Gi-Yeul Bae , and Yingzhi Lu [email protected] Authors Info & Affiliations https://doi.org/10.22541/au.176466104.46271507/v2 318 views 229 downloads Contents Abstract Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract Error monitoring has been widely characterized as a graded process that scales with the magnitude of prediction errors. Yet it remains unclear whether this rule is fixed or can be reshaped by expertise. Using a graded-deviation paradigm in table-tennis experts and novices, we examined how long-term training refines the neural coding of prediction–outcome discrepancies. Experts demonstrated higher accuracy and stability as deviation angles increased (from 0° to 30°). EEG analyses revealed that frontal-midline theta power scaled with deviation magnitude in both groups but exhibited distinct representational structures: experts showed a three-step graded organization, whereas novices displayed a binary pattern. Multivariate decoding further confirmed that only the frontal-midline region carried discriminable information about error magnitude, with decoding accuracy significantly higher in experts. These findings suggest that expertise refines the gain structure of error coding—transforming a general mismatch signal into a parametric mapping of prediction–outcome discrepancies. Frontal-midline theta thus provides a neural substrate for expertise-dependent tuning of predictive precision, demonstrating that graded error monitoring is not static but adaptively reorganized through motor expertise. 1. Introduction The ability to generate predictions about others’ actions and to monitor deviations between expected and actual outcomes is fundamental for adaptive behavior in dynamic environments (Moreau et al., 2020). Within the predictive coding framework, action prediction and error monitoring are tightly coupled through the continuous comparison of internal models with incoming sensory evidence. Empirical studies show that expert athletes not only outperform novices in anticipating others’ actions (Calvo-Merino et al., 2005; Cross et al., 2006; Wang et al., 2019; Zhao et al., 2018) but also display stronger neural differentiation when outcomes deviate from expectation (Amoruso et al., 2014; Panasiti et al., 2016; Proverbio et al., 2012) s, with response amplitudes scaling to the magnitude of the mismatch between predicted and observed events (Wang et al., 2024). These findings suggest that motor expertise sharpens the brain’s sensitivity to prediction errors. However, the computational principles governing this process remain unclear, specifically, how motor expertise shapes the neural encoding of prediction errors, as formalized in models of predicted outcome monitoring (Alexander & Brown, 2011, 2017). Within this theoretical context, motor expertise offers a powerful framework for examining how predictive precision modulates error coding. The Predicted Response Outcome (PRO) model posits that the error-monitoring system continuously compares predicted and actual outcomes, with neural activity scaling as a function of their discrepancy (Alexander & Brown, 2011, 2017). From the perspective of predictive coding and internal models (K. J. Friston et al., 2010; Kawato, 1999; Wolpert et al., 1995), expertise enhances the precision of outcome predictions through repeated sensorimotor calibration, thereby reducing uncertainty in the expected sensory consequences of others’ actions. This refinement of internal models increases the weighting assigned to prediction errors, leading to stronger differentiation of outcome discrepancies. Empirically, experts show reduced prediction uncertainty in behavioral tasks (Mann et al., 2010) and heightened modulation of error-related neural activity (Amoruso et al., 2014; Lu et al., 2020; Panasiti et al., 2016; Wang et al., 2024). Consequently, enhanced predictive precision amplifies the neural representation of even subtle deviations, whereas novices, constrained by noisier internal models, may only register larger or more salient mismatches. The influence of motor expertise on error monitoring is reflected in both regional activation patterns and the modulation of neural oscillations. Regional activation within the error-monitoring network, particularly the anterior cingulate cortex (ACC), medial prefrontal cortex (mPFC), and inferior frontal gyrus, consistently emerge during the detection of outcome deviations (Holroyd & Coles, 2002; Ullsperger et al., 2014). Meanwhile, error-related neural oscillations provide a complementary view of this process, with frontal-midline theta activity reflecting dynamic communication within this network that supports the evaluation of negative outcomes and the recruitment of cognitive control (Cavanagh et al., 2009; Trujillo & Allen, 2007; van Driel et al., 2012). These neural responses can be interpreted as computations of prediction–outcome discrepancies, consistent with the PRO model. Specifically, the amplitude of frontal-midline theta activity has been shown to scale with the magnitude of the mismatch between predicted and actual outcomes (Pezzetta et al., 2018; Spinelli et al., 2018), suggesting that the error-monitoring system encodes graded deviations rather than merely signaling the presence of an error. This graded encoding may provide the computational basis for rapid behavioral adjustment, as estimating the degree of deviation is more adaptive and efficient than binary detection in dynamic, interactive contexts. Empirically, previous work indicates that experts show stronger engagement of the action-observation network and enhanced frontal-midline theta responses when inconsistencies arise during unfolding action sequences (Abreu et al., 2012; Lu et al., 2020; Wang et al., 2019; Zhao et al., 2021). Together, these findings suggest that theta oscillations may constitute a neural substrate through which motor expertise enhances predictive precision and facilitates the fine-grained monitoring of outcome deviations. Nevertheless, whether frontal-midline theta simply indexes the salience of errors or represents the mapping between expected and actual outcomes remains unresolved. Most prior work has examined binary outcomes (correct vs. error), leaving the computational role of theta in encoding continuous prediction–outcome discrepancies unclear. Within a graded error-monitoring framework, theta activity may not merely signal that an error has occurred but encode the representational structure of the discrepancy itself (Iwane et al., 2023). Notably, stronger theta power observed in experts during action observation ( Lu et al., 2020; Wang et al., 2019; Zhao et al., 2021) could reflect either increased error salience or finer-grained encoding of prediction errors. To resolve this ambiguity, we employed multivariate pattern analysis (MVPA) to determine whether frontal-midline theta carries information about error magnitude. Unlike conventional univariate analyses that quantify overall response amplitude, MVPA can reveal whether distributed theta patterns discriminate among varying degrees of mismatch (Li et al., 2022, 2023). Such evidence would not only clarify the computational role of theta oscillations but also specify how expertise refines the neural representation of prediction errors. Demonstrating such graded decoding in the frontal-midline region would indicate that theta oscillations represent the mapping between predicted and actual outcomes, thereby identifying a key neural substrate through which motor expertise sharpens predictive computations. For instance, in table-tennis serves, deviations in ball trajectory vary continuously, requiring players to generate fine-grained predictions to detect small errors in real time. Individuals lacking such experience form less precise expectations and therefore evaluate discrepancies in a more categorical manner. Although recent work has demonstrated graded neural responses to prediction errors in experts (Wang et al., 2024), the absence of novice comparison groups leaves open whether expertise truly modifies the computational rule of error monitoring. To address this gap, the present study investigates whether motor expertise alters the gradient of error coding during action prediction. Guided by the PRO model, we hypothesize that both experts and novices follow a graded error-monitoring rule, but that experts will exhibit a steeper linear gradient reflecting higher predictive precision. Alternatively, novices may rely on a more threshold-like, all-or-none monitoring strategy, whereas experts implement a fine-grained, graded monitoring strategy mediated by the frontal-midline theta network. 2. Materials and Methods 2.1 Participants A priori power analysis (G-Power; f = 0.25, power = 0.90, α = .05, based on detecting an interaction effect) indicated N ≈ 22 for a 2×7 repeated-measures design (Faul et al., 2007). The study involved 32 professional table tennis athletes and 32 novices. Sample size was consistent with prior ERP work on expertise–deviation effects (Wang et al., 2024). The demographic characteristics of the participants are summarized in Table 1. Table 1. Descriptive statistics of participants Sample size 32 32 Sex (male : female) 11:21 15:17 Age (year) 21.72 ± 2.98 21.36 ± 2.74 Years of training (year) 0 9.41 ± 3.83 Frequency (day/week) 0 2.69 ± 1.23 Session duration (hour) 0 2.28 ± 0.72 The eligibility criteria for the professional athletes were as follows: (a) a national level 2 or higher table tennis athlete qualification; (b) at least 10 years of professional training experience; (c) at least 4 hours of practice per week for the past 3 years. The 32 novices had no prior experience in playing table tennis. All participants were recruited from Shanghai University of Sport and Shanghai Jiao Tong University. All participants were right-handed with normal or corrected-to-normal vision, and had no history of mental or other major health disorders. Informed consent was obtained from all participants prior to the experiment, and they received 100 Chinese renminbi (the equivalent of US $15) compensation following their participation. This study was approved by the Ethics Committee of Shanghai University of Sport (No. 102772020RT111). 2.2 Stimuli The stimuli consisted of self-recorded serving videos. The server was a male athlete from the China Table Tennis College, using a right-handed, horizontal grip with a backhand serve. A Canon 5D Mark III SLR camera (resolution: 1280 × 720 pixels, 30 frames per second) was positioned 40 cm to the side of the receiver’s midline to capture a total of 21 video clips. The player was instructed to serve to 7 different target locations, Region 1-7 (the horizontal width of the table was divided into these 7 regions, as shown in Figure 1A). To create 7 versions of the video material with different deviations (Wang et al., 2024), the videos were processed using Adobe Premiere software with the following steps: (a) the videos were exported as continuous images at 20 frames per second (resolution: 640 × 360 pixels); (b) audio was removed; (c) the server’s face was pixelated to exclude facial cues; (d) using the “ball-racket contact” frame as the key frame, we extracted the 18 frames preceding contact (Clip A; kinematic cues, referred to as the ”real location”) and the subsequent ball-flight frames (Clip B, referred to as the ”presented location”) after omitting the first 6 post-contact frames; Clip B comprised 11 frames. These frames were stitched together to form a video consisting of 30 images. As shown in Figure 1B, the first 19 frames with the server’s body cues indicated that the ball’s real location was in Region 1, but the following 11 frames, showing the ball’s trajectory, indicated that the presented location was in Region 7, creating a 30° deviation. If the real location was in Region 1 and the presented location was in Region 3, this resulted in a 10° deviation, and so on, resulting in a total of 7 different deviations (Deviation 1-7: 0°, 5°, 10°, 15°, 20°, 25°, 30°). After reassembling the real and presented locations, a final set of 49 reassembled videos was created (Figure 1A, the middle panel). Fig. 1 — Presentation of the seven deviation levels and the experimental procedure. A) The real location is indicated by the kinematic cues from the server’s body, while the presented location is defined by the ball’s flight trajectory. By splicing these two video segments, seven levels of deviation were created (ranging from 0° to 30°). The middle panel illustrates the conditional matrix formed by combinations of real and presented locations, and the right panel shows the response keys corresponding to the seven presented locations. B) Trial structure. Clip A depicts the server’s body cues (real location), while Clip B shows the ball’s flight trajectory (presented location). Participants were instructed to mentally predict the landing point after viewing Clip A and then respond based on the final presented location during Clip B. The onset of Clip A was used as the EEG time zero. Theta oscillations during the Clip B period (t1–t7) were analyzed, using the inter-stimulus blank screen (1650-1950 ms) as the baseline. 2.3. Procedure The experiment aimed to explore how motor expertise influences the rules of error monitoring in action prediction. Each experimental trial began with a central fixation point displayed for 1600 ms, followed by Clip A showing the server’s body cues for 950 ms. Participants were instructed to watch the clip and form a mental prediction of the ball’s landing location, without making any button response. After a 1000 ms blank screen, Clip B displayed the ball’s trajectory in the air. Participants’ task was to press a key on a keypad according to the presented location in Clip B (Figure 1B). The original keypad’s number buttons “1,” “4,” “7,” “8,” “9,” “6,” and “3” corresponded to the 7 target locations, with labels placed on the buttons to avoid incorrect responses (Figure 1A, the right panel). To ensure that participants formed their prediction after watching Clip A, detection trials were included in the experiment. In these trials, the trial was terminated at Clip A, and participants were asked to respond immediately after watching Clip A with their predicted location. The experiment consisted of 5 blocks, each containing 70 experimental trials and 15 detection trials. Each deviation condition had 50 trials. 2.4 Data and statistical analysis 2.4.1 Behavioral analysis For behavioral analyses, accuracy (ACC) and reaction times (RTs) were modeled at the single-trial level using linear mixed-effects models (LMMs) (D. Bates, Mächler, et al., 2015). Accuracy was defined as whether participants correctly pressed the key corresponding to the presented location, and reaction time was measured as the latency between the onset of Clip B and the participant’s keypress. RTs were analyzed on correct trials. Models were fitted using the lme4 package (v.1.1–34) in R, and significance of fixed effects was assessed via Satterthwaite’s approximation implemented in the lmerTest package (v.3.1–3; Kuznetsova et al., 2017). For each dependent variable, fixed effects included Group (Expert vs. Novice), Deviation (seven levels, 0°–30°), and their interaction. Novices and the deviation 1 (0°) served as reference levels. Both factors were modeled as repeated contrasts comparing each level to the reference. To account for inter-individual differences, random intercepts for participants were included. Maximal random-slopes structures were initially specified, and correlation parameters were set to zero or slopes explaining zero variance were omitted if necessary to achieve model convergence and avoid overparameterization (D. Bates, Kliegl, et al., 2015). To follow up significant interactions, we conducted simple effects analyses using the emmeans package (v.1.8.9; Lenth, 2020). Pairwise comparisons between Experts and Novices at each deviation level were computed, and p-values were adjusted using the false discovery rate (FDR) method. We report estimated marginal means (EMMs), odds ratios for ACC, or regression coefficients for RTs, standard errors, test statistics (t or z), and p-values. In addition to accuracy and reaction time, we also calculated response error (−6…+6) to quantify the precision of error monitoring in participants’ responses. Response error was defined as the signed distance between the participant’s chosen response location and the true landing location presented in the stimulus. Each trial required participants to predict the ball’s landing position by pressing one of seven keys corresponding to possible landing locations. The correct key (coded as 0) represented an accurate prediction, whereas responses to the left or right of the correct position were coded as negative or positive errors, respectively. Details are provided in Supplementary Material (Table S1, Figure S1). 2.4.2 EEG recording and preprocessing EEG data were recorded using the BrainAmp Standard system (Brain Products GmbH, Germany) with active electrodes (acticap) and processed using Brain Vision Recorder 2.0 (Brain Products GmbH, Germany). A total of 64 Ag/AgCl electrodes were placed according to the international 10-20 system, with a sampling rate of 1000 Hz. The FCz electrode was used as the online reference, and Fpz was the ground electrode. Vertical electrooculograms (VEOG) were recorded from an electrode placed 1 cm below the left eye, and horizontal electrooculograms (HEOG) from an electrode placed 1 cm to the right of the outer canthus. Before the experiment, the impedance between electrodes and the scalp was reduced to below 5 kΩ. Preprocessing of EEG data was conducted using the open-source toolbox Letswave 2017 (https://www.letswave.org/) in MATLAB R2014b. The preprocessing steps were as follows: 1) A notch filter was applied to remove 50 Hz line noise; 2) Data were filtered using a non-causal zero-phase Butterworth filter with a high-pass cutoff of 0.1 Hz and a low-pass cutoff of 30 Hz to avoid phase distortion; 3) The signals were re-referenced to the whole brain average, and the online reference was recovered to FCz; 4) Independent Component Analysis (ICA) was applied, and 2–3 components, which typically reflecting eye movements or muscle activity, were manually removed for each participant based on visual inspection; 5) Data were segmented with the onset of Clip A serving as time 0 ms. EEG epochs were extracted from –500 to 2500 ms, encompassing the pre-Clip A (–500 to 0 ms), Clip A (0–950 ms), the inter-clip interval (950–1950 ms), and Clip B (1950–2500 ms); 6) Extreme values were rejected as artifacts using a threshold of ±100 µV, resulting in 22,330 out of 22,400 trials being retained (retention rate: 99.7%); 7) The baseline correction window was set to 1650-1950 ms. This cleaned data were used for subsequent time-frequency analysis and MVPA. 2.4.3 Time-frequency analysis A continuous wavelet transform (CWT) was used for the time-frequency decomposition of the segmented EEG data (i.e., 1650 ms to 2500 ms, Clip B) for the frequency range between 1 and 30 Hz in MATLAB. A complex Morlet wavelet was selected as the mother wavelet, with a time decay parameter of 1.0 s and a center frequency of 1.5 Hz, following the recommendations for cognitive EEG analysis (Herrmann et al., 2014). These settings allowed for the time‐frequency localization of the analysis window at 1 Hz interval in frequency domain. The power spectrum of the CWT at each frequency was separately normalized with respect to its corresponding mean power during the baseline period (i.e., 1650 to 1950 ms), by which the transient changes in the power at certain frequencies could be observed. Specifically, we quantified power modulations of EEG rhythms using the event-related spectral perturbation (ERSP) method. The ERSP was computed for each time–frequency point according to: ERSP( t, f ) \(=\frac{A\left(t,\ f\right)\ -\ R(f)}{R(f)}\) × 100% Here, A(t,f) represents the single-trial spectral power at a given time point t and frequency f , and R(f) denotes the mean spectral power during the baseline period averaged across trials and time points within the baseline window. This normalization expresses the percentage change in oscillatory power relative to baseline activity, allowing comparisons across conditions and participants (Pfurtscheller & Lopes Da Silva, 1999). We selected frontal-midline regions (Fz, FCz, Cz) to detect theta oscillation activity in the 4–8 Hz band. Based on the temporal window for error monitoring emphasized in previous studies (Lu et al., 2020; Wang et al., 2024), which corresponds to the ball’s flight trajectory in Clip B, we divided the interval from 2050 to 2400 ms into seven consecutive 50-ms time windows (t1–t7). For each individual trial, we extracted the theta ERSP within each time window from the average of the frontal-midline region (the mean of the three electrode sites). Then, for each subject, we averaged the ERSP for each deviation condition across all trials. Finally, a 2 (Expertise) × 7 (Deviations) × 7 (Time windows) repeated-measures analysis of variance was conducted. The unit of ERSP magnitudes was expressed in ER%. 2.4.4 Clustering analysis To examine whether expertise influences error monitoring rules, we performed clustering analyses on the theta oscillations elicited at each deviation level in both groups. First, time–frequency analysis revealed that theta ERSP was most pronounced during the t4–t6 interval. For each subject group, we began by computing the mean θ‐ERSP at each of the seven deviation levels, resulting in a seven‐element vector of average responses. We then applied hierarchical agglomerative clustering to these vectors by first calculating all pairwise Euclidean distances between deviation‐level means. Ward’s minimum‐variance linkage criterion was used to guide the merging of clusters, producing a dendrogram that visually represents how deviation levels group together at each step (Ward, 1963; Bridges, 1966). The optimal number of clusters (K) was objectively determined using the silhouette index and the NbClust package, both applied within a range of K = 2-6. This procedure revealed distinct organizational structures across groups: experts exhibited a three-cluster solution (Deviation 1-5, 6, 7), whereas novices displayed a two-cluster solution (Deviation 1-6 vs. 7). To verify the robustness of these results, we also conducted K-means clustering on the same seven-dimensional mean theta vectors (Hartigan & Wong, 1979), specifying the number of clusters according to the optimal K obtained from the hierarchical analysis (i.e., K = 3 for experts and K = 2 for novices). The two clustering methods yielded highly consistent partitions, confirming that the seven deviation levels formed reproducible cluster structures regardless of the algorithm used. 2.4.5 Multivariate spectral decoding Given the greater sensitivity of multivariate approaches over univariate analyses, we implemented a multivariate time–frequency decoding procedure using the ERPLAB Studio toolboxes (Lopez-Calderon & Luck, 2014). Preprocessed EEG data were imported into ERPLAB Studio, and epochs were extracted from –500 ms to 3000 ms time-locked to the onset of Clip A (defined as time zero). Baseline correction was applied using the interval from 1650 to 1950 ms (i.e., 300 ms period before the onset of Clip B). Following preprocessing, EEG data were band-pass filtered at 4-8 Hz, and the analytic signal was obtained using the Hilbert transform. The instantaneous amplitude was squared to yield single-trial total power (including both evoked and induced components) rather than simply filtered voltage. For each participant, theta power was averaged within 10 ms sliding windows and concatenated across electrodes, forming a feature vector for each trial and time window. Each deviation condition (Deviation 1-7) contained 50 valid single trials after artifact rejection, ensuring equal trial counts across classes. Prior to classification, all features were z-scored within participant and time window to remove scale differences across channels. Decoding was implemented using a linear support vector machine (SVM) in a one-vs-all framework. For each subject, classification was performed separately at every time window across the entire scalp and for predefined regions of interest (ROIs; see below). To enhance generalizability, a 5-fold cross-validation scheme was applied, repeated 100 times with randomized trial assignments. Because trial counts were already balanced across the seven deviation conditions, no additional resampling was required. The chance level for classification was 1/7 ≈ 14.29%. Additionally, to specifically test whether motor expertise modulates error monitoring via midfrontal cortical mechanisms, we conducted region-specific analyses to assess the spatial distribution of predictive decoding (Bae, 2021). Classifications were conducted separately for the following regions of interest: frontal-midline (channels: FCz, Cz, Fz), anterior (channels: Fp1, Fp2, AF7, AF3, AFz, AF4, AF8, F7, F5, F3, F2, F4, F6, F8, FT7, FC5, FC1, FC2, FC4, FC6, FT8), posterior (channels: TP9, TP7, CP5, CP3, CP1, CPz, CP2, CP4, CP6, TP8, TP10, P7, P5, P3, P1, Pz, P2, P4, P6, P8), and occipital (channels: POz, PO3, PO7, PO4, PO8, O1, Oz, O2) (see Figure 5A for the spatial arrangement of electrodes). For each region, decoding results were saved and aggregated across participants. All other decoding parameters were identical to those used in the whole-scalp analysis, with only the channel selection varying by region. To evaluate statistical significance, decoding accuracy values were exported into group-level matrices and subjected to nonparametric cluster-based permutation tests (Maris & Oostenveld, 2007). For both within-group analyses (against the theoretical chance level of 1/7 ≈ 14.29%) and between-group comparisons (experts vs. novices), statistical significance was determined using two-tailed cluster-based permutation tests with 1000 iterations. This approach enabled the identification of significant temporal clusters of above-chance decoding performance while effectively controlling for multiple comparisons over time. To test the frequency specificity of this effect, we additionally performed identical decoding analyses in the alpha band (9–12 Hz); all parameters were the same as in the theta analysis. 3. Results 3.1 Behavioral results We fit a generalized linear mixed-effects model with Group (Expert, Novice) and Deviation (seven levels) as fixed effects and subjects as a random intercept, using Novice at Deviation 1 (0°) as the reference level. Given the Group × Deviation interaction, we report estimated marginal means (EMMs; emmeans) and simple-effects contrasts, and we provide the full fixed-effect coefficients in Table 2 for completeness. On the EMM scale, Experts were more accurate than Novices overall (Experts: M = .889, 95% CI [.869, .906]; Novices: M = .847, 95% CI [.821, .869]). The EMM contrast (Expert vs. Novice) was significant, odds ratio (OR) = 1.447, z = 2.716, p = .007. Critically, EMM-based simple effects confirmed a Group × Deviation modulation: Experts showed significantly higher accuracy than novices at larger deviations (Deviation 6, 25°: OR = 1.678, z = 3.208, p = .001; Deviation 7, 30°: OR = 2.062, z = 4.501, p .05). Note that the main-effect “Group” coefficient in Table 2 is not significant because it reflects the reference cell (Novice–Deviation 1) within an interaction-coded model; the aggregated EMMs and simple-effects contrasts more directly quantify the group difference across deviation levels. These results indicate that experts exhibited generally higher accuracy than novices, with group differences becoming particularly pronounced under conditions of large deviations. RTs were analyzed with a linear mixed-effects model. The overall EMMs indicated comparable RTs between groups (Experts: M = 974 ms, 95% CI [932, 1017]; Novices: M = 993 ms, 95% CI [950, 1035]; EMM contrast Expert–Novice: t = 0.606, p = .545). The Group × Deviation interaction was followed up with EMM-based comparisons. Between-group contrasts were not significant at individual deviations (all ps > .10). Within-group contrasts revealed distinct response patterns: Novices showed a significant RT increase from 25° ( M = 961 ms, SE = 22) to 30° ( M = 991 ms, SE = 22; EMM contrast 30°–25°: t = 4.309, p < .001), whereas Experts showed no difference between 25° ( M = 949 ms, SE = 22) and 30° ( M = 943 ms, SE = 22; EMM contrast 30°–25°: t = 0.969, p = .333). These findings indicate that although overall RTs were comparable between groups, novices showed a selective slowdown when facing fully inconsistent prediction-outcome events (i.e., when the ball trajectory in Clip B deviated completely from the landing position predicted based on Clip A), whereas experts maintained stable RTs across these conditions (Figure 2, Table 2). Table 2. Mixed model summary statistics for accuracy (GLMM) and reaction times (LMM), showing effects of motor expertise and deviation b (SE) z p b (SE) t p Intercept (Novice, Deviation 1, 0°) 1.647 (0.114) 14.48 < .001 994.96 (22.13) 44.96 < .001 Group (Expert vs. Novice) 0.243 (0.163) 1.49 .137 -18.73 (31.28) -0.60 .551 Deviation 2 (5°) 0.162 (0.097) 1.66 .097 -0.47 (6.81) -0.07 .945 Deviation 3 (10°) 0.548 (0.105) 5.23 < .001 8.68 (6.73) 1.29 .197 Deviation 4 (15°) 0.287 (0.100) 2.88 .004 -17.81 (6.78) -2.63 .009 Deviation 5 (20°) 0.147 (0.097) 1.51 .131 31.11 (6.81) 4.57 < .001 Deviation 6 (25°) -0.280 (0.091) -3.06 .002 -34.13 (6.95) -4.91 < .001 Deviation 7 (30°) -0.433 (0.090) -4.82 < .001 -3.53 (7.02) -0.50 .615 Group × Deviation 2 -0.074 (0.143) -0.52 .607 2.32 (9.55) 0.35 .727 Group × Deviation 3 0.086 (0.157) 0.55 .584 -8.46 (9.55) -0.88 .377 Group × Deviation 4 0.091 (0.149) 0.61 .540 -2.84 (9.49) -0.45 .651 Group × Deviation 5 0.023 (0.144) 0.16 .871 0.70 (9.46) 0.04 .299 Group × Deviation 6 0.275 (0.138) 2.00 .046 -7.05 (9.66) -0.73 .465 Group × Deviation 7 0.482 (0.137) 3.51 < .001 -30.04 (9.71) -3.09 .002 Model Formula ACC ~ Group × Deviation + (1 | Subject) RT ~ Group × Deviation + (1 | Subject) Fig. 2 — Behavioral Results. A) Accuracy (bars, left y-axis) and reaction time (lines, right y-axis) across deviation levels for Experts and Novices. Values are estimated marginal means from the GLMM (ACC) and LMM (RT); error bars denote ±SE; RTs are computed from correct trials only. B) Expert–Novice differences by deviation (Δ = Expert − Novice): bars show Δ Accuracy (left y-axis) and Δ RT (right y-axis; negative values indicate faster responses in Experts). Asterisks mark pairwise Expert–Novice contrasts that reached significance in emmeans (FDR-adjusted). * p < .05, ** p < .01, *** p < .001. 3.2 Time-frequency results To examine how motor expertise modulates the temporal dynamics of error-related neural oscillations, we analyzed the event-related spectral perturbation in the frontal-midline theta band (4-8 Hz), a 2 Expertise × 7 Deviations × 7 Time window repeated-measures ANOVA revealed the following results: 1) The main effect of Expertise was marginally significant, F (1, 62) = 3.544, p = .064, η 2 p = .054, with experts ( M = .281, SE = .024) showing greater theta oscillations than novices ( M = .217, SE = .024); 2) The main effect of Deviation was significant, F (6, 372) = 14.967, p < .001, η 2 p = .194; 3) The main effect of Time window was significant, F (6, 372) = 85.248, p < .001, η 2 p = .579, and there was no significant interaction between Expertise and Time window ( p = .200). Pairwise comparisons showed that, for both experts and novices, theta oscillations steadily increased from t1 to t6 and decreased at t7, with the peak observed between t4 and t6 (2200-2350 ms); 4) Critically, the interaction between Expertise and Deviation was significant, F (6, 372) = 4.466, p = .002, η 2 p = .067. Post hoc analysis showed that, among novices, only Deviation 1 differed from Deviation 7 (0° vs. 30°), with no significant differences between the other deviation levels, while experts showed a graded theta response with three levels for Deviations 1-5, 6, and 7 (0°-20°, 25°, 30°); 5) The interaction between Deviation and Time window was significant, F (36, 2232) = 10.764, p < .001, η 2 p = .148. Post hoc analysis indicated that the rise and fall trends for the 7 deviations across the 7 time windows were similar; 6) The three-way interaction was significant, F (36, 2232) = 2.746, p = .010, η 2 p = .042. Post hoc analysis revealed that for Deviation 1, experts were greater than novices from t1 to t4, while no differences were observed between the two groups from t5 to t7. For Deviations 2-5, no differences were observed between the groups at any time point. Importantly, for Deviations 6 and 7 (25°, 30°), experts exhibited greater theta oscillations from t1 to t7 compared to novices. These results suggest: 1) The time window for error monitoring was similar across both groups, peaking between 150-250 ms (2200-2350 ms) after the error occurred; 2) The error monitoring strategies of the two groups were different, with novices employing an all-or-none strategy, while experts used a fine-grained, gradual monitoring strategy (Figure 3). Fig. 3. — Univariate theta oscillation results. A) ERSP for deviation levels t1-t7 relative to the 1650–1950 ms baseline. B) Time courses of theta ERSP magnitude (t1–t7) for novices and experts, and their difference (expert − novice). Shaded areas represent ±1 SEM across participants. C) Heatmaps showing pairwise comparisons of theta ERSP magnitude (percentage change) across time points and deviation levels for each group and their difference (expert − novice). Diagonal cells represent mean theta values for each deviation level. * p < .05, ** p < .01, *** p < .001. 3.3 Clustering results 3.3.1 Hierarchical clustering results Based on the time–frequency results, we extracted the average ERSP values from the t4-t6 interval for clustering analysis. Hierarchical clustering of the seven mean‐theta values for each group revealed distinct patterns in novices versus experts. In novices, the dendrogram split the seven deviation levels into two clusters: deviations 1 through 6 formed a single cluster, while deviation 7 formed its own cluster. This suggests that, for novice participants, theta‐power responses at the first six graded deviations were highly similar, with a marked shift only at the most extreme deviation. In contrast, experts exhibited three well‐separated clusters. Deviations 1–5 (0°-20°) grouped together into one cluster, deviation 6 (25°) formed a second cluster, and deviation 7 (30°) formed a third cluster (Figure 4A). 3.3.2 K-means results K‐means clustering produced the same overall structure when constrained to the same number of clusters identified by the hierarchical analysis. For the novice group (K = 2), K‐means again assigned deviations 1–6 (0°-25°) to one cluster and deviation 7 (30°) to the other, confirming that novices’ theta responses were largely uniform until the largest deviation. In the expert group (K = 3), the K‐means solution closely matched the hierarchical clusters: deviations 1–5 (0°-20°) fell into one cluster, deviations 6 (25°) formed a second cluster, and deviation 7 (30°) remained alone in a third cluster. To formally assess whether the optimal number of clusters differed between groups, we conducted a validation analysis using the NbClust package (Charrad et al., 2014), which evaluates 30 clustering indices to determine the most supported solution. Consistent with the visual inspection, the majority of indices converged on a two-cluster solution for novices and a three-cluster solution for experts, confirming that the groups differed in the internal structure of theta-based responses (Figure 4B). Fig. 4 — Clustering results. A) Hierarchical clustering of mean theta values (t4–t6) across the seven deviation levels for novices and experts. Dendrograms illustrate similarity relationships among deviations, with color-coded heatmaps showing average theta magnitude within each cluster. Experts exhibited a three-cluster structure (0°-20°, 25°, and 30°), whereas novices showed a two-cluster structure (0°-25° and 30°). B) Mean theta magnitude as a function of deviation level and group. Cluster assignments (C1–C3) were determined using K-means analysis. 3.4 Multivariate spectral decoding results To explore the differences in sensitivity to deviations and error monitoring strategies between the two groups, we conducted the spectral MVPA without predefining electrode sites or time windows to examine possible changes in space, time, and frequency of error-related neural signatures. Specifically, the classifier was trained to distinguish among the seven deviation levels, which reflected the degree of mismatch between the predicted landing location (from Clip A) and the actual landing location (in Clip B). In the whole-scalp analysis, decoding accuracy exceeded the chance level (1/7) in both groups, with significant time windows ranging from 2120 to 2440 ms in the novice group ( p < .001, cluster-corrected) and 2140 to 2500 ms in the expert group ( p < .001). However, no significant between-group differences were observed, suggesting that both groups demonstrated sensitivity to deviation information at the global level, albeit without detectable differences in decoding magnitude (Figure 5B). In the anterior region, decoding accuracy was above chance from 2160 to 2360 ms in novices ( p < .001) and from 2160 to 2500 ms in experts ( p < .001). Similar to the whole-scalp results, no significant group differences emerged (Figure 5C). In contrast, the frontal-midline—a region commonly associated with predicted error monitoring (Holroyd & Coles, 2002; Ullsperger et al., 2014)—showed a distinct group-related modulation. Both groups exhibited above-chance decoding accuracy (novices: 2200–2320 ms, p < .001; experts: 2160–2500 ms, p < .001). Critically, a significant between-group difference was identified from 2320 to 2420 ms ( p = .033). This finding suggests that midfrontal theta activity may be a neural correlate of the superior error monitoring sensitivity observed in experts (Figure 5D). In the posterior region, both groups displayed strong above-chance decoding performance throughout the late time window (2140–2500 ms, p < .001), but no group differences reached significance. Similarly, in the occipital region, decoding accuracy was above chance in both groups (novices: 2160–2500 ms, p < .001; experts: 2120–2500 ms, p < .001), without significant between-group effects. Taken together, these findings highlight the frontal-midline as a key neural substrate for expertise-related enhancement in error monitoring (Figure 5E, F). Complementary alpha-band decoding did not reveal any significant expertise-related differences across regions, confirming that the midfrontal theta effect uniquely accounts for the expertise modulation (Figure 6). Fig. 5 — Multivariate spectral decoding results in the theta band. A) Panel shows the corresponding electrode selections on the scalp map, with colors matching those used in each decoding plot. B-F) Panels display time-resolved decoding accuracy for five electrode regions: whole-brain, anterior, frontal-midline, posterior, and occipital, respectively. Solid and dashed lines indicate expert and novice groups, respectively, with shaded bands representing ±1 SEM. Horizontal bars mark time windows with significant above-chance decoding within each group (cluster-corrected, p < .05). The dashed horizontal line denotes the chance level (1/7). The vertical dashed line at x = 1950 ms indicates the onset of Clip B, when the ball trajectory became visible. A significant between-group difference (expert > novice) was observed only in the frontal-midline region from 2320 to 2420 ms (panel B, p = .033). Fig. 6 — Multivariate spectral decoding results in the alpha band. A) Panel shows the corresponding electrode selections on the scalp map, with colors matching those used in each decoding plot. B-F) Panels display time-resolved decoding accuracy for five electrode regions: whole-brain, anterior, frontal-midline, posterior, and occipital, respectively. Solid and dashed lines indicate expert and novice groups, respectively, with shaded bands representing ±1 SEM. Horizontal bars mark time windows with significant above-chance decoding within each group (cluster-corrected, p < .05). The dashed horizontal line denotes the chance level (1/7). The vertical dashed line at x = 1950 ms indicates the onset of Clip B, when the ball trajectory became visible. No significant between-group differences were observed in any region. 4. Discussion The present study investigated how motor expertise reshapes the neural dynamics of error monitoring during action prediction, focusing on frontal-midline theta activity as a key mechanism supporting predictive evaluation. Using a graded-deviation paradigm that systematically varied trajectory congruency from fully consistent to fully inconsistent, we found that experts were more accurate than novices, with group differences becoming pronounced at larger mismatches. Univariate analyses of frontal-midline theta power revealed distinct clustering structures across expertise levels: hierarchical and k-means clustering consistently showed that experts exhibited a three-step graded profile (0°-20°, 25°, 30°), whereas novices displayed a near-binary split (0°-25° vs. 30°). Multivariate decoding further confirmed that only the frontal-midline region yielded a group difference in decoding accuracy (2320–2420 ms), indicating that frontal-midline theta carries more discriminative information about error magnitude in experts. Together, these findings identify frontal-midline theta oscillations as the neural substrate through which motor expertise refines graded error monitoring, transforming general mismatch signals into parametrically organized representations of error magnitude. At the behavioral level, both accuracy and reaction patterns indicated that experts were more effective than novices in handling prediction–outcome mismatches, particularly when deviations became large. This advantage suggests that long-term motor experience facilitates the rapid detection and correction of incongruencies between predicted and actual events (Cross et al., 2006; Kemmerer, 2021; Pierella et al., 2019). Accuracy differences emerged primarily at larger deviations, and experts maintained stable reaction times whereas novices exhibited additional slowing at the most extreme mismatch (deviation 7). This pattern implies that experts treat moderate and large discrepancies as comparably salient errors, preventing further delay, whereas novices require maximal mismatches to surpass their monitoring threshold. Such threshold-shift patterns align with the PRO model, which proposes that error monitoring is engaged only when prediction uncertainty exceeds a critical level (Arrighi et al., 2016; Jonker et al., 2021; Silvetti et al., 2013). High-precision priors in experts enable early, efficient detection, while low-precision priors in novices delay the recruitment of corrective mechanisms. The reaction-time peak observed at deviation 5 likely reflects the spatial configuration of response keys rather than genuine cognitive difficulty, as the cross-side (180°) keypress mapping introduces motoric interference independent of monitoring (Rosenbaum, 1980; Proctor & Vu, 2006). At the neural level, frontal-midline theta oscillations provided converging evidence for expertise-dependent differences in error monitoring (Lu et al., 2020; Wang et al., 2024; Zhao et al., 2021). Across all participants, theta power increased systematically with deviation magnitude, peaking around 2200–2350 ms following the onset of outcome presentation, consistent with the canonical time window for error evaluation reported in previous work (Cohen & Donner, 2013; Levy et al., 2023; Töllner et al., 2017). Importantly, experts exhibited overall stronger theta modulation than novices, suggesting that motor expertise enhances the dynamic range of neural responses to prediction–outcome discrepancies. To further characterize how these neural responses were internally organized across deviation levels, we conducted clustering analyses on the frontal-midline theta power. The results revealed distinct grouping structures between the two groups: experts showed a three-step graded profile (0°-20°, 25°, 30°), whereas novices displayed a near-binary split (0°-25° vs. 30°). Such graded encoding accords with previous findings that the brain continuously scales its evaluative responses to the size of prediction errors (Anguera et al., 2009; De Bruijn et al., 2003; Omedes et al., 2015; Spinelli et al., 2018; Vocat et al., 2011) and with predictive coding accounts suggesting that expertise reduces internal noise and enhances the precision weighting of error signals (Ludolph et al., 2017; McNamee and Wolpert, 2019; Oh and Schweighofer, 2019). This graded clustering pattern revealed that experts differentiated outcome levels with finer granularity, particularly for larger deviations, whereas novices showed a more binary distinction between correct and incorrect outcomes. This pattern suggests that expertise does not simply enhance overall sensitivity but reorganizes how prediction–outcome discrepancies are internally categorized (Chen et al., 2020; Musco et al., 2023). Experts appear to compress minor deviations into a single representational category, reducing computational noise and preserving efficiency, while expanding the representational distance for salient mismatches that demand corrective adjustment (Feldman & Friston, 2010; Friston, 2010). This adaptive partitioning accords with prior evidence that expert athletes flexibly modulate their error sensitivity depending on task relevance and context (Mann et al., 2010; Müller & Abernethy, 2012; Wilken et al., 2025). For instance, while few studies have directly compared sensitivity to small versus large perturbations, recent work shows that athletes develop enhanced neural responsiveness and adaptability after extensive training (Wilken et al., 2025). Such evidence supports the interpretation that long-term training shapes how prediction errors are scaled, selectively amplifying responses to deviations with higher informational value. This finding further refines the PRO framework by showing that the sensitivity of this scaling is modulated by motor expertise. Experts’ long-term training allows their internal models of action–outcome contingencies to encode finer distinctions within the prediction space (Bates et al., 2005; Pierella et al., 2019), leading to distinct responses even for moderate mismatches, as reflected in the emergence of three neural clusters (1–5 / 6 / 7). This expertise-dependent recalibration suggests that the gain function governing prediction–outcome comparisons is plastic rather than fixed, adaptively tuned through experience to optimize the precision of error monitoring in dynamic contexts. Importantly, this interpretation extends the traditional notion of neural efficiency beyond resource minimization (Neubauer & Fink, 2009). Experts are not merely processing prediction errors with fewer neural resources but employing a more strategic form of precision optimization: selectively suppressing minor discrepancies while enhancing sensitivity to meaningful violations (Hung et al., 2004). Such strategic precision tuning reflects an experience-dependent reorganization of predictive coding parameters and provides an efficient computational mechanism for adaptive evaluation under uncertainty. Multivariate decoding further substantiated this interpretation by demonstrating that only the frontal-midline region carried discriminable information about error magnitude, and that decoding accuracy was significantly higher in experts. The decoding window (2320–2420 ms), occurring after outcome presentation but before motor response, indicates that this theta activity supports evaluative computations rather than simple response correction (Herweg et al., 2020; Mas-Herrero & Marco-Pallarés, 2014). Together, these results identify frontal-midline theta as the locus through which motor expertise reorganizes predictive evaluation, transforming a domain-general control signal into a graded, experience-dependent representation of error magnitude. Thus, frontal-midline theta serves as a neural bridge linking predictive computation and expertise-dependent plasticity in error monitoring, providing a mechanistic substrate for the refinement of adaptive control through long-term training. Crucially, the MVPA results showed that this expertise-dependent graded encoding was localized specifically to the frontal-midline region, rather than broadly distributed across the scalp. This spatial specificity extends previous findings that expertise enhances frontal-midline theta power by revealing that the pattern of its internal representational structure also changes with experience. The midfrontal cortex, particularly the ACC, has long been identified as the central hub of the error-monitoring system, integrating bottom-up conflict signals with top-down control demands (Ullsperger et al., 2014). Our findings suggest that motor expertise enhances the computational precision of this hub: instead of generating binary “error/no-error” responses, the ACC encodes a parametric map of prediction–outcome discrepancies, allowing for more nuanced evaluative adjustments. This mechanism aligns with the view that during prediction–outcome mismatches, motor expertise enhances the activation of both the ACC and the AON (Abreu et al., 2012; Wang et al., 2019). However, whether the functional coupling between these two regions is modulated by expertise remains an open question for future research (Moreau et al., 2020). While the present findings provide converging neural and behavioral evidence for expertise-dependent graded error monitoring, several limitations warrant mention. First, the keypress paradigm used to simulate table-tennis landing positions necessarily simplified the spatiotemporal and interactive richness of real prediction contexts, and response execution may have introduced non-predictive motor delays. Future studies could adopt continuous tracking or immersive action observation paradigms to capture prediction–outcome updating in real time. Second, deviation magnitude was systematically manipulated, yet identical physical mismatches may carry distinct semantic meanings (i.e., within-side versus cross-side deviations), which could differentially engage evaluative mechanisms. Finally, although the present results identify frontal-midline theta as a key representational substrate, they do not directly test how expertise-related changes in predictive precision are implemented across the broader error-monitoring network. Future research integrating causal or connectivity-based approaches could address how long-term motor experience recalibrates network-level computation of prediction errors. References Abreu, A. M., Macaluso, E., Azevedo, R. T., Cesari, P., Urgesi, C., & Aglioti, S. M. (2012). Action anticipation beyond the action observation network: A functional magnetic resonance imaging study in expert basketball players. European Journal of Neuroscience , 35 (10), 1646–1654. https://doi.org/10.1111/j.1460-9568.2012.08104.x Alexander, W. H., & Brown, J. W. (2011). Medial prefrontal cortex as an action-outcome predictor. Nature Neuroscience , 14 (10), 1338–1344. https://doi.org/10.1038/nn.2921 Alexander, W. H., & Brown, J. W. (2017). The Role of the Anterior Cingulate Cortex in Prediction Error and Signaling Surprise. Topics in Cognitive Science , 11 (1), 119–135. https://doi.org/10.1111/tops.12307 Amoruso, L., Sedeño, L., Huepe, D., Tomio, A., Kamienkowski, J., Hurtado, E., Cardona, J. F., Álvarez González, M. Á., Rieznik, A., Sigman, M., Manes, F., & Ibáñez, A. (2014). Time to Tango: Expertise and contextual anticipation during action observation. NeuroImage , 98 , 366–385. https://doi.org/10.1016/j.neuroimage.2014.05.005 Anguera, J. A., Seidler, R. D., & Gehring, W. J. (2009). Changes in performance monitoring during sensorimotor adaptation. Journal of Neurophysiology , 102 (3), 1868–1879. https://doi.org/10.1152/jn.00063.2009 Arrighi, P., Bonfiglio, L., Minichilli, F., Cantore, N., Carboncini, M. C., Piccotti, E., Rossi, B., & Andre, P. (2016). EEG theta dynamics within frontal and parietal cortices for error processing during reaching movements in a prism adaptation study altering visuo-motor predictive planning. PLoS ONE , 11 (3), e0150265. https://doi.org/10.1371/journal.pone.0150265 Bae, G. Y. (2021). Neural evidence for categorical biases in location and orientation representations in a working memory task. NeuroImage , 240 , 118366. Bates, A. T., Patel, T. P., & Liddle, P. F. (2005). External behavior monitoring mirrors internal behavior monitoring. Error-related negativity for observed errors. Journal of Psychophysiology , 19 (4), 281–288. https://doi.org/10.1027/0269-8803.19.4.281 Bates, D., Kliegl, R., Vasishth, S., & Baayen, H. (2015). Parsimonious mixed models. arXiv . https://arxiv.org/abs/1506.04967 Bates, D., Mächler, M., Bolker, B. M., & Walker, S. C. (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software , 67 (1). https://doi.org/10.18637/jss.v067.i01 Bridges, C. C., Jr. (1966). Hierarchical cluster analysis. PsychologicalReports,18 (3),851854. Calvo-Merino, B., Glaser, D. E., Grèzes, J., Passingham, R. E., & Haggard, P. (2005). Action observation and acquired motor skills: An fMRI study with expert dancers. Cerebral Cortex , 15 (8), 1243–1249. https://doi.org/10.1093/cercor/bhi007 Cavanagh, J. F., Cohen, M. X., & Allen, J. J. B. (2009). Prelude to and resolution of an error: EEG phase synchrony reveals cognitive control dynamics during action monitoring. Journal of Neuroscience , 29 (1), 98–105. https://doi.org/10.1523/JNEUROSCI.4137-08.2009 Charrad, M., Ghazzali, N., Boiteau, V., & Niknafs, A. (2014). NbClust: an R package for determining the relevant number of clusters in a data set. Journal of Statistical Software , 61 (6), 1–36. http://www.jstatsoft.org/ Chen, Y. H., Chang, C. Y., Huang, S. K., & Yen, N. S. (2020). Nonlinear engagement of action observation network underlying action anticipation in players with different levels of expertise. Human Brain Mapping , 41 (18), 5199–5214. https://doi.org/10.1002/hbm.25186 Cohen, M. X., & Donner, T. H. (2013). Midfrontal conflict-related theta-band power reflects neural oscillations that predict behavior. Journal of Neurophysiology , 110 (12), 2752–2763. https://doi.org/10.1152/jn.00479.2013 Cross, E. S., Hamilton, A. F. de C., & Grafton, S. T. (2006). Building a motor simulation de novo: Observation of dance by dancers. NeuroImage , 31 (3), 1257–1267. https://doi.org/10.1016/j.neuroimage.2006.01.033 De Bruijn, E. R. A., Hulstijn, W., Meulenbroek, R. G. J., & Van Galen, G. P. (2003). Action monitoring in motor control: ERPs following selection and execution errors in a force production task. Psychophysiology , 40 (5), 786–795. Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods , 39 (2), 175–191. Feldman, H., & Friston, K. J. (2010). Attention, uncertainty, and free-energy. Frontiers in Human Neuroscience , 4 . https://doi.org/10.3389/fnhum.2010.00215 Friston, K. (2010). The free-energy principle: A unified brain theory? In Nature Reviews Neuroscience (Vol. 11, Issue 2, pp. 127–138). https://doi.org/10.1038/nrn2787 Friston, K. J., Daunizeau, J., Kilner, J., & Kiebel, S. J. (2010). Action and behavior: A free-energy formulation. Biological Cybernetics , 102 (3), 227–260. https://doi.org/10.1007/s00422-010-0364-z Hartigan, J. A., & Wong, M. A. (1979). Algorithm AS 136: A K-Means Clustering Algorithm. Journal of the Royal Statistical Society. Series C (Applied Statistics) , 28 (1), 100–108. Herrmann, C. S., Rach, S., Vosskuhl, J., & Strüber, D. (2014). Time-frequency analysis of event-related potentials: A brief tutorial. Brain Topography , 27 (4), 438–450. https://doi.org/10.1007/s10548-013-0327-5 Herweg, N. A., Solomon, E. A., & Kahana, M. J. (2020). Theta oscillations in human memory. Trends in Cognitive Sciences , 24 (3), 208–227. Holroyd, C. B., & Coles, M. G. H. (2002). The neural basis of human error processing: Reinforcement learning, dopamine, and the error-related negativity. Psychological Review , 109 (4), 679–709. https://doi.org/10.1037/0033-295X.109.4.679 Hung, T. M., Spalding, T. W., Maria, D. L. S., & Hatfield, B. D. (2004). Assessment of reactive motor performance with event-related brain potentials: Attention processes in elite table tennis players. Journal of Sport and Exercise Psychology , 26 (2), 317–337. https://doi.org/10.1123/jsep.26.2.317 Iwane, F., Sobolewski, A., Chavarriaga, R., & Millán, J. D. R. (2023). EEG error-related potentials encode magnitude of errors and individual perceptual thresholds. IScience , 26 (9). https://doi.org/10.1016/j.isci.2023.107524 Ward Jr, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American statistical association , 58 (301), 236-244. Jonker, Z. D., van der Vliet, R., Maquelin, G., van der Cruijsen, J., Ribbers, G. M., Selles, R. W., Donchin, O., & Frens, M. A. (2021). Individual differences in error-related frontal midline theta activity during visuomotor adaptation. NeuroImage , 245 . https://doi.org/10.1016/j.neuroimage.2021.118699 Kawato, M. (1999). Internal models for motor control and trajectory planning. Current Opinion in Neurobiology , 9 (6), 718–727. Kemmerer, D. (2021). What modulates the mirror neuron system during action observation? Multiple factors involving the action, the actor, the observer, the relationship between actor and observer, and the context. Progress in Neurobiology, 205 , 102128. Kuznetsova, A., Brockhoff, P. B., & Christensen, R. H. B. (2017). lmerTest Package: Tests in Linear Mixed Effects Models. Journal of Statistical Software , 82 (13), 1–26. https://doi.org/10.18637/JSS.V082.I13 Lenth R (2020) emmeans: Estimated marginal means, aka least-squares means. R package version 1.4.6. Levy, A., Enisman, M., Perry, A., & Kleiman, T. (2023). Midfrontal theta as an index of conflict strength in approach-approach vs avoidance-avoidance conflicts. Social Cognitive and Affective Neuroscience , 18 (1), nsad038. https://academic.oup.com/scan/article/18/1/nsad038/7231216 Li, Q., Wang, J., Li, Z., & Chen, A. (2022). Decoding the Specificity of Post-Error Adjustments Using EEG-Based Multivariate Pattern Analysis. Journal of Neuroscience , 42 (35), 6800–6809. https://doi.org/10.1523/JNEUROSCI.0590-22.2022 Li, Q., Yin, S., Wang, J., Zhang, M., Li, Z., Chen, X., & Chen, A. (2023). Not all errors are created equal: decoding the error-processing mechanisms using alpha oscillations. Cerebral Cortex , 33 (13), 8110–8121. https://doi.org/10.1093/cercor/bhad102 Lopez-Calderon, J., & Luck, S. J. (2014). ERPLAB: An open-source toolbox for the analysis of event-related potentials. Frontiers in Human Neuroscience , 8 , 213. https://doi.org/10.3389/fnhum.2014.00213 Lu, Y., Yang, T., Hatfield, B. D., Cong, F., & Zhou, C. (2020). Influence of cognitive-motor expertise on brain dynamics of anticipatory-based outcome processing. Psychophysiology , 57 (2), 1–13. https://doi.org/10.1111/psyp.13477 Ludolph, N., Plöger, J., Giese, M. A., & Ilg, W. (2017). Motor expertise facilitates the accuracy of state extrapolation in perception. PLoS ONE , 12 (11). https://doi.org/10.1371/journal.pone.0187666 Mann, D. L., Abernethy, B., & Farrow, D. (2010). Action specificity increases anticipatory performance and the expert advantage in natural interceptive tasks. Acta Psychologica , 135 (1), 17–23. https://doi.org/10.1016/j.actpsy.2010.04.006 Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG-data. Journal of Neuroscience Methods , 164 (1), 177–190. https://doi.org/10.1016/j.jneumeth.2007.03.024 Mas-Herrero, E., & Marco-Pallarés, J. (2014). Frontal theta oscillatory activity is a common mechanism for the computation of unexpected outcomes and learning rate. Journal of Cognitive Neuroscience , 26 (3), 447–458. https://doi.org/10.1162/jocn_a_00516 McNamee, D., & Wolpert, D. M. (2019). Internal Models in Biological Control. Annual Review of Control, Robotics, and Autonomous Systems , 2 (1), 339–364. https://doi.org/10.1146/annurev-control-060117 Moreau, Q., Candidi, M., Era, V., Tieri, G., & Aglioti, S. M. (2020). Midline frontal and occipito-temporal activity during error monitoring in dyadic motor interactions. Cortex , 127 , 131–149. https://doi.org/10.1016/j.cortex.2020.01.020 Müller, S., & Abernethy, B. (2012). Expert Anticipatory Skill in Striking Sports: A Review and a Model. Research Quarterly for Exercise and Sport , 83 (2), 175–187. https://doi.org/10.5641/027013612800745059 Musco, M. A., Zazzera, E., Paulesu, E., & Sacheli, L. M. (2023). Error observation as a window on performance monitoring in social contexts? A systematic review. Neuroscience and Biobehavioral Reviews, 147 , 105077. https://doi.org/10.1016/j.neubiorev.2023.105077 Neubauer, A. C., & Fink, A. (2009). Intelligence and neural efficiency. Neuroscience and Biobehavioral Reviews, 33 (7), 1004–1023. https://doi.org/10.1016/j.neubiorev.2009.04.001 Oh, Y., & Schweighofer, N. (2019). Minimizing precision-weighted sensory prediction errors via memory formation and switching in motor adaptation. Journal of Neuroscience , 39 (46), 9237–9250. https://doi.org/10.1523/JNEUROSCI.3250-18.2019 Omedes, J., Iturrate, I., Minguez, J., & Montesano, L. (2015). Analysis and asynchronous detection of gradually unfolding errors during monitoring tasks. Journal of Neural Engineering , 12 (5). https://doi.org/10.1088/1741-2560/12/5/056001 Panasiti, M. S., Pavone, E. F., & Aglioti, S. M. (2016). Electrocortical signatures of detecting errors in the actions of others: An EEG study in pianists, non-pianist musicians and musically naïve people. Neuroscience , 318 , 104–113. https://doi.org/10.1016/j.neuroscience.2016.01.023 Pezzetta, R., Nicolardi, V., Tidoni, E., & Aglioti, S. M. (2018). Error, rather than its probability, elicits specific electrocortical signatures: A combined EEG-immersive virtual reality study of action observation. Journal of Neurophysiology , 120 (3), 1107–1118. https://doi.org/10.1152/jn.00130.2018 Pfurtscheller, G., & Lopes da Silva, F. H. (1999). Event-related EEG/MEG synchronization and desynchronization: Basic principles. Clinical Neurophysiology, 110 (11), 1842–1857. https://doi.org/10.1016/S1388-2457(99)00141-8 Pierella, C., Casadio, M., Mussa-Ivaldi, F. A., & Solla, S. A. (2019). The dynamics of motor learning through the formation of internal models. PLoS Computational Biology , 15 (12), e1007118. https://doi.org/10.1371/journal.pcbi.1007118 Proctor, R. W., & Vu, K. P. L. (2006). Stimulus-response compatibility principles: Data, theory, and application . CRC press. Proverbio, A. M., Crotti, N., Manfredi, M., Adorni, R., & Zani, A. (2012). Who needs a referee? How incorrect basketball actions are automatically detected by basketball players’ brain. Scientific Reports , 2 . https://doi.org/10.1038/srep00883 Rosenbaum, D. A. (1980). Human Movement Initiation: Specification of Arm, Direction, and Extent. Journal of Experimental Psychology: General , 109 (4), 444–474. Silvetti, M., Seurinck, R., & Verguts, T. (2013). Value and prediction error estimation account for volatility effects in ACC: A model-based fMRI study. Cortex , 49 (6), 1627–1635. https://doi.org/10.1016/j.cortex.2012.05.008 Spinelli, G., Tieri, G., Pavone, E. F., & Aglioti, S. M. (2018). Wronger than wrong: Graded mapping of the errors of an avatar in the performance monitoring system of the onlooker. NeuroImage, 167 , 1–10. https://doi.org/10.1016/j.neuroimage.2017.11.019 Töllner, T., Wang, Y., Makeig, S., Müller, H. J., Jung, T. P., & Gramann, K. (2017). Two independent frontal midline theta oscillations during conflict detection and adaptation in a Simon-type manual reaching task. Journal of Neuroscience , 37 (9), 2504–2515. https://doi.org/10.1523/JNEUROSCI.1752-16.2017 Trujillo, L. T., & Allen, J. J. B. (2007). Theta EEG dynamics of the error-related negativity. Clinical Neurophysiology , 118 (3), 645–668. https://doi.org/10.1016/j.clinph.2006.11.009 Ullsperger, M., Fischer, A. G., Nigbur, R., & Endrass, T. (2014). Neural mechanisms and temporal dynamics of performance monitoring. Trends in Cognitive Sciences , 18 (5), 259–267. https://doi.org/10.1016/j.tics.2014.02.009 van Driel, J., Richard Ridderinkhof, K., & Cohen, M. X. (2012). Not all errors are alike: Theta and alpha EEG dynamics relate to differences in error-processing dynamics. Journal of Neuroscience , 32 (47), 16795–16806. https://doi.org/10.1523/JNEUROSCI.0802-12.2012 Vocat, R., Pourtois, G., & Vuilleumier, P. (2011). Parametric modulation of error-related ERP components by the magnitude of visuo-motor mismatch. Neuropsychologia , 49 (3), 360–367. https://doi.org/10.1016/j.neuropsychologia.2010.12.027 Wang, Y., Lu, Y., Deng, Y., Gu, N., Parviainen, T., & Zhou, C. (2019). Predicting domain-specific actions in expert table tennis players activates the semantic brain network. NeuroImage, 200 , 482–489. https://doi.org/10.1016/j.neuroimage.2019.06.035 Wang, Y., Zhao, Q., Shan, X., & Lu, Y. (2024). Coping strategies and brain activity in table tennis players faced with discrepancies between predicted and actual outcomes. Journal of Sports Sciences , 42 (8), 665–675. https://doi.org/10.1080/02640414.2024.2358291 Wilken, S., Böttcher, A., Beste, C., Raab, M., & Hoffmann, S. (2025). The neural basis of sensorimotor expertise: Investigating the role of theta oscillations and the superior parietal lobule using repetitive transcranial magnetic stimulation (rTMS). Cerebral Cortex , 35 (6). https://doi.org/10.1093/cercor/bhaf158 Wolpert, D. M., Ghahramani, Z., & Jordan, M. I. (1995). An internal model for sensorimotor integration. Science, 269 (5232), 1880–1882. https://doi.org/10.1126/science.7569931 Zhao, Q., Lu, Y., Jaquess, K. J., & Zhou, C. (2018). Utilization of cues in action anticipation in table tennis players. Journal of Sports Sciences , 36 (23), 2699–2705. https://doi.org/10.1080/02640414.2018.1462545 Zhao, Q., Wang, Y., Chen, Y., Wang, Y., Zhou, C., & Lu, Y. (2021). Expertise influences congruency monitoring during action observation at the motor level. Social Cognitive and Affective Neuroscience , 16 (12), 1288–1298. https://doi.org/10.1093/scan/nsab078 Information & Authors Information Version history V1 Version 1 02 December 2025 V2 Version 2 12 December 2025 Copyright This work is licensed under a Non Exclusive No Reuse License. Keywords error monitoring frontal-midline theta motor expertise multivariate pattern analysis predicted error Authors Affiliations Yiheng Chen 0009-0006-7467-1677 Shanghai University of Sport View all articles by this author Guanghui Zhang 0000-0003-0134-3614 Liaoning Normal University View all articles by this author Yinyue Wang Shanghai University of Sport View all articles by this author Gi-Yeul Bae Arizona State University Department of Psychology View all articles by this author Yingzhi Lu [email protected] Shanghai University of Sport View all articles by this author Funding Information National Natural Science Foundation of China 32471131, 32071089 Yingzhi Lu Metrics & Citations Metrics Article Usage 318 views 229 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation Yiheng Chen, Guanghui Zhang, Yinyue Wang, et al. Motor expertise reshapes the representational structure of graded error monitoring in midfrontal theta. Authorea . 12 December 2025. DOI: https://doi.org/10.22541/au.176466104.46271507/v2 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.176466104.46271507/v2","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9ffc0e0cea3458d3',t:'MTc3OTQ1NTM2Mg=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();