An fMRI study of error monitoring in Montessori and traditionally-schooled children

The study investigates whether and how children’s error monitoring and its neural underpinnings are influenced by pedagogical experience, comparing students from Montessori and traditional schools. Error monitoring—detecting outcomes that violate expectations and adapting behavior—is fundamental to learning, improves through childhood into adolescence, and involves cingulate regions, especially the ACC. Montessori pedagogy emphasizes self-directed recognition and constructive engagement with errors, often via immediate, self- or peer-guided correction, whereas traditional pedagogy typically provides teacher-driven, often delayed feedback. The authors aimed to test whether these differing experiences shape children’s behavior and brain activity/connectivity during error monitoring in a math task. Hypotheses: (1) Across groups, incorrect versus correct responses would elicit increased activity along the cingulate gyrus, consistent with adult data; (2) Group differences would emerge: Montessori students would show higher activation/connectivity in error-monitoring regions (ACC, medial frontal cortex), whereas traditionally-schooled students would show effects in memory-related regions (e.g., hippocampus).

Prior work links error monitoring to surprise/expectation violation, self-regulation, and adaptive behavior, with developmental maturation in cingulate networks (particularly ACC) and shifts in ACC connectivity across childhood and adolescence. Most neuroimaging studies have been in adults, showing distinct neural signatures for error monitoring versus conflict/reward and highlighting social influences on error processing. Educationally, Montessori emphasizes autonomous, process-oriented engagement with errors, while traditional approaches emphasize correctness and recall. Previous behavioral/EEG work by the authors suggests traditionally-schooled children react more strongly to detecting incorrect responses (potentially reflecting lower self-regulation), and adult studies often tag errors as negative/aversive. Together, the literature suggests pedagogical context could shape error-monitoring processes, especially in the 8–12-year developmental window when cingulate maturation is active.

Design: Cross-sectional fMRI study comparing Montessori and traditionally-schooled children during a math error-monitoring task with feedback. Participants: 37 children (18 females; 8–12.3 years) enrolled in Montessori or traditional schools from kindergarten or for at least 3 years (for youngest). Exclusions: sickness (n=1 Montessori), braces (n=1 traditional), high dyslexia/dyscalculia (n=1 Montessori), excessive motion (n=2; one per group). Final N=32 (16 Montessori, 16 traditional; 17 female; age 8–12.3 years; mean ± SD = 9.98 ± 1.25). Mostly right-handed. Ethics approved; parental consent and child assent obtained. Group comparability measures: Non-verbal intelligence (Raven’s Progressive Matrices), self-reported anxiety (STAI-Y), working memory (digit-letter span), standardized mathematical skills, family SES, parental report of child’s math affect, home physical environment, parents’ perceived life stress, and home pedagogical environment. No significant between-group differences after FDR correction. Task: Novel math proofreading paradigm during fMRI. Each trial: start cue (1000 ms), math problem (addition/subtraction) with proposed solution (correct/incorrect) for 3000 ms during which participants judged right/wrong via button press, feedback (“correct”/“incorrect”) 2000 ms, jittered fixation 2000–3000 ms. 64 trials in 8 blocks; inter-block interval 14,000 ms. Responses later than 3000 ms were labeled as misses and excluded from imaging analyses. Reward images were shown during feedback for both correct and incorrect trials (reward effects not analyzed here). Behavioral analyses: Correlations validating task with standardized math performance and parental math affect; ANOVA on accuracy (response type: correct/incorrect/missed × group), ANOVA on RT (response type: correct/incorrect × group), post hoc Tukey tests, and efficiency metric = RT / proportion correct with independent t-test. MRI acquisition: Siemens 3T Prisma-Fit, 64-channel head coil. Structural T1 MPRAGE (1 mm isotropic). Functional EPI with SMS (TR=1000 ms, TE=30 ms, 64 axial slices, 2 mm isotropic, flip angle 80°, matrix 100×100, FOV=200 mm, SMS factor 4, GRAPPA 2). One session of 740 volumes (7 dummy discarded), 12 min 26 s total. Preprocessing: SPM12. Realignment, slice timing correction, coregistration to T1, normalization to MNI, 8-mm FWHM smoothing. Motion quality control; subjects with >20% volumes exceeding 3 mm/3° excluded (n=2). First-level GLM: Modeled the 4-s period covering task and feedback following the cue. Contrasts: correct vs incorrect responses; 6 motion parameters as nuisance; high-pass filter 128 s. Second-level analyses: Mixed-design ANOVA with within-subject factor Response (correct vs incorrect) and between-subjects factor Pedagogy (Montessori vs traditional). Cluster-level FDR p<0.05 (voxel p<0.001; cluster >30 voxels). Functional connectivity: PPI analyses with seed ROIs from activation results. Seeds: from main effect of response (left middle frontal cortex x=-30,y=28,z=44; ventral anterior precuneus x=-10,y=-46,z=46) and from main effect of pedagogy (right medial prefrontal cortex x=6,y=64,z=8; anterior middle cingulate/left ACC x=-2,y=52,z=-2; right cuneus x=22,y=-90,z=8). Spherical ROIs (8-mm radius). Second-level t-tests for response type and pedagogy. Threshold p<0.001 uncorrected (exploratory).

Behavioral:

• Groups were comparable on demographics, cognition, affect, SES, and home environment; marginally more girls in the traditional group (p=0.08), gender effects non-significant (p=0.18).
• fMRI task performance correlated with standardized mathematical skills (r=0.56, n=32, p<0.001) and with parental report of child’s math affect (r=0.38, n=32, p=0.038).
• Response pattern: Correct responses were more frequent than incorrect or missed, F(60,2)=22.88, p<0.001. Group×response interaction: F(60,2)=3.78, p=0.028. Groups did not differ in correct rates (Ptukey=1.0); Montessori had higher incorrect rates, traditional missed more trials (t(60)=3.05, Ptukey=0.038).
• Reaction time: Montessori faster (M=1719 ms, SD=405) than traditional (M=2060 ms, SD=351), F(30,1)=6.55, p=0.016, η²p=0.18. No main effect of response type and no response×group interaction on RT.
• Efficiency (RT/proportion correct) higher in Montessori (M=3.26, SD=1.31) vs traditional (M=2.46, SD=0.81), t(30)=2.09, p=0.046, d=0.74. Neural activation (ANOVA):
• Main effect of response (Correct > Incorrect): Increased activity in bilateral posterior cingulate cortex (PCC) and left precuneus; middle frontal gyrus; left inferior temporal gyrus; left cerebellar Crus I and Crus II.
• Main effect of pedagogy (Montessori > Traditional; irrespective of response): Increased activation in left middle occipital cortex and right cuneus; right superior parietal lobule; medial prefrontal cortex with extension into left ACC. No interaction. Functional connectivity (PPI):
• Across groups, for correct vs incorrect responses, precuneus and mid-frontal seeds showed stronger connectivity to the left insula.
• Group-specific patterns: • Traditional: For correct trials, stronger connectivity between seeds (left ACC, right medial prefrontal cortex, right cuneus) and right hippocampus; additionally, right medial prefrontal seed to right putamen. • Montessori: For incorrect trials, stronger connectivity between left ACC seed and right middle/superior frontal cortices and left orbitofrontal cortex. • No regions showed greater connectivity for correct trials in Montessori or for incorrect trials in Traditional.

The study shows that children aged 8–12 display greater neural activation for correct than incorrect responses across PCC/precuneus, middle frontal, inferior temporal, and cerebellar regions, contrary to typical adult findings where errors often elicit greater responses. This pattern may reflect developmental differences in the salience and learning utility of correct responses in late childhood, potentially emphasizing associative integration of successful procedures. Importantly, pedagogical exposure was associated with both behavior and neural measures: Montessori students responded faster, missed fewer trials, and showed greater activation in visual, parietal, and frontal regions implicated in math and executive/attentional control. Connectivity analyses revealed divergent strategies: Traditionally-schooled children increased ACC/mPFC/cuneus connectivity with hippocampus (and mPFC-putamen) after correct responses, suggesting memory-based reinforcement of correct solutions; Montessori students increased ACC connectivity with frontal control regions after errors, consistent with a more exploratory, self-corrective approach to processing mistakes. These findings indicate that schooling context can modulate the development of error-monitoring systems in the brain, with implications for fostering process-oriented learning behaviors and adaptive self-regulation.

This study provides initial neuroimaging evidence that pedagogical experience relates to children’s error monitoring and its neural correlates. Children, unlike adults, showed stronger activations to correct than incorrect responses, and Montessori versus traditional schooling was associated with distinct functional connectivity patterns: Montessori students engaged frontal control networks after errors, whereas traditionally-schooled students engaged hippocampal memory systems after correct responses. Behaviorally, Montessori students were faster, more efficient, and missed fewer trials with comparable accuracy. These results suggest that daily pedagogical practices may shape the maturation of error-monitoring networks and learning strategies. Future work should use larger, longitudinal, and randomized or quasi-experimental designs, test domain generality beyond math, and include adults to map developmental trajectories and the role of schooling history.

• Non-randomized, cross-sectional comparison; potential selection/confounding factors (e.g., family or motivation) cannot be fully ruled out despite extensive controls.
• Modest sample size (N=32) limits power and generalizability.
• No adult comparison group; cannot determine if observed activation patterns are child-specific or persist into adulthood.
• Missed trials were excluded from imaging analyses; reasons for misses (e.g., fatigue, delayed response, uncertainty) were not differentiated.
• Connectivity analyses were exploratory with uncorrected voxel-wise thresholds.