The evidence base for metacognition training

Meta-analysis of 67 studies. Metacognitive training produced a 0.63 standardised improvement in academic performance — larger than homework, class-size reduction, or most tutoring interventions. Trained students outperformed 73% of untrained peers.

Controlled study of structured metacognitive training. Effect size of 1.11 — one of the largest observed in the cognitive training literature. Trained participants outperformed approximately 86% of untrained controls.

Development and validation of the Metacognitive Awareness Inventory (MAI). Established the two-factor structure of metacognition — Knowledge of Cognition and Regulation of Cognition — that underpins most subsequent research.

Coined the term "metacognition" and established its original taxonomy. Defined metacognitive knowledge (person, task, strategy) and metacognitive experiences as distinct components. The bedrock paper for all subsequent work.

Examines how metacognition drives self-regulated learning. Demonstrates that learners who monitor their own understanding and adjust strategies accordingly consistently outperform those who do not, independent of ability level.

Predictive validity study showing metacognitive skills measured in Grade 3 significantly predict mathematics performance in Grade 4, even after controlling for prior math scores. Establishes metacognition as a stable forward predictor.

Comprehensive review establishing metacognition as a predictor of academic achievement above and beyond intelligence. Argues for metacognition as the "second pillar" of learning, independent of IQ.

Landmark study showing people saying they are "99% certain" on general knowledge questions are wrong approximately 40% of the time. Established overconfidence as a systematic, measurable bias rather than an occasional error.

Foundational calibration research establishing the methodology for measuring probability calibration. Demonstrated that calibration is a stable individual trait that varies systematically across domains and can be trained.

Good Judgment Project. Across 3,000 forecasters over multiple years, structured metacognitive practice — tracking predictions, reviewing errors, updating beliefs — produced consistent 14% improvements in forecasting accuracy above baseline.

Distinguished three distinct types of overconfidence: overestimation (thinking you did better than you did), overplacement (thinking you did better than others), and overprecision (being too confident in the accuracy of your beliefs). MindFrame measures all three.

Comprehensive treatment of systematic biases in human inference, including the tendency toward overconfidence. Established that cognitive biases are features of normal reasoning, not defects of irrational individuals.

Demonstrated that calibration is a measurable, trainable skill rather than a fixed trait. Feedback-based training on confidence accuracy produced significant improvements that generalised beyond the specific domain trained.

Developed the Metacognitive Model of anxiety and depression. Core insight: emotional distress is driven not by negative thoughts themselves but by metacognitive beliefs about those thoughts (e.g., "worrying is protective", "rumination helps me solve problems"). The model that underpins MCT.

Meta-analysis comparing Metacognitive Therapy (MCT) to Cognitive Behavioural Therapy (CBT). MCT outperformed the gold standard with g = 0.69. Trained patients scored better than approximately 75% of CBT patients. Largest head-to-head comparison to date.

Clinical trial of Metacognitive Therapy for generalized anxiety disorder. Demonstrated significant symptom reduction with effect sizes exceeding those typically seen in CBT trials for the same population.

Validation of the Metacognitions Questionnaire (MCQ-30), which measures beliefs about worry, cognitive monitoring, and the need to control thoughts. Widely used in clinical and research contexts.

Examines the relationship between metacognitive beliefs and rumination. Negative beliefs about the uncontrollability of rumination are stronger predictors of depression severity than the content of the rumination itself.

Thinking, Fast and Slow. Synthesises decades of research into System 1 (fast, intuitive) and System 2 (slow, deliberate) thinking. Metacognition is the mechanism by which System 2 monitors and overrides System 1 errors — the cognitive layer MindFrame trains.

Demonstrated the bias blind spot: people reliably judge themselves as less biased than average, and are resistant to updating this belief even when presented with direct evidence. Metacognitive training specifically targets this pattern.

Showed that individual differences in rational thinking are real and measurable, and are not fully explained by IQ. Identified actively open-minded thinking — a metacognitive disposition — as a key predictor of reasoning quality.

Analysis of framing effects and how awareness of them does and does not reduce their impact. Suggests that knowing about a bias is insufficient — structured metacognitive practice is required for reliable debiasing.

Established "actively open-minded thinking" (AOT) as a measurable cognitive style linked to better reasoning outcomes. AOT is characterised by seeking disconfirming evidence, updating beliefs, and resisting cognitive entrenchment — all metacognitive skills.

Meta-analysis of 69 studies on EQ and job performance. Found modest predictive validity (ρ = 0.23) — but this largely disappeared when controlling for general mental ability and personality traits.

Meta-analytic review confirming that EQ's predictive validity for job performance is near-zero after controlling for GMA and Big Five personality. EQ's apparent predictive power largely reflects overlap with existing constructs, not a distinct capability.

Systematic critique of EQ as a scientific construct. Argues that ability-based EQ is a form of social intelligence already captured by existing measures, and that mixed models of EQ conflate personality traits with cognitive abilities.

Meta-analysis of EQ training programmes. Found modest average effects (d ≈ 0.35) that typically decayed rapidly after training ended. Significantly weaker and less durable than metacognitive training effects.

Comprehensive review of EQ science, distinguishing ability models from mixed models. Notes the scientific case for ability EQ is stronger than for mixed models, but acknowledges the incremental validity problem remains unresolved.