1/48
Looks like no tags are added yet.
Metacognition Definition
A family of metacognitive processes that monitor or control other (first order) cognitive processes & are unified by their self-referential quality.
Metacognition vs. Related Constructs
1. Introspection
> Involves conscious self-report BUT MC includes implicit behaviourally-indexed monitoring
2. Executive Function
> Involves top-down control (inhibition, switching, updating) BUT MC specifically refers to self-referential evaluation
3. Self-Regulation
> Involves coordinating thought, emotion behaviour towards goals BUT MC is only one component of self-regulation
Historical Origin on Metacognition
1. "Feeling-of-Knowing" (60s)
> Monitoring in the absence of retrieval (i.e. you know the answer but can't remember)
2. Metamemory & Metacognition (70s)
> Flavell's ed psych of how people assess memory/learning
Monitoring-Control Framework (Nelson & Narens)
Metacognition is two-tier/a feedback loop
First Order = Object Level (Actual cognitive processes)
Second Order = Meta-Level (Self-Model)
Flow of Info
Object > Meta = Monitoring Judgements
Meta > Object = Control Strategies
Four Metacognitive Monitoring Judgements
1. Ease of Learning
"How easy is this to learn?" (made before study)
2. Judgement of Learning
"How well did I learn this?" (made during/after study)
3. Feeling of Knowing
"I can't recall it but can recognise it"
(known after retrieval failure)
4. Confidence
"How likely is it that my answer is correct?"
(made after response)
NOTE Monitoring is only useful if it drives adaptive control: Self-awareness without change is just good monitoring with poor control
Four Metacognitive Control Strategies
1. Allocation: Spend more time on items judged less well-learned
2. Strategy Selection: Switch strategies when ineffective
3. Termination: Stop when judgements of learning satisfied
4. Response Withholding: Don't answer if low confidence
Measures of Metacognitive Accuracy
1. Calibration = Bias
> Match between mean confidence & mean accuracy
> E.G. 80% confidence but 60% accurate = overconfident by 20%
NOTE Captures systematic bias NOT whether confidence discriminates correct from incorrect
2. Resolution = Accuracy-Confidence Correlation
> Gamma correlation used to determine resolution: whether confidence predict accuracy trial-by-trial
> BUT gamma bad with ceiling/floor effects & conflates bias & sensitivity
3. Sensitivity (Meta d') = Knowing When Wrong/Right
> A signal-detection measure of how well confidence discriminates correct from incorrect responses
NOTE Resolution and sensitivity are conceptually the same but resolution has poorer measurement validity.
Signal Detection Theory
Test consists of target presence/absence judgements
> Type 1 Judgment = stimulus decisions (object level)
> Type 2 Judgement = Confidence rating (meta-level)
> Criterion = Level of evidence required to make stimulus decision
> Type 1 sensitivity (d') = How easy it is to tell absent from present
> Type 2 sensitivity (meta-d') = How well confidence can tell correct from incorrect
Metacognitive Efficiency
M-Ratio (Meta-d' / d') Calculates metacognitive sensitivity while controlling for type-1 performance to reduce expertise confound in individual differences research.
Conceptually Ratio represents the amount of type-1 information accessible to metacognition since confidence is affected by domain knowledge.
Developmental Trajectory of Metacognitive Ability
3-4Y Implicit monitoring develops before conscious report
5-8y Monitoring develops before control
Middle Childhood Vocab & strategy knowledge develops alongside underlying competence of metacognitive control. By 10-12, strategy & vocab are fully developed (knows what to study more, can allocate time)
NOTE Implicit & explicit metacognition can dissociate.
Ageing & Metacognitive Ability (Adulthood)
Ageing pattern suggests metacognition is varies by age and is not just a reflection of task performance!
1. Decline = Selective!
> Semantic "Feeling of knowing" is preserved
> BUT episodic/procedural/perceptual efficiency declined with age
2. Calibration Worsens (Overconfidence!)
> Increased overconfidence in older adults.
3. Neurodegenerative Disease Impact
> Alzheimer's = massive impairment (anosognosia)
Neural Correlates of Metacognition
Anterior PFC = MC Sensitivity
Makes sense because PFC involved in executive control
Anterior Insular = Confidence & Interoceptive Signals
Makes sense because it is the salience network
Precuneus = Self-Referential Processing
Anosognosia
Shows that monitoring is a separable system NOT a side-effect of performance!
Presentation Stroke/Alzheimer's: Unaware of clear motor/memory deficit (metacognition is impaired while object-level cognition remains partially intact).
Metacognitive Profile in Psychiatric Conditions
Metacognition can CONTRIBUTE to clinical symptoms NOT just correlated with them!
1. Anxiety/Depression
> Reduced confidence but increased sensitivity (depressive realism = bias-sensitivity dissociation)
2. Schizophrenia
> Impaired metacognitive efficiency even when first-order performance matches health control (can't distinguish reality/delusions)
3. OCD
> Inflated doubt (can't trust own perception!)
NOTE Wells' metacognitive therapy: Pathology is maintained by metacognitive beliefs (e.g. "worrying keeps me safe")
Effect of Expertise on Metacognition
Metacognition uses domain-general & domain-specific signals! Expertise does not reliably transfer metacognitive ability to other domains!
Dunning-Kruger Fiasco Poorest performers on a task show the largest level of overconfidence BUT possible explained by regression to the mean?
Nuanced Framing Low performers have less reliable monitoring signals, so their metacognitive estimates are naturally "noisier" (experts show better calibration because they can better judge their performance/ability).
Difficulties Studying Individual Differences in Metacognitive Ability
Measures are often noisy at individual level
Demand characteristics in self-report measures
Reactivity effects: Naturally improve if overconfident or worsen if underconfident OR become more conservative
Metacognition confounded by strategy/motivation
"Fake it till you make it" is a strategy
Circularity problem
Inferring metacognition from behaviour, then using it to explain behaviour
Metacognitive Monitoring/Control Models
1. Discrepancy Reduction
> Study until judgement of learning reaches criterion then move on - drives effort towards what you are least confident with.
2. Region of Proximal learning
> Prioritise items that are partially known "happy medium" difficulty - concentrates effort where gradient is steepest.
BOTH Assume monitoring is accurate (e.g. judgements of learning can be wrong!)
Delayed-JOL Effect
> Delay increases accuracy of judgements of learning vs. immediately after study
> Supports cue-utilisation theory: delay makes JOL based on failed retrieval attempts, whereas immediate JOLs are based on processing fluency
Illusions of Competence
Example of when metacognition fails!
Results from using incorrect cues to judge learning
1. Rereading > Mistaken as sign of learning
> Fluent re-experience produces less durable retention than retrieval
2. Large font size > Mistaken as easy to remember
> Inflates JOL but unrelated to recall
3. Presentation clarity > Mistaken as easy to comprehend
> Inflates confidence without improving understanding
4. Answer visible (foresight) > Mistaken as learning
> Overestimates one's ability to retrieve memory
Retrieval Practice / Testing Effect
Monitoring quality affects downstream control BUT beliefs about best strategies don't always lead to uptake!
Better Signal = Better Calibration = Better Study Decisions
Retrieval is a better diagnostic monitoring signal than rereading because you find out where you can produce the answer, not just whether it feels familiar
Beliefs Shape Resource Allocation + Strategy Selection
> BUT frequency of use vs. perceived efficacy complicates this (self-testing remains underused even when students know it is better than rereading)
Social & Communicative Functions of Confidence
Confidence is a Post-Decisional Control Signal!
> Overconfidence = premature commitment (exploit)
> Under-confidence = excessive information-seeking (explore)
Confidence has Social Functions!
> Read by other people (non-verbal) & used to weight advice
> Persuasive even when unreliable (e.g. unreliable)
> Social calibration = accuracy as a signal to others about reliability of info
Empirical Evidence
> Dyads can outperform individuals on a perceptual task ONLY when confidence is shared & well-calibrated
> Group performance suffers when the loudest, most confident voice dominates.
NOTE Appropriate deference to expert is threatened by egocentric discounting (tendency to under-weight advice relative to own opinion)
Assess Metacognitive Training & Clinical Interventions
Is Training Learning or Just Self-Awareness?
>Uncertain whether improved monitoring leads to better learning or just better self-knowledge.
> BUT research suggests giving feedback on metacognitive judgements does improve confidence-accuracy calibration AND transfers to unrelated tasks
Three Programs Targeting Metacognition
1. Metacognitive Therapy
> RCTs show effective for GAD/depression
2. MCT for Schizo
> Targets overconfidence in incorrect beliefs & biases that maintain delusional thinking
3. Metacognitive Interpersonal Therapy
> Developed for personality disorders: targets patient's model of their own & others' minds.
Role of Metacognition in Emotion Regulation
Selecting & adjusting a regulation strategy involves monitoring affective state!
E.G. Gross' process model of emotion regulation implicitly involves metacognition
E.G. Wells' metacognitive therapy (believing worrying about things is helpful leads to maintenance of pathological anxiety).
Why is it Difficult to Define Intelligence?
1. Lay Conceptions Distinguishing implicit theory from scientific constructs
2. Constructs vs. Measures Intelligence is a latent ability inferred from test performance
3. Ability vs. Achievement Accumulated knowledge is not the same as ability
4. Multi-Dimensional Is intelligence a single general ability or family of distinct abilities?
Major Structural Models of Intelligence
1. Spearman's Two-Factor Theory
> Intelligence = G + S (a task-specific factor)
2. Cattell-Horn
> Fluid (Gf) = Novel problem-solving/abstract reasoning (culture-fair but still shaped by experience, peaks early & declines with age)
> Crystallised (Gc) = Accumulated semantic knowledge (culture-dependent, maintained or increases with age)
3. CHC Three-Stratum Hierarch
> G | 10 broad abilities (inc. Gf + Gc) | 70+ narrow abilities
Critically explain what G is
True Meaning of G
> A diverse cognitive battery which is remarkably robust across tests & a strong predictor of achievement, job performance, health & longevity
> G is a statistical extraction mean to represent whatever is common across diverse cognitive tests
G is not.. A brain mechanism, fixed at birth, the entirety of intelligence or a claim that intelligence reduces to 1 no.
Positive Manifold Scores on cognitive tasks tend to be positively correlated with on another
Explanation 1 Sampling Theory
> Tests which sample overlapping processes correlate & this overlap = g
Explanation 2 Mutualism
> Cognitive processes start out unrelated but reinforce each other throughout develop (e.g. having a good memory increases processing increasing reasoning)
NOTE The positive manifold is a fact BUT a single causal g is one of several explanations
Alternative Intelligence Models
Alternative frameworks have made important points about breadth/context but do not receive strong empirical support!
1. Gardner's Multiple Intelligences
> 7 types (linguistic, logic-math, spatial etc.
+ Good face validity/influence on education BUT...
- More a philosophy, conflates abilities with interests, basis in prodigy case studies = low generalisability
2. Sternberg's Triarchic Theory
> Proposes intelligence is not so narrow; rather, consists of analytical, creative & practical domains.
+ Practical domain correlated with g & evokes importance of adapting intelligence to real-world context! BUT...
- Hard to operationalise, 3 domains too correlated to be distinct, tests sample a narrow range of contexts.
3. Emotional Intellignece Models (Ability & Mixed)
> Ability has incremental validity
> Mixed/Trait highly correlated with personality & emotional stability
Development of Intelligence Tests
First Sensory/Motor but Binet is a legend!
Galton & Cattel Sensory & motor measures/"Mental Tests" which were poor predictors of intellectual achievement
Binet Actually used complex cognitive tasks
> Key innovation = assess high-order thinking!
> First scale ordered by item difficulty, but revised scale organised by age-level & mental-level procedure (task assigned to specific age level if at least 75% of kids could successfully complete)
Ratio & Deviation IQ
Stern/Terman's Ratio IQ
> Mental Age / Chronological Age x 100
> Good for kids but not adults (age standards lose meaning after mid-teens)
> Uneven scaling: 4yo with mental age 6 has IQ 150 but 10yo with mental age 12 has IQ 120
Deviation IQ Addresses limitation of ratio IQ: not based on mental age but how you score compared to others in same age group
> E.G. 115 score = 1 SD above average if mean is 100 & SD =15
> Reflects relative performance
> Allows age-appropriate comparison across lifespan
> More statistically robust
> Enables norm-referenced interpretation of scores
Describe the Structure of WAIS & Raven's
Weschler Scales Family of intelligence/cognitive abilities test used across lifespan
1. Deviation IQ
2. Sub-scales (separates verbal from performance scales to recognise intelligence is not monolithic!)
3. Point scales instead of age scales
Raven's Matrices A non-verbal test of fluid intelligence/abstract reasoning
> E.G. pick the piece that completes visual pattern
> Used cross-culturally because it minimises language demands BUT abstract pattern completion is itself culturally situated.
Psychometric Properties of Intelligence Tests
Good Reliability & Validity!
Reliability
1. Test-retest (great, typically 90+)
2. Stability (relatively stable from childhood)
3. Validity (strong criterion validity for academic/occupational outcomes but content validity is theory-dependent)
4. Invariance (contested whether test measures same construct across different groups)
Predictive Validity of Intelligence Tests
> Large correlation between IQ with academic achievement (strongest predictor!)
> Moderate correlation with job performance (estimated vary by method/study era)
> Reliable but small correlation with health/longevity
NOTE IQ does not predict everything equally! It is most predictive when tasks are cognitively complex (the more complex the task, the more it draws on g & the more it separates high scores from low - g is better predictor of complex jobs than simple)
Flynn Effect
Observation that intelligence increasing (~3 points per decade), with gains largest on fluid intelligence/Raven's tests
> Too fast to be genetic
> Candidate Guesses = nutrition, reduced disease, more schooling, greater exposure to abstract thinking, smaller families
> Overall Implication = IQ reflects environment not just innate capacity!
NOTE Plateau in Scandinavian countries: Ceiling? Changing environment? Compositional change in tested populations?
Explain the Logic of Heritability Estimates
H2 Proportion of variance in a trait within a population attributable to genetic variation
> Twin/adoption studies estimate heritability of IQ in adults .50-.80 (rises with age)
NOTE Population statistic not personal!
NOTE Does not account for environmental factors (e.g. nutrition) OR between-group causes!
Explain the GxE Interplay Intelligence
Nature & Nurture are entangled (even causal sequence)!
GxE Correlation
> Consists of passive (genes + home), active (niche-picking) + evocative (curious kids change environment by drawing stimulating activities from parents)
> Lower heritability in low-SES families & high heritability in high-SES families suggest impact of welfare
NOTE Wilson Effect: Heritability rises from childhood to adulthood (adults can select their environment & express "genetic dispositions")
NOTE GWAS shows massive polygenicity each with tiny effect (predicting only 7-10% but smaller for within-family scores)
Group Differences in Intelligence
Within-group heritability NOT the same as b/w-group cause! High within-group heritability IS compatible with environmental b/w-group difference!
1. Racial/Ethnic
> Average USA Black-White gap is 10-15 pts
2. Sex
> Near-zero average difference in g between men & women
> Reliable differences in specific abilities (women = verbal, men = spatial rotation)
> Greater variance among males suggests culture/socialisation x biology interaction
3. SES
> Strong SES-IQ correlation, but causal direction runs both ways
Myth of Hereditarian Hypothesis of Intelligence
1. No genetic mechanism for group differences has been identified
2. Flynn Effect shows IQ scores are environmentally malleable
3. Social-structural & Test Confounds (e.g. poverty, test familiarity, stereotype threat)
4. Adoption studies are limited
5. Debate intertwined with scientific racism
Implications of Stereotype Threat on Test Interpretation
True effect is smaller than initially thought BUT scores are not pure measures of ability & can be confounded by testing context!
Definition Performance suffers when test framed as diagnostic of stereotype-related ability (due to anxiety, cognitive load, self-monitoring).
Evidence Dramatic initial finding contrasted by recent meta-analyses + showing many boundary conditions
Psychometric Test Bias & Ethical Test Fairness
Psychometric Test Bias
1. Measurement Bias
> Does test measure same underlying construct in same way across groups?
> Tested using measurement invariance & differential item functioning
2. Predictive Bias
> Does a given score predict an external criterion similarly across groups?
> Tested using regression slopes, intercepts & validity coefficients.
Evidence Modern intelligence tests show great measurement equivalence & similar predictive validity across groups (but not uniform - depends on test/groups)
Ethical Test Fairness Test can be psychometrically unbiased BUT STILL unfair!
> If it measures something shaped by unequal opportunity
> Used to amplify inequality
> Accurately predicts a criterion that is itself biased
Cultural Loading vs. Cultural Bias
> All tests are culturally loaded but not necessarily biased (can still predict outcomes equally across groups
> Raven's Matrices are "culture-free" but don't entirely eliminate loading
Ability vs. Mixed Model EI
Ability Model of EI
Set of cognitive abilities for processing emotional information (processing, reasoning, problem-solving applied to emotional context).
Four branches, hierarchically organised where lower branches feed higher ones
1. Perceiving (Faces, voices, images)
2. Using (Facilitate thinking)
3. Understanding (Vocab, emotion change/blend)
4. Managing (Regulate own & others)
Mixed Models of EI (Trait / Emotional Self-Efficacy)
A. Goleman (Includes traits/motivations/empathy/social skills)
B. Bar-On (Includes interpersonal skills, adaptability, mood)
Ability vs. EI Measures & Incremental Validity
Ability model supported as "new construct" given modest/small correlation with g/personality, BUT trait EI has near zero correlation with g & large correlation with personality!
Ability = Performance Based (MSCEIT)
> Consensus scoring - majority not always "correct" BUT expert scoring still subjective
> "Using" emotions is psychometrically weak
> Perceiving & Using = Experiential; Understanding & Managing = Strategic
> Correlates modestly with g (.10-.30)
Trait EI = Self-Report (TEIQue)
> Self-reported ability not actual ability (hard to interpret)
> Easy to administer, good internal consistency/face validity
Incremental Validity
> Ability = Small but non-trivial incremental validity over G (predicts interpersonal/emotional domains)
> Trait EI = Predicts wellbeing, but adds little once neuroticism/extraversion are controlled
Relationship Between EI, Metacognition & Emotion Regulation
Emotion Management significantly overlaps with metacognitive control of affect!
Meta-level monitors > affective object level > exert control or regulation
E.G. Alexithymia (difficulty identifying own emotions/impaired emotional monitoring) can predict autism & other disorders
Thorndike's Theory (Social Intelligence)
Point Measurement failure!
> High face validity/intuitive appeal BUT tests of social judgement correlate substantially with g
> Poor discriminant validity when controlling for g = no distinct social intelligence factor
Summary
> (1920s) Proposed 3 intelligences: abstract, mechanical, social
> Social = Ability to understand & manage people & act wisely in human relations
Is Theory of Mind a Distinct Intelligence?
Better to frame as "social cognitive abilities" (plural!) than social intelligence (singular!)
Argument
> Dissociations & specific neural mechanisms suggest so (autism, Phineas Gage)
> Measurement validity problem indicated by high correlation with G
> Arguably, a set of narrow abilities (emotion recognition, empathy, social learning, pragmatics) which share a social content domain
NOTE Autism point: Baron-Cohen's E-S framework (systematising vs. empathising cognitive styles distinct from general intelligence as in autism)
Sternberg's Practice Intelligence / Tacit Knowledge
Tacit Knowledge Acquired through experience vs. being explicitly taught ("knowing how" not "knowing that")
> E.G. Navigating office politics, managing people, prioritising under pressure
> Inventory involves tests of situational judgement
> Can predict job performance
Critique
Methodological: TKI's actually do correlate with g in larger studies
Scoring: Experts disagree & strategies can be domain-specific
Conceptual: Conflates intelligence with experience & personality. Is tacit knowledge intelligence or accumulated expertise + personality + g?
Logical: A low IQ correlation is insufficient to establish new intelligence: many specific skills correlate modestly with g without being a form of intelligence
NOTE Ceci's Bio-Ecological Model = cognition is always situated by biological potential x context interaction
> E.G. Brazilian street kids can perform complex mental arithmetic but fail equivalent problem in school format (performance is context-dependent!)
Assess whether social & practical intelligence are distinct from g
No & that's okay! IQ doesn't measure everything that matters for life success
> Real-world competence = g + domain knowledge + personality + motivation + opportunity. No separate intelligence needs to be hypothesised
> New forms of intelligence needs to have distinct measurement, incremental predictive validity & theoretical coherence!
Creativity
Defined by novelty & usefulness!
> Novelty is not enough - usefulness distinguishes creativity from randomness
> Big-C = eminent, field-changing creativity
> Little-C = everyday creativity (most measurements target this)
> Open-personality signature, own neural patterns BUT partly dependent on G & not a single unitary faculty
Measures of Creativity
1. Divergent Thinking = Dominant paradigm (scored on fluency, flexibility, originality & elaboration)
A. Guildford's Alternative Uses
B. Torrance Tests (verbal battery hypotheticals/product improvement, figural, humour)
2. Convergent Thinking
A. Remote Associates (Find word linking 3 cues (cottage / Swiss / cake = cheese)
B. Consensual Assessment (Experts rate product design); High validity!
NOTE Measurement critiques:
> Divergent-thinking is a PROXY; only modestly predict creative achievement
> Fluency confound (originality increases simply with more responses)
> Scoring originality by statistical rarity is sensitive to sample & the scorer
> Real creativity also required convergent thinking (picking the good idea!)
Threshold Hypothesis Creativity
Classic Claim = IQ + creativity correlate up to 120 IQ!
> Recent evidence that intelligence contributes to creativity, real threshold is fuzzy
Creativity-Intelligence Relationship Needs...
> Openness to experience,
> Intrinsic motivation
> Deep domain knowledge
> Environment willing to recognise/adopt contribution
Neuroscience of Creative Cognition
Creativity is the product of network dynamics!
Default Mode Network = Spontaneous idea generation
Executive Control Network = Evaluation & refining