Metacognition Flashcards
1
Q
What is metacognition
A
Processes involved in monitoring/controlling performance on a task
2
Q
Levels of cognitive operations
A
- Object level (includes your actions/behaviours)
- Meta-level (mental simulation that keeps track of what’s going on at the object-level)
- It’s a process model of a flow of information
- Monitoring process –> (object-level to meta-level) , monitors our cognitive operations and results in subjective experience or feeling
- Control process –> (meta-level to object level), subjective experience from monitoring determines current and future cognitive operations
3
Q
Using these levels - E.g. how tall is/was the world’s tallest person?
A
- So at the object level you search your memory
- While you are doing this search, you are also monitoring (object-meta level) whats going on, so you might come up with some possible candidate answers/subjective feelings about possible answers
- you also, based on subjective experiences, exert control over the answer you give and modify behaviour to achieve goal
4
Q
Measures of monitoring
A
- Ease of learning - Before you learn the items, how difficult will it be to learn each item?
- Judgement of learning - for each of the items you have learned, how likely is it that you will recall them later?
- Feeling-of-knowing - for items that you are not able to recall, how likely is it that you would recognise them from among a set of possible items? (strong FOK –> TOT phenomenon (tip of tongue))
- Confidence - for items that you recall/recognise, how likely is it that your answer is correct?
5
Q
Measures of control
A
- Self-paced study time
- Response time (recognition and recall)
- Quantity of information reported
- Verdicality/correspondence of info reported (how accurate
- Grain-size of info reported (level of detail)
6
Q
Examples of metacognition in action- studying for exams
A
- you have to decide when you have learned enough
- monitoring: Judgements of learning
- Control: amount of time spent studying
7
Q
Does monitoring work? can people accurately monitor cognitive processes?
A
Hart (1965) - FOK judgements predict performance on recognition test
Underwood (1966): EOL Judgements predict recall performance
8
Q
Hart (1965)
A
- FOK judgements
- Recall-judgement-recognition paradigm
- 50 or 75 general knowledge questions e.g. who wrote the ‘tempest’?
- If participant fails to recall an answer, could you recognise the answer if we present you with a set of possible answers? YES/NO or ratings 1-6
- Multiple-choice recognition test of all items.
- Is accuracy higher for ‘FK items’?
- FK items that they felt they would know, MC was higher than FK items participants felt they would not know
9
Q
Typical approaches to memory
A
- Typical laboratory paradigm
- Storehouse metaphor
10
Q
Typical-learning paradigm
A
List-learning experiment
- memory performance measured by % of items recalled/recognised
11
Q
Storehouse metaphor: quantity reiented
A
- Memory as a storage place
- Contents as discrete elements
- Memory assessed in input-bound manner - how much of the input is correctly remembered
- Forgetting is a loss of elements
12
Q
Correspondence approach to memory
A
Everyday memory research - e.g. eyewitness report of a crime. Memory performance measured by faithfulness to past event
Correspondence metaphor: accuracy-oriented
- memory is about past events
- focus on accuracy of report of original event
- content is important
- memory is assessed in output-bound manner - begin with output, examine accordance with input
- forgetting is a loss of correspondence
13
Q
Measuring memory accuracy - EXAMPLE
A
- Ss asked to study a list of words
- Later given a list of cue words and asked to write a word from the previously studied list.
- Performance: 8 word correct, 2 words incorrect, 2 blanks
- ‘Quantity’ = 8/12 = 67%
- ‘Accuracy’ = 8/10 = 80%
- The latter approach is more important for eyewitness memory, etc
14
Q
Recall-recognition paradox
A
What is it?
- Recognition is better in lab. Recall is better in eyewitness studies.
Why does it occur?
- Memory property under consideration: Accuracy vs. quantity
- Response option: Forced vs. free
- Test format: recall vs. recognition
15
Q
Koriat & Goldsmith
A
a model of monitoring and control in memory
found: test format (recall vs. recognition) - did affect quantity, did not affect accuracy (much) response option (free vs. forced) - did not affect quantity - did affect accuracy
16
Q
Implications for free-report memory performance
A
- Within the frame work, free-report memory performance depends on 4 factors:
1. overall retention - Amount of correct information that can be retrieved.
2. Monitoring effectiveness - the extent to which assessed probabilities differentiate correct vs. incorrect candidate answers
3. Control sensitivity - the extent to which volunteering or withholding information is based on monitoring output
4. Report criterion setting - Above which answers are volunteered
- below which answers are withheld
17
Q
Assumptions about memory performance
A
- At retrieval, we can’t change the amount of information in memory
- but we can enhance accuracy of report by choosing what to volunteer or withhold - Based on subjective confidence
- Reasonable but imperfect monitoring effectiveness –> quantity-accuracy tradeoff
18
Q
Quantity/accuracy trade off & monitoring effectiveness
A
….
19
Q
Basis for metacognitive judgements
A
- Direct Access View
- Inferential View
20
Q
Direct Access View
A
Judgements are made on the basis of features of the targets that can be accessed or retrieved.
- we can directly index the strength of our memory
21
Q
Inferential View
A
Judgements are based on a host of cues & clues.
- generated by act of learning or remembering
- from knowledge-specific information about own memory
- e.g. ease of retrieval.
22
Q
Problems with Direct Access
A
Implications of direct access
- Variables that increase memory strength should also increase metacognition judgements. e.g. allow longer study duration –> increase memory strength and JOL
- There should be no possibility that weakly stored information (as measured by actual later performance on a memory test) should be rated as being more recallable that strongly stored information
23
Q
Implications of the inferential view
A
- Metacognitive judgements are not based directly on memory traces - variables can affect memory and metamemory differently
- There may be situations where there is a dissociation between metacognitive judgements and memory performance
24
Q
Memory Implantation
A
- how might these false autobiographical memories occur in people?
Three step recipe
1. Convince people that the target event is plausible
2. Convince people that they experienced the target event
3. Have people reinterpret images and narratives as memories of target event
25
Gary, Manning, Loftus & Sherman (1996)
One study that has supported step 2: Imagination inflation
- imagining event --> inflated confidence that it occurred.
- Pre-tested Ss on confidence that childhood events had happened, e.g. broke a window with your hand.
2 weeks later:
- Asked Ss to imagine some of the (unlikely) events
- Including familiar places, people & things in imagined event
- Answered questions about the imagined event.. and then re-tested confidence
26
Gary, Manning, Loftus & Sherman (1996) results
- For most, confidence scores didn't change
- for the scores that did increase, the imagined ones changed more often
- Further research could should more imagining --> more confidence, more sensory modalities involved --> more confidence
27
Direct Access view vs. Inferential, who wins?
Inferential. People use other cues, like how quickly it comes to mind, not just drawing on direct access.
28
Metcalfe, Schwartz, & Joaquim (1993)
- Are FOK judgements based on cue familiarity or partial retrieval of target?
Experiment 2
- Paired associate task
study word pairs: cue-target (A-B), if cue is paired with more than one target, recall the last target
Test- given cue, try to recall target. For incorrect items, FOK judgement (1-100). Forced-choice recognition test.
29
Kelley & Lindsay
- Fluency & confidence - e.g. what was buffalo bill's last name? - Cody (correct), Hickock (related, incorrect) or Melville (unrelated)
- General knowledge questions, rate confidence
- Manipulated fluency by pre-exposure to answers
- Studied list of names before test: 1/3 correct answers, 1/3 related but incorrect, 1/3 unrelated fillers.
- Prior exposure increased : Rates of responding with those answers, speed of answers, confidence.
- similar effects were obtained when told that: list contained related but incorrect answers.. boosted fluency --> illusions of knowing. Caused incorrect answers to be retrieved quickly and believed to be accurate
30
Confidence as viewed by researchers
- Many findings indicate that confidence can be changed by factors unrelated to memory strength.
- Survey of eyewitness researchers-87% 'confidence is not a good predictor of accuracy' why?
- Most eyewitness studies use one set of stimulus materials for all witnesses (restricting variability in encoding and test conditions)
- Most eyewitness researchers have looked at the confidence accuracy relationship using point-biserial correlation
30
Confidence as viewed by researchers
- Many findings indicate that confidence can be changed by factors unrelated to memory strength.
- Survey of eyewitness researchers-87% 'confidence is not a good predictor of accuracy' why?
- Most eyewitness studies use one set of stimulus materials for all witnesses (restricting variability in encoding and test conditions)
- Most eyewitness researchers have looked at the confidence accuracy relationship using point-biserial correlation
31
Measuring the Confidence-Accuracy relationship
- point-biserial correlation
| - Confidence accuracy correlation
31
Measuring the Confidence-Accuracy relationship
- point-biserial correlation
| - Confidence accuracy correlation
32
Point-biserial correlation
Many studies just use point-biserial correlation, but what's the problem with this?
- cannot 'see' the full relationship between each level of confidence and accuracy
- the correlation provides no info about the extent to which participants over or underestimate the probability that their decisions were correct
32
Point-biserial correlation
Many studies just use point-biserial correlation, but what's the problem with this?
- cannot 'see' the full relationship between each level of confidence and accuracy
- the correlation provides no info about the extent to which participants over or underestimate the probability that their decisions were correct
33
A better way -
Confidence-Accuracy Calibration.
- plot subjective confidence (measure on 0-100% or 50-100% scale) against proportion correct
- Calibration statistic - ranges from 0 to 1, 0=perfect, 1= hopeless
- over/underconfidence statistic - ranges from -1 to +1, negative=underconfident
- Normalised resolution index - Ranges from 0 to 1, how we confidence discriminates correct from incorrect responses.
33
A better way -
Confidence-Accuracy Calibration.
- plot subjective confidence (measure on 0-100% or 50-100% scale) against proportion correct
- Calibration statistic - ranges from 0 to 1, 0=perfect, 1= hopeless
- over/underconfidence statistic - ranges from -1 to +1, negative=underconfident
- Normalised resolution index - Ranges from 0 to 1, how we confidence discriminates correct from incorrect responses.
34
Factors affecting calibration
- Decision type: positive vs. negative decisions
- Reflection: questionnaire --> reflect on encoding & ID test conditions
- Hypothesis Discornfirmation: produce reasons for incorrect decision
34
Factors affecting calibration
- Decision type: positive vs. negative decisions
- Reflection: questionnaire --> reflect on encoding & ID test conditions
- Hypothesis Discornfirmation: produce reasons for incorrect decision
35
Single-dimensional model & two-dimensional model - Predictions of models?
- The two models make different predictions about accuracy-confidence scatterplots.
- E.g. plot reflecting single dimension, points fall on one curve; overlapping lines. Two conditions just affect strength --> same accuracy & confidence scores in both conditions
- e.g. plot reflecting >1 dimension, 2 curves separated; dont overlap. Accuracy is determined only by strength --> same accuracy, confidence is determined by strength & certainty --> confidence higher for rehearsal
36
Experiment 3: Study-Test Luminance
- do 'bright' test conditions lead to more confidence that is warranted by the effect on accuracy?
- Study bright - test bright/ study dim - test dim/ study bright - test dim/ study dim - test bright. This might sometimes be the case for eyewitnesses
Results - Confidence increased in bright-test conditions more than was warranted
37
A model of monitoring and control in memory - We can manipulate - ?
Situational demands/pay offs - Report option & Accuracy incentive
- Assuming good monitoring effectiveness, free report allows increased accuracy with little/no cost to quantity
- interviewer can set accuracy incentive
38
Control - Witnesses can control the information they provide through:
- Control over report option: Volunteering information or saying 'dont know'
- Control over grain size: Fine detail vs. general information
39
Weber & Brewer (2008)
- Conflicting goals: truth vs. whole truth.
- Role of confidence in regulation of grain-size.
- Experiment 2 - Video of bank robbery, then 2 phases:
Phase 1 - e.g. what colour was the robbers hair? Fine: specific colour, coarse: shade (light/dark)
Phase 2: Choose one as their final response, or say 'I don't know'
- Ss could simultaneously regulate reporting & grain-size. Answers withheld in phase 2 were less accurate than those volunteered.
Extension of Koriat & Goldsmith model - If confidence in fine-grained candidate answer did not pass the report criterion, might retrieve less-detailed answer or withhold response.
40
Recommendations
Look at the trade off between accuracy and informativeness when comparing interview techniques.
Most benefit comes from allowing control over report option and grain-size
41
Metcalfe & Kornell (2005)
Control of study time
-Region-of-proximal-learning-framework
First eliminate items believed to be mastered --> usually negative correlation between JOL & study time.
Then people study easiest items then harder items.
Support of this framework.
- translate english word --> spanish
- studied word pairs, JOLs, initial recall test
- for incorrect items, could choose some for restudy
- final test
- -> when eliminated items that Ss already knew, people chose to restudy items with high JOL (i.e. easiest remaining items)
42
Metacognitive Beliefs
- Insufficiently sensitive to: Benefit of spaced study.. cramming (massed study)
43
Kornell & Bjork (2008)
- Self-regulated study, flash cards
- when to 'drop' cards for info thought to be learned.
- depends on effective monitoring and control strategy
- Dropping items --> spend more time on other items. But dropping items...
- prevents revisiting items, some may actually be in the 'region of proximal learning'
- decreases space of repetitions for remaining cards
Experiment 1 - flawed decision about not revisiting items. Studied 2 lists of English-Swahili translations, 10 mins on each list. Cycled through flash cards, could drop cards for one list. Cued recall test, immediately or after 1 week.
- found: accuracy was impaired by allowing participants to drop cards!
Experiment 2 - impaired accuracy was not due to poor monitoring. Some Ss made JOLs whenever they dropped an item. Poor decisions about dropping - dropped items they thought they may not remember at test. UNDERWEIGHTED benefit of additional study & spacing of study trials.
Experiment 3 - Explicitly tested the number of times Ss correctly recalled an item before dropping. Poor decisions about dropping. Dropping items after 1 correct response - UNDERWEIGHTED benefit of additional study and spacing of study trials. Accuracy increased with more correct recalls before dropping.
44
What circumstances ensure testing will be diagnostic of criterion performance?
- Delay after initial learning phase - 'delayed JOL effect'
- If students are trained to assess memory with delayed tests of the to-be-remembered info, they may be better at regulating self-paced study.
45
Mitchum & Kelly (2010) Strategies in RAPM
- Constructive Matching - Look at problem, try to think of answer then inspect alternatives.
- Response elimination: look at response options and try to eliminate incorrect alternatives.
- using different strategies should lead to different cues being available for monitoring
- constructive matching = more and different quality cues -> enhance monitoring
46
Mitchum & Kelly (2010) experiment 1
- 55 participants completed RAPM - first screen showed matrix only - second screen showed matrix and response options
- Rated confidence after each item
- Post-experiment questionnaire: Indicated which strategy they used.
RESULTS
- constructive matching (i.e. time spent examining matrix) - weakly related to task performance
- confidence for those relying more heavily on constructive matching was better calibrated.
Experiment 2 - one group could use own strategy, other group had to generate an alternative. RAPM task performance was the same. Better calibration with constructive strategy.
