Confidence in value-based choice

Confidence in value-based choice De Martino Fleming Garrett Dolan

Tag layer of the DHSS Meta-Tagging Project. The article text is reproduced from the staged copy named in sourceDesc and is not altered; the tags are held separately in standOff so that overlapping and nested spans are preserved.

Confidence in value-based choice 2013 Nature Neuroscience 10.1038/nn.3279 De-Martino2013-confidence-in-value-based-choice De-Martino2013-confidence-in-value-based-choice.txt, the project's staged copy f7b30141bb8dafe530cb5765f3a718888336f5e8 not-redistributable neuroscience

the paper's own words

the tag layer; a sentence written by the tagger about the paper, always anchored to a quotation but not itself a quotation

0 spans located by searching the stored text for the quote's exact words, allowing any whitespace between them how subjective confidence in a choice relates to the value comparison that produced it, in behaviour and in the brain Quotation withheld because this record is not cleared for redistribution. whether confidence is a read-out of noise in the value comparison, separable from the value difference itself Quotation withheld because this record is not cleared for redistribution. dissociates confidence from value difference, shows a race model explains their joint relation to accuracy and reaction time, locates both in vmPFC, and predicts individual metacognitive ability from vmPFC-rRLPFC coupling Quotation withheld because this record is not cleared for redistribution. mechanism Quotation withheld because this record is not cleared for redistribution. decision confidence in value-based choice Quotation withheld because this record is not cleared for redistribution. fmri fmri Quotation withheld because this record is not cleared for redistribution. mechanisms Confidence arises because evidence for each option accumulates stochastically during value comparison, so trial-to-trial noise makes the final gap between accumulators (and hence confidence, speed and accuracy) vary even at fixed initial value difference. Quotation withheld because this record is not cleared for redistribution. mechanisms Inter-trial noise in the comparison process drives the joint pattern of confidence, reaction time and accuracy. Quotation withheld because this record is not cleared for redistribution. mechanisms Confidence computed in vmPFC becomes reportable by being read out by a hierarchically distinct second-order network in rRLPFC. Quotation withheld because this record is not cleared for redistribution. mechanisms Noise added at the read-out stage degrades the correspondence between reported confidence and actual accuracy, producing individual differences in metacognition. Quotation withheld because this record is not cleared for redistribution. in the race model confidence is the gap between the two accumulators when one crosses threshold, so it tracks trial-to-trial noise rather than the starting value difference Quotation withheld because this record is not cleared for redistribution. the same noise makes a trial faster, more accurate and more confident at once, which is why the three covary without any of them causing the others Quotation withheld because this record is not cleared for redistribution. because confidence follows the stochastic path rather than the initial difference, the model itself explains why value and confidence come apart Quotation withheld because this record is not cleared for redistribution. a second-order region reads confidence out of the comparison process, and how faithfully it does so varies between people Quotation withheld because this record is not cleared for redistribution. Only a small, experimentally controllable subset of the factors that could influence confidence was measured. what this limits: the account of what drives confidence: WE REPORT A LIMITED SET OF THESE FACTORS ... FOR WHICH WE COULD EXERCISE GOOD EXPERIMENTAL CONTROL Quotation withheld because this record is not cleared for redistribution. The design cannot adjudicate between confidence emerging in vmPFC and confidence being computed elsewhere and fed into vmPFC; that would need a high-temporal-resolution method such as MEG. what this limits: the direction of the vmPFC-rRLPFC relationship: RESOLVING THIS POSSIBILITY IS BEYOND THE DESIGN OF THE CURRENT STUDY AND WILL REQUIRE OTHER TECHNIQUES WITH HIGH TEMPORAL RESOLUTION Quotation withheld because this record is not cleared for redistribution. Alternative functional mappings from model confidence to reported confidence were not explored. what this limits: the model-to-report mapping: OTHER FUNCTIONAL FORMS FOR THE LINK BETWEEN MODEL CONFIDENCE AND REPORTED CONFIDENCE ARE POSSIBLE, BUT WE DO NOT INVESTIGATE THESE HERE Quotation withheld because this record is not cleared for redistribution. the authors state that only a limited set of the factors affecting confidence could be controlled what this limits: the same scope as limitations[0] - other internal and external factors ARE LIKELY TO AFFECT SUBJECTIVE CONFIDENCE and were not measured Quotation withheld because this record is not cleared for redistribution. the direction of the vmPFC-rRLPFC relation is not resolvable by fMRI; confidence could be computed elsewhere and imported what this limits: the same scope as limitations[1], with the alternative stated: confidence may be CODED ELSEWHERE ... AND IS THEN COMMUNICATED TO vmPFC Quotation withheld because this record is not cleared for redistribution. 8 of 28 participants were excluded, including one for not varying their confidence ratings enough to estimate the measure of interest what this limits: the sample every metacognitive estimate rests on, which is conditioned on the measure itself: PARTICIPANTS WERE ONLY INCLUDED IF THEY USED A SUFFICIENT RANGE OF CONFIDENCE RATINGS (STANDARD DEVIATION > 0.8) TO ALLOW ESTIMATION OF METACOGNITIVE ABILITY Quotation withheld because this record is not cleared for redistribution. value is measured after scanning by a separate bidding task, so choice and value come from different sessions what this limits: any joint use of choice and value: value was elicited AT THE END OF THE SCANNING SESSION by a separate BDM bidding task, so the two come from different sessions and instruments Quotation withheld because this record is not cleared for redistribution. the BDM incentive-compatible auction, used to elicit each participant's willingness-to-pay as an objective measure of item value Quotation withheld because this record is not cleared for redistribution. the idea that confidence is a second-order metacognitive judgement measurable separately from the choice itself, imported into value-based choice Quotation withheld because this record is not cleared for redistribution. the second-order framing of confidence and the requirement to measure it separately from choice Quotation withheld because this record is not cleared for redistribution. the BDM auction, used to make stated values incentive-compatible Quotation withheld because this record is not cleared for redistribution. the race model, imported from perceptual choice into value-based choice Quotation withheld because this record is not cleared for redistribution. each item pair was presented twice with the left-right position counterbalanced eliminates: that a choice or a confidence rating is driven by which side of the screen an item appeared on · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. confidence for repeated pairs was correlated across repetitions with the value difference partialled out eliminates: that apparent confidence stability is just the value difference reappearing - that subjects are re-reading the same easy choice rather than holding a stable confidence · outcome (stated by the tagger, no quotation): held Quotation withheld because this record is not cleared for redistribution. confidence here is NOT risk: every option has a known outcome, so this is certainty about having picked the better item, not uncertainty about what will happen Quotation withheld because this record is not cleared for redistribution. choice confidence (second order, about the comparison) versus bid confidence (first order, about each item's value) - two senses of one word, measured separately and adjudicated by model comparison Quotation withheld because this record is not cleared for redistribution. metacognitive accuracy is operationalised as the CHANGE in logistic slope between low and high confidence trials, not as calibration or correlation Quotation withheld because this record is not cleared for redistribution. the slope of the logistic fit is read as choice accuracy AND as noise in the choice process - one number, two descriptions, used interchangeably Quotation withheld because this record is not cleared for redistribution. value here means willingness-to-pay elicited by a BDM auction after scanning, not a rating and not a preference order Quotation withheld because this record is not cleared for redistribution. claim_type: mechanism a second reading reaching the same value. Recorded: mechanism — note: a second reading reaching the same value from a different sentence. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. method: fMRI during binary snack choice with trial-by-trial confidence report, values elicited afterwards by an incentive-compatible bidding procedure, logistic model comparison and a race model a second reading DIFFERING from the recorded value. Recorded: fmri — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. phenomenon: decision confidence: subjective certainty in having chosen the better of two options a second reading DIFFERING from the recorded value. Recorded: decision confidence in value-based choice — note: a second reading, verbatim-verified, disagreeing with the value on the record. Not applied - a person decides. Quotation withheld because this record is not cleared for redistribution. builds-on → the race model of evidence accumulation and decision confidence (refs 7, 21) Quotation withheld because this record is not cleared for redistribution. builds-on → Insabato et al.'s two-layer neural network account of confidence read-out (ref 18) Quotation withheld because this record is not cleared for redistribution. extends → earlier studies that defined decision confidence in terms of the factors governing choice (rat orbitofrontal firing rates, ref 7; human vmPFC fMRI, ref 8) Quotation withheld because this record is not cleared for redistribution. critiques → earlier studies that defined confidence through the factors governing choice, which cannot separate confidence from value differs in: THE AXIS IS HOW CONFIDENCE IS DEFINED. 'Quotation withheld' The earlier work is not wrong about what it measured - its definition makes one question unaskable, which is a critique of scope rather than of result. Quotation withheld because this record is not cleared for redistribution. extends → the perceptual decision-making tradition of measuring confidence independently of choice Quotation withheld because this record is not cleared for redistribution. extends → race models of decision confidence, applied here to value-based rather than perceptual choice Quotation withheld because this record is not cleared for redistribution. supports → the account of vmPFC as a dynamic accumulator of choice values Quotation withheld because this record is not cleared for redistribution. supports → two-layer architectures in which a second-order network reads out first-order performance Quotation withheld because this record is not cleared for redistribution. generalises → rRLPFC's role in metacognition from perceptual to value-based decisions Quotation withheld because this record is not cleared for redistribution. confidence —distinct-from→ risk confidence risk strength of the relation: 1 · note: CRITERION, stated: every option here determines a known outcome, so there is nothing to be risky about Quotation withheld because this record is not cleared for redistribution. choice confidence —distinct-from→ bid confidence choice confidence bid confidence strength of the relation: 1 · note: CRITERION, and it is TESTED: model comparison shows the second-order measure modulates choice and the first-order one does not Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes '2nd level confidence' and '1st level confidence in the item values', not 'choice confidence' / 'bid confidence'. Quotation withheld because this record is not cleared for redistribution. confidence —operationalised-as→ accumulators confidence accumulators strength of the relation: 1 · note: in the model confidence IS the gap between accumulators at threshold - a definition, not a correlate Audit 2026-09-28: DOUBTFUL - read against its span and kept, but confidence was modelled in a race model of accumulators; operationalised-as is loose. Quotation withheld because this record is not cleared for redistribution. metacognitive accuracy —operationalised-as→ logistic regression metacognitive accuracy logistic regression strength of the relation: 1 · note: the change in slope between low and high confidence trials; a different measure from calibration or meta-d prime Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'slope of the logistic fit', not 'logistic regression'. Quotation withheld because this record is not cleared for redistribution. willingness-to-pay —operationalised-as→ BDM willingness-to-pay BDM strength of the relation: 1 Quotation withheld because this record is not cleared for redistribution. difference in value —operationalised-as→ willingness-to-pay difference in value willingness-to-pay strength of the relation: 1 Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span writes 'value for individual items', not 'difference in value'. Quotation withheld because this record is not cleared for redistribution. noise in the choice process —operationalised-as→ logistic regression noise in the choice process logistic regression strength of the relation: 0.8 · note: the SAME slope is called choice accuracy and noise in the same sentence; a record joining on either word joins on one number Quotation withheld because this record is not cleared for redistribution. EVIDENCE CORRECTED (RE-GROUNDED) · confidence —distinct-from→ difference in value confidence difference in value EVIDENCE CORRECTED: RE-GROUNDED. The old span reported a variance figure without naming confidence. This sentence names both and separates them in the paper's own construction - the same region 'NOT ONLY ... BUT ALSO the confidence'. — strength of the relation: 1 · note: CRITERION: value difference explains only 17.7% of confidence variance, and that residual is what the whole design rests on Quotation withheld because this record is not cleared for redistribution. confidence —causes→ reversed confidence reversed status: PREDICTIVE, NOT DEMONSTRATED CAUSAL: 'initial confidence was a significant negative predictor of choice reversal' in a logistic regression. A regression coefficient is not an intervention. · strength of the relation: 0.8 · note: SIGN: negative. Lower initial confidence predicts changing your mind on the repeat Audit 2026-09-28: DOUBTFUL - read against its span and kept, but confidence negatively predicted reversal; the edge carries no direction. Quotation withheld because this record is not cleared for redistribution. logistic regression logistic regression Quotation withheld because this record is not cleared for redistribution. Bayesian information criterion Bayesian information criterion Quotation withheld because this record is not cleared for redistribution. likelihood ratio test likelihood ratio test Quotation withheld because this record is not cleared for redistribution. ANOVA ANOVA Quotation withheld because this record is not cleared for redistribution. t-test t-test Quotation withheld because this record is not cleared for redistribution. partial correlation partial correlation Quotation withheld because this record is not cleared for redistribution. general linear model general linear model Quotation withheld because this record is not cleared for redistribution. multiple-comparison correction multiple-comparison correction Quotation withheld because this record is not cleared for redistribution. standard error of the mean standard error of the mean Quotation withheld because this record is not cleared for redistribution. autocorrelation autocorrelation Quotation withheld because this record is not cleared for redistribution. Bayesian information criterion Bayesian information criterion Quotation withheld because this record is not cleared for redistribution. likelihood ratio test likelihood ratio test Quotation withheld because this record is not cleared for redistribution. logistic regression logistic regression Quotation withheld because this record is not cleared for redistribution. partial correlation partial correlation Quotation withheld because this record is not cleared for redistribution. family wise error family wise error Quotation withheld because this record is not cleared for redistribution. one-sample t-test one-sample t-test Quotation withheld because this record is not cleared for redistribution. general linear model general linear model Quotation withheld because this record is not cleared for redistribution. Subjective confidence was only weakly related to the objective difference in value between the two options, so the two could be dissociated. Quotation withheld because this record is not cleared for redistribution. Trials on which participants reported higher confidence showed higher choice accuracy, i.e. a steeper logistic slope relating value difference to choice. Quotation withheld because this record is not cleared for redistribution. Reaction time showed main effects of both value difference and confidence, with no interaction. Quotation withheld because this record is not cleared for redistribution. The regression model in which choice confidence modulates the effect of value difference on choice fitted better than models using confidence in the individual item bids. Quotation withheld because this record is not cleared for redistribution. The confidence-accuracy relationship was stable across scanning sessions and was not explained by trial-to-trial fluctuations in attention. Quotation withheld because this record is not cleared for redistribution. Choices made with low initial confidence were more likely to be reversed when the same item pair was presented again. Quotation withheld because this record is not cleared for redistribution. The same vmPFC region carried both a value-difference signal and a confidence signal. Quotation withheld because this record is not cleared for redistribution. Value and confidence had additive but non-interacting effects on vmPFC activity. Quotation withheld because this record is not cleared for redistribution. Right rostrolateral prefrontal cortex encoded reported confidence but not the difference in value. Quotation withheld because this record is not cleared for redistribution. The strength of vmPFC-rRLPFC functional coupling predicted how well a participant's confidence tracked their own accuracy. Quotation withheld because this record is not cleared for redistribution. value difference accounted for only 17.7% of the variance in confidence Quotation withheld because this record is not cleared for redistribution. high confidence trials showed a steeper value-to-choice slope, so confidence tracks accuracy Quotation withheld because this record is not cleared for redistribution. value difference and confidence both had main effects on reaction time, with no interaction Quotation withheld because this record is not cleared for redistribution. the model in which choice confidence modulates value difference beat four rivals on BIC Quotation withheld because this record is not cleared for redistribution. that model beat the canonical value-only model in 19 of 20 participants individually Quotation withheld because this record is not cleared for redistribution. 14.7% of choices reversed on repeat, and reversed choices had had markedly lower initial confidence Quotation withheld because this record is not cleared for redistribution. vmPFC activity was modulated by both value difference and confidence Quotation withheld because this record is not cleared for redistribution. rRLPFC tracked confidence but not value difference Quotation withheld because this record is not cleared for redistribution. vmPFC-rRLPFC coupling scaled with confidence and predicted metacognitive accuracy across people Quotation withheld because this record is not cleared for redistribution. LOAD-BEARING NULL: no value-by-confidence interaction in vmPFC, which is what supports their theoretical independence Quotation withheld because this record is not cleared for redistribution. CONTROL NULL: no effect of session and no session-by-confidence interaction, so the relation is stable across the hour Quotation withheld because this record is not cleared for redistribution. CONTROL NULL: confidence was not predicted by the preceding five trials' confidence, so it is not drifting attention Quotation withheld because this record is not cleared for redistribution. CONTROL NULL: mean confidence did not differ between first and second presentation of the same pair Quotation withheld because this record is not cleared for redistribution. CONTROL NULL: the vmPFC confidence response was not a categorical response to errors Quotation withheld because this record is not cleared for redistribution. confidence for a repeated item pair correlated with itself after partialling out value difference, in 19 of 20 subjects Quotation withheld because this record is not cleared for redistribution. DECLARED NOT EVIDENCE: the vmPFC bar plot is shown to clarify the pattern, with inference carried out in a separate model Quotation withheld because this record is not cleared for redistribution. DECLARED NOT EVIDENCE: the between-subject scatter plot is illustrative only and not the basis of the statistic Quotation withheld because this record is not cleared for redistribution. choice confidence choice confidence Quotation withheld because this record is not cleared for redistribution. difference in value (DV) difference in value (DV) Quotation withheld because this record is not cleared for redistribution. metacognitive accuracy metacognitive accuracy Quotation withheld because this record is not cleared for redistribution. psychophysiological interaction (PPI) psychophysiological interaction (PPI) Quotation withheld because this record is not cleared for redistribution. metacognitive noise metacognitive noise Quotation withheld because this record is not cleared for redistribution. confidence confidence Quotation withheld because this record is not cleared for redistribution. race model race model Quotation withheld because this record is not cleared for redistribution. metacognitive accuracy metacognitive accuracy Quotation withheld because this record is not cleared for redistribution. BDM BDM Quotation withheld because this record is not cleared for redistribution. second-order second-order Quotation withheld because this record is not cleared for redistribution. signed difference in value signed difference in value Quotation withheld because this record is not cleared for redistribution. choice reaction time choice reaction time Quotation withheld because this record is not cleared for redistribution. Signal in vmPFC Signal in vmPFC Quotation withheld because this record is not cleared for redistribution. change in slope change in slope Quotation withheld because this record is not cleared for redistribution. choice confidence choice confidence Quotation withheld because this record is not cleared for redistribution. difference in value difference in value Quotation withheld because this record is not cleared for redistribution. reaction time reaction time Quotation withheld because this record is not cleared for redistribution. metacognitive accuracy metacognitive accuracy Quotation withheld because this record is not cleared for redistribution. 17.7% — share of the variance in confidence ratings explained by the unsigned value difference, showing the two are largely independent Quotation withheld because this record is not cleared for redistribution. 20 — number of participants whose fMRI and behavioural data entered the final analysis Quotation withheld because this record is not cleared for redistribution. 340 — total binary food choices each participant made in the scanner, across 4 sessions Quotation withheld because this record is not cleared for redistribution. 19 out of 20 — participants for whom the confidence-modulated choice model beat the value-only model by a likelihood ratio test Quotation withheld because this record is not cleared for redistribution. Model 1: 5424; Model 2: 4995; Model 3: 5388; Model 4: 5498; Model 5 5291 — BIC scores of the five competing logistic choice models, lowest (best) for the confidence-modulated Model 2 Quotation withheld because this record is not cleared for redistribution. 14.7 % — proportion of choices that were reversed when the identical item pair was shown a second time Quotation withheld because this record is not cleared for redistribution. F (1,19) = 39.75; P < 0.0001 — main effect of confidence on choice accuracy when the logistic fit was split by scanning session Quotation withheld because this record is not cleared for redistribution. 210.6 ± 72.4 SD; 340.2 ± 53.5 SD — mean confidence (arbitrary units) on choices later reversed versus choices later repeated Quotation withheld because this record is not cleared for redistribution. F (1,19) = 5.1, p<0.05; F (1,19) = 7.6, p<0.05; F (1,19) = 0.7, p>0.5 — vmPFC main effects of value and of confidence with no value-by-confidence interaction Quotation withheld because this record is not cleared for redistribution. 17.7% — variance in confidence explained by unsigned value difference - the residual is what the design exploits Quotation withheld because this record is not cleared for redistribution. 14.7% — choices reversed when the same pair reappeared Quotation withheld because this record is not cleared for redistribution. 19 of 20 — participants for whom the confidence-modulated model beat the value-only model Quotation withheld because this record is not cleared for redistribution. 28 down to 20 — participants recruited against participants analysed, after four exclusion rules Quotation withheld because this record is not cleared for redistribution. standard deviation > 0.8 — the spread of confidence ratings a participant had to show to be included at all Quotation withheld because this record is not cleared for redistribution. 340 — choices per participant, each pair seen twice in mirrored positions Quotation withheld because this record is not cleared for redistribution.

The full text of this article is not redistributable, so it is not included here. The evidence sentences are withheld with it; the tag layer and its grounding status remain. The article itself is at https://doi.org/10.1038/nn.3279.