nature neurOSCIenCe VOLUME 16 | NUMBER 1 | JANUARY 2013 1 0 5
a r t I C l e S
The subjective confidence we have in our decision-making, and that
of others, has far-reaching consequences. For example, the recom-
mendations of a financial advisor who expresses high confidence in
a particular investment option will carry more weight than one who
is ambivalent. An expression of doubt in or caution concerning a par-
ticular course of action can lead one to question or revisit a previous
decision. Previous work has established that the vmPFC has a central
role in computing the value of potential choice options
1–5
, with activ-
ity in this region reflecting the dynamic evolution of a value com-
parison
6
. However, this work has focused exclusively on the choice
process, without considering the subject’s level of confidence in the
decision. Consequently, it is unknown how a process of value com-
parison, instantiated in vmPFC, relates to subjective confidence.
Previous studies have reported neural correlates of decision con-
fidence in brain regions associated with a value representation. For
example, firing rates in rat orbitofrontal cortex
7
and functional
magnetic resonance imaging (fMRI) signal in human vmPFC
8
show
graded changes as perceptual decisions become more difficult.
However, as these studies delineate confidence in terms of factors gov-
erning choice, they are unable to tease apart the relationship between
trial-to-trial subjective confidence and decision value. In contrast, the
field of perceptual decision-making has noted that confidence can
be measured independently of the choice process itself
9,10
, where it
is conceptualized as reflecting a ‘second-order’ metacognitive evalu-
ation. Critically, dissociating confidence from other features of the
decision process requires acquisition of separate measures of choice
and confidence
11
.
Here we implement such an approach to dissociate value and confi-
dence during decision-making and to identify their respective neural
substrates. We collected trial-by-trial estimates of decision confidence
while healthy volunteers chose between pairs of snack items. We also
measured the subjective value of each snack item by means of a standard
incentive-compatible bidding procedure. This allowed us to dissociate
confidence from value, and in so doing provide evidence that confi-
dence reflects an assessment of choice accuracy.
To explore systematic relationships between confidence, accuracy,
choice and reaction time, we modeled our data using a variant of a
race model
7,12
(one of a larger class of dynamic models of decision-
making
13
). This model predicts that subjective confidence reflects
the stochastic accumulation of evidence during the value comparison
process. As is consistent with this prediction, we show that the same
anatomical region in ventromedial prefrontal cortex (vmPFC) not only
reflects a difference in value between available options, but also the
confidence associated with a value comparison process. Finally, we
show that individual differences in participants’ abilities to relate con-
fidence to decision performance is linked to increased functional con-
nectivity between vmPFC and rostrolateral prefrontal cortex (RLPFC),
a region previously shown to function in metacognitive appraisal
14
.
RESULTS
We scanned twenty hungry participants while they made choices
between food items that they could consume later (Fig. 1a). After
making each choice, participants reported the degree of confidence in
their decision (choice confidence). Note that confidence, or certainty,
in the present study is conceptually distinct from risk, in that each
choice determined a known outcome. Confidence here reflects the
degree of subjective certainty in having made the best choice, which
equates to choosing the higher valued item. To establish value for indi-
vidual items, we asked participants at the end of the scanning session
to place a bid for each food item using a standard incentive-compat-
ible procedure, the Becker-DeGroot-Marschak (BDM) mechanism
15
.
BDM is widely used in behavioral economics and neuroeconomics
to elicit nonstrategic reservation prices, also known as willingness-
to-pay. In this phase subjects were required to state their maximum
willingness-to-pay for each food item (see Online Methods). Several
studies have shown that this mechanism reliably elicits goal values
1
Psychology and Language Sciences, University College London, London, UK.
2
Wellcome Trust Center for Neuroimaging, at University College London, UK.
3
Division
of the Humanities and Social Sciences, California Institute of Technology, Pasadena, California, USA.
4
Center for Neural Science, New York University, New York, New
York, USA.
5
Department of Experimental Psychology, University of Oxford, Oxford, UK.
6
These authors contributed equally to this work. Correspondence should be
addressed to B.D.M. (b.martino@ucl.ac.uk) or S.M.F. (fleming.sm@gmail.com).
Received 6 August; accepted 15 November; published online 9 December 2012; doi:10.1038/nn.3279
Confidence in value-based choice
Benedetto De Martino
1–3,6
, Stephen M Fleming
2,4–6
, Neil Garrett
1
& Raymond J Dolan
2
Decisions are never perfect, with confidence in one’s choices fluctuating over time. How subjective confidence and valuation of
choice options interact at the level of brain and behavior is unknown. Using a dynamic model of the decision process, we show that
confidence reflects the evolution of a decision variable over time, explaining the observed relation between confidence, value, accuracy
and reaction time. As predicted by our dynamic model, we show that a functional magnetic resonance imaging signal in human
ventromedial prefrontal cortex (vmPFC) reflects both value comparison and confidence in the value comparison process. Crucially,
individuals varied in how they related confidence to accuracy, allowing us to show that this introspective ability is predicted by a
measure of functional connectivity between vmPFC and rostrolateral prefrontal cortex. Our findings provide a mechanistic link between
noise in value comparison and metacognitive awareness of choice, enabling us both to want and to express knowledge of what we want.
npg
© 2013 Nature America, Inc. All rights reserved.
1 0 6 VOLUME 16 | NUMBER 1 | JANUARY 2013 nature neurOSCIenCe
a r t I C l e S
that are used by the decision maker to guide choice
16–18
. Participants
also provided a rating of their confidence in each bid (bid confi-
dence). Participants’ bids for the leftmost items were then subtracted
from bids for the rightmost items to calculate a signed difference
in value (DV) between each pair of items, which was then entered
into a logistic regression to predict the probability that the subject
chose the rightmost item on each trial (Fig. 1b). In line with previous
studies
2,19
, we found that DV was a reliable predictor of participants’
choices, with the slope of the logistic regression being a measure of
choice accuracy, or noise in the choice process
20
.
Choice, confidence and reaction time
Unsigned |DV| only accounted for an average of 17.7% of the vari-
ance in participants’ confidence ratings (r = 0.42 ± 0.19, s.d.). This
partial independence between confidence and |DV| allowed us to ask
whether confidence reflects changes in choice accuracy (the selec-
tion of items with higher subjective value). By splitting our logistic
regression fit into high- and low-confidence trials, we showed that
higher confidence was consistently associated with increased choice
accuracy (Fig. 1b,c and Supplementary Fig. 1). This effect of confi-
dence on choice was also reflected in reaction time (RT), with main
effects of both |DV| and confidence (both P < 0.001) but no interac-
tion (
Fig. 1d). The three-way relationship between |DV|, confidence
and RT is plotted in Figure 1e. We recognize that other factors (inter-
nal and external) besides |DV| and RT are likely to affect subjective
confidence. We report a limited set of these factors (Supplementary
Table 1) for which we could exercise good experimental control.
Using logistic regression, we next compared models of the interaction
between confidence and value comparison. Choice confidence, unlike
DV, is in itself not a predictor of choice (right or left item) but instead
refers to accuracy of the decision. We thus expected choice confidence
to modulate the link between DV and choice. Model 1 predicted choice
using DV alone; model 2 included choice confidence (that is, confi-
dence at the decision time) as a modulator of DV (DV × confidence);
models 3–5 examined whether bid confidence (that is, confidence at
the bid time) could explain additional variance in the link between DV
and choice (see Online Methods). In accordance with our predictions,
model 2 provided a better account (that is, lower Bayesian information
criterion (BIC)) of participants’ choices than the other four models
(Fig. 2a), as shown by the difference in BIC relative to model 2: model
1, 214.6; model 3, 196.2; model 4, 251.7; model 5, 111.9. Furthermore,
model 2 was a better fit than the canonical model 1 in 19 of 20 par-
ticipants as assessed by a likelihood ratio test (
α
= 0.05). This analysis
confirms that a critical modulator of choice accuracy is second-order
confidence arising in the context of the comparison process (model 2)
as opposed to first-order confidence in the item values (models 3–5).
Stability of confidence over time
We next examined whether the relationship between confidence and
choice was stable over time. Splitting the logistic regression analysis into
separate sessions revealed a robust main effect of confidence (F
1,19
=
39.75; P < 0.0001) but a nonsignificant main effect of session (F
3,57
=
0.3; P = 0.7) and a lack of interaction between session and confidence
(F
3,57
= 0.13; P = 0.9; Supplementary Fig. 2). To examine whether local
fluctuations in attention affected confidence, we constructed a serial
autocorrelation regression model that predicted the current confidence
rating from the confidence ratings given on the immediately preceding
five trials, in addition to |DV|. None of the autocorrelation coefficients
reached group-level significance (all t < 1.2, P > 0.27). Together these
results indicate that confidence is a stable predictor of choice accuracy
and that it does not reflect local changes in attention.
As each item pairing was presented twice (once in each spatial con-
figuration), it was also possible to examine the relationship between
confidence ratings given for identical choice pairs. As confidence is
partly determined by absolute difference in value (|DV|, which does
not vary across choice pairs), we expected some stability purely driven by
DV. Thus, to address this question, we computed the partial correlation
between first and second confidence ratings, controlling for DV. There
was no significant difference between mean confidence ratings for the
first and second presentations of the same item pairs (t
19
= −0.64, P =
0.53). For 19 of 20 subjects, there was a significant partial correlation (P <
0.05) between confidence ratings for repeated item pairs after controlling
for the influence of |DV|, indicating stability in confidence for judgments
of particular item pairs that cannot be accounted for by |DV| alone.
Choice
a b
c d e
*
£0
10
***
**
**
***
1.2
1
0.5
0
–0.5
Confidence
(mean z-score)
2
3
4
1.0
0
8
6
Slope
RT (s)
4
2
Low High
Low High Low High
1 2
Reaction time (quantile)
3 4
DV
Confidence
0
£1 £2 £3
fMRI taskPost-scan task
Confidence rating
Confidence rating
Low confidence
High confidence
1
2.5 s
Bid (BDM)
3.5 s
2 3 4 5 6
1 2 3 4 5 6
1.0
0.5
P(chose right item)
0
–2 –1 0
Value right item–value left item
1 2
Low confidence
High confidence
All data
DV
(quantile)
Figure 1 Task and behavioral results. (a) fMRI
task (top): subjects were presented with a choice
between two snacks and were then required to
choose (2.5 s) one item to consume at the end
of the experiment. After each choice, subjects
indicated their level of confidence in having
made a correct decision (choice confidence).
Post-scanning task (bottom): subjects were
presented with each item individually and had
to submit a bid to buy each item. After each bid,
they were asked to rate their level of confidence
in having provided a correct bid price (bid
confidence). (b) Probability of choosing the
item on the right as a function of DV (that is, bid
price) between the two items (logistic fit) for an
exemplar subject (see Supplementary Fig. 2 for
all individual subjects). Dashed line, all choices;
black line, low-confidence choices; gray line,
high-confidence choices. The red double-headed
arrow indicates the increase in choice accuracy
(change in slope) for high- versus low-confidence
trials used in the between-subjects analyses
(Figs. 4b and 5b). (c) The slope of the logistic
fit is systematically higher (sharper) in high-
confidence compared to low-confidence trials (**P < 0.005; ***P < 0.0001). (d) Average RT as a function of confidence and |DV|. (e) Heat map showing
mean z-scored confidence (color bar) across subjects, as a function of subject-specific |DV| and RT quantiles. Error bars represent s.e.m.
npg
© 2013 Nature America, Inc. All rights reserved.
nature neurOSCIenCe VOLUME 16 | NUMBER 1 | JANUARY 2013 1 0 7
a r t I C l e S
We then examined whether choices were stable over time. On aver-
age, 14.7% of choices (± 5.7% s.d.) were reversed on the second pres-
entation. Choices that would be subsequently reversed were associated
with significantly lower initial confidence than those that would sub-
sequently be repeated (in arbitrary units: reversal confidence = 3.11
± 0.72 (s.d.); repetition confidence = 4.40 ± 0.54 (s.d.); t
19
= 12.1,
P < 10
−10
). In a logistic regression model predicting subsequent
reversal from both |DV| and initial confidence, initial confidence
was a significant negative predictor of choice reversal (mean stand-
ardized regression coefficient −0.99 ± 0.40 (s.d.); one-sample
t-test t
19
= −11.2, P < 10
−9
). These data support a hypothesis that low
confidence is associated with subsequent changes of mind.
Race model
Our best-fitting regression model suggested
that confidence reflects accuracy in a value
comparison. This led us to explore in more
detail the precise mechanism by which confi-
dence and value interact during the decision
process. We adapted a race model
12,21
wherein
evidence in favor of each of the options (the
snacks presented on the left and right sides
of the screen) is accumulated over time and
the decision is made on the basis of the first
option to reach a threshold (
Fig. 2b). In this
model, confidence is defined as the absolute
difference between the two accumulators at
decision time (∆e). Such a model predicts
that when ∆e is large, then choice accuracy
is increased, reflected by a sharper slope in
the logistic regression (Fig. 2c). Thus, the
race model neatly accounts for an increase in
choice accuracy we observed behaviorally in
the high-confidence condition (Fig. 1b and
Supplementary Table 2). Furthermore this
model predicts a decrease in RT when either
|DV| or ∆e are increased (Fig. 2d), as seen in
the behavioral data (Fig. 1d). The intuition is
that, even within a particular level of initial DV,
inter-trial noise in the value comparison process results in some trials
having greater final DV values (higher confidence) than others. Such
decisions will tend to be made more quickly, be more accurate and be
associated with higher confidence (Fig. 2e). Indeed, this predicted rela-
tionship among RT, |DV| and confidence closely matched the behavioral
data (Fig. 1e). Finally, since the model predicts that confidence reflects
the stochastic evolution of a value comparison process, it will only be
weakly related to initial DV. This feature of the model provides a parsi-
monious explanation for why DV and confidence are dissociable in our
behavioral data.
Confidence and value in vmPFC
We next hypothesized that if choice confidence is an emergent prop-
erty of a value comparison process, the same brain regions involved
e
1
2
3
4
1 2
Reaction time (quantile)
Confidence
3 4
1.0
0.5
0
–0.5
–1.0
5,800
a
5,500
5,200
4,900
0
Model 1
BIC
Model 2
Model 3
Model 4
Model 5
b
0
100
Threshold
∆e
50
Evidence (a.u.)
0
40 80
Time (a.u.)
Right item
Left item
120
c
–0.2
1.0
0.5
P(chose right item)
0
–0.1 0
Value right item–value left item
0.1 0.2
Small ∆e
Large ∆e
d
250
200
0
Reaction time (a.u.)
DV
Low
High
Low
High
Small ∆e
Large ∆e
DV
(quantile)
Figure 2 Computational model. (a) Comparison of regression models. Plotted are BIC scores (model 1: 5,424;
model 2: 4,995; model 3: 5,388; model 4: 5,498; model 5: 5,291). Smaller numbers indicate a better model fit.
See Results and Online Methods for details of each model. (b) Dynamic (race) model of value comparison for an
example trial. Evidence in favor of each option accumulates over time, with a choice in favor of one or other option
being made when threshold is reached. In this model, decision confidence is derived from the absolute difference
between the two accumulators at the time of the decision (∆e). a.u., arbitrary units. (c–e) Model predictions. (c) When
∆e is large (that is, high confidence) choice accuracy is predicted to increase, reflected by a sharper curve in the logistic
regression. (d) RT values are predicted to decrease when either |DV| or ∆e increase. (e) Matrix representing how model
confidence changes across |DV| and RT quantiles. Note the similarity between the model predictions and behavior (Fig. 1c–e).
a
Percentage signal change in vmPFC
Confidence
t-score
DV
0.10
Low confidence
High confidence
DV
*
*
0.05
3.6
3.22.8
y = 52
y = 47
b
c
0
–0.05
–0.10
Low High Low High
A
A R
R
Figure 3 vmPFC. (a) Brain activity in precuneus and vmPFC. vmPFC (Montreal Neurological
Institute (MNI) space coordinates (x, y, z) = (12, 56, 4); circled) correlating with increases in DV
between the two items presented (P < 0.05, FWE corrected at cluster level). Left image, sagittal
section; right image, coronal. (b) Brain activity in precuneus and vmPFC (12, 47, −11) correlating
with increases in subjective confidence (P < 0.05 FWE corrected at the cluster level). (c) Signal in
vmPFC (6-mm sphere centered at the peak of the DV main effect (12, 56, 4)), showing main effects
of DV and degree of confidence in the absence of an interaction. The plot (extracted from GLM 2;
see Online Methods) is shown only to clarify the signal pattern in vmPFC (that is, lack of interaction
between confidence and DV) and to confirm statistical inference (from GLM 1) regarding the main
effects of DV and confidence. *P < 0.01. Error bars represent s.e.m.
npg
© 2013 Nature America, Inc. All rights reserved.
1 0 8 VOLUME 16 | NUMBER 1 | JANUARY 2013 nature neurOSCIenCe
a r t I C l e S
in value-based decision-making should also represent subjective con-
fidence in a value estimate. In other words, if a brain region involved
in value comparison is implementing a process akin to a race model
6
,
then activity in that region should be modulated by both initial |DV|
and noise (confidence) on that trial. To test this hypothesis, we con-
structed a general linear model (GLM) of our fMRI data in which
each trial was modulated by two parametric regressors: |DV| and
confidence orthogonalized with respect to |DV|. We show that activ-
ity in vmPFC was indeed modulated by both value and confidence
(Fig. 3a,b and Supplementary Table 3; P < 0.05 family-wise error
(FWE) corrected at cluster level). This pattern is consistent with the
established function of this region in encoding goal-values
1,2
and
with our hypothesis that this region also represents the confidence
associated with a value comparison.
We next investigated whether |DV| and confidence interacted in
vmPFC by splitting the model into high- and low-confidence trials,
both parametrically modulated by |DV|
(
Fig. 2c). This analysis showed main effects
of |DV| and confidence in vmPFC but no interaction between them
(2 × 2 ANOVA with factors value, confidence: main effect of value
F
1,19
= 5.1, P < 0.05; main effect of confidence F
1,19
= 7.6, P < 0.05;
interaction F
1,19
= 0.7, P > 0.5) (Fig. 3c). The absence of an inter-
action at the neural level is consistent with a theoretical independence
between value and noise in the choice process, such that one can have
high confidence in a low-value choice and vice versa. Furthermore,
the pattern across conditions closely resembles that seen for RT
values (Fig. 1d) providing convergent evidence that vmPFC activity
is tightly linked to behavior. We also confirmed that the response
to confidence was not driven by a categorical response to errors
8
(Supplementary Fig. 3).
Confidence in right rostrolateral prefrontal cortex
A key question is how confidence-related information represented
in vmPFC becomes available for self-report. One computationally
plausible hypothesis is a hierarchical model wherein confidence in a
comparison process is ‘read out’ by an anatomically distinct second-
order network
22–24
. Right rostrolateral prefrontal cortex (rRLPFC) is
a likely candidate, as this region is implicated in metacognitive assess-
ments of perceptual decisions
9,14,25
. Consequently, we tested whether
this region acts more generally in metacognitive appraisal by enabling
explicit report of confidence in a value comparison.
We first established that rRLPFC tracked changes in reported con-
fidence but did not code for DV (Fig. 4a,b, Supplementary Fig. 4 and
Supplementary Table 3; P < 0.005, small-volume corrected (SVC)),
as expected for a region providing a readout of decision confidence.
a
c
b
y = 42
t-score
2.8 3.2 3.6
R
y = 47
R
Low confidence
High confidence
n.s.
–0.10
Percentage signal change in rRLPFC
n.s.
5
DV
Choice accuracy (high confidence–low confidence)
Subjects
Parameter estimates in rRLPFC
(high confidence–low confidence)
0.05
0
–0.05
–0.10
4
3
2
1
0
–1
–2
–3
–4
–4 –3 –2 –1 0 1 2 3 4 5 6
Low LowHigh High
Figure 4 RLPFC. (a) Brain activity in rRLPFC correlating with decreases
in subjective confidence (P < 0.005, small-volume FWE corrected).
Coronal section; R, right. (b) Signal in rRLPFC (6-mm sphere MNI space
coordinates (x, y, z) = (39, 41, 16)) showing a main effect of confidence
but not DV. The plot (extracted from GLM 2; see Online Methods) is shown
only to clarify the signal pattern in rRLPFC (that is, absence of main effect
of DV). n.s., not significant. (c) Between-subjects regression analysis
considering the change in choice accuracy (slope of the logistic fit) between
low- and high-confidence trials (see red double-headed arrow in Fig. 1b) as
a covariate for confidence-related activity in rRLPFC (peak (x, y, z) = (27,
44, 16); P < 0.05, small-volume FWE corrected). The scatter plot is not
used for statistical inference (which was carried out in the SPM framework);
it is shown solely for illustrative purposes. Error bars represent s.e.m.
PPI
A
y = 47
y = 56
R
2.8 3.2
t-score
3.6
5
1.0
R
a
b
4
3
3 4 5
2
2
1
1
0
0
–1
–1
–2
–2
Choice accuracy (high confidence–low confidence)
Subjects
–3
–3
–4
–4
0.8
0.6
0.4
Parameter estimates in vmPFC
PPI analysis with rRLPFC seed
Parameter estimates in vmPFC
PPI connectivity with rRLPFC
(high confidence–low confidence)
0.2
0
Low High
Confidence
Figure 5 Connectivity analysis. (a) PPI analysis.
Left: three-dimensional rendering, radiological
orientation. Right: sagittal section. vmPFC (circled
in black in right panel) shows increases in
connectivity with a region of rRLPFC (6-mm sphere
(x, y, z) = (39, 41, 16); blue, left panel) previously
identified as being modulated by confidence
(vmPFC peak (x, y, z) = (9, 50, −11); R, right;
A, anterior. P < 0.05, small-volume FWE
corrected). 3D rendering made in mricron
(radiological orientation). R, right. A, anterior.
(b) Between-subjects regression analysis considering
the increase in choice accuracy (see red double-
headed arrow in Fig. 1b) between high-confidence
and low-confidence conditions as a covariate for
the modulation of connectivity (vmPFC peak (x, y,
z) = (15, 56, −5); P < 0.05, small-volume FWE
corrected). Coronal section. The scatter plot was
not used for statistical inference (which was carried
out in the SPM framework); it is shown solely for
illustrative purposes. Error bars represent s.e.m.
npg
© 2013 Nature America, Inc. All rights reserved.
nature neurOSCIenCe VOLUME 16 | NUMBER 1 | JANUARY 2013 1 0 9
a r t I C l e S
DISCUSSION
Here we show that decision confidence emerges from a value com-
parison process in vmPFC and that this region is in turn accessed
by rRLPFC to enable a subjective assessment of confidence. Our
neural findings are consistent with previous evidence showing that
choice difficulty is coded by vmPFC in humans and analogous OFC
neurons in rodents
7,8
. There is also an established body of work
showing that this brain area represents the expected value of an out-
come
1–6
. However, as previous studies defined confidence in terms
of factors governing choice, they were unable to tease apart the
relationship between value and confidence. Our results go beyond
these studies by dissociating subjective confidence from DV. In so
doing, we demonstrate that neural activity in the same anatomical
region represents both variables, suggesting that confidence and
DV are separate behavioral manifestations of the same underlying
decision variable.
Choice confidence can be seen to emerge from the dynamics of
noisy accumulators in the race model
7,12,21
, leading to dual effects
of DV and RT on confidence
27
. The race model has previously been
proposed to account for decision confidence in perceptual decision-
making. In keeping with recent research efforts that have incorpo-
rated dynamic models into the field of economic decision-making
28
,
we find that this model captures several features of the relationship
between choice, RT and confidence in a value-based choice model.
The separation between confidence and BDM values in the present
study provides a new perspective on how an underlying decision vari-
able can be fractionated into distinct behavioral components. Given
that DV and confidence had independent effects on vmPFC activity,
this result provides convergent support for the idea that vmPFC acts
as a dynamic accumulator of choice values
6
. Our findings also accord
with a theoretical Bayesian scheme in which uncertainty, or precision,
is an inherent property of the neural code
29–31
.
A central problem for computational models of metacognition is
how confidence information is read out for appraisal and communica-
tion to others. It has been proposed
22,24
that such a computation can
be achieved by a two-layer neural network architecture, in which the
second-order network receives information about the performance of
the first-order network and uses this information to generate reports
of confidence. Our fMRI data can be interpreted in this framework
∆e
∆e
100
Threshold
50
a
b
0
A
R
1 1.0
�
conf
= 0 �
conf
= 0.1 �
conf
= 0.3
0.5
Low confidence
High confidence
Low confidence
High confidence
Low confidence
High confidence
0
1.0
0.5
0
0
–0.2 0 0.2 –0.2 0 0.2 –0.2 0 0.2
Value right item–
value left item
Value right item–
value left item
Value right item–
value left item
P(chose right item)
0 40 80
Right item
Left item
120
Metacognitive
access
Time (a.u.)
Evidence (a.u.)
�
conf
Figure 6 Schematic of network relating
confidence to subjective report. Summary of the
relationship between our computational model
and neuroimaging analyses. (a) Confidence in the
decision (∆e) emerges from the value comparison
process instantiated in vmPFC. (b) To reach
metacognitive awareness (and be reported by
the participant), this information is transferred
to rRLPFC. The parameter
σ
conf
governs the
noise in the readout of ∆e (that is, decision
confidence). If
σ
conf
is zero, the information
about confidence (∆e) is uncorrupted, resulting
in a pronounced shift in the choice accuracy
between high-confidence and low-confidence
trials (red double-headed arrows). As the level
of metacognitive noise increases (higher values
of
σ
conf
) the shift between the two curves (low
and high confidence) diminishes. Differences
in
σ
conf
account for the inter-subject variability
in metacognitive reportability we observed
behaviorally. a.u., arbitrary units.
We next harnessed individual differences in metacognition to provide
a more stringent test for the role of rRLPFC. We defined an indivi-
dual’s metacognitive accuracy as the change in choice accuracy (slope
of the logistic fit) between low- and high-confidence trials (Fig. 1b).
We reasoned that if rRLPFC acts in the metacognitive appraisal of
confidence, activity in this region and/or its coupling with vmPFC
should predict this change in slope across individuals. To test our
first prediction, we entered change in slope as a between-subjects
covariate in the whole-brain analysis of confidence-related activity,
finding that this parameter significantly modulated the response to
confidence in rRLPFC (P < 0.05; SVC for multiple comparisons).
In other words, participants manifested a neurometric-psychometric
match between their behavioral and neural responses to change in
confidence level (Fig. 4c).
Metacognitive access: interaction between vmPFC and rRLPFC
To test our second prediction, that these two regions are part of the
same functional network (in the context of our task), we performed
a psychophysiological interaction (PPI) analysis using rRLPFC
as a seed (Fig. 5a). This analysis revealed a robust modulation of
connectivity between rRLPFC and vmPFC (P < 0.05 small-volume
FWE corrected) by confidence level (Fig. 5a,b). Furthermore, the
strength of connectivity between these two regions also predicted
metacognitive accuracy across subjects (vmPFC; P < 0.05; SVC for
multiple comparisons) (Fig. 5b). Thus, both activity in rRLPFC itself
and its coupling strength with vmPFC influenced the degree to which
confidence was effectively read out for metacognitive report.
How might this readout process relate to our computational model
of confidence? Intuitively, if reported confidence is a noisy facsimile
of the confidence inherent in a decision process, the relationship
between confidence and behavior will weaken and metacognitive
accuracy will decrease
26
. We were able to modify the race model,
introduced previously, to account for the inter-subject variability in
metacognitive reports observed experimentally. We introduced an
additional parameter (
σ
conf
) governing the noise in the read-out of ∆e
(that is, decision confidence) computed during the value comparison.
Variation in this parameter captured variability in the change in slope
between high- and low-confidence conditions, despite overall choice
accuracy remaining equal (Fig. 6). Together with our imaging results,
this analysis suggests that rRLPFC may indeed mediate variability in
reported confidence (see Fig. 6 and Discussion).
npg
© 2013 Nature America, Inc. All rights reserved.
