Trading mental effort for confidence in the metacognitive control of value-based decision-making

  1. Douglas G Lee  Is a corresponding author
  2. Jean Daunizeau  Is a corresponding author
  1. Sorbonne University, France
  2. Paris Brain Institute (ICM), France
  3. Institute of Cognitive Sciences and Technologies, National Research Council of Italy, Italy
  4. Translational Neuromodeling Unit (TNU), ETH, Switzerland
24 figures and 3 additional files

Figures

The metacognitive control of decisions.

First, automatic processes provide a ‘pre-effort’ belief about option values. This belief is probabilistic, in the sense that it captures an uncertain prediction regarding the to-be-experienced …

The expected impact of allocated resources onto value representations.

Left panel: the expected absolute mean difference E[|Δμ(z)||z] (y-axis) is plotted as a function of the absolute prior mean difference μ0 (x-axis) for different amounts z of allocated resources (color code), …

Experimental design.

Left: pre-choice item rating session: participants are asked to rate how much they like each food item and how certain they are about it (value certainty rating). Center: choice session: …

Accuracy of model postdictions and out-of-sample predictions.

The mean within-subject (across-trial) correlation between observed and predicted/postdicted data (y-axis) is plotted for each variable (x-axis, from left to right: choice confidence, spreading of …

Three-way relationship between RT, value, and value certainty.

Left panel: Mean standardized regression weights for |ΔVR0| and VCR0 on log-RT (cst is the constant term); error bars represent s.e.m. Right panel: Mean z-scored log-RT (y-axis) is shown as a …

Three-way relationship between subjective effort rating, value, and value certainty.

Same format as Figure 5.

Impact of consequential and penalized conditions on effort-related variables.

Left panel: log-RT: mean standardized regression weights (same format as Figure 4 – left panel, cons = ‘consequential’ condition, pena = ‘penalized’ condition). Right panel: subjective effort …

Three-way relationship between choice confidence, value, and value certainty.

Same format as Figure 5.

Three-way relationship between change of mind, value, and value certainty.

Same format as Figure 5.

Three-way relationship between spreading of alternatives, value, and value certainty.

Same format as Figure 5.

Three-way relationship between value certainty gain, value, and value certainty.

Same format as Figure 5.

Appendix 1—figure 1
Quality of the analytical approximation to P-.

Upper left panel: the Monte-Carlo estimate of P- (color-coded) is shown as a function of both the mean μ-4,4 (y-axis) and the variance σ20,4 (x-axis) of the parent process x~Nμ,σ2. Upper right panel: analytic …

Appendix 1—figure 2
The β-effect: MCD-optimal effort and confidence when effort has no impact on the value difference.

MCD-optimal effort (left) and confidence (right) are shown as a function of the absolute prior mean difference μ0 (x-axis) and prior variance σ0 (y-axis).

Appendix 1—figure 3
The γ-effect: MCD-optimal effort and confidence when effort has no impact on value precision.

Same format as Appendix 1—figure 2.

Appendix 1—figure 4
MCD-optimal effort and confidence when both types of effort efficacy are operant.

Same format as Appendix 1—figure 2.

Appendix 1—figure 5
Comparison of simulated and estimated MCD parameters.

Left panel: estimated parameters (y-axis) are plotted against simulated parameters (x-axis). Each dot is a Monte-Carlo simulation and different colors indicate distinct parameters (blue: efficacy …

Appendix 1—figure 6
Relationship between choices, pre-choice value ratings, and choice confidence.

Left panel: the probability of choosing the item on the right (y-axis) is shown as a function of the pre-choice value difference (x-axis), for high- (blue) versus low- (red) confidence trials. The …

Appendix 1—figure 7
Relationship between pre-choice value ratings, choice confidence, and response times.

Left panel: response times (y-axis) are plotted as a function of low- and high- |ΔVR0| (x-axis) for both low- (red) and high- (blue) confidence trials. Error bars represent s.e.m. Right panel: A …

Appendix 1—figure 8
Correlation between pupil size and subjective effort ratings during decision time.

Left panel: Mean (± s.e.m.) correlation between pupil size and subjective effort (y-axis) is plotted as a function of peristimulus time (x-axis). Here, epochs are co-registered w.r.t. stimulus onset …

Appendix 1—figure 9
Gaze bias for low- and high-effort trials.

Mean (± s.e.m.) gaze bias is plotted for both low- (left) and high- (right) effort trials.

Appendix 1—figure 10
Accuracy of RT postdictions.

Left panel: The mean within-subject (across-trial) correlation between observed and postdicted RT data (y-axis) is plotted for each model (gray: MCD, blue: DDM1 and DDM2); error bars depict s.e.m. …

Appendix 1—figure 11
Accuracy of out-of-sample change of mind postdictions.

Same format as Appendix 1—figure 10.

Appendix 1—figure 12
Comparisons of MCD model with linear and saturating γ-effects.

Left panel: The mean within-subject (across-trial) correlation between observed and postdicted data (y-axis) is plotted for dependent variable (x-axis, from left to right: choice confidence, …

Appendix 1—figure 13
Comparisons of MCD and model-free postdiction accuracies.

The mean within-subject (across-trial) correlation between observed and postdicted data (y-axis) is plotted for each variable (x-axis, from left to right: choice confidence, spreading of …

Additional files

Download links