When this article was published in Psychological Medicine, Figures 2A, Figure 3A, G, and H used incorrect colours in the legends and labels.
The correct versions of these figures can be found on following pages.
(a) Underconfidence in relation to Anxiety symptom scores increases with post-decision time (z-scored within each dataset and task). (b) Conversely, underconfidence in relation to Gender decreases with post-decision time. (a), (b) Lines show marginal effects of two-way interactions of individual difference factors with post-decision time regressed upon z-scored mean confidence. Dots show regression predicted confidence data tiled into six time bins. (c), (d) The difference in slopes obtained from the same regression models between higher and lower Anxiety scores (c) and women and men (d). (c) Increasing underconfidence in relation to Anxiety symptoms is observed in the combined dataset, as well as individually in Experiments 1 and 2. (d) Decreasing underconfidence in relation to Gender is observed in the combined dataset, as well as in Experiment 1. Error bars show 95% bootstrapped confidence intervals. **p < .01, *p < .05.
Figure 2. Long description
Panel A is a line graph titled Anxiety score. The x-axis is Time z-scored from minus 2 to 2. The y-axis is Confidence z-scored from minus 0.2 to 0.2. A green line representing Lower anxiety shows a linear increase in confidence over time. A red line representing Higher anxiety shows a linear decrease in confidence over time. Shaded areas indicate 95 percent confidence intervals.
Panel B is a line graph titled Gender. The axes are the same as Panel A. A green line representing Men shows a steep linear decrease in confidence over time. A red line representing Women shows a slight linear increase in confidence over time, resulting in the lines crossing near the center.
Panel C is a bar chart showing Slope difference higher minus lower anxiety across five datasets: Combined, Exp 1, Exp 2, Exp 3, and Exp 4. All bars are below the zero line, indicating a negative slope difference. The Combined bar is marked with two asterisks.
Panel D is a bar chart showing Slope difference women minus men across the same five datasets. All bars are above the zero line, indicating a positive slope difference. The Combined bar is marked with one asterisk. Error bars in both C and D represent 95 percent bootstrapped confidence intervals.
Three accounts of how confidence distortions could be generated for the paradigmatic case of underconfidence in relation to simulated Anxiety symptoms. (a) D1: Accumulative – distorted accumulation of evidence over time after making a decision. (b) D2: Additive – the criterion for reporting higher confidence is shifted to a higher level of evidence. (c) D3: Multiplicative – evidence is more gradually converted into extreme confidence values. (d)–(f) Difference in regression slopes relating post-decision time to confidence between higher and lower Anxiety scores (as in Figure 1) for a range of model parameters similar to fitted parameters from observed data (Supplementary Figure 2 shows similar plots with a larger range of simulated parameters). (d) Negative values show that for D1, distortion increases with time. (e) For D2, distortion decreases with time. (f) For D3, distortion can increase, decrease, or not change with time, depending on other parameters. Model fitted confidence (open triangles) depicted besides observed data from Experiments 1 and 2 combined (filled circles), split by (g) higher and lower Anxiety scores, and (i) women and men (similar plots for Experiments 2 and 3 shown in Supplementary Figure 4). The model can capture the opposite relationships between confidence distortion and time observed for individual differences in Anxiety symptoms and Gender. Regression slopes for four parameters recovered from the model, V-bias (drift rate/accumulation distortion), A-bias (additive distortion), M-bias (multiplicative distortion), and V-ratio (metacognitive efficiency), in relation to (h) Anxiety scores, and (j) Gender. For (g)–(h), error bars show 95% confidence intervals across participants.
Figure 3. Long description
The figure is organized into panels A through J.
Row 1: Computational Models
* Panel A (D 1: Accumulative): A drift diffusion model plot where evidence accumulates toward a choice boundary ‘a’. Post-choice, a red line (higher anxiety) shows a slower, distorted accumulation (V-bias) compared to a green line (lower anxiety), leading to lower confidence.
* Panel B (D 2: Additive): Shows a shift in the criterion for confidence. The red curve is shifted rightward relative to the green curve, indicating an A-bias where more evidence is required for higher confidence.
* Panel C (D 3: Multiplicative): Shows a change in the slope of the confidence mapping. The red curve is more gradual than the green curve, representing an M-bias.
Row 2: Simulated Slope Differences
* Panel D: A bar chart showing negative slope differences (higher minus lower) that decrease linearly from -0.05 to -0.18 as simulated Beta upon V-bias increases from -0.1 to -0.25.
* Panel E: A bar chart showing positive slope differences that increase from 0.02 to 0.06 as simulated Beta upon A-bias increases.
* Panel F: A bar chart showing fluctuating slope differences around zero for simulated Beta upon M-bias.
Row 3: Experimental Data and Parameter Recovery
* Panel G: A line graph of confidence (z-scored) versus post-decision time. Lower anxiety scores (green circles) remain stable around 0.1, while higher anxiety scores (red circles) are lower, around -0.1.
* Panel H: A bar chart of regression slopes for model parameters. V-bias is negative (p < .0001), A-bias is positive (p < .0001), M-bias is near zero, and V-ratio is slightly negative (p = .010).
* Panel I: Similar to G, split by gender. Men (green circles) show higher confidence than women (red circles) across post-decision time.
* Panel J: Parameter recovery for gender shows a significant positive A-bias (p = .025), with other parameters (V-bias, M-bias, V-ratio) not reaching significance.
The Publisher apologises for this error.
Open access
