Article Figures & Data
Figures
- Figure 1.
Design for Experiments 1–4. A, In Experiments 1 and 2, subjects performed a spatial two-alternative forced-choice task. On each trial, two patches of visual noise simultaneously appeared to the left and right of fixation. One of these patches contained an embedded sinusoidal grating. Subjects first indicated whether the left or right patch contained the grating and then rated decision confidence on a scale of 1–4. Trial duration was determined by subject response time. B, Experiment 3 was similar to Experiments 1 and 2, except that in even-numbered blocks of trials (partial type 2 blocks), subjects were not required to rate confidence for the first half (50 trials) of the block. A written cue appeared above fixation on all trials where subjects were required to rate confidence. Trial duration was fixed to be 2.533 s. In Experiment 4, subjects wagered points rather than rating confidence, such that they won or lost the number of points wagered depending on the accuracy of the left/right decision. Subjects were also provided with feedback about wagering performance after each block.
- Figure 2.
Individual results from Experiments 1–4 reveal a dissociation between SDT expectations and the relationship between changes in d′ and changes in meta-d′. A, Results from Experiment 1 displaying between-subject correlation of changes in perceptual and metacognitive performance and simulations based on SDT expectations. We computed the change in d′ and meta-d′ between the first and second halves of all blocks (i.e., Δd′ = d′second block half − d′first block half; Δmeta-d′ = meta-d′second block half − meta-d′first block half) and found that these measures were inversely related, in stark contrast to SDT expectation. This suggests a trade-off effect whereby maintenance of perceptual performance comes at the expense of maintenance in metacognitive performance, and vice versa. B–D, Data from Experiments 2–4. As in Experiment 1, the relationship between changes in d′ and meta-d′ failed to match SDT expectation.
- Figure 3.
Brain regions selected for voxel-based morphometry analysis. Two regions of interest in aPFC were selected for analysis on the basis of positive correlations with metacognitive efficiency (meta-d′/d′). These regions were identified in a previous analysis of the data, conducted by McCurdy et al. (2013), and are consistent with previous findings relating metacognitive sensitivity to aPFC gray matter volume (Fleming et al., 2010). To obtain the most robust estimate of aPFC volume, we combined both aPFC clusters to produce an average volume, as described by McCurdy et al. (2013). The peak voxel coordinate for left aPFC is [−12, 54, 16]. The peak voxel coordinate for right aPFC is [32, 50, 7]. Both survived cluster familywise error correction. This figure is adapted from McCurdy et al. (2013).
- Figure 4.
Perception and metacognition as a function of aPFC volume. A median split analysis revealed that subjects with lower aPFC volume (A) tended to experience decreases in d′ and increases in meta-d′ (task type × time × aPFC volume, p = 0.03), contrary to the pattern of subjects with higher aPFC volume (B). This suggests that the between-subject inverse relationship between changes in d′ and meta-d′ may be partially accounted for by individual differences in aPFC volume. Error bars represent within-subjects SEs (Morey, 2008).
- Figure 5.
Model of the relationship between aPFC volume and changes in perceptual and metacognitive performance. A–C, Correlation analyses from Experiment 2 reveal significant positive correlations between d′ across block halves (p < 0.001) (A), meta-d′ and d′ within block halves (p < 0.001) (B), and aPFC volume with first-half meta-d′ (p = 0.001) and second-half d′ (p = 0.03) (C), after removing variation attributable to first-half d′. (Lines of best fit for both correlations overlap.) D, A schematic representation based on the correlations exhibited in A–C.
- Figure 6.
Results for Experiments 3 and 4. A, B, Mean perceptual (d′) and metacognitive (meta-d′) performance over time. When subjects were not required to place metacognitive judgments in the first block half (partial type 2 blocks), perceptual vigilance increased (block type × time interaction, p = 0.002), but metacognition in the second block half, as measured by meta-d2′/d2′, was not affected (block type, p > 0.4). Error bars represent within-subjects SEs (Morey, 2008). C, Resource account of findings. The results of Experiments 3 and 4 can be understood in terms of the model derived from Experiment 2. By relieving subjects of the requirement to place metacognitive judgments in the first block half, aPFC resources normally dedicated to initial metacognitive performance may have been spared for the separate task of maintaining perceptual vigilance.
- Figure 7.
Signal detection theory simulations of the relationship between changes in perceptual and metacognitive sensitivity. A, Basic SDT model. In a series of SDT simulations closely matching the properties of Experiments 1–4, changes in d′ and meta-d′ across block half are strongly positively related. Displayed is a contour plot based on the two-dimensional histogram of Δmeta-d′ versus Δd′ for all simulated subjects in all simulated experiments. The white line is the line of best fit to simulated data; gray dashed lines are lines of best fit from data in Experiments 1–4. B, SDT model with aPFC adjustment. We adjusted the outcomes of the initial SDT simulation so as to conform to the empirically observed relationships between aPFC volume, Δd′, and meta-d1′/d1′ in Experiment 2 (see Materials and Methods for details). This substantially weakened the relationship between Δd′ and Δmeta-d′ in the simulated data, as demonstrated by a more circular contour plot and smaller slope for the line of best fit. C, Distributions of correlation coefficients for Δd′ and Δmeta-d′. Across 2000 simulated experiments, the basic SDT model yielded correlation values consistently higher than those observed in Experiments 1–4 (one-tailed p values of 0.002, 0.026, 0.001, and 0.004). The adjusted model incorporating the aPFC findings from Experiment 2 yielded a distribution of correlations more closely in line with the data (one-tailed p values of 0.067, 0.383, 0.043, and 0.161).
