Microlongitudinal design and undirected associations between depression and metacognitive confidence.

(a) Design of the response scale for EMA item and summary of observations used for analysis. 14 depressive symptoms were assessed twice daily (morning and evening) on an 11-point visual analogue scale, while metacognitive confidence was assessed bidaily in the evenings for 8 weeks. (b) Meta-mind procedure. For each assessment, players are instructed to navigate the ship to the side that has more dots. At the start of each meta-mind round (2 rounds with 20 trials each) players are shown a screen with no stimuli and then two icons with varying dot densities then appear at the top of the screen. After the dots disappear, participants move the spaceship and select the icon containing more dots. At the end of each trial, players are asked to rate their local confidence in the accuracy of their choice. After 20 trials, players are asked to evaluate their global confidence in the overall accuracy for the round. (c) Between-person and contemporaneous associations. Plots display results from multilevel autoregressive models examining the relationship between overall depression and its symptoms, with metacognitive local (diamond) and global (circle) confidence. Between-person partial correlation coefficients (in navy, ‘Trait’) indicate the association between the mean of each EMA item and the mean of confidence, controlling for temporal (autoregressive and cross-lagged) effects as well as age, gender, and education. Within-person contemporaneous partial correlation coefficients (in blue, ‘State’) indicate how well each EMA item relates to metacognitive confidence in the same time window. Vertical lines reflect the averaged 95% confidence interval. Post hoc analyses of individual symptom level associations were corrected for multiple comparisons using the Benjamini-Hochberg adaptive linear step-up procedure. Filled shapes indicate significant results using an alpha level of P < .05.

Trait depression weakens the carry-over from local to global confidence, and this operates via a moderated-mediation pathway.

(a) Cross-level interaction between trait depression/symptoms (between-person) and lagged local confidence (within-person deviation) predicting global confidence. Point estimates of standardised beta interaction coefficients from Bayesian mixed-effects models for total depression and each symptom, with horizontal lines indicating the 90% credible interval. Filled shapes indicate 90% credible intervals are outside of 0. (b) Moderated mediation path diagram for total depression. Trait total depression interacts with within-person lagged local confidence to predict within-person concurrent local confidence (Path a) and in turn global confidence (Path b). The same interaction term has a total effect on global confidence (path c) and a direct effect controlling for concurrent local confidence (Path c’). Values under each path reflects the standardized path coefficients with 90% credible intervals. ΔACME above the mediator denotes the index of moderated mediation, i.e., the difference in the indirect effect (a x b) at ±1 SD of trait total depression. (c) Exploratory moderated-mediation test. Plots show the difference in the posterior distributions of index of moderated mediation (ΔIMM) at ±1 SD of trait depression. Negative ΔIMM values indicate weaker indirect effects at higher depression. Fill indicates the Bayesian evidence ratio (Savage–Dickey Bayes factor), with lighter greens denoting stronger evidence. For total depression and symptoms (except for tiredness), ΔIMM < 0 with strong–extreme Bayesian support, consistent with depression blunting local-confidence autocorrelation and, in turn, its integration into global confidence.

Computational accounts of distorted post-decision evidence accumulation for the case of under-confidence in relation to depression.

(a) v-bias: a distorted accumulation of evidence over time. A negative bias in post-decisional drift rate (red) results in lower evidence favouring the decision made, resulting in lower confidence. (b) a-bias: a distorted shift in confidence criterion. An additive shift in the confidence criterion (red) - represented by a lateral shift in the sigmoid function translating evidence into bounded confidence, results in the same amount of evidence translating into lower confidence. (c) m-bias: a multiplicative shift in confidence criterion. A multiplicative shift in the confidence criterion (red) – represented by a flatter sigmoid function, results in a slower rate of change from accumulated evidence into bounded confidence. (d) Bootstrapped mixed effects models predicting depression from a-bias. Ridgelines show bootstrapped standardised beta coefficient distribution for parameter a-bias predicting depression, with 5th/50th/95th quantile lines. Shading of the ridges indicates frequentist significance. Analyses included all computational parameters and controlled for age, gender, and education as covariates. Symptom-level associations (but not total depression) were corrected for multiple comparisons.

Reliability and validity checks for depression EMA and confidence ratings.

(a) Bivariate correlations. Between-person Spearman’s rank correlation coefficients between non-detrended total depression scores in the morning (second week) and total QIDS-SR-16 scores assessed on assessment day 14. (b) Distribution of the sum of all depression EMA scores (i.e. total depression) across each participant’s first 14 evening assessment. Total depression scores show zero-inflation. (c) Average total depression over each participant’s first 14 evening assessments. There is minimal drift in average total depression scores in 162 participants across assessments. The error bars represent standard error. Internal consistency of total depression (d), local confidence (e), and global confidence (f) ratings for each participant’s first 14 assessments ordered by their means. ICC analyses were performed using evening EMA assessments of 162 participants with a total of 2268 observations.

Mean EMA Responses and Confidence Ratings

Parameter recovery of simulated data generated by randomly drawn parameters

Between-person EMA Associations with Local Confidence

Between-person EMA Associations with Global Confidence

Contemporaneous EMA Associations with Local Confidence

Contemporaneous EMA Associations with Global Confidence

Lagged EMA Predicting Local Confidence (Forward)

Lagged EMA Predicting Global Confidence (Forward)

Lagged Local Confidence Predicting EMA (Reverse)

Lagged Global Confidence Predicting EMA (Reverse)

Same-level Contemporaneous Bayesian Interaction Predicting Global Confidence

Same-level Trait Bayesian Interaction Predicting Global Confidence

Same-level Lagged Bayesian Interaction Predicting Global Confidence

Cross-level Contemporaneous Bayesian Interaction Predicting Global Confidence

Cross-level Lagged Bayesian Interaction Predicting Global Confidence

Trait Depression Moderation on Local Confidence Autoregression

Index of Moderated Mediation on Global Confidence for Depression at ±1 SD

Direct Effect of Interaction on Global Confidence (Path c’)

Depression Associations with Computational Parameters

Demographic and descriptive statistics for the main samples

Mean Responses in EMA Questionnaires and Confidence

Weakly Informative Priors

PSIS-LOO Cross Validation of Mediation Model of Path a