Dense repeated sampling reveals impaired hierarchical integration of metacognitive confidence that scales with depression

Reviewer #1 (Public review):

Summary:

The authors investigate how depressive symptoms relate to metacognitive confidence across multiple levels of a metacognitive hierarchy. Participants completed twice-daily assessments of depressive symptoms and bi-daily assessments of a perceptual confidence task for eight weeks. The study replicates prior findings that depression is associated with lower confidence and extends these findings by suggesting that trait depression weakens the temporal persistence of local confidence signals, thereby impairing their accumulation into global confidence.

The study addresses an important question in computational psychiatry: how disturbances in confidence contribute to persistent negative self-beliefs in depression. The longitudinal design and repeated assessments represent advances over prior cross-sectional studies. The attempt to bridge momentary confidence fluctuations with broader self-beliefs through a hierarchical metacognitive framework is novel.

The principal contribution is the finding that trait depression moderates the lagged relationship between local and global confidence. The authors interpret this as evidence that positive fluctuations in local confidence decay more rapidly in individuals with higher depression, limiting their integration into higher-order confidence beliefs. This contribution is potentially important.

However, several aspects of the interpretation warrant caution. The moderated mediation analysis remains correlational and does not establish that impaired local confidence persistence causally produces global under-confidence. Alternative explanations, including stable individual differences in response styles, latent third variables, or measurement properties of the confidence scales, remain plausible. The manuscript occasionally adopts language suggesting mechanistic or causal conclusions that exceed the inferential scope of the analyses.

The temporal resolution of the study also complicates interpretation. The absence of cross-lagged effects between depression and confidence may reflect a mismatch between the timescales over which mood and metacognition interact. The authors acknowledge this possibility, but it substantially limits conclusions regarding temporal precedence.

Furthermore, the sample size is not justified, and the sample differs considerably from populations typically studied in depression research. Participants were older, predominantly female, and self-selected citizen scientists with low and relatively stable depression scores. Consequently, it remains unclear whether the observed dynamics generalise to clinically depressed populations, where symptom severity and variability may differ substantially.

Overall, this is a thoughtful and technically sophisticated study that provides valuable new data on the temporal organisation of confidence in relation to depression. The central findings are interesting and likely to stimulate future work, although the mechanistic interpretations would benefit from greater caution.

Strengths:

(1) Innovative use of dense longitudinal sampling to investigate metacognitive processes.

(2) Large number of repeated observations per participant and good adherence over eight weeks.

(3) Integration of EMA, multilevel vector autoregression, Bayesian modelling, and computational modelling.

(4) Novel proposal that depression weakens the persistence of local confidence signals and their integration into global confidence.

(5) Careful consideration of local versus global metacognitive processes.

Weaknesses:

(1) The causal and mechanistic claims may exceed what can be inferred from the data.

(2) No justification is provided for the sample size, and the sample is older, predominantly female, and largely non-clinical, limiting generalisability.

(3) The sampling intervals may not be optimally suited to detect temporal relationships between mood and confidence.

(4) Several modelling decisions require additional justification and sensitivity analyses.

(5) The moderated mediation framework assumes a temporal ordering that cannot be conclusively established.

Reviewer #2 (Public review):

Summary:

Phon-Amnuaisuk and colleagues address an important question in cognitive psychology and computational psychiatry: how does the established relationship between depression and metacognitive under-confidence unfold over time? The study is grounded in a hierarchical view of metacognition, in which local confidence in individual decisions contributes to global estimates of performance. To test this, the authors used an intensive longitudinal design in which participants repeatedly reported depressive symptoms and completed a gamified perceptual decision-making task over eight weeks. This allowed them to examine whether depressive symptoms and confidence fluctuate together within individuals, whether one predicts the other over time, and whether trait depression alters the way local confidence is carried forward and integrated into global confidence. The main findings are that higher trait depression is associated with lower local and global confidence, that within-person mood fluctuations show limited temporal precedence over confidence at the two-day timescale, and that trait depression is linked to weaker temporal persistence of local confidence and reduced carry-over into later global confidence.

Strengths:

A major strength of the study is its repeated-measures design across a large sample. Participants completed up to 28 metacognitive task sessions over eight weeks, alongside repeated ratings of depressive symptoms. This allows the authors to separate stable between-person differences from within-person changes over time, which is a clear advantage over standard cross-sectional studies. The analytical strategy is also appropriate: the authors use multilevel vector autoregressive models to examine temporal, contemporaneous, and between-person associations; Bayesian ordinal models to test interactions; and computational modelling to examine how confidence is formed.

The strongest results concern stable individual differences. Participants with higher average depressive symptoms reported lower local and global confidence. While this pattern is consistent with prior work showing reduced confidence in depression, the computational model further suggests that higher depression is associated with a more conservative confidence criterion: these participants required more evidence before reporting high confidence. This adds nuance by suggesting that under-confidence in depression may not reflect poorer metacognitive sensitivity, but rather a bias in how confidence is reported.

The study also proposes a novel temporal account. Trait depression was associated with weaker autocorrelation of local confidence across sessions, which in turn reduced the extent to which local confidence carried over into later global confidence. This indirect pathway was supported by a moderated mediation analysis and was consistent across most individual depressive symptoms. The finding is theoretically interesting and supported by strong statistical evidence.

Weaknesses:

The main limitation concerns the interpretation of the temporal and mechanistic claims. The strongest effects are observed at the trait level, whereas the within-person temporal associations between mood and confidence are weak or absent. The study therefore provides stronger evidence that people with higher average depressive symptoms are generally less confident than evidence that momentary changes in depressive mood drive later changes in confidence, or vice versa.

The sample also constrains the conclusions. Depression scores were strongly concentrated near the lower end of the scale, and the final sample was highly selected, with many of the initial 976 participants excluded because they did not complete enough task sessions. This means that the study may be better suited to detecting stable individual differences than dynamic mood-confidence processes.

The temporal spacing of the metacognitive assessments is a further constraint. Because the metacognition task was administered every two days, the cross-lagged analyses could only test mood-confidence dynamics across this interval. If depressive mood and confidence influence each other over shorter timescales, such effects may have been missed. The absence of cross-lagged effects should therefore be interpreted with caution: it shows that temporal precedence was not detected at the two-day lag in this sample, but it does not rule out shorter-term directional effects or effects in more symptomatic clinical populations.

Related to this, the manuscript moves between several related but distinct terms - "mood," "depressive mood," "depression," "depressive symptoms," and "trait depression" - without always clarifying whether these are intended as interchangeable or as conceptually distinct constructs. This matters for a study whose central claims concern temporal precedence and trait-versus-state distinctions, and it is compounded by a sample with generally low depressive symptom levels, where the boundary between transient low mood and a trait-like depressive disposition is harder to draw.

Author response:

Reviewer #1:

We thank the reviewer for their comments. They raised an issue with the correlational nature of the analysis, suggesting alternative explanations, including stable individual differences in response styles, latent third variables, or measurement properties of the confidence scales. We agree with the reviewer’s comments that a latent third variable may confound our findings and higher-order metacognitive beliefs (which we did not assess) are a possible candidate. We will broaden our discussion to include other plausible confounds that may jointly relate to depression and confidence dynamics. Regarding stable individual differences in response styles, our analysis decomposed confidence into within and between-person components and standardised the within-person component by each participant’s own variability. This allowed the interaction and moderated mediation analyses to separate within-person associations from stable between-person differences (e.g., range of confidence scale used; Epskamp et al., 2018). With regards to measurement properties, we will conduct additional analyses investigating whether trait depression is associated with altered response patterns (e.g., non-linear or more variable mapping of latent evidence into confidence reports).

The reviewer also noted that the temporal resolution we selected may not be optimal to measure co-fluctuation of depression and confidence. We agree this remains a challenge for the depression research (Jamalabadi et al., 2025; Tamm et al., 2024), and metacognitive confidence research (da Fonseca et al., 2023; Wright et al., 2024). We chose the interval as it allowed us to balance retention and data quality across an 8-week study (Eisele et al., 2022) and address a gap in the literature – frequent but shorter (~1-week) EMA-based assessments of existing metacognitive confidence studies (da Fonseca et al., 2023; Wright et al., 2024) precludes insights into the dynamics of mood and metacognitive confidence across longer timescales. We have explored one-day and half-day lagged associations but did not find significant effects across most symptoms for both local and global confidence. This was omitted from the original manuscript as the analyses could not symmetrically control for local/global confidence (as these were only assessed bi-daily) but will be included in the revised manuscript’s supplement.

The reviewer noted that the temporal ordering of the moderated mediation analysis cannot be conclusively established. We agree that there is a lack of research investigating alternative orderings. We selected the local-to-global ordering because prior work has similarly modelled and demonstrated global confidence estimates as a cumulative integration of local confidence estimates across the block (Cavalan et al., 2023; Katyal et al., 2025; Lee et al., 2021; Rouault et al., 2019, 2022). Nevertheless, we modelled an alternative process (i.e. whether global confidence interacted with trait-level depression in predicting subsequent local confidence) but did not find a significant frequentist interaction effect. We will elaborate on our rationale for the current ordering and include analyses of this alternative process in our revised manuscript and supplement.

Finally, we agree with the reviewer’s comments that the sample constraints generalisability. The revised manuscript will more clearly discuss the generalisability of our findings to clinical populations and the implications of self-selection and limited within-person variability in depressive symptoms for interpretation and future work. We will also tone down the causal language in the manuscript.

Reviewer #2:

We thank the reviewer for their comments. We agree with the reviewer that our strongest effects live at the trait level, whereas our cross-lagged findings provided insufficient evidence for depression driving underconfidence or vice versa (at least with a two-day lag). However, our moderated mediation finding centres on a different angle - trait level depression moderating how within-person confidence is integrated into more global beliefs. Nevertheless, we agree that the findings provide a plausible account for why underconfidence (and possibly metacognitive beliefs) remain persistent in depression, but not how either underconfidence or depression arose in the first place. We will make this clearer in our revised manuscript.

Consistent with reviewer #1’s comments, we agree that the non-clinical nature of sample does constrain generalisability and inference. We will conduct a sensitivity analysis with a larger sample (more lenient inclusion criteria) and discuss its implications in more detail in our revised manuscript. We also agree with the reviewer’s comments that our findings do not preclude the possibility of temporal precedence and will further clarify in our revised manuscript with reference to shorter time lags. The reviewer noted that our use of terminologies (e.g., “depression”, “depressive symptoms”, “depressive mood”) was not clearly defined. Our revised manuscript will make this point clearer whilst also making the usage of terminologies consistent.

References

Cavalan, Q., Vergnaud, J.-C., & de Gardelle, V. (2023). From local to global estimations of confidence in perceptual decisions. Journal of Experimental Psychology: General, 152(9), 2544–2558. https://doi.org/10.1037/xge0001411

da Fonseca, M., Maffei, G., Moreno-Bote, R., & Hyafil, A. (2023). Mood and implicit confidence independently fluctuate at different time scales. Cognitive, Affective, & Behavioral Neuroscience, 23(1), 142–161. https://doi.org/10.3758/s13415-022-01038-4

Eisele, G., Vachon, H., Lafit, G., Kuppens, P., Houben, M., Myin-Germeys, I., & Viechtbauer, W. (2022). The Effects of Sampling Frequency and Questionnaire Length on Perceived Burden, Compliance, and Careless Responding in Experience Sampling Data in a Student Population. Assessment, 29(2), 136–151. https://doi.org/10.1177/1073191120957102

Epskamp, S., Waldorp, L. J., Mõttus, R., & Borsboom, D. (2018). The Gaussian Graphical Model in Cross-Sectional and Time-Series Data. Multivariate Behavioral Research, 53(4), 453–480. https://doi.org/10.1080/00273171.2018.1454823

Jamalabadi, H., Koosha, T. A., Stocker, E., Jansen, A., Ebner-Priemer, U. W., Proppert, R. K. K., Rieble, C. L., Tutunji, R., & Fried, E. I. (2025). Optimizing the frequency of ecological momentary assessments using signal processing. Psychological Medicine, 55, e358. https://doi.org/10.1017/S003329172510264X

Katyal, S., Huys, Q. J., Dolan, R. J., & Fleming, S. M. (2025). Distorted learning from local metacognition supports transdiagnostic underconfidence. Nature Communications, 16(1), 1854. https://doi.org/10.1038/s41467-025-57040-0

Lee, A. L. F., de Gardelle, V., & Mamassian, P. (2021). Global visual confidence. Psychonomic Bulletin & Review, 28(4), 1233–1242. https://doi.org/10.3758/s13423-020-01869-7

Rouault, M., Dayan, P., & Fleming, S. M. (2019). Forming global estimates of self-performance from local confidence. Nature Communications, 10(1), 1141. https://doi.org/10.1038/s41467-019-09075-3

Rouault, M., Will, G.-J., Fleming, S. M., & Dolan, R. J. (2022). Low self-esteem and the formation of global self-performance estimates in emerging adulthood. Translational Psychiatry, 12(1), 1–10. https://doi.org/10.1038/s41398-022-02031-8

Tamm, J., Takano, K., Just, L., Ehring, T., Rosenkranz, T., & Kopf-Beck, J. (2024). Ecological Momentary Assessment versus Weekly Questionnaire Assessment of Change in Depression. Depression and Anxiety, 2024, 9191823. https://doi.org/10.1155/2024/9191823

Wright, A. C., Palmer-Cooper, E., Cella, M., McGuire, N., Montagnese, M., Dlugunovych, V., Liu, C.-W. J., Wykes, T., & Cather, C. (2024). Experiencing hallucinations in daily life: The role of metacognition. Schizophrenia Research, Hallucinations: Neurobiology and Patient Experience, 265, 74–82. https://doi.org/10.1016/j.schres.2022.12.023