Low self-esteem and the formation of global self-performance estimates in emerging adulthood

The study investigates whether low self-esteem is associated with domain-general biases in constructing self-beliefs about performance in non-social contexts without external feedback. Prior work showed that low self-esteem alters the formation of momentary self-worth from social feedback, but it was unclear if similar anomalies extend to non-social self-performance estimates. Within a hierarchical metacognitive framework, self-esteem may act as a global prior that shapes task-level self-performance estimates formed over minutes. The authors hypothesized that individuals with low self-esteem might underestimate their abilities despite comparable objective performance, reflecting a disconnection between local task evaluations and global self-esteem. Understanding this relationship could illuminate cognitive building blocks relevant to mental health and inform interventions for conditions linked to altered self-esteem, such as depression.

The paper builds on literature linking self-esteem to mental health outcomes, notably depression and anxiety, and on research into the construction of self-beliefs at multiple abstraction levels. Prior studies showed low self-esteem individuals exhibit altered updating of self-worth in response to social feedback and that decision confidence contributes to self-performance estimates in the absence of feedback. A hierarchical metacognitive framework suggests global self-esteem may provide priors for local confidence and task-level self-estimates. Related work documents: (i) associations between self-esteem and neural responses to social evaluation; (ii) relationships between depressive/anxious symptoms and metacognitive alterations; (iii) that overall confidence on perceptual tasks correlates with self-esteem independent of depressive symptoms; and (iv) that learning from positive feedback may be attenuated in depression. These strands motivate testing whether biases in self-performance estimation extend beyond social domains and are reflected as global underestimation in low self-esteem individuals.

Participants: From the NSPN cohort (N = 2402; ages 14–24 at baseline), 62 participants were recruited based on Rosenberg Self-Esteem Scale (RSES) scores in the bottom (0–12) or top (27–30) deciles; after exclusions (chance performance, invariant ratings, comprehension failure), N = 57 remained (29 low self-esteem, 28 high), ages ~21, gender-matched. Groups were not matched on depression/anxiety; strong correlations existed between self-esteem and depression (MFQ) and trait anxiety. Ethics approval obtained; participants compensated.

Task design: Participants completed 30 learning blocks of two interleaved perceptual tasks (dot numerosity comparison) identified by color cues. Each block had 2, 4, 6, 8, or 10 trials per task (learning duration). Task factors: Difficulty (easy: +60 dots vs difficult: +24 dots) and Feedback (veridical feedback vs no-feedback). Six possible task pairings across these factors were presented in pseudo-random order. Trial sequence: task cue (1200 ms), stimuli (300 ms), response (untimed), highlight (300 ms), feedback screen (1500 ms; Correct/Incorrect or blank), ITI (600 ms). Stimulus side randomized.

Measures of self-performance estimates: (1) End-of-block task choice: participants chose the task they believed they performed better on, to be used for a subsequent 6-trial test block (no feedback) for bonus. Choice keys differed from perceptual responses. (2) End-of-block task ratings: participants rated perceived accuracy for each task on a continuous 50–100% scale (ticks at 60/70/80/90%), using the mouse; no time limits.

Statistical analyses: Three 2×2×2 repeated-measures ANOVAs examined effects of Feedback (present/absent) and Difficulty (easy/difficult) as within-subject factors, and Self-Esteem group (high/low) as between-subject factor, on (1) objective performance, (2) task choice (arcsine square-root transformed proportions), and (3) task ratings. Analyses were replicated using past (recruitment) self-esteem status. To test sensitivity to block-wise fluctuations in objective performance beyond difficulty, logistic (for task choice) and linear (for ratings) regressions were run per task pairing, predicting end-of-block measures from the absolute difference in objective performance between tasks, with Rosenberg score and its interaction with performance difference as regressors (z-scored). Fixed-effects models treated participants as fixed due to few blocks per pairing. To assess learning duration effects, logistic/linear regressions predicted task choice/rating differences from learning duration, self-esteem, and their interaction, per pairing.

• Objective performance: Main effect of Difficulty, with better performance on easy tasks (F(1,56) ≈ 472.7, p ≈ 1.1×10^-28). No main effect of Feedback (F(1,56)=0.622, p=0.434). No main effect of Self-Esteem (F(1,56)=1.675, p=0.201). One interaction: Difficulty × Self-Esteem (F(1,56)=5.174, p=0.027), driven by slightly worse performance by low self-esteem participants on easy tasks; pairwise comparisons were not significant across individual conditions (all p≥0.07).
• Task choice (self-performance estimate 1): Main effects—Difficulty (F(1,56)=108.8, p=1.2×10^-14) and Feedback (F(1,56)=93.8, p=1.7×10^-13); trend-level Difficulty × Feedback interaction (F(1,56)=3.81, p=0.056). No main effect of Self-Esteem (F(1,56)=0.295, p=0.59) and no interactions with Self-Esteem (all p>0.33).
• Task ratings (self-performance estimate 2): Main effects—Difficulty (F(1,56)=211.7, p=1.7×10^-20) and Feedback (F(1,56)=139.9, p=9.7×10^-17); significant Difficulty × Feedback interaction (F(1,56)=35.6, p=1.8×10^-7). Crucially, main effect of Self-Esteem (F(1,56)=5.92, p=0.018): low self-esteem participants gave consistently lower ratings across feedback and difficulty conditions despite similar objective performance.
• Sensitivity to block-wise performance fluctuations: Difference in objective performance significantly predicted task choices for all pairings (p≈1.56×10^-7) except Easy-No-Feedback vs Difficult-No-Feedback (β=0.083, p=0.49); no interactions with Self-Esteem (all β<0.25, p>0.21). For ratings, performance difference significantly predicted rating differences across all pairings (β>0.016, p<0.0023); no interactions with Self-Esteem for five of six pairings; one exception: Easy-No-Feedback vs Difficult-Feedback showed a significant interaction (β=0.016, p=0.022), indicating greater influence of performance difference among high self-esteem participants; no main effect of Self-Esteem in that model (β≈-0.011, p=0.088).
• Learning duration: Generally no effect on task choices (4/6 pairings non-significant; all β<0.21, p>0.099). Exceptions: Difficult-Feedback vs Difficult-No-Feedback showed decreasing sensitivity with duration (β=-1.49, p=0.013); Easy-No-Feedback vs Difficult-No-Feedback showed a significant effect (β=-1.6, p=0.009) with an interaction with Self-Esteem (β=-1.04, p=0.026). Ratings: No effect for 5/6 pairings (β<0.012, p>0.16); exception: Easy-Feedback vs Easy-No-Feedback (β=-0.018, p=0.033). No interactions between learning duration and Self-Esteem for ratings (|β|<0.012, p>0.15).
• Replication across self-esteem status definitions: Analyses using past (recruitment) self-esteem yielded virtually identical results, indicating stability of the observed associations.

Findings demonstrate that low self-esteem is associated with a generalized underestimation of self-performance in a non-social perceptual task, as reflected in lower subjective task ratings, despite comparable objective accuracy. This supports a hierarchical metacognitive account in which global self-esteem serves as a prior that shifts overall confidence or global self-performance estimates without impairing sensitivity to task-specific information (difficulty, feedback) or to block-level performance fluctuations. The absence of self-esteem effects on binary task choices suggests that global baseline shifts in self-evaluation can cancel out when choosing between two tasks, whereas continuous ratings reveal the global underestimation bias. The small Difficulty × Self-Esteem interaction on objective performance does not account for the robust rating differences, implying a metacognitive rather than performance-driven origin. These results replicate that self-performance estimates integrate difficulty, feedback, and objective performance fluctuations and extend prior work by linking these processes to a global disposition relevant to mental health. The pattern aligns with literature showing associations between self-esteem and overall confidence, and with theories that depressive/anxious symptomatology modulates processing of positive feedback; however, in this dataset, the mechanisms by which feedback updates are integrated did not significantly differ by self-esteem for most conditions. The results suggest that individuals with low self-esteem retain intact mechanisms for constructing self-performance estimates from local evidence but exhibit a global negative bias in evaluation, potentially contributing to or reflecting vulnerability to mental health problems.

The study shows that emerging adults with low self-esteem consistently underrate their task performance relative to high self-esteem peers, even when objective performance is equivalent, revealing a domain-general negative bias in global self-performance estimates. Self-performance estimates remained sensitive to task difficulty, feedback, and performance fluctuations across groups, supporting a hierarchical model where self-esteem acts as a global prior shifting overall evaluation without disrupting local metacognitive integration. Future research should use longitudinal designs to track how global self-performance estimates develop and influence self-esteem over longer timescales, examine broader age ranges and clinical samples, disentangle the roles of co-occurring depressive and anxious symptoms, test potential non-linear relationships across the self-esteem spectrum, and probe mechanisms of feedback-based updating (especially responses to positive feedback) and uncertainty/precision in self-evaluations.

• Fixed-effects regression models (necessitated by few blocks per pairing per participant) limit generalizability to the population level.
• Groups differed on depression and anxiety measures, reflecting real-world comorbidity; thus, effects cannot be unequivocally attributed to self-esteem independent of these symptoms.
• Limited age range (emerging adults) constrains developmental generalization.
• Perceptual task with minimal prior familiarity enhances internal validity but may limit generalization to domains with stronger priors (e.g., memory, academic tasks) or social contexts.
• No direct measure of precision/uncertainty in task ratings; potential changes with learning duration were not assessed.
• Small interaction between Difficulty and Self-Esteem on objective performance raises, but does not confirm, the possibility of subtle effort or expectation effects.
• Binary task choices may be insensitive to global baseline shifts in self-evaluation, potentially masking some self-esteem effects.