Abstract
Depression has been associated with a negative attribution style – increased self-attribution of negative outcomes and decreased self-attribution of positive outcomes. This project aimed to investigate this computationally, assessing how causal attributions impact learning, and if different attribution patterns relate to transdiagnostic anxious-depressive and other mental health symptoms. An online study (N=454) was conducted where participants learned which of two coloured mines was rewarded more often (80/20%), across 3 conditions in which a hidden agent could intervene on some trials (30%) and cause: losses (adversarial), rewards (benevolent), or either outcome (random condition). After seeing the outcome, participants judged whether the hidden agent had intervened. Mental health questionnaires were also completed.
Results showed an overall tendency for an optimistic bias in causal attributions and learning, which was modulated by condition [interaction effect on attributions: F(2,882) = 1129, p<.001; learning rates: F(1.4,611) = 1799, p<.001]. Specifically, self-attributions were higher for positive compared to negative outcomes in the random condition, an effect increased in the adversarial condition, and reversed in the benevolent condition. A Bayesian model showed higher learning rates for positive compared to negative outcomes in the adversarial condition, and the reverse in the benevolent condition. Preliminary analyses suggest that anxious-depressive symptoms were correlated with increased self-attribution for negative outcomes (r=.093, p=.05), but not with attribution for positive outcomes. A similar pattern for negative outcomes was seen for social withdrawal and compulsivity and intrusive thought (CIT) symptoms, with CIT also linked to decreased self-attribution for positive outcomes. Further model-based analyses are ongoing to disentangle how causal attribution influences learning and relationships with mental health.
Results showed an overall tendency for an optimistic bias in causal attributions and learning, which was modulated by condition [interaction effect on attributions: F(2,882) = 1129, p<.001; learning rates: F(1.4,611) = 1799, p<.001]. Specifically, self-attributions were higher for positive compared to negative outcomes in the random condition, an effect increased in the adversarial condition, and reversed in the benevolent condition. A Bayesian model showed higher learning rates for positive compared to negative outcomes in the adversarial condition, and the reverse in the benevolent condition. Preliminary analyses suggest that anxious-depressive symptoms were correlated with increased self-attribution for negative outcomes (r=.093, p=.05), but not with attribution for positive outcomes. A similar pattern for negative outcomes was seen for social withdrawal and compulsivity and intrusive thought (CIT) symptoms, with CIT also linked to decreased self-attribution for positive outcomes. Further model-based analyses are ongoing to disentangle how causal attribution influences learning and relationships with mental health.
| Original language | English |
|---|---|
| Publication status | Unpublished - 2025 |
| Event | Computational Psychiatry Conference - Tübingen, Germany Duration: 14 Jul 2025 → 17 Jul 2025 |
Conference
| Conference | Computational Psychiatry Conference |
|---|---|
| Country/Territory | Germany |
| City | Tübingen |
| Period | 14/07/25 → 17/07/25 |
