Work

It Depends On The Working Memory: Testing Correlations Between The Reinforcement Sensitivity Theory And Reinforcement Learning

Year: 2026
Type: preprint
Abstract: Reinforcement Sensitivity Theory (RST) is an influential framework explaining individual differences in sensitivity to reward and punishment. The theory's empirical definitions require that self-report and behavioral measures of Reinforcement Learning (RL) correlate. Recent experimental research has shown that RL is influenced by working memory (WM) load, challenging RST assumptions. The aim of this pre-registered study was to evaluate the relationship between self-report RST measures and performance on an experimental RL task that isolates operant conditioning from the impact of WM. A sample of 110 participants (aged 18–30) completed the RST-PQ, the BIS/BAS Scale, the Barratt Impulsiveness Scale (BIS-11), and the DASS-21. In the laboratory, they responded to the computerized Reinforcement Learning/Working Memory (RLWM). Behavioral performance was analyzed through Pearson/Spearman correlations, mediation analyses, and three computational models (Classic RL, RLWMi, and RLWMi + Deduction) were fitted to the data to estimate latent learning parameters. Only the Fun Seeking subscale of the BIS/BAS Scale predicted learning, both in the behavioral measure under WM-controlled conditions (β= .190, p = .033) and, with greater magnitude, in the computationally modeled learning-rate parameter (β= .233, p = .020), derived from the best-fitting model (RLWMi + Deduction; AIC = 634.151). Contrary to expectations, WM did not mediate this relationship in either condition. Complementary analyses revealed that Behavioral Approach System (BAS) subfactors differentially correlate with distinct processes: while Fun Seeking predicted the positive reinforcement parameter but not perseverative errors, Reward Reactivity predicted perseverative errors but not learning rate, and Reward Interest showed marginal associations with both. These results suggest that different personality traits map onto distinct computational mechanisms of reward- and punishment-based learning, independently of WM, and that computational modeling detects these associations with greater sensitivity than raw behavioral performance.
Language: English
Open Access status: green
OpenAlex ID: W7236244016
Cites:
Cited by:
Related to: 10