Rumination Derails Reinforcement Learning With Possible Implications for Ineffective Behavior

Abstract

How does rumination affect reinforcement learning—the ubiquitous process by which people adjust behavior after error to behave more effectively in the future? In a within-subjects design ( N = 49), we tested whether experimentally manipulated rumination disrupts reinforcement learning in a multidimensional learning task previously shown to rely on selective attention. Rumination impaired performance, yet unexpectedly, this impairment could not be attributed to decreased attentional breadth (quantified using a decay parameter in a computational model). Instead, trait rumination (between subjects) was associated with higher decay rates (implying narrower attention) but not with impaired performance. Our task-performance results accord with the possibility that state rumination promotes stress-generating behavior in part by disrupting reinforcement learning. The trait-rumination finding accords with the predictions of a prominent model of trait rumination (the attentional-scope model). More work is needed to understand the specific mechanisms by which state rumination disrupts reinforcement learning.

Associations of rumination, behavioral activation, and perceived reward with mothers’ postpartum depression during the COVID-19 pandemic: a cross-sectional study

“…Recent laboratory studies on rumination and reward perception showed that rumination disrupted reinforcement learning (adjusting behavior after error to behavior), thereby impairing learning adaptive behavior and promoting stress-generating behavior; however, participants were not in their perinatal period ( 44 ).…”

“…Mothers with their partner's support (n = 374) impairing learning adaptive behavior and promoting stressgenerating behavior; however, participants were not in their perinatal period (44). Rumination might impair concentration and contingency attention associated with adaptive behavior, suggesting that rumination reduces sensitivity to contextual details (45).…”

The challenge of learning adaptive mental behavior.

“…We compared a variety of computational models that might plausibly capture learning and choice processes in the task in terms of HITCHCOCK AND FRANK model comparison (i.e., fit to task data while penalizing for model complexity) and model validation (i.e., ability to capture key features of the data; "Method"). In both experiments, the most successful model (Figure 3) was a Q-learning RL model that learned values of the actions in each state (Q-values) through trial and error, but where the Q-values decayed (Brown et al, 2018;Collins et al, 2014;Hitchcock et al, 2022;Katahira & Toyama, 2021;Niv et al, 2015) over trials toward a default prior via a decay parameter, w,…”

Back to the future: Progressing memory research in eating disorders

“…These changing, state‐based processes are particularly relevant given robust links between affective dysregulation and stress reactivity with EDs (Chami et al, 2019; Lavender et al, 2015; Wonderlich & Lavender, 2017). For instance, in a recent study in individuals with elevated depressive symptoms, researchers used a within‐subjects design to explore the influence of state‐based depressive rumination on reinforcement learning performance, finding that performance was impaired by engagement in rumination (Hitchcock, Forman, et al, 2022). Given the relevance of rumination and other forms of repetitive negative thinking in EDs (Palmieri et al, 2021), this type of manipulation may increase external validity of findings as well as inform more dynamic, context‐dependent models of procedural memory in EDs (Hitchcock, Fried, & Frank, 2022).…”