Computational Characterization of Information Processing in Impulsivity

Abstract

Impulsivity has been theorized as a multidimensional construct, necessitating focused research into determining how these dimensions manifest together and separately. This work aims to uncover how information processing, particularly in learning and decision-making, may be susceptible to rapid and unplanned responding. While reinforcement learning (RL) has traditionally sat outside the umbrella of impulsivity, sequentially chaining prediction errors could be vulnerable to any assortment of impulse-control valuation and exploratory influence. We propose that impulse-control RL, such as in probabilistic and delayed learning, may have relationships with other impulsivity dimensions, such as temporal discounting (TD) or cognitive control (CC). Participants (N = 210) completed five online cognitive tasks (RL: two-armed probabilistic and delayed bandits, CC: Go/No-Go and flanker, and TD: delay discounting task). The two-armed delayed bandit was designed to experimentally isolate the effect of delay learning and required an explicit estimation of time delay. We simultaneously fit a hierarchical framework with the five drift diffusion models that estimated covariance parameters within and across tasks. Within this framework, a Rescorla-Wagner and hyperbolic discounting model scaled their respective value differences into evidence accumulators for probabilistic bandit and discounting tasks, respectively. We find multiple points of support with some unexpected results for our hypotheses. First, probabilistic and delayed reward learning showed positive relationships in exploratory behavior toward optimal RL policies. Consistency in RL learning behaviors also appeared to relate to TD and CC. Relationships with self-report metrics were predominantly weak, but impulsivity measured by self-report showed a possible influence on stay-rate of the delayed bandit. Taken together, our findings suggest that impulsivity might share related mechanisms of information processing across a battery of tasks.