Exploring computational models of the Iowa Gambling Task: Age-related findings from the most valued model

• In Event: Poster Session 05
  In Poster Session: PS 05 Topic - Cognitive Processes

Abstract

The Iowa Gambling Task (IGT) is a reliable tool for assessing developmental differences in decision-making. While overall performance (number of advantageous minus disadvantageous choices) increases linearly with age (Almy et al., 2018), this metric fails to take into account specific decision-making processes that lead to behavioral changes from adolescence into adulthood. Computational models propose specific parameters that account for these processes (e.g., working memory), potentially allowing us to better understand why advantageous decisions increase during adolescence. However, most existing work on computational models has been conducted with adults and/or adult patient populations, with limited application to adolescents. An open question is which computational model of IGT performance best explains the decisions of adolescents and whether the model parameters reflect developmental changes. The recently published Value and Sequential Exploration (VSE) model (Ligneul, 2018) was shown to perform optimally across various model comparison metrics relative to five of the most widely implemented computational models of IGT performance. The VSE accounts for exploratory behavior in a parsimonious fashion that may be particularly crucial for capturing developmental changes in decision-making, given that exploratory patterns change across the lifespan (Beitz et al., 2014; Cassotti et al., 2014). The VSE model proposes five parameters that relate to sensitivity to value, memory of deck selections, tendency to explore unfamiliar decks, the rate at which exploration tendencies change based on exploration behavior, and consistency of decision-making. To better understand processes that contribute to adolescent decision-making, we compared the performance of the VSE to five widely cited computational models among a sample of 187 typically-developing individuals aged 9-23 who completed the standard IGT.

We replicated the findings of Ligneul (2018) and observed that the VSE model fit best, as evidenced by lower goodness of fit values across three different metrics (Figure 1). Additionally, the VSE model demonstrated the highest accuracy for fitted and simulated participant choices (as produced by in silico agents) relative to the other models (Figure 2). Importantly, correlations between the VSE model parameters and overall performance closely matched the reported correlations between model parameters and overall performance reported by Ligneul (2018). Thus, the VSE model appears optimal for describing IGT-based decision-making processes in this developmental sample. Across the five parameters, a tendency to explore unfamiliar decks decreased with age (rho = -.42, p < .001) whereas consistency also slightly decreased with age (rho = -.18, p = .01). Given a moderate negative association between exploring unfamiliar decks and overall performance (rho = -.21, p = .005) this pattern suggests that continued exploration and failure to establish a deck preference may prevent learning of which decisions are advantageous. Indeed, a pattern of selecting from all 4 decks consecutively remained above chance levels across the task’s duration for the sample on average. Relative to adult samples, this sample demonstrated lower decision-making consistency. These findings represent an important step in elucidating age-related differences in adolescent decision-making. Future work will analyze how these processes change longitudinally with repeated attempts, as well as in concert with brain development.

Authors

• Brandon Almy, University of Minnesota - Twin Cities
  Presenting Author

• Joanne Jung, University of Minnesota- Twin Cities
  Non-Presenting Author

• Paul Collins, University of Minnesota
  Non-Presenting Author

• Monica Marie Luciana, University of Minnesota–Twin Cities
  Non-Presenting Author