Isomorphic decisional biases across perceptual tasks

Abstract

Humans adjust their behavioral strategies to maximize rewards. However, in the laboratory, human decisional biases exist and persist in two alternative tasks, even when this behavior leads to a loss in utilities. Such biases constitute the tendency to choose one action over others and emerge from a combination of external and internal factors that are specific for each individual. Here, we explored the idea that internally-mediated decisional biases should stably occur and, hence, be reflected across multiple behavioral tasks. Our experimental results confirm this notion and illustrate how participants exhibited similar choice biases across days and tasks. Moreover, we show how side-choice behavior in a two alternative choice task served to identify participants, suggesting that individual traits could underlie these choice biases. The tasks and analytic tools developed for this study should become instrumental in exploring the interaction between internal and external factors that contribute to decisional biases. They could also serve to detect psychopathologies that involve aberrant levels of choice variability.

Two‐stage reinforcement learning task predicts psychological traits

“…Next, we quantified a moving average of the participants' % correct sidechoices (the blue line corresponds to group averaged choices, span = 40 trials, Figure 2H). As expected, participants exhibited various learning rates based on how they used the feedback information (Pulcu & Browning, 2017;Trevino et al, 2021). We grouped the learning curves by reinforcement conditions and found no net left versus right bias (Kruskal-Wallis multicomparison test, p < .001; Figure 2H).…”

“…Thus, the BL algorithm computes a threedimensional density function of r, v, and k (the rate of change of v) for every trial. r (or v) is a weighted mean of the marginal distribution integrated by v (or r) and k. This ideal-observer model takes the reward rate from previous observations and transforms the agent's belief about the task's state by updating representations of reward and volatility through multivariate probability distributions (Behrens et al, 2007;Trevino et al, 2021). A complete breakdown of the BL equations and methods can be found in the original publication's supplementary material (Behrens et al, 2007).…”

“…(A) Tables illustrate the seven response variables (i.e., the psychological traits, PTs, enclosed in a dotted rectangle) and the 24 predictors extracted from the reversal-learning task (tea green background). We combined predictors with the PTs and applied a multivariable linear regression model to extract empirical slopes (Trevino et al, 2021). (B) The high cross-correlation coefficients across pairs of predictors (left panel) and response variables (right panel) reveal that many of these parameters are not fully independent from each other.…”

“…Greater distances imply stronger differences between regression coefficients. (D) Cumulative match curves calculated using different sets of predictors, with their strength identified using a step-wise regression learner (Trevino et al, 2021). Stronger predictors render a more accurate classification.…”

“…To further explore the link between the behavioral factors from the reversal-learning task (i.e., the predictors) and the PTs (i.e., the response variables), we employed a MLRM (Figure 5A). We collected a total of 24 behavioral metrics extracted from the reversal-learning task (i.e., a multimetric behavioral profile), including the stationary and transition entropies (Trevino et al, 2021). Using correlation analysis, we found that many predictors and response variables were not entirely independent (Figure 5B).…”