Estradiol has recently started gaining interest as a modulatory agent of human cognitive functioning. A proposed mechanism by which modulation occurs is a change in dopamine availability. As the implications of the latter in reinforcement learning have been well documented and because estradiol might alter the cognitive processes involved in the successful execution of a reinforcement learning task, we have explored whether administering an acute dose (3 mg) of estradiol in healthy young men indeed affected learning in a probabilistic reinforcement learning task. By using a double-blind, placebo-controlled design, we assessed group differences in learning rates by employing a Q-learning model. A hierarchical Bayesian learning model (the hierarchical Gaussian filter) was used in addition as an alternative that was predicted to be better at explaining the observed behaviour. Namely, the hierarchical Gaussian filter enables a more precise and complex interpretation of behavior by modeling beliefs and yielding parameter estimates related to those beliefs. Both models were compared using two commonly used model selection criteria (i.e. the Bayesian Information Criterion and the Akaike Information Criterion) and by assessing their model accuracy. Furthermore, subjective reports of their beliefs about the task in the form of single point estimates and dynamic, time-dependent estimates were used at the end of the experiment as a way of validating the second model's values. We did not find a clear effect of estradiol administration in the reinforcement learning task. In contrast to our expectations, the Q-learning model fit the data better. Finally, we have shown that the model values statistically significantly correlated with subjective reports in some of the computed correlations, giving further weight to pairing subjective reports with model values and the validity of the model estimates of the hierarchical Gaussian filter themselves.
|
