The relationship between anxious traits and learning about changes in stochasticity and volatility.

Anxiety is known to alter learning in uncertain environments. Experimental paradigms and computational models addressing these differences have mainly assessed the impact of volatility, with more highly anxious individuals showing a reduced adaptation of learning rate in volatile compared to stable environments. Previous research has not, however, independently assessed the impact of both changes in volatility, i.e., reversals in reward contingency, and changes in stochasticity (noise) in the same individuals. Here, in an original online study (Experiment 1; N = 80) and a pre-registered replication attempt (Experiment 2; N = 160), we use a simple probabilistic reversal learning paradigm to independently manipulate the level of volatility and noise at the experimental level in a fully orthogonal design. We replicate previous studies showing general increases, irrespective of anxiety levels, in positive learning rate (Experiment 1) and negative learning rate (Experiments 1 and 2) for high compared to low volatility, but here only in the context of low noise. Across both experiments, there was an interaction between volatility and noise on behaviour, with more win-stay responses for high compared to low volatility under low noise, but similar or fewer win-stay responses for the same comparison under high noise. The impact of anxious traits presented differently across experiments; in Experiment 1, increases in lose-shift responses in high versus low noise conditions scaled with level of anxious traits, whereas in Experiment 2, there was a full interaction between volatility, noise and anxious traits on win-stay behaviour. These anxiety-related lose-shift or win-stay differences were reflected in their corresponding negative and positive reinforcement learning rate parameters, respectively. Experiment 2 represents a more robust set of results with a larger sample size, balanced gender representation, and extended block order balancing. These findings suggest that changes in both sources of uncertainty - stochasticity and volatility - should be carefully considered when investigating learning and how learning is shaped by anxiety.