Individuals who suffer from anxiety focus disproportionately on the potential occurrence of future negative outcomes and whether or not they can be averted. Attempts to understand this have led to suggestions that anxiety may be linked to difficulty in estimating the probability or severity of future negative events or in combining these estimates when choosing between actions
Anxious individuals are reported to be highly intolerant of situations characterized by uncertainty as to what will happen or which course of action should be followed. They are more likely to endorse finding such situations distressing, leading to a sense of immobilization. One possible explanation for this is that individuals prone to anxiety may have difficulty estimating outcome likelihood when there are sources of uncertainty complicating the action-outcome relationship.
In the context of decision-making models, a number of alternate forms of uncertainty are recognized. Here, we focus on two types:
- One source of uncertainty is produced by noise in the relationship between actions and outcomes, such as occurs if an action only leads to a given outcome on a proportion of the occasions it is performed.
- A second source of uncertainty is produced when the underlying causal structure is non-stationary, or volatile, for example, when action-outcome contingencies switch and an action that was primarily associated with a given outcome becomes predominantly associated with another.
If unexpected or ‘surprising’ outcomes are caused by noise, then current action choices are optimally determined by averaging over the outcomes of many previous actions. In contrast, if surprising outcomes are caused by a change in action-outcome contingencies in a volatile environment, then only the most recent events should be used to guide action choice.
In terms of formal reinforcement accounts of learning, a higher learning rate should be implemented when the environment is volatile than when it is stable. Computational studies of decision-making reveal that healthy volunteers do indeed adapt their learning rate in response to changes in environmental volatility. Specifically, the behavior of participants is consistent with faster updating of action-outcome contingencies in volatile than in stable environments. This change in learning behavior occurs even when changes in environmental volatility are not explicitly cued and has been shown to closely track an optimal Bayesian decision-making strategy.
Individuals prone to anxiety might either show:
- a general deficit in updating action-outcome estimates following unexpected aversive outcomes or
- a specific problem with adjusting the speed of updating (that is, learning rate) to reflect the stability or volatility of the current environment.
Recent Pavlovian fear conditioning findings suggest that trait-anxious individuals struggle to adjust fear down regulation to reflect changes in stimulus-stimulus contingencies between contexts. If individuals prone to anxiety have a particular difficulty in processing contingencies that change over time or between contexts, we might predict that they will show a deficit in using changes in environmental volatility to infer whether or not action-outcome contingencies have changed, and to alter their behavioral choices.
Statistical regularities in the causal structure of the environment enable us to predict the probable outcomes of our actions. Environments differ in the extent to which action-outcome contingencies are stable or volatile. Difficulty in being able to use this information to optimally update outcome predictions might contribute to the decision-making difficulties seen in anxiety.
- Human participants low in trait anxiety matched updating of their outcome predictions to the volatility of the current environment, as predicted by a Bayesian model.
- Individuals with high trait anxiety showed less ability to adjust updating of outcome expectancies between stable and volatile environments. This was linked to reduced sensitivity of the pupil dilatory response to volatility, potentially indicative of altered norepinephrinergic responsivity to changes in this aspect of environmental information.
Such a deficit would mean difficulty in differentiating contexts or periods of time in which unexpected aversive outcomes should be ignored as chance events and those in which unexpected aversive outcomes are likely to signal a change in action-outcome contingencies, and as such should lead to a change in action. This could potentially explain why individuals at trait risk for anxiety show:
- intolerance of uncertainty
- poor decision-making
…in real-life settings in which the relationship between alternate courses of actions and avoidance of future negative events is often uncertain.….when attempting to avoid aversive outcomes, high trait-anxious individuals would be less able than low trait-anxious individuals to adjust their updating of action-outcome contingencies in response to changes in environmental volatility.
…the difference in learning rate between stable and volatile task blocks would be reduced in high trait-anxious individuals.
Recent findings indicate that pupil dilation may track important changes in the causal statistics of the environment, including changes in environmental volatility…trait anxiety–related deficits in differential learning between volatile and stable task blocks would be accompanied by a reduced pupil dilatory response to environmental volatility.
Consistent with our predictions:
- high trait-anxious individuals showed a specific deficit in adjusting learning rate in response to changes in environmental volatility in our aversive learning task.
- This was associated with a reduced pupil response to trial-wise estimates of environmental volatility.
- These results provide evidence that trait vulnerability to anxiety is associated with impoverished use of environmental statistics, especially that pertaining to environmental volatility, to determine the extent to which to update action-outcome contingencies when attempting to avoid aversive outcomes.
- This may represent a core deficit underlying impoverished decision-making in individuals at elevated risk of developing anxiety disorders.
…low trait-anxious individuals were able to adjust their learning appropriately…, learning (that is, updating outcome estimates) fast when the world was fast-changing and slowly when the world was stable. By contrast, high trait-anxious individuals were not able to learn in this flexible manner, instead learning similarly in both blocks…This suggests that high trait-anxious individuals did succeed in updating outcome expectancies following surprising outcomes, but that they were unable to modulate this on the basis of the volatility of the current environment.
…Participants with low levels of trait anxiety altered their learning rate between the stable and volatile blocks to an equivalent degree to the ideal Bayesian learner. As trait anxiety levels increased, participants diverged increasingly from the optimal change in learning rate between stable and volatile blocks described by the Bayesian learner model, showing reduced adaptation of learning rate.
As outlined earlier, when learning rate is high, recent trial outcomes inform behavioral choice to a greater extent. Consistent with elevated trait anxiety being associated with reduced adaptation of learning rate, high trait-anxious individuals showed less adjustment between volatile and stable blocks in their use of recent trial outcomes to inform choices than low trait-anxious individuals.
RESPONSES TO SURPRISING OUTCOMES ARE NOT MODULATED BY TRAIT ANXIETY.
…In the reinforcement learning literature, it has been reported that, when an outcome is surprising, participants typically show slowed decision-making on the subsequent trial. Consistent with this, we observed that, across participants, surprising outcomes were indeed associated with slowed choice reaction times on the subsequent trial. However, the effect of surprise on participants’ reaction times was not modulated by participant trait anxiety. This finding is consistent with high trait-anxious participants being specifically insensitive to changes in environmental volatility rather than simply unable to process surprising outcomes.
Trait Anxiety Modulates The Pupil Response To Volatility
As reported above, we observed that high trait-anxious individuals showed impoverished adjustment of learning rate between the stable and volatile blocks of the aversive learning task employed here, but did not show impoverished adjustment of choice reaction times following surprising outcomes.
This is consistent with a specific deficit in the use of environmental volatility to adjust action-outcome updating and, through this, to guide decision-making…high trait-anxious participants show a reduced pupil dilatory response to environmental volatility alone…Elevated trait anxiety was associated with a decreased mean pupil response to volatility…
We further addressed the possibility that high anxious individuals’ pupillary response does in fact track environmental volatility, but in a manner less like the Bayesian learner… an inverse relationship between trait anxiety and the effect of volatility, but not surprise, on the post outcome pupil response.
Our results suggest that people are generally able to rationally adapt their learning about aversive outcomes on the basis of whether action-outcome associations are volatile or stable. However, individuals with high trait anxiety show an impoverished ability to do this. This difficulty in using information about the stability of action-outcome contingencies to correctly judge whether or not to repeat an action that has led to an unexpected aversive outcome may well lead high trait-anxious individuals to engage in poor decision-making.
It might also result in aversive outcomes being experienced as less predictable and less avoidable. This could in turn lead to an increase in anxiety-related symptomatology, and potentially even be involved in the onset or maintenance of anxiety disorders.
Non-luminance–related changes in pupil size are argued to reflect, among other influences, the activity of central arousal systems…Activity of this system is held to be closely linked to the processing of environmental volatility and the use of this to guide learning. Consistent with this, we observed that:
- participants showed a significantly greater increase in pupil dilation following trial outcomes in which environmental volatility was high.
- Furthermore, the degree to which participants’ post-outcome pupil dilatory response tracked environmental volatility was significantly correlated with the degree to which they adjusted their learning rate between stable and volatile blocks.
- Of greatest pertinence to our study, trait anxiety modulated this pupil response to volatility, with high trait-anxious participants showing a smaller effect of volatility on post-outcome pupil dilation.
….individuals with high levels of trait anxiety appear to respond normally to the experience of unexpected aversive outcomes, but are unable to utilize higher order statistical information present in the distribution of these outcomes to modulate the updating of action-outcome contingency estimates. These findings also indicate that any anxiety-related deficit in the norepinephrinergic control of learning mechanisms would have to be fairly specific in nature.
In conclusion, our findings reveal that:
- trait vulnerability to anxiety is associated with a deficit in the use of higher order statistics about the causal structure of adverse environments to guide decision-making.
- High trait-anxious individuals did not differ from low trait-anxious individuals in their mean learning rate or in their behavioral or pupillary response to surprising adverse outcomes.
- Nor did they show altered preferences for minimizing shock probability versus shock magnitude.
- Instead, their pattern of decision-making was indicative of a selective difficulty with differentially updating action-outcome contingencies as a function of whether the current environment was stable or volatile.
- Our pupilometry data also confirmed a specific insensitivity to environmental volatility and raised the possibility that this might reflect impoverished modulation by environmental volatility of activity in the locus coeruleus–norepinephrine system.
In everyday life, determining whether, given the current context, an unexpected negative outcome is probably a chance event or something likely to occur again if the action that led to it is repeated may be essential to personal relationships and work place judgments.
A deficit in this aspect of learning may have an important maintaining, or even etiological, role in the anxiety experienced by high trait-anxious individuals. We have taken a step toward elucidating this deficit, and hope to have illustrated how computational models can be integrated with behavioral and pupillometry analyses to begin to identify the mechanisms underlying disrupted decision making in high trait-anxious individuals.