Imagine this: you are walking down the street when all of a sudden you see someone walking with a big dog by their side. You’re terribly afraid of dogs, so instead of walking past the dog, you cross the street. Why? Because, according to a recent study by Mkrtchian, Aylward, Dayan, Roiser & Robinson (2017), anxious people have an avoidance bias. This means they are more likely to avoid the things they’re afraid of.
In their study, the authors recruited 58 people without and 43 people with symptoms of mood and anxiety disorders. The participants did a computer task which had safe and threat blocks. What made the threat blocks so threatening, was that participants were of a risk of getting a shock. The safe and threat block were recognizable by their colour and a written message on the screen.
During a block, participants either had to perform a ‘go’ action to get a reward or to avoid punishment or perform a ‘no-go’ action to get the reward or avoid losing. When the block was a threat block, participants got a shock about 1/3rd of the time. So, in half of the trials, the performed action was the same as the action tendency (go/approach for winning the reward, and no-go/avoid the loss). In the other half, the performed action was not the same as the tendency (go/approach for avoiding loss and no-go/avoid for winning the reward).
You can look at the figure below from their paper to see what it looked like.
The authors set up seven parameterized reinforcement learning models that should fit the behaviour of the subjects. The models consisted of the parameters action (go/no-go), stimulus (go win, go avoid, no-go win, no-go avoid) and reinforcement (+1 for reward, -1 for punishment, 0 no feedback), which updated for each trial per participant. In the end, this means you get a sort of formula that represents the bias a participant has towards approaching or avoidance.
After this, they used a statistical procedure to fit the parameters across all subjects and conditions. They tested four different population distributions.
The winning reinforcement model had a parameter for approach bias, avoidance bias and separate speeds of learning about rewards and punishments. The best fitting population distribution was a single population distribution, so only one distribution for all participants and conditions. This implies that there is not enough evidence to suggest that the anxious and control groups come from different populations.
Having their predictive model with parameters of the winning model, they ran it as if the computer was each individual subject. The authors performed permutation tests on the data that came out of this. These tests showed that there was an increased reliance on the avoidance bias parameter in the anxious group, with a greater increase in the avoidance parameter under the threat condition versus the safe condition in the anxious group, compared to the control group. There was no difference between the groups in the safe condition.
So, in summary, these results suggest that people with mood and anxiety symptoms show a higher reliance on avoidance, especially during stressful situations. It is important to understand this underlying habit of people with mood and anxiety disorders because treatment can then target clear mechanisms instead of symptoms. For example, exposure therapy combined with behavioural training to overcome avoidance seems to be effective, but it is unclear why this is the case. The findings of this study suggest that behavioural training encourages the override of the avoidance bias during the action selection.
There is still much to explore when it comes to avoidance behaviour. For instance, the authors suggest looking at if training can overcome the bias to go avoid trials so that people would approach when their natural tendency is to avoid. This could improve treatment and be of huge value for public health.