worry as planning

Study synopsis

A secondary purpose of the understanding how humans infer when to initiate planning is to understand whether and how this is disrupted in chronic worry. Indeed, those who worry bring online aversive planning in situations where it only harms them. This may be the result of poor meta-control over the planning process. As discussed in the Delayed Planning post, the task we used to study this phenomenon involved participants learning the cognitive map displayed below by randomly starting at a specific image (e.g., car) and being instructed to take a specific action (left or right). This action deterministically transitioned them to a subsequent image, reflected in the black arrows. Participants learned these 1-step transitions and were quizzed on them. They were told they could then use this knowledge to plan and win money in a later phase of the task. Note, all variations of this experiment presented here are pilot studies (Study 1 with time pressure n=14; Study 2 without time pressure n=16; Study 3 n=10).

Planning depth determines when to start planning

Participants were told that they would always start by deciding between the images at the top of the map – here the trident or the planet. Their decision was based on an instructed reward they could plan for. For example, they could be told 400 points is hiding behind the image of the snorkel, and they can plan to get to the snorkel from the very first decision. As you see below, if the reward is at the snorkel, they NEED to start planning from their 1st of 4 decisions, because if they don’t choose the planet, they cannot arrive at the snorkel. Therefore, the snorkel has a planning depth of 4, requiring all 4 decisions to be planned out starting from the very first decision. When participants took an action, they arrived at a new image determined by the map structure, and could continue executing a plan or initiating a new plan.

Each goal had a different planning depth, such as the tophat which had a planning depth of 2 and only required planning to begin at the 3rd decision.

How we incentivized delayed planning

For each of the 4 decisions participants faced during a planning trial, they were given the option to have the computer randomly choose their action for them, which would gift them 100 points. Thus, if they didn’t need to plan at a given stage, they could give up control and win points for doing so. The predicted pattern for giving up control optimally, which is based on when you need to initiate planning, is depicted below. As an example, if the planning depth is 3, you should give up control at the very first decision because you only need to plan at the second decision. Importantly because we thought it may take participants a number of trials with a given goal to infer when to initiate planning, we had participants plan for each goal 20 separate times. Goals were instructed in random order.

Failure to delay planning under time pressure

Remarkably, when participants were under time pressure, which in our task meant that the longer the took the less points they could get, they failed to delay planning when they should.

Worriers fail to plan under time pressure

Perhaps even more interesting, for subjects the reported they worried frequently, they failed to plan successfully at all under these stressful conditions. Here we measured if participants arrived at the instructed goal.

Humans infer when to start planning

When we removed time pressure, we saw individuals in general successfully inferred when to initiate planning.

Chronic worry is associated with impaired inference to delay planning

A motivating clinical hypothesis that we investigated is whether individuals that worry frequently plan too much. Intuitively we know such individuals are constantly planning how to avoid dangerous futures even when it isn’t helpful to do so, given how exhausting planning can be. We show here the first experimental evidence that chronic worry is associated with over-planning (specifically, a failure to optimally delaying planning). We created a continuous score, shown on the y-axis of the graph below, which quantifies the degree to which participants delayed control when they should and engaged control when they should. The correlation (r = -0.63) replicates a correlation we found above in the more complicated 4-decision task.