Measuring individual differences in the depth of planning

Abstract

While making plans, people have to decide how far out into the future they want to plan: days, months, years, or even longer. Overly short-sighted planning can harm people's well-being in important life domains, such as health, finances, and academics. While self-report scales exist to measure people's planning, people's answers to such questions may be distorted by their desire to make a good impression and conform to norms and expectations. Here, we introduce a method for objectively quantifying people's propensity to plan into the future. Our method combines a process-tracing method with Bayesian inverse reinforcement learning to measure how prone an individual is to plan multiple steps ahead. To infer this from a person's process-tracing data, our method inverts a new resource-rational model of individual differences in planning. This model assumes that subjective planning costs are captured by a cost function with two parameters: a mental effort cost and a planning depth cost. Upon showing that our model of planning explains individual participants' planning behavior better than the best previous models, we validate our method on simulated data and real data from a large online experiment where the cost of planning was manipulated within participants. Our results show that our method can infer individual differences in the planning depth cost. Our model provides a mechanistic account for why some people plan too shortsightedly. The subjective planning costs inferred by our method can be used as an objective, non-self-report measure of individual differences in people's propensity to plan into the future.

What are the mechanisms underlying metacognitive learning?

“…This observation, combined with the high proportion of non-learning models, suggests that there is still room for improvement. Furthermore, planning incurs cognitive costs above and beyond the cost of acquiring information Felso, Jain, & Lieder, 2020;Callaway et al, 2022). Therefore, further work can improve our models by incorporating these additional costs into the reward signals that the models learn from.…”

Naturalistic reinforcement learning

“…Notably, this task also incorporated a predator agent whose actions were incorporated into the tree search simulations, highlighting how this approach can incorporate social information into the decision-making process. Another recent study further exploited this approach to demonstrate that individuals differ in the subjective cost of planning, explaining variability across individuals in planning depth [ 74 ]. Providing further support for the utility of this type of planning strategy in naturalistic environments, one study showed through simulation that tree search planning algorithms are especially advantageous in large, open environments where the agent has an extended visual range, characteristics of many of the terrestrial environments that humans inhabit [ 75 ].…”

Wagers for work: Decomposing the costs of cognitive effort

“…Tasks which subjects demanded the most incentives to complete (5)(6)(7) or which subjects tended to avoid in favor of other tasks with equivalent rewards (4,8,9) are considered most effortful. Some costly aspects of these tasks are external, like time on task (10)(11)(12) or the complexity of the cognitive model required by the task (13)(14)(15)(16), but other costs arise from the internal operations necessary to realize external actions.…”