Background

Imagine you have an important meeting in the morning and are about to take your usual route to work. Just then, one of your colleagues suggests an alternative shortcut that might be faster. You are now faced with a decision: adopt the newly learned path or stick with the one you know well? It is conceivable that many of us would choose to take the more familiar route, despite the potential advantages of the new option. This choice is neither irrational nor necessarily suboptimal. Rather, it draws upon our proceduralized knowledge accumulated through repeated experience—knowledge that encompasses not only navigation but also the ability to anticipate and respond to disruptions such as traffic congestion, road closures, or detours. In this context, choosing the familiar route, even if it may entail slightly longer travel time, represents an adaptive decision-making strategy to ensure timely arrival at work. Here, the road “shortcut” stands in contrast to the mental “shortcut” people typically rely on.

Mental shortcuts in decision-making—commonly referred to as heuristics—have been widely observed across various decision contexts. They highlight the essence of real-world decision-making under limited time and knowledge as the selection of the most adaptive option based on available resources, rather than the theoretically optimal choice. Consistent with this idea of ecological rationality, previous studies have identified a cost-benefit trade-off, where individuals may allocate additional cognitive resources to engage in a more controlled, calculative decision-making process—only when the benefits of identifying the optimal option are high and/or the cognitive cost of doing so is low. This trade-off motivates our current work to understand the transition mechanism between heuristic and deliberative systems.

Frequency Effects

In two recent studies from our lab, we observed frequency effects in a binary-outcome choice task, where more frequently presented options were preferred over more rewarding, but less frequently presented alternatives. Importantly, this effect emerged only when reward probability differences were small; it was absent when one option was clearly superior. We reasoned that in binary-outcome choice tasks, there was likely substantial uncertainty concerning the true reward rate for each alternative. As a result, computing precise expected values (EVs) may become prohibitively demanding, especially in the face of only marginal potential benefits, which prompts individuals to use reward frequency as a heuristic proxy for value. We therefore postulated that frequency effects may arise from a distinct, heuristic “gist-based” system—one that approximates value using reward frequency when precise EV estimation is either unavailable or overly costly. If this interpretation is correct, then reducing uncertainty should weaken or eliminate the influence of frequency accordingly.

Current Study

To test this idea, we used a task with continuous rewards drawn from normal distributions centered around the same mean reward rates used in our prior binary-outcome studies. We first matched the reward standard deviations to those derived from the binomial distribution in the binary task. This condition was the high variance condition. We then systematically reduced the variance by half and by one-fourth to create moderate and low variance conditions, respectively.

We predicted reduced, or no frequency effects in the moderate and low variance conditions, and an intact frequency effect in the high variance condition. Our results aligned perfectly with our predictions: participants preferred the more frequently rewarded but less valuable option in the high variance condition, showed no significant preference in the moderate condition, and preferred the objectively more valuable option in the low variance condition. This proportional shift in preference based on reward variance strongly supports our speculation that uncertainty drives the strategic transition between the heuristic frequency-based system and the deliberative value-based system.

To formally capture this mechanism, we developed a computational dual-process model that incorporates separate cognitive systems—value-based and frequency-based systems—modeled by a Gaussian and a Dirichlet distribution, respectively. Uncertainty within each system is quantified as the differential entropy of its respective distribution, with higher decision weight given to the system showing lower entropy (i.e., lower uncertainty). In this way, the model operationalizes uncertainty as the key determinant of system selection. Notably, this dual-process model substantially outperformed standard reinforcement learning models in predicting participant behavior. As traditional models often overlook the role of uncertainty or fail to translate it into strategy switching, our model uniquely captures the uncertainty-driven strategic adaptation underlying each decision.

Summary

Our findings offer new insight into how people dynamically balance “gist-based” and value-based reasoning. The results suggest that individuals do not simply follow fixed decision rules, but instead engage in active strategic switching based on perceived uncertainty. By formalizing this process into a distributional dual-process model, we provide a principled account of how cognitive systems allocate control over decision-making systems—offering a fresh perspective on the computational mechanisms that shape real-world decision-making.