Modeling Dimensions of Individual Variation in Adaptive Foraging Decisions

PROJECT SUMMARY/ABSTRACT Persistence toward prospective rewards is a critical element of normative real-world decision making. Equally important is the ability to disengage from goals that have diminished in value. Deficits in regulating goal- directed behavior are associated with impulsivity-related traits.

The present project will develop and test computational models to account for individual differences in the context-appropriate calibration of persistence.

We will employ a willingness-to-wait task paradigm in which human decision makers are given repeated opportunities to persist voluntarily toward delayed monetary rewards in a foraging-like environment.

The distribution of uncertain delay durations in the paradigm can be experimentally manipulated to create an environment in which either high or low persistence is advantageous.

Previous results from the same paradigm have shown that, on average, decision makers tend to adjust their behavior appropriately for their environment.

However, substantial differences across individuals have been observed in (1) overall levels of behavioral persistence, (2) the consistency of behavior while the environment remains stable, and (3) flexible adaptation when the environment changes.

We hypothesize that inter-individual heterogeneity can be accounted for in terms of individual differences in the latent parameters of behavior-generating computational models.

We further hypothesize that individual-specific parameter estimates will be proximally associated with dimensional trait measures of impulsivity.

We will test the hypotheses by implementing two novel theoretical models of adaptive persistence toward delayed rewards.

The first, a statistical learning model, hones an internal representation of reward timing on the basis of experience and produces adaptive persistence decisions using a planning mechanism.

The second, a motor preparation model, produces responses in a habit-like manner at times when responses have been cued in the past.

The internal structure of each model will be refined using task data from 160 community-based healthy volunteers, and parameter estimates will be tested for associations with trait variables. The models will then be validated and compared using an independent confirmatory sample (n = 400).

The two target models will be compared to one another, to a null model, and to an existing reinforcement learning model.

The results will establish a basis for future back-translational research in non-human model systems, given that the experimental foraging task is experience-based and non-linguistic.

It will also establish a basis for future studies examining the computational basis of dimensional constructs relevant to psychopathology in clinical populations.