Improving the Reliability of Computational Analyses: Model-Based Planning and Its Relationship With Compulsivity

Area of Science:

• Computational psychiatry
• Neuroscience
• Reinforcement learning

Background:

• Computational models offer insights into decision-making processes and neural substrates.
• Model-based planning in reinforcement learning is linked to transdiagnostic compulsivity.
• The reliability of computational model-derived measures is not well-established.

Purpose of the Study:

• To assess the reliability of model-based planning measures in patients with compulsive disorders.
• To investigate the impact of different analysis approaches on the reliability of these measures.
• To determine if reliability improvements generalize to larger, independent datasets.

Main Methods:

• Assessed within- and across-session reliability of model-based planning.
• Evaluated effects of model estimation, parameterization, and data cleaning.
• Tested reliability in patient (n=38) and large generalization samples (n=541, 111, 1413).

Main Results:

• Analysis approaches significantly influenced reliability, ranging from 0 to over 0.9.
• Robust model-estimation accounting for hierarchical parameters most impacted reliability.
• Reliability improvements generalized and reduced required sample sizes.

Conclusions:

• Computational psychiatry measures, like model-based planning, can reliably assess decision-making.
• Methodological choices in data analysis are critical for ensuring measure reliability.
• Accurate estimation of complex models from limited data is essential for reliable computational psychiatry.