Sensitivity Analysis of Computational Models in Psychology: A Tutorial
Year: 2026
Type: preprint
Abstract: There has been an increasing interest in formal models in psychology for understanding psychological systems and informing interventions. However, methods and tools to understand and analyse these models are largely unfamiliar and rarely used in psychological research. This paper provides an introduction to global Sensitivity Analysis (SA), a technique from computational science that can be used to quantify the influence of model parameters on an output variable. An SA allows to identify the most relevant parameters for causing specific changes in a model output, such as suicidal thoughts or frequency of panic attacks. Here we focus on the variance-based Sobol SA and the distribution-based PAWN methods. The utility of these techniques is demonstrated using an existing computational model of suicidal thoughts in a didactic tutorial-like fashion, supplied with (annotated) code in R. The results are connected to empirical literature and implications for model evaluations are highlighted. To facilitate the application of SA, we aggregate the entire process into a workflow and code pipeline that can be easily applied to other models. The introduced methodology will help to optimize and simplify the modelling process and increase interpretability in psychological research. We also identify and discuss further extensions and research directions such as model calibration.
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7161268958