Abstract

Abstract for: Bridging Causal Loop Diagrams and Quantitative Simulation: A Data-Informed Framework for Estimating Latent Social Dynamics

Introduction Translating causal loop diagrams (CLDs) into empirically parameterized simulation models remains a challenge in system dynamics, particularly in socio-behavioral systems. In these systems, key variables are often latent, and available data are typically repeated cross-sectional rather than longitudinal. We develop a methodological framework to empirically calibrate quantitative system dynamics models that preserves CLD structure through the systematic integration of qualitative knowledge and cross-sectional data.
Approach Our framework extends CLDs from qualitative representations of feedback structure to quantitative constraints within a discrete-time linear dynamical system. Sign and sparsity restrictions derived from the CLD restrict the admissible interaction matrix, while pathway strengths are estimated to jointly match time-varying patterns in repeated cross-sectional data and the covariance structure implied by the discrete Lyapunov equation. This formulation constitutes a constrained dynamical inference problem.
Results We apply this framework to an HIV prevention case study using repeated cross-sectional data from the American Men’s Internet Survey. Early results show the approach generates trajectories consistent with population-level patterns while preserving CLD-derived feedback relationships. Estimated pathway strengths suggest geographic heterogeneity in latent dynamics and a consistent inhibitory effect of healthcare engagement on anticipated healthcare stigma. These results remain exploratory but align with patterns observed in the literature.
Discussion Our findings suggest that constrained dynamical inference may offer a scalable methodological framework for translating qualitative CLDs into empirically informed dynamic models. Rather than specifying mechanistic equations a priori, it can infer latent relationships consistent with both data and feedback structure. In doing so, it enables analysis of socio-behavioral systems under uncertainty. Ongoing work focuses on validation, extension to nonlinear dynamics, and tangible applications to intervention and policy analysis.

Introduction	Translating causal loop diagrams (CLDs) into empirically parameterized simulation models remains a challenge in system dynamics, particularly in socio-behavioral systems. In these systems, key variables are often latent, and available data are typically repeated cross-sectional rather than longitudinal. We develop a methodological framework to empirically calibrate quantitative system dynamics models that preserves CLD structure through the systematic integration of qualitative knowledge and cross-sectional data.
Approach	Our framework extends CLDs from qualitative representations of feedback structure to quantitative constraints within a discrete-time linear dynamical system. Sign and sparsity restrictions derived from the CLD restrict the admissible interaction matrix, while pathway strengths are estimated to jointly match time-varying patterns in repeated cross-sectional data and the covariance structure implied by the discrete Lyapunov equation. This formulation constitutes a constrained dynamical inference problem.
Results	We apply this framework to an HIV prevention case study using repeated cross-sectional data from the American Men’s Internet Survey. Early results show the approach generates trajectories consistent with population-level patterns while preserving CLD-derived feedback relationships. Estimated pathway strengths suggest geographic heterogeneity in latent dynamics and a consistent inhibitory effect of healthcare engagement on anticipated healthcare stigma. These results remain exploratory but align with patterns observed in the literature.
Discussion	Our findings suggest that constrained dynamical inference may offer a scalable methodological framework for translating qualitative CLDs into empirically informed dynamic models. Rather than specifying mechanistic equations a priori, it can infer latent relationships consistent with both data and feedback structure. In doing so, it enables analysis of socio-behavioral systems under uncertainty. Ongoing work focuses on validation, extension to nonlinear dynamics, and tangible applications to intervention and policy analysis.