An Entropy-initiated Coupled-Trait ODE Framework for Modeling Longitudinal Cohort Dynamics

TL;DR

This paper presents a novel entropy-initiated coupled-trait ODE framework for modeling longitudinal cohort dynamics, effectively capturing population trends with low-dimensional, interpretable models validated across multiple datasets.

Contribution

It introduces an information-theoretic dynamical system that uses entropy measures to initialize and inform low-dimensional cohort models, offering a transparent alternative to complex machine learning methods.

Findings

Successfully reproduces cohort trajectories from Swedish data

Demonstrates generalization across different cohorts and measurement instruments

Achieves stable out-of-sample forecasting performance

Abstract

This work introduces a minimal, information-theoretic dynamical framework for modeling longitudinal cohort data using an entropy-initiated system of coupled-trait ordinary differential equations (ECTO). For each survey wave, item-level Likert responses are compressed into a normalized Shannon entropy index that summarizes cross-sectional dispersion; this index is used to initialize the low-dimensional state variables of the autonomous ODE system. ECTO then tracks the interactions among a primary trait-like state, a secondary coupled state, and a latent environmental-stress component through phenomenological terms representing generic self-limitation, trade-offs, and feedback. Using data from the Swedish Adoption/Twin Study on Aging (SATSA), the framework reproduces broad cohort-level trajectories and is evaluated with leave-one-wave-out forecasting and comparisons against simple statistical baselines. A second longitudinal dataset of U.S. dental student data provides an external validation test, demonstrating that low-dimensional dynamics initialized from entropy measures can generalize across cohorts with different measurement instruments, demographic compositions, and timescales. Across both datasets, ECTO achieves stable out-of-sample performance, indicating that major cohort-level trends can be captured without assuming complex latent-variable models or time-varying causal inputs. Entropy here functions as a compact summary of population heterogeneity rather than a dynamical driver, and the coupled ODEs supply an interpretable alternative to high-dimensional or black box machine-learning approaches. This framework establishes a concise, transparent method for linking information-theoretic preprocessing with cohort-level dynamical modeling and provides a foundation for future multivariate or multi-cohort extensions.