Approximating individual interactions in compartmental system dynamics models

Abstract

AbstractThis paper introduces an approach known as a “pair approximation” developed in physics and mathematical biology that captures some of the effects of network structure in a compartmental system dynamics (CSD) model without substantially increasing model complexity or computational efficiency. We describe the approach using a familiar example: the standard SIS model from epidemiology that underpins many models from the marketing and innovation diffusion literature. We show that the pair approximation better approximates the results of an agent‐based simulation than the standard CSD. The model provides insights regarding the epidemic threshold, with implications for halting the spread of a disease or encouraging the spread of a product, that cannot be obtained from the standard CSD. We further illustrate the technique through the development of two new CSD models: a model of social contagion, SIS2, that allows for “social recovery,” and a model of technology adoption with local network effects.Copyright © 2018 System Dynamics Society