Formalizing the Role of Agent-Based Modeling in Causal Inference and Epidemiology

Abstract

Calls for the adoption of complex systems approaches, including agent-based modeling, in the field of epidemiology have largely centered on the potential for such methods to examine complex disease etiologies, which are characterized by feedback behavior, interference, threshold dynamics, and multiple interacting causal effects. However, considerable theoretical and practical issues impede the capacity of agent-based methods to examine and evaluate causal effects and thus illuminate new areas for intervention. We build on this work by describing how agent-based models can be used to simulate counterfactual outcomes in the presence of complexity. We show that these models are of particular utility when the hypothesized causal mechanisms exhibit a high degree of interdependence between multiple causal effects and when interference (i.e., one person's exposure affects the outcome of others) is present and of intrinsic scientific interest. Although not without challenges, agent-based modeling (and complex systems methods broadly) represent a promising novel approach to identify and evaluate complex causal effects, and they are thus well suited to complement other modern epidemiologic methods of etiologic inquiry.

Policy, Work Organization and Sleep Health and Safety of Commercial Drivers: Introducing a Complex Systems Paradigm

“…Upstream distal factors-those which dictate work organization, for example, such as labor policies related to deregulation and driver compensation-are neglected in reductionist-oriented mental models, as are broad spatiotemporal influences and outcomes. Guided by such narrow mental models, corresponding etiological models place causal factors into "silos" to be modeled and analyzed in the pursuit of specific cause-effect relationships [22,26].…”

“…Traditional statistics are inherently reductionist [29], and characteristics of many real-world problems conflict with these analyses. Not only are most linear statistics limited to a single level of influence [29], their assumptions regarding cause-effect relationshipsincluding that these relationships are linear and independent [26]contradict reality.…”

Impacts of licensed premises trading hour policies on alcohol‐related harms

“…They can be used to conduct experiments not possible in the real world with the aim of identifying the most acceptable and effective policy responses in particular contexts, and importantly, provide a mechanism for estimating the impacts, both beneficial and adverse, of a policy option into the future 17. Agent‐based models (ABMs) are a class of computational model that provide powerful tools for simulating human behaviour due to their ability to capture the interacting influences of individual characteristics, social networks, local context, and the broader economic and policy environment 18 that are important drivers of alcohol behaviours and risk of harms. ABMs are underpinned by complexity theory, with their architecture drawing on both theory and empirical data as they attempt to create a virtual representation of real world systems in order to better understand the nature and consequences of human behaviour.’ In practical terms, agents (individuals) in the model are given key characteristics and exposures that reflect those in the real world.…”

Policy instruments to control Amazon fires: A simulation approach

“…Part of the richness of the results is very hard to reduce to refutable hypotheses that may guide policy evaluation. As producing such hypotheses is one of our main goals, we opted for a causal inference approach to simulation analysis (Marshall and Galea, 2014). This means focussing on comparing policy outcome variables in baseline and policy scenarios, rather than exposing the plethora of patterns the variables describe across time and space.…”

“…Simulation models are different from models with analytic (pen-and-paper) solutions in that they do not necessarily yield identification. Non-linearity and stochasticity, coupled with endogeneization of most variables, makes it hard to track the causes of the observed behavior of the main variables (Marshall and Galea, 2014). This difficulty grows with realism (El-Sayed et al, 2012, Cederman and Giradin, 2007, Townsley and Birks, 2008.…”