Invited commentary: Agent-based models for causal inference—reweighting data and theory in epidemiology

Abstract

The relative weights of empirical facts (data) and assumptions (theory) in causal inference vary across disciplines. Typically, disciplines that ask more complex questions tend to better tolerate a greater role of theory and modeling in causal inference. As epidemiologists move toward increasingly complex questions, Marshall and Galea (Am J Epidemiol. 2015;181(2):92-99) support a reweighting of data and theory in epidemiologic research via the use of agent-based modeling. The parametric g-formula can be viewed as an intermediate step between traditional epidemiologic methods and agent-based modeling and therefore is a method that can ease the transition toward epidemiologic methods that rely heavily on modeling.

Keywords: agent-based models; causal inference; parametric g-formula.

Figure 1.

The relative position of several scientific disciplines along the causal inference spectrum according to the relative weights of data and theory.