A Review and Synthesis of Multi-level Models for Causal Inference with Individual Level Exposures

Purpose of reviewMulti-level models are ways to model data using multiple levels of information. Here, we provide a narrative review some of the relevant literature on how multi-level models can interface with causal inference for individual level exposures.Recent findingsMuch of this discussion focuses on clarifying and synthesizing some of the complex ideas in the literature. We discuss how multi-level models can be seen as approximate ways to relax some of the identifying conditions of causal inference.There are significant gaps in the literature on causal inference with multi-level models, but we list some published approaches for further guidance. We close with some practical advice on when multi-level models might be best utilized for causal inference and how they might be used in ways that go beyond simply interpreting their (potentially highly conditional) parameters.

Automated summary

Multi-level models are used to model data using multiple levels of information. This review examines how multi-level models can be used for causal inference with individual level exposures. The review clarifies and synthesizes complex ideas in the literature and discusses how multi-level models can relax some identifying conditions of causal inference. However, there are gaps in the literature on causal inference with multi-level models, but some published approaches are provided for further guidance. Practical advice is given on when to use multi-level models for causal inference and how to go beyond interpreting their parameters.