simcausal R Package: Conducting Transparent and Reproducible Simulation Studies of Causal Effect Estimation with...

Area of Science:

• Causal Inference
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Complex longitudinal data present challenges in causal inference.
• Time-dependent confounding, selection bias, and monitoring processes complicate real-world data analysis.
• Existing tools may lack flexibility for diverse causal inference scenarios.

Purpose of the Study:

• Introduce the simcausal R package for specifying and simulating complex longitudinal data.
• Provide a flexible tool for transparent and reproducible causal inference simulation studies.
• Address common challenges in real-world causal inference, including time-dependent confounding and selection bias.

Main Methods:

• Utilizes non-parametric structural equation models for data structure specification.
• Enables simulation of counterfactual data under static, dynamic, deterministic, or stochastic interventions.
• Allows concise expression of complex functional dependencies between numerous time-dependent nodes.

Main Results:

• The simcausal package supports the simulation of various interventions, including treatment regimens and censoring events.
• It facilitates the computation of key causal quantities like average treatment effects and marginal structural model coefficients.
• Demonstrated applicability by replicating results from two published simulation studies.

Conclusions:

• simcausal offers a robust and flexible platform for causal inference simulation studies.
• The package simplifies the analysis of complex longitudinal data with time-dependent confounding and selection bias.
• It enhances the transparency and reproducibility of causal inference research.