simulations/README.md

In benkeser/intermed:

R/intermed/`simulations

Simulation studies for manuscript "Nonparametric estimators of mediation effects with multiple mediators"

Author: David Benkeser and Jialu Ran

Description

This repository houses simulation code for replicating simulation results. There are five simulations. Each simulation is run at four sample sizes on one thousand simulated data sets.

The code is organized using a Makefile with package management via the renv package.

The general work flow for simulations is as follows.

1. load project environment

Install the renv package in R (e.g., via install.packages). From the simulations folder, synchronize the package environment as follows.

Rscript -e "renv::restore()"

This command will install packages from the project lockfile.

1. make the simulation parameters

make output/parm.RData

This creates the .RData file that defines the parameters for simulation jobs.

1. make the simulation results

make simulation_results

This command will only work on a slurm cluster -- it submits 20k jobs using sbatch. Some modification to the Make recipe may be needed depending on cluster configuration.

If a cluster is not available, then it should be straightforward to follow the logic of the code to reproduce the results using some other form of high throughput computing.

1. merge the simulation results

The simulation_results recipe generates 20k output files that need to be merged.

make merge_results

1. make the figures

make figures

This makes all figures presented in the main text and supplement.

1. make the heatmaps illustrating mediator distributions

make heatmaps

G-computation simulation

The code for executing the G-computation simulation included in the supplement is included in a separate repository.

benkeser/intermed documentation built on Dec. 19, 2021, 8:43 a.m.