This repository contains the data and scripts needed to reproduce results from the manuscript by Grosser, Lotspeich, and Garcia (2023+). This manuscript introduces the novel method "actice correction for error in imputation" (ACE imputation), which adjusts for the errors incurred when imputing censored covariates.
The ACEimpute package, which corrects for imputation error, can be found in this repo here. The imputeCensoRd package, which implements the conditional mean imputation approaches from the paper, can be found in its own repo here.
# Run once: install.packages("devtools")
devtools::install_github(repo = "Tanya-Garcia-Lab/ACEimpute/ACEimpute")
library(ACEimpute)
devtools::install_github(repo = "Tanya-Garcia-Lab/Imputing-Censored-Covariates/imputeCensoRd")
library(imputeCensoRd)
The scripts in this repo are coded to run 5 replication of each simulation setting for demonstration purposes. In the manuscript listed above, all simulation settings were run with 1,000 simulations each.
Tables 1 and 4. Simulation results comparing i) restricted maximum likelihood estimation (REML) with the full data, ii) conditional mean imputation (with a correctly specified imputaiton model) plus REML, and iii) conditional mean imputation (with a correctly specified imputaiton model) plus ACE imputation to correct for imputation error.
- Script (run simulation)
- Script (make table)
- Data (simulation replicates)
- Data (simulation results)
- Tables (manuscript Table 1 and Table 4)
Tables 2 and 5. Simulation results comparing i) restricted maximum likelihood estimation (REML) with the full data, ii) conditional mean imputation (with a misspecified imputaiton model) plus REML, and iii) conditional mean imputation (with a misspecified imputaiton model) plus ACE imputation to correct for imputation error.
- Script (run simulation)
- Script (make table)
- Data (simulation replicates)
- Data (simulation results)
- Tables (manuscript Table 2 and Table 5)
Figure 2. Plot power curves comparing sample sizes for a clinical trial based on estimates from i) complete case analysis, ii) conditional mean imputaiton plus REML, and iii) ACE imputation
