afttest: Model Diagnostics for Accelerated Failure Time Models

The R package afttest is intended to be a user-friendly supplementary package to the base package aftgee.

Features

In addition to the R interface, afttest provides a C++ header-only library integrated with Rcpp, which allows the construction of spline basis functions directly in C++ with the help of Rcpp and RcppArmadillo.

Installation of CRAN Version

You can install the released version from CRAN.

install.packages("afttest")

Development

The latest version of the package is under development at GitHub. If it is able to pass the automated package checks, one may install it by

if (!require(devtools)) install.packages("devtools")
devtools::install_github("WoojungBae/afttest")

Performance

The implementation of the main functions has been written in C++ with the help of the Rcpp and RcppArmadillo packages. The computational performance has thus been boosted.

Replication Materials & File Manifest

As per the computational demands of the Monte Carlo simulations, the replication materials for the empirical results are divided into the following self-contained files.

1. Main Analysis (Fast Execution)

• afttest_replication_reduced_script.R This is the main replication script for the manuscript. It reproduces the results reported in the paper that are intended for routine verification on a standard laptop or desktop computer. In particular, it generates the empirical analysis, summary outputs, and the reduced simulation results reported in the article.

• afttest_replication_script.R

• Description: This is the primary script containing the code to reproduce the main empirical data analyses, summary statistics, and all Figures presented in the manuscript (e.g., the Primary Biliary Cirrhosis dataset analysis).

• Execution Time: Fast (Runs on a standard desktop in minutes).

2. Table 1: Computational Time Simulation (Long-Running)

• package_sim2_hpc_TIME.R & batchAFTsim2_hpc_TIME.sbatch
• Description: These files generate the data required to reproduce Table 1 (Average running times comparing the asymptotic linear approximation vs. standard resampling).
• Execution Time: [WARNING: EXTREMELY LONG-RUNNING] These simulations are highly computationally intensive. The provided .sbatch file is configured for execution on a High-Performance Computing (HPC) cluster (e.g., University of Florida HiPerGator).

3. Table 2: Empirical Rejection Rates Simulation (Long-Running)

• package_sim2_hpc.R & batchAFTsim2_hpc.sbatch
• Description: These files generate the data required to reproduce Table 2 (Empirical Type I error and power/rejection rates).
• Execution Time: [WARNING: EXTREMELY LONG-RUNNING] Similar to Table 1, this script involves thousands of Monte Carlo replications and resampling paths. It is designed to be submitted to an HPC cluster via the provided .sbatch file.

Important Note on the Simulations Reported in the Paper

Due to the journal’s time and computational constraints, the simulation results reported in the paper were based on a reduced set of Monte Carlo runs and parameter settings.

Simulations reported in the paper

• Table 1: Only the omnibus test was considered, with sample sizes n = 100 and n = 500, each based on 10 simulation runs, comparing linApprox = TRUE versus linApprox = FALSE.
• Table 2: For each combination of omnibus, link, and covForm, simulations were run with sample sizes n = 100 and n = 500, each based on 100 simulation runs, with gamma = 0.0 and 0.5. Results report the empirical Type I error and power for linApprox = TRUE.

Extended Simulations Available in the Replication Folder

Running the scripts in the replication folder produces a broader set of simulation results than those reported in the paper.

• Table 1: For each combination of omnibus, link, and covForm, simulations are run with sample sizes n = 100 and n = 500, each based on 1000 simulation runs.
• Table 2: For each combination of omnibus, link, and covForm, simulations are run with sample sizes n = 100 and n = 500, each based on 1000 simulation runs, with gamma = 0.0, 0.1, 0.2, 0.3, 0.4, and 0.5. Results report the empirical Type I error and power for linApprox = TRUE.

Instructions for Execution

1. Ensure all files are in the same working directory.

2. To reproduce the main text examples and figures: Open R or RStudio, set your working directory to this folder, and source the main script:

   source("afttest_replication_script.R")

3. To reproduce Tables 1 and 2 (HPC environment): If you have access to a Slurm-based HPC cluster, submit the batch jobs directly from the terminal:

   sbatch batchAFTsim2_hpc_TIME.sbatch
   sbatch batchAFTsim2_hpc.sbatch

   Note: To run the simulation scripts locally for verification, open the respective R scripts, drastically reduce the sim_per_file parameter, and execute them in your R console.

Reference

Bae, W., Choi, D., Yan, J., Kang, S. (2026). Diagnostics for semiparametric accelerated failure time models with R Package afttest. Journal of Statistical Software, to appear, arxiv.org/abs/2511.09823.

Choi, D., Bae, W., Yan, J., and Kang, S. (2024). A general model-checking procedure for semiparametric accelerated failure time models. Statistics and Computing, 34(3) 117.

Bae, W., Choi, D., Yan, J., Kang, S. (2022). afttest: Model Diagnostics for Accelerated Failure Time Models. CRAN, https://CRAN.R-project.org/package=afttest.

Chiou, S., Kang, S., and Yan, J. (2014). Fitting accelerated failure time model in routine survival analysis with R package aftgee. Journal of Statistical Software, 61(11): 1–23.