Practical Guidelines for Goodness-of-Fit Testing for Right-Censored Data
Abstract
In survival studies, right censoring often prevents the observation of subjects' complete outcomes, complicating inference about the underlying distribution. Evaluating whether a given distribution adequately fits right-censored data is therefore essential. However, guidance on the relative performance of available goodness-of-fit tests under different data conditions remains limited. In this work, we present a comparative assessment of three classical goodness-of-fit tests adapted for right-censored data: Kolmogorov-Smirnov, Anderson-Darling, and Cramér-von Mises. Through an extensive simulation study, we investigate how factors such as the proportion of right-censored observations, sample size, and distributional symmetry influence the statistical power of each test. All procedures were implemented using the R package GofCens, and power estimates were obtained from extensive replicated simulations. The results identify the settings in which each test exhibits superior performance, providing practical, evidence-based recommendations for test selection in survival analysis. These recommendations are summarized as user-oriented guidelines and complemented by a concise graphical cheat sheet, enabling practitioners to quickly identify the most suitable test for their specific application.
Related
- Data-Matrix-Based Overall Goodness-of-Fit Testing for Confirmatory Factor Analysis: In-Sample Prediction Error for Observed Variables
- R-estimation with right-censored data.
- SurvFM enables tabular foundation models for right-censored survival prediction
- Optimal subsampling approach for Cox models under large-scale length-biased right-censored data
- Multi-Absorbing Phase-Type Distributions for Right-Censored Competing Risks Data
BibTeX
@misc{gomezcatala2026practical,
title = {{Practical Guidelines for Goodness-of-Fit Testing for Right-Censored Data}},
author = {Arnau Gómez-Català and Guadalupe Gómez Melis and Klaus Langohr},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.37513},
eprint = {2610.37513},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.37513},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Practical Guidelines for Goodness-of-Fit Testing for Right-Censored Data
AU - Arnau Gómez-Català
AU - Guadalupe Gómez Melis
AU - Klaus Langohr
PY - 2026
DA - 2026/10/27
PB - arXiv
DO - 10.48550/arxiv.2610.37513
UR - https://arxiv.org/abs/2610.37513
AB - In survival studies, right censoring often prevents the observation of subjects' complete outcomes, complicating inference about the underlying distribution. Evaluating whether a given distribution adequately fits right-censored data is therefore essential. However, guidance on the relative performance of available goodness-of-fit tests under different data conditions remains limited. In this work, we present a comparative assessment of three classical goodness-of-fit tests adapted for right-censored data: Kolmogorov-Smirnov, Anderson-Darling, and Cramér-von Mises. Through an extensive simulation study, we investigate how factors such as the proportion of right-censored observations, sample size, and distributional symmetry influence the statistical power of each test. All procedures were implemented using the R package GofCens, and power estimates were obtained from extensive replicated simulations. The results identify the settings in which each test exhibits superior performance, providing practical, evidence-based recommendations for test selection in survival analysis. These recommendations are summarized as user-oriented guidelines and complemented by a concise graphical cheat sheet, enabling practitioners to quickly identify the most suitable test for their specific application.
ER -