A reproducible simulation dataset and benchmark for parametric competing-risks inference under censoring and latent...

This article describes a simulation dataset for evaluating parametric competing-risks inference under censoring and latent dependence. The dataset was generated in Python and contains 648,000 replicate records from 303 block-specific scenarios, representing 291 unique statistical conditions. The design includes exponential, Weibull and Gompertz lifetime models, two, three and five competing causes, and sample sizes ranging from 20 to 200. Type-I, Type-II and hybrid censoring are considered. A separate dependence block uses Clayton, Frank and Gumbel copulas with Kendall's tau values of 0.25, 0.50 and 0.75, with inference performed under a working-independence assumption. Replicate records contain estimates, test statistics and availability indicators for Wald, likelihood-ratio and score tests. The release provides 303 Parquet partitions, scenario and parameter registries, summaries of bias and root mean squared error, Monte Carlo uncertainty estimates, and bootstrap coverage checks. A representative subject-level sample contains 35,250 rows. Supporting materials include an executed notebook, figures, a field-level data dictionary, validation records and file checksums. Recorded identifiers and deterministic seeds support regeneration of individual simulated studies. Researchers can use the data to evaluate statistical procedures across the documented conditions, examine test availability and boundary cases, reproduce benchmark calculations, and develop teaching examples.