Simulation dataset on power and size of Wald, likelihood-ratio, and score test statistics in parametric competing risks models under hybrid censoring

This data article describes a simulation dataset generated for parametric competing risks models (CRM) with K = 2 or K = 3 independent competing causes of failure under hybrid (Type I/II combined) censoring — a scheme that combines a time limit T 0 (Type I) and a failure-count limit r (Type II), common in accelerated life testing and clinical trials. The dataset was produced by Monte Carlo simulation in Python 3.11 ( R = 5,000 replications; base seed 42; independent parallel streams). Three parametric families are included: exponential, Weibull (α = 2), and Gompertz. The experimental grid covers n ∈ {50, 100, 200, 500} and β ∈ {0.25, 0.50, 0.75, 1.00}. For each combination the dataset records: (i) empirical bias and RMSE of the MLE together with the theoretical prediction from the Fisher information identity I ( θ 0 ) = β I M ( θ 0 ) [ 1 ]; (ii) empirical Type I error and power of the Wald ( W n ), likelihood-ratio ( L R n ), and score ( S n ) statistics at α = 0.05; (iii) non-centrality parameter λ* and theoretical power; (iv) asymptotic and bootstrap 95 % CI coverage at n ∈ {30, 50, 100}; (v) Type I error under Clayton copula dependence; (vi) required minimum sample size n*. A diagnostics file and an application to the survival::lung dataset ( n = 228) are also included. All 16 files are deposited in Mendeley Data (DOI: 10.17632/mv42gshrcv.2). This dataset enables researchers to: (i) benchmark new statistical methods for competing risks under hybrid censoring without rerunning all simulations; (ii) empirically validate theoretical Fisher information predictions from I ( θ 0 ) = β I M ( θ 0 ) ; (iii) plan sample sizes for clinical trials and reliability tests using the n * tables; and (iv) reproduce or extend the simulation framework via the provided Python and R scripts.

Ҳали таржима қилинмаган