Random Number Generation and Inferences for Two-Stage Life-Testing Data

Provides random number generation, estimation, and hypothesis testing for two-stage life-testing experiments in which units begin on a baseline stress and, at a fixed or random change point, some or all surviving units move to an elevated stress ("stage life testing"), following the cumulative-exposure model of Laumen and Cramer (2019), "Stage life testing", . Twelve baseline lifetime distributions are supplied (Weibull, Power Lindley, Exponentiated Lindley, Exponentiated Exponential, Marshall-Olkin Exponential, Lomax, Exponentiated Teissier, Log-Logistic, Chen, Log-normal, Gamma, and Power Muth/Power Teissier) together with three parameter-sharing cases (common, shape1, shape11). The package implements data simulation under both the fixed-proportion (typeP) and fixed-count (typeM) rules and the simple-step-stress special case, maximum likelihood estimation using a hybrid Nelder-Mead and BFGS optimisation routine, asymptotic (Wald) and non-parametric/parametric bootstrap confidence intervals (including log-scale variants that guarantee a positive interval), a stage-1-observation planning probability, Kolmogorov-Smirnov goodness-of-fit testing with a parametric-bootstrap correction for estimated parameters, likelihood-ratio and information criteria for nested/non-nested model comparison, and diagnostic plots comparing fitted and empirical distributions, including the cumulative distribution function (CDF), survival function, probability-probability (P-P), and quantile-quantile (Q-Q) plots.

stagelife

stagelife implements the stage life testing (SLT) model of Laumen & Cramer (2019), Naval Research Logistics 66:632-647 -- a two-stage life-testing model that unifies progressive censoring with fixed censoring times and step-stress accelerated life testing under the cumulative-exposure model -- together with a suite of simulation, estimation, and validation tools.

Features

• 12 baseline lifetime distributions: Weibull, Power Lindley, Exponentiated Lindley, Exponentiated Exponential, Marshall-Olkin Exponential, Lomax, Exponentiated Teissier, Log-Logistic, Chen, Log-normal, Gamma, and Power Muth.
• 3 parameter-sharing cases for the elevated-stage shape parameter: shape1 and shape11 (the paper's own models) and common (a shared-shape extension).
• Data generation under the random-withdrawal design (rslt()), with both fixed-proportion and fixed-count selection rules; the classical simple-step-stress design is available as a special case.
• Estimation: maximum likelihood via fit_slt(), with standard coef(), vcov(), logLik(), AIC(), BIC(), and confint() methods.
• Inference: asymptotic (Wald) and bootstrap (non-parametric and parametric) confidence intervals, via slt_bootstrap_ci().
• Experimental design: slt_prob_stage1() computes the probability of observing at least one elevated-stage failure, for planning a test.
• Goodness of fit and model comparison: a Kolmogorov-Smirnov test with a parametric-bootstrap correction (slt_gof_ks()), likelihood- ratio tests between nested cases (slt_lrt()), and an AIC/BIC comparison table across fits (slt_model_comparison()).
• Diagnostics: fitted CDF, survival, P-P, and Q-Q plots.
• 4 bundled real step-stress data sets for examples.

Installation

# From the built tarball:
# install.packages("remotes")
remotes::install_local("stagelife_0.1.0.tar.gz")

# Or directly from GitHub:
remotes::install_github("vksharma-bhu/stagelife")

Quick start

library(stagelife)

theta0 <- c(0.05, 1.5); theta1 <- c(0.20, 1)
tau1 <- slt_tau("weibull", theta0, p = 0.5)
d <- rslt(200, "weibull", theta0, theta1, tau1, prop_transfer = 1)

fit <- fit_slt(d, family = "weibull", case = "shape1", tau1 = tau1)
summary(fit)
confint(fit)
plot_cdf(fit)

See vignette("stagelife-intro", package = "stagelife") for the full workflow, including goodness of fit, model comparison, and a real-data example.