Abstract

This study proposes two new methods of goodness-of-fit (GOF) tests for progressively Type-II censored data from any continuous distribution. For the first method, we transform the original censored sample into an approximately independent and identically distributed normality complete sample, and perform a complete sample GOF test for normality thereafter based on the empirical cumulative distribution function (ECDF). This method merely requires one table of critical values for all the distributions. For the second method, we propose a parametric bootstrap GOF test based on test statistics proposed by Pakyari and Balakrishnan. This method does not require data transformation, but directly uses the observed censored sample to the GOF test. This proposed approach does not require some tables for critical values, which are constructed using parametric bootstrap samples. We estimate the power of the two proposed methods for several well-known parameter distributions, and compare the power of parametric bootstrap method with that of Pakyari and Balakrishnan through various censoring schemes. Simulation results reveal that two new methods both possess good power properties in detecting departure from the null distribution, and the parametric bootstrap method provides as good or better power than the method of Pakyari and Balakrishnan. Lastly, the proposed methods are applied to two real data sets from engineering reliability aspect to prove their practical versatility.