Anderson-Darling and Cramer-Von Mises Based Goodness-Of-Fit Tests for the Weibull Distribution with Known Shape Using Normalized Spacings

Two new goodness-of-_t tests are developed for the three-parameter Weibull distribution with known shape parameter. These procedures eliminate the need for estimating the unknown location and scale parameters prior to initiating the tests and are easily adapted for censored data. This is accomplished by employing the Anderson-Darling A2 s and Cram_er-von Mises W2 S statistics based on the normalized spacings of the sample data. Critical values of A2 s and W2 s are obtained for common signi_cance levels by large Monte Carlo simulations for shapes of 0.5(0.5)4.0 and sample sizes of 5(5)40 with up to 40% censoring (type II) from the left and/or right. An extensive Monte Carlo power study is also conducted to compare the two tests with each other and with their prominent competitors. The competitors include another spacings test, Z_, and the modi_ed Kolmogorov-Smirnov (KS), Cram_er- von Mises (W2) and Anderson-Darling (A2) EDF tests. The power results indicate that no one test is superior in all situations. When the alternatives considered are tested against a skewed Weibull null distribution, A2 s and W2 s achieve considerably higher power than the other EDF tests, but do not perform as well as Z_. On the other hand, when the null distribution is symmetric, Z_ loses all of its power, while A2 s and W2 s yield power comparable to the other EDF tests. Results also show A2 s generally outperforms W2 s, and for these reasons, A2 s is the preferred test for the three-parameter Weibull with known shape.