Discussion of the Papers by Olteanu and Freeman, and Genshel and Meeker

Click to increase image sizeClick to decrease image size

Notes

¹Thoman et al. (1969) contains factors to reduce the bias in MLE Weibull parameter beta very similar to the MLE-RBA corrections in The New Weibull Handbook, Chap. 6.

²Liu (1997) is the most extensive and important study of these two distributions using several different methods and both Weibull and log-normal data sets with and without random suspensions. Liu made three major conclusions: (1) median-rank regression, X on Y, is more accurate then MLE except for very large sample sizes; (2) the two best measures of goodness of fit are the p value or r ² and the likelihood ratio test; (3) with 20 failures or less the two-parameter Weibull is the distribution of choice.

³Wenham (1997) is an internal confidential document that recommends MRR over MLE for type 1 testing.