Publication date: 05/24/2021

Notes Regarding Median Rank Regression

When there are no censored rows and no Weight column, median rank regression (MRR) for Weibull parameter estimates is available in the Life Distribution platform. The following conditions must be met before MRR estimates appear in the Life Distribution report:

This Life Distribution platform preference is selected: Report Median Rank Regression Based Weibull Parameter Estimates When There Are No Censored Observations.

There are no censored observations in the analysis. In other words, all observations represent failure times.

There is no Weight column specified.

If all of the above conditions are met, the Parametric Estimate - Weibull report is modified to show MRR estimates as well as the usual maximum likelihood estimates (MLE). The column headings in the parameter estimates table are prefixed with MLE or MRR to denote which estimation method was used to produce each column of estimates. There are no confidence intervals available for the MRR estimates.

Caution: The use of MRR estimates is not recommended. Median rank regression uses least squares estimation to produce estimates, instead of maximum likelihood. There are no commonly accepted methods to apply least squares estimation to censored data. Further, Genschel and Meeker (2010) show that the MLE method is more precise. They provide simulation results that are based on the fact that the true estimates are known, but in reality, the accuracy of MRR estimates is unknown. Because MRR is a loosely defined and ad hoc estimation procedure, different software packages produce different MRR estimates for the same data.

Want more information? Have questions? Get answers in the JMP User Community (community.jmp.com).