Rank Regression in Weibull

[Sharon Randall]

[Originally Posted: 3/31/00-- Transferred by ReliaSoft Moderator]

From my past experiences with Weibull analysis, the terms "Rank Regression" and "Maximum Likelihood" have always been used when referring to parameter estimations. However, after newly using your Weibull++ software I see there are two ways of performing Rank Regression - RRX and RRY. Is there a rule of thumb on which one to use? I have tested 6 parts to failure and need to determine a B10 life. Due to variations in data, the B10 life is significantly different when running a 2-parameter Weibull RRX versus RRY. Also, since I am determining a B-10 life, is it correct to assume that I need to use the Type 1 Percentile Confidence Bounds? Thanks.

[Adamantios Mettas]

[Originally Posted: 4/3/00-- Transferred by ReliaSoft Moderator]

The rule of thumb is to use RRX when dealing with Life Data. This is because when using RRX the horizontal (x-axis) distance between the point and the line is minimized. Since your data is on the x-axis, RRX is more appropriate. On the other hand, if you have interval data, there is uncertainty on the x-axis (as to when a failure occur), therefore minimizing the vertical (y-axis) distance between the points and the line is more appropriate and RRY is used. However, the best rule of thumb is to use the method which gives the best fit on your data (especially in the region where you are making predictions).

For the bounds on the B10 life, you are absolutely correct. You should use Type 1 bounds (percentile).