Life distribution - reliability projection

[#1]

Hello,

I have a set of time to even data and when I fit the life distribution, JMP provides the best distribution as Frechet with the following plot. The points seem to trend up and not nicely scattered around the line. How do I interpret this? This is supposed to be the best fit per JMP's choice.

[#2 · Accepted solution]

The best fit is not necessarily a good fit. The other fits are worse. There is a table of all the fits for comparison using criteria such as AICc. JMP presents the fits in ascending order of AICc (smaller is better).

Can you show this table for comparison? Can you show plots for the other fits that you entertained?

[#3]

Hi Mark,

Thank you and here is the table you asked for:

lognormal plot:

Weibull plot:

Cheers,

[#4]

I noticed that the data plotted is not part of the current row selection in the data table. Is it possible that you have excluded rows that are plotted here? That row state would affect the fit a lot.

[#5]

I noticed the probability values are extremely small. I also notice, there is a long horizontal segment at the end of the nonparametric estimate. I also have a feeling that it is probable that there is unlikely to have failures between 0 and 100.

Based on those observations, I suggest looking into two special types of distributions: Threshold distributions and Defective Subpopulation distributions. They are labelled as TH distributions, e.g. TH Lognormal, TH Weibull, etc. And DS distributions.