distribution parameters mismatch between Competing Risk Mixture and SubSample in Life Data platformHi All,
I use Life Distribution platform from Reliability and Survival to fit distribution to reliability results.
To distinguish between failure causes I used 2 Weibull distributions in Competing Risk Mixture of Life Distribution platform (see graphs below and attached file).
Portion 1 value (marked with yellow) from the Competing Risk Mixture report species part of reliability results related to 1st Weibull distribution.
In Life Distribution platform Weibull distribution was fitted to [ 1 - (Portion 1) ]% subset of reliability results.
As far as I understand Weibull parameters of 2nd distribution in Competing Risk Mixture report and subset have to match, but they are actually not (marked with yellow).
Why Weibull parameters between 2nd distribution in Competing Risk Mixture report and subset are different?
OtherRe: distribution parameters mismatch between Competing Risk Mixture and SubSample in Life Data platformWhen you fit Competing Risk Mixture, the data meant that there is not information about failure cause.
When you fit modified data and treat it as a subset due to the second cause, you provide concrete information about failure cause.
Different data, different models, so different results.
I also see how you modify the data. You consolidated the first 25 rows into one row with weight = 26. I guess that you want to use that single row to represent failures due to the first cause. Assuming that you are correct about the causes, the way of modifying the data is not correct. Take second row as an example, it fails between [0.25,0.5] due to the first cause. You should have modify it as [0.5, MISSING], so that is treated as a right censored observation due to second cause, which means that if it would have failed due to the second cause, it would have happened sometime after 0.5.
Regardless, even you modify the data as I described, the results won't be the same anyway. The information are different, models are different, results will be different.
Re: distribution parameters mismatch between Competing Risk Mixture and SubSample in Life Data platformDear peng_liu,
Thank you for quick and clear answer.
I made suggested changes to split reliability data to two failure modes, where split point is Portion 1 value from Competing Risk Mixture report.
The Weibull α and β distribution parameters of group 2 (see graph below) got same values as Weibull α and β distribution parameters from subset in my post above.
P.S.
Is there way to do Competing Risk Mixture in Fit Life by X platform?
