Reliability - fiter data on probability information<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />Hello everybody,
A potential JMP User (Aledia) would like to do some filtering on reliability platform. he has the following data
the fit is not really good. he would like to filter depending the Y-axis, it means that the filtering will be associated to Probability information (so remove value below 0.05 and upper than 0.99
what would be the good way to do that? (it seems that he is able to do that in other software??)
a script would be ok if it is needed.
Thank a lot for your suggestion and have a joyful chrismast break!
Florence
Re: Reliability - fiter data on probability informationAs it is using an exclusion approach it is the same behavior ; the probabilities change:
[cid:image004.jpg@01D13D78.C9B04F80]
De : malcolm.moore1
Envoyé : mercredi 23 décembre 2015 11:44
À : Florence Kussener
Objet : Re: - Reliability - fiter data on probability information
JMP User Community<>
Reliability - fiter data on probability information
reply from malcolm.moore1<> in JMP SE Community - View the full discussion<>
Re: Reliability - fiter data on probability informationWill option 2 (based on mixture portion) work for your user? This solution requires sorting the data in ascending order for failure time, which I forgot to mention. Otherwise, it may be a case of selecting points from the probability plot (or event plot) then excluding and re-running the analysis.
Malcolm
Re: Reliability - fiter data on probability informationThank a lot Malcolm to continue the exploration of my issue…
In fact all approach with filtering are not good as they change the distribution of the data and the probability.
For your first case, you can observe this below.
On the left, with a filter, on the right with no filter. You can see that the selected points don’t have the same probability as some have been excluded (it also excludes the points with no estimation (censored data)).
Florence
De : malcolm.moore1
Envoyé : mercredi 23 décembre 2015 10:57
À : Florence Kussener
Objet : Re: - Reliability - fiter data on probability information
JMP User Community<>
Reliability - fiter data on probability information
reply from malcolm.moore1<> in JMP SE Community - View the full discussion<>
Re: Reliability - fiter data on probability informationHere are a couple of suggestions Florence:
1) From the non-parametric estimates report do a right mouse click and copy the Kaplan-Meier estimates and paste these to a new column in the data table then filter on the low probability estimates (and high probability estimates) to exclude the points that do not belong to the failure distribution of interest.
2) A less subjective approach would be to use the mixture portion estimate from the mixed life distribution estimates to define the portion(s) not belonging to the distribution of interest, then exclude the corresponding number of rows. For example, in the two mixture distribution case in the file "Mixture Demo.jmp" there are two distributions and the first 24.7% of rows belong to the first distribution and the remaining rows belong to the second distribution. Multiplying by the total number of rows (400 in this case) gives the first 99 rows belonging to first distribution and therefore the first 99 rows should be excluded to focus life time analysis on the second distribution.
Malcolm
Re: Reliability - fiter data on probability informationI discuss with the customer who is not happy with this workaround.
He wants to keep a simple model (only lognormal/weibull/normal), without mixture.....
any other suggestion?
De : malcolm.moore1
Envoyé : mardi 22 décembre 2015 16:16
À : Florence Kussener
Objet : Re: - Reliability - fiter data on probability information
JMP User Community<>
Reliability - fiter data on probability information
reply from malcolm.moore1<> in JMP SE Community - View the full discussion<>
Re: Reliability - fiter data on probability informationThis looks like a mixture failure distribution fitting problem to me Florence rather than the need to down-weight the points that don't correspond to the predominant failure distribution. Try sharing the attached example with your user to see how fitting a mixture of failure distributions corresponds to their needs. Or if user has identity of different failure causes then you might try competing causes failure distribution.
Malcolm
Re: Reliability - fiter data on probability informationNo, the issue is not to detect the outlier, the issue is not having them used by the fit (but not filtering them because in this case, the distribution and probability will be changed.) a robust fit like in Fit Y by X would be great or a weight variable. Or???
De : malcolm.moore1
Envoyé : mardi 22 décembre 2015 15:24
À : Florence Kussener
Objet : Re: - Reliability - fiter data on probability information
JMP User Community<>
Reliability - fiter data on probability information
reply from malcolm.moore1<> in JMP SE Community - View the full discussion<>
Re: Reliability - fiter data on probability informationFlorence,
Would the Explore Outliers option in Cols > Modeling Utilities be a viable alternative?
Malcolm
Re: Reliability - fiter data on probability informationThis approach is not good as if I apply a filter, I change the distribution of the point and the probability. it is not something that I want to have. I just want to apply a fit on a local range, but without changing the distribution.
Florence
Re: Reliability - fiter data on probability informationMy first approch would be to create a column with the row proportion and apply a filter on it.
But perhaps there is something easier I don't know.
Florence
