nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

JMP 16.2 Life Distribution Weibull Distribution Alpha/Beta Generation

WoHNY

I am trying to generate the Weibull Distribution alpha and beta (shape & scale) values of a large data set containing tens of thousands of rows and have those values in a data table. Other than the Analyze > Reliability and Survival > Life Distribution portal is there another way to generate those values?

Accepted answers

Mark_Bailey

Right-click on the values in the estimates table and select Make Into Data Table. If you have a grouping variable in the By analysis role, then select Make Combined Data Table instead.

Is this result what you want?

All comments

Mark_Bailey

Why do you need another way if the Life Distribution platform estimates the Weibull model parameters?

WoHNY

Mark: Thank you for responding. How would I get the data that is generated in the Parametric Estimate - Weibull column format into a data table? Another point is, it takes quite a while to generate those numbers. Maybe my question should have been worded as, are there Formula functions available to generate those parameters?

Mark_Bailey

Right-click on the values in the estimates table and select Make Into Data Table. If you have a grouping variable in the By analysis role, then select Make Combined Data Table instead.

Is this result what you want?

peng_liu

To save estimates in the report to a data table, try the following: Right click in the Parametric Estimate part, choose "Make into Data Table".

For your second question. There is a JSL function which can estimate Weibull parameters without going through Life Distribution. Following is the screenshot of it from Scripting Index. But I am not sure whether it is what you want.

When data is large, most of the time spent on estimating Weibull in Life Distribution for this size of data is on constructing the non-parametric estimate and plotting it. There are some situations that you may be able to speed it up a little bit.

If your data can be summarized, you should summarize your data first. For example, if you have multiple observations with the same value, instead of each observation occupying a distinct row, you may want to consolidate those observations in the same row, and put up a count to record how many they are. Then feed that count column to the Freq in the launch dialog. If you are not familiar how that can be done, check out Summarize Your Data
If your data does not have censoring, you may want to consider using the Distribution platform, which can fit Weibull as well if there is no censoring. In JMP 17, the upcoming one, Distribution platform will support Limit of Detection, which cover certain types of censoring schemes. So you may want to try Distribution platform, even if you have censored observations, when JMP 17 becomes available.
You may want to try the JSL function that I mentioned above, if you have some JSL programming skill. The function allows arbitrary censoring schemes, just like what are supported in Life Distribution, but by passing estimating the non-parametric estimate using all data, if you call the function appropriately by specifying the last argument to use a sub-sample. The function will definitely no draw the non-parametric estimate, which will save time.

However, you may still want to check out the plotted non-parametric estimate. You may want to know whether Weibull is a good fit, based upon the non-parametric estimate. Here is one example which indicates the parametric fit is appropriate: Example of the Life Distribution Platform

In my experience, when the data is large, usually none of these parametric distributions can fit data well.

WoHNY

Mark: I should have caught that about the right click but not seeing a red triangle available in the Parametric Estimate - Weibull section I was at a loss. Thank you for pointing this out. That solves the part about getting this tabulated data into a data table.

WoHNY

Peng: Thank you for your response. Both you and Mark pointed out the right click which I missed. Thank you for that. Regarding the JSL, I do have some JSL scripting background. You have given me some things to review and dig deeper on. Regarding the number of rows I indicated, I should have been clearer. For each part we test 20 sites and we have thousands of parts. So say the data set has 1500 parts with 20 data points each. This would result in a data set having about 30k rows. So there are a limited number of measurements per part at 20 but many parts. This is what is slowing the output down.

I will look into your recommendations and see if I can implement in my situation. Thank you again.