How to find the common denominator (failure) among builds with different parts?

Question

Hi there,

I have a large data table containing "builds" which are comprised of 10 different "parts". Many parts are found in multiple builds. Each build is represented by a number, and whether the build passed or failed a test. Each part is represented by a unique number.

Is there any way I can use jmp to statistically find out which "part" has a high probability for being the reason for failed builds? Thank you in advance

Mark_Bailey · Answer

I would also suggest using Analyze > Modeling > Partition as a way to find the most likely predictors. This type of model is very robust and should find the large effects if they exist.

Bill_Worley · Answer

Hi jackprada,

JMP Fit Model under the Analyze platform is a great place to start.  Use your Pass/Fail criteria as your "Y" variable which will give you a Nominal Logistic fit.  Your "X(s)" would either be your builds or your parts depending on how your data table is set up.

Based on your description, you will likely have to run the fit several times of the different builds. Saving the Probability Formulas from each fit model as you go.  You can then build a combined prediction profiler that will potentially allow you to determine the "part" that is the culprit in your failed builds.  The tricky part is making sure you select Expand Intermediate Formulas when you build your profiler.  If you don't do that things will likely look very messy.

Would it be possible to supply an example data table?  Making sure to anonymize your data as much as possible.  We can give it a try and send it back with results.