I ran a DOE to find out which manufacturing factors affect the fatigue life of a material (as well as other properties).
Since the response variable of interest is survival data with censored data, I am using the Parametric Survival platform in JMP 14 to analyze the data. The Parametric Survival platform doesn’t support the Assess Variable Importance option or a profiler as far as I can figure out.
I would like to determine which factors have the biggest influence on life and the settings that maximize life. I’m at a loss as to how do this without the Assess Variable Importance option or the profiler. Can anyone recommend the approach I should take to do this?
Would it be sufficient to assess the effect sizes visually in the profiler?
I think I’m starting to see why JMP removed Assess Variable Importance and the Profiler from the Parametric Survival platform. In this platform you are not predicting a response so much as you are simply fitting a non-linear function, specifically a probability distribution. For similar reasons the profiler doesn’t make sense either.
At this point I have a probability distribution that is a function of ‘n’ of my factors. What I need to do now is pick a measure of reliability for my problem and maximize it. For example, if my goal is to maximize the cycles before 5% of my samples fail, then I need maximize the inverse cumulative distribution function for p = 0.05.
Now I just need to figure out how to do that in JMP.
Use the Quantile Profiler. Set the Desirability Function to Maximize (default). Set the Probability to 0.05 and lock it. Then maximize the desirability.
The Quantile Profiler is a good solution if you have JMP Pro. I'll probably have to export the distribution equations and use an outside program.