Limit yield to max 100% in profiler by using other variables - Trindade?I have a customer asking me about how to use some combination of variables (such as # passing, # failing, and/or total parts) rather than yield in a model so that the outcome variable can't be higher than 100%. He said he saw something on this - maybe Dave Trindade did it? Specifically, he doesn't want to limit the profiler (or put bounds, etc.), he wants to use a different set of outcome variables so that they can't be over the real values. Does this ring a bell for anyone? He wants to implement this methodolgy, so I was trying to find an example he could follow.
Re: Limit yield to max 100% in profiler by using other variables - Trindade?Thanks Clay - Exactly what I was looking for - Generalized Linear Models with 2 inputs for regular JMP customers!
Re: Limit yield to max 100% in profiler by using other variables - Trindade?Maybe something like this:
Using a formula like this:
xxx = :X1 + :X2;
If( xxx >= 1, 1, xxx );
Re: Limit yield to max 100% in profiler by using other variables - Trindade?I agree with Mark, sounds like a logistic regression problem. Both the generalized linear model and genreg platforms can do it, the key is to specify the number of trials and the number of events as columns. I've attached a simulated example.
Re: Limit yield to max 100% in profiler by using other variables - Trindade?Laura:
Since you mention some counting type inputs to the 'new variable' definition idea...I'm thinking maybe a new derived variable such as First Pass Yield might be what your customer is after? For discrete manufacturing and continuous manufacturing processes we used this often at Kodak to talk about the health of a process. Here's the classic definition of First Pass Yield:
https://en.wikipedia.org/wiki/First_pass_yield
Some of FPY's attractive features are it's bounded by zero and 1. And for managers it's a real easy number to grasp...zero is really, really bad...1 is really, really good. I can talk you through some examples of how we used it at Kodak if you'd like to learn more? Give me a holler.
Re: Limit yield to max 100% in profiler by using other variables - Trindade?Sounds like logistic regression to me. Response should be categorical. JMP takes care of the rest.
