How Do I Plug in Data and Have JMP Determine Relationships Between Factors and Responses

JoshE87

I'm still very new to JMP, but I've figured out (with help) how to have JMP show me polynomial responces between responces and factors. The issue is that I know from my own plots in excel that some factor/responses have polynomial relationships, some have exponential and still others have logarithmic realtionships. Sometimes, a single factor will influence several responces in two or more of these ways.

I do not understand how to have JMP determine which relationship the various factors and responses. It seems like I already have to know the relationships before I begin analyzing the data to have it be able to generate an accurate prediction. Is that the case? Also, I do not really know which factors compound on each other to create a larger response, and it seems like if I just plug all responses in to check, the system overloads and says that none (or a very large number of them that includes factor interactions that I know do matter to various responses) influence the responses, as shown below. 

 

Then I can delete several of the least likely interactions and suddenly the rest will snap to show significance. This is also shown below. 

I'm confused as to how I can get JMP to tell me which ones matter, rather than me having to guess. 

 

Also, as far as getting Logs and exponents to work, I see that I can select a factor and "transform" it into a Log or exponent or whatever, but then adding it to the "Factors" box just leads the program to tell me that that data doesn't exist. One or two times, I got the program to compile anyway, and then I realized from dragging the prediction graphs around that changing the log(x) factor did not change the value of the x factor that I was testing to see if it had a logarithmic relationship with any of the responces.  

Basically, I want to throw a ton of data that I don't quite understand the relationships between into JMP and have it spit out a decently accurate prediction model. 

I hope I've made my series of questions clear enough. I'm a scientist/engineer, not a statistician, so I realize that these questions might be frusterating to some JMP users/staff members.  Anyway, thank you very much for reading this and providing help, should you choose to!

jules

Hi @JoshE87,

As you've discovered, most statistical modeling involves making a lot of choices, not simply about what factors to include when modeling a response, but choices about the specific nature of the relationship between the factors and response you are willing and able to estimate. Your choices determine what functional forms are estimated, and those choices also impact what else can be estimated in the model. That said, certain kinds of predictive modeling (and data mining) can help you uncover relationships with fewer choices upfront. But, that flexibility can be a double-edged sword. All of this is the topic of entire classes, and I don't think I'll be able to answer all your questions about how to model your data, but I do think there are some great next steps you can take to help!  My #1 suggestion is to watch several of the Mastering JMP videos on modeling. These will give you a sense of not just how to set these designs up in JMP, but also how to interpret the output, and how to make choices in each of the different modeling platforms. Here are some that are in line with your needs:

• Specifying and Fitting Models
• Data Mining and Predictive Modeling 
• Discovering and Predicting Patterns Using JMP® Pro Neural Network Models (which might help you with your question of fitting a flexible model *without* knowing ahead of time what form it will take)
• Building Better Predictive Models Part 1: Understanding the Principles
• Building Better Predictive Models Part 2: Implementing Models
• Using JMP Pro to Build Models Using Generalized Regression Variable Selection Techniques
• Advanced Mastering JMP: Using Generalized Regression in JMP Pro to Create Robust Linear Models

 

Overall though, if as you say, you want to "throw a ton of data that I don't quite understand the relationships between into JMP and have it spit out a decently accurate prediction model," it sounds like what you really want is data mining/predictive modeling tools. The tools under Analyze > Predictive Modeling will help you model your response with far fewer choices upfront than standard linear modeling using Analyze > Fit Model. But, like all tools, it's important to have a good sense of what you're doing, and the Mastering JMP videos I mentioned are a good place to get started. After that, look into general coverage of those techniques online, and maybe even JMP Training on specific topics like predictive modeling. There's a lot to learn to be sure, but it's all worth it. 

 

I hope this helps you a bit along your way, and I hope you have some fun learning all the new and powerful things you're going to be able to do with JMP!

@jules