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; } } }

Modeling Method Decision Tree One Page Type Guide?

Peter_Bartell

At a Dr. Is In event with a customer/JMP user told/asked me, "I only use standard least squares for all my modeling work because it's the only technique I know. What, when or why should I consider other techniques?" I tried to give her a short tutorial on the Fit Model Personality drop down sequence for each entry in the dropdown. She was taking notes furiously. What I was wondering is if somebody has a one or two page kind of guide that looks something like a tree diagram or some such decision construct that would help someone choose the most appropriate modeling method for a given type of response, factor, model construct, etc. Any ideas? Maybe appropriate content for the Statistics Knowledge Portal?

Accepted answers

mia_stephens

What about the "Data Analysis Assistant"? This is a concept we're developing for the Statistical Thinking course. Run the attached script. It'll ask some questions, and point to potential methods. It's still in the concept phase and is not fully developed, but will give the general idea.

TT Selector V10.jsl

All comments

MikeD_Anderson

Did you check in the ed team’s resources? They’ve got a bunch of excellent one pagers for lots of topics.

M

Peter_Bartell

@MikeD_Anderson by '...ed team's resources...' do you mean the JMP Learning Library? Yeah, I did check in the correlation and regression and data mining sections. It's just kind of cumbersome because someone would have to read/open each pdf file, or watch the video, and then create their own one page guide that ties/differentiates the various methods together.

Hmmm...maybe there's a JMP blog in my future if nobody has already created this wheel?

MikeD_Anderson

No the PDFs are in the *Academic* resources (sorry, couldn’t remember the exact term it was under) section of the home page: https://www.jmp.com/en_us/learning-library.html

Looks like here are a couple in there that might fit the bill. You can also download the lot in a single file using the last link on the page.

M

Dan_Obermiller

I can see where this would be extremely popular. I can also see this being a real challenge.

Here's the scenario: I have a categorical response of Yes/No with several possible predictors. What model approach should I choose?

My choices are quite large (just a few off the top of my head):

Logistic Regression

Generalized Linear Model with a Binomial Distribution and Logit link function.

Generalized Linear Model with a Binomial Distribution and Probit link function. (+ probably others)

Discriminant analysis

Partition model

Neural network

There are a few questions that could rule out some of these alternatives, and some are equivalent. But so much of the choice will depend on the data and what you actually see and judgements need to be made. Plus, it could be a disadvantage if you "miss" a technique in the document.

JerryFish

Didn't Minitab try to do something like this with their "Design Assistant"? (Maybe I've got the name wrong... hard to remember, and I don't have a copy of MT.) They pushed it pretty hard, but as a (former) user I didn't get much out of it.

@Potcner, with your Minitab background, do you have any thoughts?

MikeD_Anderson

Ish - but, as you said, it wasn’t very good. I think the goal here is to help people understand, at a 10,000’ level, the different types of modeling methods and what they are generally good for - a framework to build a more detailed and nuanced understanding on.

M

mia_stephens

TT Selector V10.jsl

Peter_Bartell

@Dan_Obermiller I agree a challenge indeed. The other thing I'd worry about is the naive person would just use ONLY the method(s) that ends up at the end of their particular path through the decision tree. And I myself shout from the rooftops 'We want a useful model...' so why not try several different techniques and see which one helps best with solving the practical problem at hand.

Peter_Bartell

I like this script @mia_stephens. It's a good way to point somebody in the right direction AND integrates JMP into the conversation.

mia_stephens

Cool, thanks! If you think we're on the right path, and that this sort of structure will work, we'll keep going. It's not very elegant, but I think does the job. Perhaps the design/structure can be revisited once we've done the hard part of compiling the details. I've attached as an add-in if you want to share (with the caveat that it is under development).

Data Analysis Assistant.jmpaddin