mixed model - clarification of terminology and concepts

[#1]

Hi,

I'm learning about the mixed model platform in JMP and I have a few questions that mix the theory with JMP's way of doing it. I'd like to ask for some help to explain as the way that the model is entered is crucial. The documentation is very technical and sparse. I've watched both mastering JMP videos on mixed models already, and they give examples but there's a lot left out. I'm finding it difficult to make the relationship between the theoretical aspects of the statistics and how to "make it happen" in JMP. The concept of random slopes and random intercepts is clear but I'm not sure I understand what is the way to implement it in JMP, or the difference to "just" a random effect.

1) the repeated covariance structure and the repeated variable are clear. But, when does the "subject" need to be used in this part and what does it represent?

2) I understand that "nesting" is the multilevel structure. 

3) What is the difference between setting a variable as random vs setting it as "nest as random coefficients"? and why NEST as random coefficients, and not just random coefficients? I guess I'm struggling to understand the difference between the random effect and the random intercepts and random slopes. 

4) nonlinear: If I have an effect that varies over time non-linearly, for examples goes up and down, my understanding is that I should use the variable as a self-crossed effect (i.e. X*X) to have the quadratic effect, correct? these can in turn be nested, or be random effects? 

5) can a given variable may have fixed and random effects? so should I add it to both columns? 

6) Finally, what about non-parametric? I guess I'd have to use the GLM and use the random effects attribute? Where would the random intercepts and slopes be in this case?

Thanks,

Uriel.

[#2 · Accepted solution]

Hi,

First of all, Mixed Models are not simple. Most people struggle with these concepts. I've done a Master's in Applied Statistics including a module on Mixed Models and I still struggle with the concepts.

There are lots of posts about Mixed Models on the JMP Community so I would encourage you to look through those. I suggest that you look for answers to each of these questions individually rather than trying to get answers to all of your questions in one go.

I also encourage you to look at some of the other resources around Mixed Models. This and this blog post by Jian Cao should be useful.

I suggest that you post data for an example of a problem that you are trying to model. This will make it easier for Community members to help you.

That said, let's see if we can answer some of your questions now...

1. This is in one of Jian's blog posts where it is used to specify Patient. When you have repeated measures in a stacked structure you need to tell JMP the column that identifies the "unit of measurement", that is, the "thing" on which you have taken repeated measures. Again, it would help if you can provide an example from your work.

2. Correct.

3. Jian's first blog post has a good example. If you nest random coefficients then you can model having a different slope for an effect for each level of the random effect. (As well as a different intercept for each level of the random effect.) E.g. a different slope for Yield vs Moisture for each level of Variety.

4. X*X would be a described as the "quadratic" effect and would allow you to model a curvilinear relationship between Y and X. Yes, you can specify these as nested random effects.

5. Again, the example from Jian's first blog post is useful here. Moisture is specified as both a fixed effect and as a random effect nested within Variety.

6. I am not sure exactly what you mean by non-parametric here. I think you are maybe asking about situations that are not standard linear models. Generalized Linear Mixed Models are not currently possible in JMP (or JMP Pro). You can add your vote and comments to the JMP Wish List here.

I hope this all helps.

Phil