Random Intercept Cross Lagged Panel Model

[#1]

Hi,

I am attempting to run a random intercept cross lagged panel model using longitudinal data with 3 waves. On the Multivariate Methods manual in Chapter 8, pg 212, it says:

Cross Lagged Panel Model: This shortcut also enables you to account for measurement error by specifying latent variables when the model includes at least three time points. In addition, you can add a random intercept to distinguish between-person effects from within-person effects, which produces more accurate autoregressive and cross-lagged estimates in the presence of trends.

However, I cannot figure out how to add a random intercept. How do I add it? Thanks kindly for your assistance.

 

[#2]

Hi @OmegaOkapi125 

Excellent question!  There are a few details to clarify... first, depending on the version of JMP Pro you're using, there will be different ways to achieve this. Version 19.0 makes it easier (though the upcoming version 19.1  is even easier--so be sure to download it when released).

I see your variable names have "time values" encoded in them. However, those values make me wonder if you have different time gaps between repeated measures? This makes the model specification a bit more involved.

For the time being, I'm including here a video of the simplest case (equally spaced repeated measures), but please let me know which version you're using and the time spacing so I can provide a better answer.

[#3]

Hi Laura,

Thanks for replying so promptly, and thanks for the video, this was amazingly helpful! 

Edition - You called it, I was running 18.2, but I just upgraded to 19.0.4. (Eagerly awaiting 19.1)

Time Gaps - You are correct, the time between the measures is not uniform. From baseline, the next measure is 3 years later, and the next is 2 years later (5 yrs from baseline)

Unfortunately when I try to run with the random intercept checked, I get an error. After replacing the random intercepts (for 19.0.4 version), running the model crashes the program. Here are screen shots:

 

[#4 · Accepted solution]

Ah!  Thanks so much for sharing that!  I'll be sure to look into it. For now, the easiest way to avoid this error is to rename your variables so the time codes are at the end of each variable name... such as "EB Grade 5." Though you can double click on the names to edit them, I think the fastest is to go to the "Cols" menu and then Column Names > Recode Column Names...

In terms of the unequal spacing... do you intend to test the equality of autoregressive and cross-lagged effects over time? If so, I'd suggest specifying "phantom variables." I'm writing a blog post to explain in more detail how to specify these variables in our SEM platform--I should be able to publish it in just a day or two (I'll come back to link it here). If you don't place equality constraints in the model, then you'll obtain estimates that describe effects from baseline to 3 years later, and then from 3 years to 5 years later.

[#5]

Hi Laura,

This worked!! Thanks so much, I truly appreciate it! 

Yes, I see, that would improve the integrity of my analysis and I am eagerly (patiently) awaiting your blog post.

[#6]

Hi @OmegaOkapi125,

Apologies for the delay on the blog post... should be available soon! Meanwhile, I made another video to illustrate how to specify the RI-CLPM with phantom variables for your particular example. The blog post will provide more context but this should get you going for now. Note that I'm specifying the model from scratch without the Model Shortcuts menu. In the future, I intend for the Model Shortcuts to do this for our users automatically. Moreover, phantom variables commonly have variances fixed to one but it can also be zero. You can play around with either specification to make sure you get a proper solution.

[#7]

The blog post describing phantom variables in the context of autoregressive models is here:

https://community.jmp.com/t5/JMPer-Cable/Phantom-Variables-in-the-Structural-Equation-Models-Platfor...

[#8]

This is so very helpful, I really appreciate it! The video was also informative, and I could easily run the model for my own dataset. Thanks for taking the time to post!