Latent Growth Curve Models

[#1]

Hello JMP Community:

Where can I find documentation on Latent Growth Curve (LGC) Models?  In Chapter 8 of Multivariate Methods documentation, LGC is mentioned as one of the capabilities in JMP15 via the Multivariate --> SEM platform.  However, I can't find any examples of how to do this in JMP15.

Can anyone help?

Thanks

Narayanan

[#2]

@Narayananit is an option in the Miltivariate platform in JMP Pro 15.  When you are in the SEM platform, it is an option in Model Selection dropdown.  You can also specify the model from scratch using the specification boxes and verify the model in the Diagram view.  I have ran a few LGC models using the platform, it is fast with relatively big datasets. 

[#3]

I was looking for an example in JMP15 documentation for LGC. Is it buried somewhere?

[#4]

Hi @Narayanan ,

A quick search did not reveal a specific example but it did outline the steps involved as noted in Overview of Structural Equation Models

@LauraCS @sheila_loring can we add a specific example of LGC in a update to the documentation?

Stan

[#5]

The 15.1 doc deadline is near. If we don't have time to add the example by then, we'll add it in the first quarter of next year. @Narayanan, I will let you know when the example is published. Thank you for your request! Sheila Loring Documentation Manager

[#6 · Accepted solution]

@Narayanan,

Unfortunately we don't have an example in the doc but it should come soon! We'll need to start by adding a sample data table that's appropriate for a LGC. 

For now, here are a few guidelines for Latent Growth Curves:

(1) The repeated measures data should be in "wide" format

(2) The most common LGCs are:

     (2.a) "no-growth" model, which is an Intercept-only model. That is, a flat line is fit but we allow for variability in the intercept such that an estimate of each individual's "flat" line can be obtained.

     (2.b) "Linear Growth" model, which has an Intercept and Slope factor, each with variances estimated to enable figuring out how much each individual departs from the average line ("departs" in terms of intercept and slope) --so we can estimte lines of best fit for every individual in the data.

    (2.c) "Quadratic Growth" model, similar to the linear but we add a Quadratic slope as well. Now the trajectory can take a non-linear form.

(3) Each of the LGCs describe in #2 can be specified automatically by using the "Model Shortcuts" drop-down menu.

(4) When fitting all 3 alternatives, the "Model Comparison" table can be used to determine the best fitting model.

(5) It's also a good idea to use a parallel (or spaghetti) plot to visualize the trajectories.

(6) Variances of Intercept, linear slope, and quadratic slope can be fixed to zero if deemed necessary.

JSL Script

I'm attaching a script that should help as an example.

- The first part simulates data for 100 individuals who have been repeatedly measured on four occasions in equal-sized intervals.

- I then convert the data into wide format to be appropriate for SEM

- To visualize trajectories, I use a parallel plot

- I then launch the SEM platform

If you're incline to run the script, please:

(1) click on Model Shortcuts, select the first option, and click Run.

(2) click again in Model Shortcuts to select the second option and click Run.

(3) Finally, select the third option and click Run.

The data-generating model was a Linear Growth Curve, which is the reason why that model fits well and doesn't produce any warnings (such as negative variances). Negative variances in LGCs are sometimes symptoms of overfitting.

I hope this helps! Longitudinal data analysis is such fun! =)

~Laura

P.S. @G_M So glad you've been using the platform and find these models to run well!

[#7]

This is what I was looking for. I look forward to an extended example of LGC in the next documentation update.

Thanks