Partial Regression Plots of Model

78 views
Skip to first unread message

Somers Smott

unread,
Aug 27, 2018, 6:21:39 AM8/27/18
to lavaan
Hi there,

A couple of PhD students and I have been inquired by our adviser on how to create partial regression plots of models using lavaan. He wants to be able to visualize how well the variables fit the model, and not just by a p-value or the model fit variable. 

When we asked elsewhere, we were told to remove the link we questioned from the model, find the predicted scores for the two ends, and plot those residuals to the original data. This doesn't seem like it would be the "partial" regression, though, since the whole model isn't being plotted.

Any advice?

Thank you!!
Somers Smott

Terrence Jorgensen

unread,
Aug 28, 2018, 6:24:23 AM8/28/18
to lavaan
Any advice?

This article sounds right up your alley:


Terrence D. Jorgensen
Postdoctoral Researcher, Methods and Statistics
Research Institute for Child Development and Education, the University of Amsterdam

Somers Smott

unread,
Aug 28, 2018, 8:43:21 AM8/28/18
to lavaan
Hi there!

Thank you for this helpful article. We actually just learned about using d-sep with our model selection process, but we've used piecewiseSEM so it's helpful to know there are even more options.

The problem is, we want to visualize residuals and partial regression plots on the final model, not the previous models we are just kind of trying out to find the fit. These packages will show the plots on sections of the model, but not while relating them to the entire final model. 

Mostly, is this even possible? Without breaking apart the model and entering parts of it..

Thank you!!
Somers

Terrence Jorgensen

unread,
Aug 28, 2018, 8:49:00 AM8/28/18
to lavaan
SEM is a bunch of simultaneous regression models, so breaking your model into smaller parts should still be possible to use that other software.  If the estimates change, you could fix the parameters at their estimated values from the full model, then "fit" the smaller part of the model to visualize.  Not sure if I'm understanding exactly what you're trying to do, though.  Could you post a path diagram of the full model, and explain which residuals you want to investigate?
Reply all
Reply to author
Forward
0 new messages