Shawn -- thanks for your nice words. If you are a total glutton for punishment, Dan and I have a full 3-day online workshop in SEM that is freely available -- see centerstat.org for details. Good luck with your work -- patrick
I had Mark Appelbaum when I was in UNC psych grad program. He was good, but you make things so exceptionally clear. What a great teacher and how lucky the students are to have someone who is such a good instructor!
Paul -- thanks for your incredibly kind message. I really appreciate it. You were so fortunate to have Mark in class -- he is one of my heroes. I don't know if you saw, but he tragically passed away earlier this spring -- what a loss for the field. Take care -- patrick
I have a question about modification indices and correlated residuals. In my model, it looks like if I allow two error variances (across latent variables) to freely covary, my model will substantially improve. I remember hearing that you should only consider allowing residuals to freely covary within the same LV. The two observed variables in question could be viewed as theoretically similar (one is emotion dysregulation - impulsivity, the other is conflict engagement). What is recommended to address this issue? I noticed that if I drop one of the variables completely, the model improves but not sure if this is ethical.
Hi Alexandra -- thanks for the note. That's a tricky question. Although correlated residuals can be used to quickly improve model fit when there is not strong theoretical support for their inclusion, at the same time a correlated residual might make perfect theoretical sense and you would do well to include this in your model. I personally don't think it is an issue of ethics as long as you clearly communicate what you are doing to the reader. If you include a correlated residual based on an MI, simply articulate this to the reader and justify why this was included. Hope this helps -- patrick
@@vutungkhachhang3162 Look into modification indices. Consider respecifying your model if there are theoretically justifiable parameters that modification indices suggest are contributing to poor model fit.
Hi -- thanks for your note. Usually if a CFI and TLI are equal to 1.0 and the RMSEA is equal to 0, that means that the model chi-square is smaller than the model degrees-of-freedom. This is perfectly fine -- it's not a problem at all -- it simply reflects that the model fits extremely well. This sometimes happens when you have a small sample size (because the model chi-square is directly a function of sample size), but by and large it simply reflects a well fitting model. Good luck with your work -- patrick
@@centerstat Thanks for your answer. It's very helpfull! I have a large sample, and it's a very simple model. I would have liked to make it more complex but when I add variables then my indices decrease... It is thus better simple than complex!