satorra bentler correction

86 views

CFAlavaansatorra-bentler

Skip to first unread message

Ana Inés

unread,

May 28, 2019, 7:03:12 AM5/28/19

to lavaan

Hello!

I am conducting a CFA with laavan, and I got poor fit indexes. In Hu&Bentler 1999 I read that N<250 could be a cause of overrejection of the model (which could be my case, N=218) and suggest using Satorra&Bentler correction.

I have downloaded the package SBSdiff, but have no idea how to run nor how to interpret it.

Any help?

Thank you very much in advance!

Terrence Jorgensen

unread,

May 31, 2019, 7:00:34 AM5/31/19

to lavaan

I am conducting a CFA with laavan, and I got poor fit indexes. In Hu&Bentler 1999 I read that N<250 could be a cause of overrejection of the model (which could be my case, N=218) and suggest using Satorra&Bentler correction.

This is a general question (not about the lavaan software) that can be posted to SEMNET:

http://www2.gsu.edu/~mkteer/semnet.html

Fit indices are not very informative. If your chi-squared test is significant and you want to know why, inspect the correlation residuals to see which correlations your model has trouble reproducing/predicting.

resid(fit, type = "cor")

If any are > 0.1 in absolute value, most would agree that is a substantial failure of the model, so you could brainstorm about why, and about how your model could be improved. It is possible all the residuals are small, in which case the chi-squared test is reflecting many insubstantial prediction errors that you could argue are not of practical significance. Any single global fit index is incapable of providing such detailed information, and no cutoffs work in general.

https://doi.org/10.1207/s15328007sem1103_2

http://dx.doi.org/10.1037/met0000152

I have downloaded the package SBSdiff, but have no idea how to run nor how to interpret it.

I don't see what this has to do with fit indices. It provides a chi-squared difference test if you use other software (e.g., Mplus). But lavTestLRT() provides this automatically when you use robust estimation.

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam