Greg Homish posted on Monday, February 26, 2007 - 8:22 am
A colleague has conducted a structure equation model using data from 3 time points. He has been told by a reviewer that the cross-sectional association between error terms should be excluded from the model, and to focus simply on the longitudinal associations. We are confused why this approach would be better than also modeling the cross-sectional effects. I would greatly appreciate any comments regarding this situation.
By cross-sectional error terms I assume you mean that you have more than one dependent variable per time point. If so, I think the most natural approach is to let those residuals correlate freely. So without knowing more about this situation, I agree with your confusion.
Greg Homish posted on Wednesday, February 28, 2007 - 7:11 am