Hi, As the user guide said, in multiple group analysis, intercepts, thresholds, and factor loadings are held equal across groups as default. But does this also apply to second-order factor model?
I have a second-order factor model: three second-order factors and each second-order factor has two first-order factors (each first-order factor has multiple observed variables). I have no problem constraining the first-order factor loadings across groups but have difficulty holding the second-order factor loadings equal across groups.
These should be held equal as the default. You should not have to do anything to hold them equal. Please send the output and your license number to firstname.lastname@example.org.
Thomas Eagle posted on Thursday, December 04, 2014 - 9:14 am
I am having trouble fitting a multiple group model with 2 2nd order factors. I can fit the two groups separately just fine with the model structure I have, but when I attempt to fit them using the default multiple group mode setup I get non-convergence and the error is related to the mean of one of the 2nd order factors (183 = alpha of 2nd order factor SpRelInt)
THE STANDARD ERRORS ... THE MODEL MAY NOT BE IDENTIFIED... PROBLEM INVOLVING THE FOLLOWING: Parameter 183, Group NOT_ENGAGED: [SPRELINT]
THE CONDITION NUMBER IS -0.348D-14.
THE ROBUST CHI-SQUARE...
I thought the means of 2nd order factors were held constant across groups. I know I have to free up, or fix some values somewhere. When I tried to fix the mean of SpRelInt to zero the error message remains the same.
Perhaps you don't fix the intercepts of the 1-st order factors to zero in your analysis.
Thomas Eagle posted on Thursday, December 04, 2014 - 5:33 pm
I figured out what I was doing wrong. When I set the 2nd order factor means/intercepts to zero for BOTH 2nd order factors the model converged nicely. I thought that was done automatically. This way I could leave the 1st order factor intercepts free. My bad. Thank you! You are always so patient and helpful!
But for scalar invariance, I think you would want the means of the second-order factors estimated in all but one group so you can compare them. That then requires that the first-order factor intercepts are zero in all groups and the factor indicator intercepts are held equal across all groups.