Fitting a multilevel factor analysis for binary responses
Posted: Fri Nov 12, 2010 4:13 pm
Dear MLWin Users,
I have been trying to fit a multilevel factor model for binary responses (using the logit link). I have followed the example of the continuous case (with some minor adjustments - e.g. link function ). However when I move from fitting the multivariate model for binary responses using IGLS and change the estimation method to MCMC, the model fails to run and gives the Warning:
MCMC Error 1301: Factor analysis model require block-diagonal variance matrices.
Yes indeed when I looked at the equations window I find that there is a variance matrix which is not diagonal. The problem is that MLwin advises me that I should revert to IGLS to rectify this. But when I go back to IGLS the variance matrix would have disappeared.
Can somebody please advise me on how to properly fit a multilevel factor analysis for binary responses?
Regards,
Luwis
I have been trying to fit a multilevel factor model for binary responses (using the logit link). I have followed the example of the continuous case (with some minor adjustments - e.g. link function ). However when I move from fitting the multivariate model for binary responses using IGLS and change the estimation method to MCMC, the model fails to run and gives the Warning:
MCMC Error 1301: Factor analysis model require block-diagonal variance matrices.
Yes indeed when I looked at the equations window I find that there is a variance matrix which is not diagonal. The problem is that MLwin advises me that I should revert to IGLS to rectify this. But when I go back to IGLS the variance matrix would have disappeared.
Can somebody please advise me on how to properly fit a multilevel factor analysis for binary responses?
Regards,
Luwis