Message/Author 

Xu, Man posted on Thursday, October 30, 2008  12:40 am



I am interested in the SEM only in the first level; however, if I need to take into account of the multilevel structure, is multilevel SEM the best way to do it, even if I am not really interested in level 2 results? (I assume that the SEM model on the second level should have the same model specification as the first level except for using group level indicators of the level 1 indicators for level 2 factors, constraining loadings to be the same as corresponding level 1 factors) Thank you! 


It is not likely that the between specification would be the same as the within specification. Usually there are fewer factors on between than within. You can explore the betweenlevel factor structure using TYPE=TWOLEVEL EFA. 

Xu, Man posted on Wednesday, November 05, 2008  9:11 am



Thank you for your help. I guess I am trying to get a similar thing as in multilevel regression random intercept models. Could there be an analogue in multilevel SEM where the level 1 intercepts of each indicators are free for each cluster instead of having to specify something on the higher level? 


You can specify your SEM model on level 1 and covariances among the random intercepts on level 2. See Example 9.6. The between part would be: %BETWEEN% y1 WITH y2 y3 y4; y2 WITH y3 y4; y3 WITH y4; These WITH statements are covariances among the random intercepts. 

Xu, Man posted on Tuesday, November 11, 2008  3:36 am



Is allowing covariance at level 2 a strategy for increasing fit? Can I just leave the between bit empty? 


If you do not want to specify a particular model on the between level, it probably makes sense to have a saturated model. 

Back to top 