

Multilevel CFA null, independence, ... 

Message/Author 


Hello, I would like assistance in setting up the syntactic code for a multilevel CFA. I plan to run three models. Using Hox’s (2002) terminology  I am after code for independence, null and maximal models. For the maximal model I need to know how to specify a saturated between level model. For the null model I need the code to fit the withinlevel model to the within and between levels simultaneously with constraints across all estimates from the within and betweensources. For the independence model, I need the specify the models such that the variances of all the variables in the betweenlevel covariance matrix are partitioned into betweenlevel and withinlevel variance, and that in addition, the withinlevel model is imposed on the within level process but that the betweenlevel processes are modelled not to covary at all. My syntax is given below with the specification of the withinlevel measurement model. ANALYSIS: TYPE is TWOLEVEL; ESTIMATOR = WLSMV; ITERATIONS 1000; CONVERGENCE = 0.00005; Model: %within% F1 BY ans12 ans13 ans14 ans15; F2 BY ans16 ans17 ans18; F3 BY ans19 ans20 ans21; F4 BY ans22 ans23 ans24; %between% <<<help>>> Most grateful for any assistance. Kind regards, Jonathon 


My second post ever, Actually I think I have the betweenlevel saturated model worked out, simply this is %between% ans12ans23 with ans24 ans12ans22 with ans23 etc etc etc...until ans12 with ans13 any assistance with the null and maximal models would be appreciated. Kind regards, Jonathon Little 


I am not familiar with the Hox terms you mention, so below are conjectures. Your saturated setup is correct. It sounds like you said in your earlier message that what you call the saturated model is the maximal model. When you say the null model I assume you mean the H0 model  the factor model you are interested in. See UG ex 9.6 for setting up such a model. It sounds like you want to add equality constraints across levels  such constraints are described in the UG, chapter 16. The remaining model is then what you call the independence model, where the crux seems to be to have the betweenlevel covariance matrix have zero offdiagonal elements. This is accomplished by saying %between% ans12ans24; 

Back to top 

