finnigan posted on Friday, June 19, 2009 - 6:54 am
I am looking at a five factor model with 56 survey items answered on a five point likert scale. Data were collected from 130 people who are nested within 25 organisations which are nested within 5 geographical regions. I am trying to understand if there is an organisational impact on the individual level factor loadings arising from organisational membership. Would a multilevel CFA help me to identify the impact of an organisational context on individual level factor structure and item loadings?
Given the above is it possible to rum a multilevel CFA?
I am attempting to fit a similar set of twolevel CFAs to 1181 cases collected from 22 villages. Most CFAs posit one or two factors at each level with 6-16 items per factor. Items have three ordered categories, so I specify them on the CATEGORICAL statement. Depending on the estimator (I've tried WLSMV & MLR) and item sets I am using, I am receiving various warnings and errors, more so than convergence (though occasionally I do get convergence for some of the one-factor models). I think this may be due to the relatively small number of villages at the between level. I imagine this problem (small number of level-2 units) is encountered frequently in twolevel CFA work. Do you have a set of recommended best practices or suggested remedies I could try to obtain convergent solutions in this situation? Thanks.
We have found that the factor structure on the between level is most often simpler than on the within level. Try one factor on the between level with the residual variances fixed at zero. See the following paper which is available on the website:
Muthén, B. (1994). Multilevel covariance structure analysis. In J. Hox & I. Kreft (eds.), Multilevel Modeling, a special issue of Sociological Methods & Research, 22, 376-398.