Controlling for other variables in a CFA

[Alex Hodgkiss]

Hi,

I'm running a CFA using five spatial ability measures as observed variables, loading onto a single factor. I also have a measure of verbal ability which I want to use as a proxy for general ability. In another similar paper I read, they referred to specifying models using the residualised covariance matrix after partialling out vocabulary. I ran this as a MIMIC model, with the factor regressed onto the verbal measure ( Factor ~ verbal), but this doesn't seem to adjust the factor loadings a great deal. Do I need to regress the individual spatial tasks separately onto the verbal measure within the model, in order to partial out the variance and/ or is this possible? Or is there another approach?

Many thanks

Alex

[Terrence Jorgensen]

A residualized covariance matrix implies that the observed variables (not common factors) were regressed on the covariates.  When you do this, you should get what you want.  I would also set conditional.x = TRUE, so that the effects of covariates are partialed out before fitting the model, as you intend.

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

UvA web page: http://www.uva.nl/profile/t.d.jorgensen

[Alex Hodgkiss]

Great. Many thanks!