Hello, I am working with a noncongeneric CFA model . N=112, 12 indicators, residual variances from one and two and three factor EFAs are in excess of .85 and as high as .95 for many indicators). Fitting a 1,2, and 3 CFA, 12 indicators on a factor 1, 5 on factor 2, 4 on factor 3. Indicators on factors 2 and three cross load on factor 1. No correlated residual indicator variances and factors are orthogonal. First run, set scale of latents by fixing loading of an indicators to 1, latent variances freely estimated. No convergence, negative variance for the second latent, zero variance for the third. Selection of indicators with fixed loadings was arbitrary. Second, run, set scale of latents by fixing their variances to 1, loadings for all indicators freed . No convergence, small residual indicator variance. Once this was set to 0, model converges to proper solution. Negative variance on last model still remains a concern. Am aware that for a congeneric model, holding all else oonstant, models are equivalent irrespective of method used to set the scale of the latent. My reading indicates that this may or may not be the case for noncongeneric models. In my situation, could the scaling of the latent have led to model misspecification accounting for failure to converge and the negative and zero latent variances? Thank you!
I don't think the problem has anything to do with the metric. Usually when setting the metric using a factor loading has problems, changing to fixing the factor variance to one solves it. It is usually caused by the factor loading selected being negative for example. It sounds like you have a difficult model. If you send the output and your license number to firstname.lastname@example.org, we can take a look at it.