I am analyzing longitudinal data. The NEO-FFI was administered two times. So I have five latent variables for each time point. I use three parcels to measure each factor. The parcels were build using item to construct matching as proposed by Little et al. I want to establish measurement invariance. N is about 170. So now here is the problem:
I get the error message: PSI ist not positively definite. The Problem is about the factor Aggreableness at timepoint 2. I looked in Tech4 and found Correlations greater than one.
Solutions I tried: I set the covariance between measurement error of each parcel at time 1 with the same parcel at time 2 free. So I capture specific variance of each parcel. This brings better model fit and the correlations of latent variables get smaller than one. But the error message stays the same.
I also fixed most covariances of factors from time one with factors of time 2 to zero. This did not help.
Next I omitted some factors from the model and found, that it works fine for some models with only 3 factors. With 4 latent variables I get the error message.
So I thought the problem might be linear dependency.
So what can I do about that? Is ommitting some factors the only way?