There must be studies with different factorial models in the non-clinical sample that can be proposed, although the most recommendable you can use both groups in only one cfa, to later perform the factorial invariance. You must take into account the distribution of the sample. In case you do not comply with the multivariate distribution, it is recommended that you can remove the extreme data so that your adjustment indices in both groups are adequate. Another option is the error covariance between items from the modification indexes that also allow to improve the cfa models.

Hi Katrine,

your model does not converge which can have a ton of reasons. Some preliminary recommendations:

a) Change to R / lavaan where it is easy to handle non-normality by using various corrections (see lavaan.org for tutorials. I know switching software is a step but it is worth it.

b) Please give more info about your model, the involved variables, question wordings, potential differences on the variances (which can be a problem), and present the path diagram. It may be that your variables are not only non-normal but have completely different distribution (e.g., Poisson).

Best,

Holger

It is very important to perform a test for normality of the dilution in the samples. Test a hypothesis about the differences between the studied characteristics. If you have control groups do a hypothesis test and stop them. If there are values for the samples that are very different from the others, then these extreme values must be justified - the reasons for them. On this basis, it may be necessary to remove them or replace them with the average value, etc.