I have two questions regarding how to perform an LPA analysis.

I have now created several models with different specifications of the variance-covariance matrix and with some specifications such as the Mplus standard specification "Variances are constrained to be equal across profiles and no covariances allowed" or even with the specification "Covariances between indicators are allowed, but both covariances and variances are equal across classes" I get a warning message when I create the model with two or more classes , and with other specifications like "Variances are allowed to vary between classes but no covariances allowed" I get no warning message. Does this mean that these specifications just don't fit my data - or how should I best handle this? I think the warning message refers to the fact that for these two specifications, only 6 people are placed in the one latent class. In other specifications I do not get such a small latent subgroup.

The warning message is: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE VALID FOR SOME PARAMETERS BECAUSE OF A NON-POSITIVE DEFINITIVE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO INITIAL VALUES, BUT MAY ALSO BE AN INDICATION OF NON-IDENTIFICATION OF THE MODEL. THE CONDITION NUMBER IS -0.102D-15. PROBLEM WITH THE FOLLOWING PARAMETER:

Parameter 8, %C#1%: [ GROAT1 ]

Thanks in advance!