Anne Chan posted on Wednesday, December 29, 2010 - 2:47 pm
Hello! I run a LPA on three variables. Comparing 1 to 3 classes solution, the 3 classes solution is the best and it works perfectly fine.
However, when I go for the 4 classes solution. There is a error message:
"THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS -0.514D-18. PROBLEM INVOLVING PARAMETER 11.
From the output file, I see that 1 group only have 1 case in it.
May I ask what does this means? Is that the 4 classes solution does not fit the data?
Can I still continue my project with the 3-class solution in this case?
If your 4-class solution has its best loglikelihood value replicated several times so you believe you have reached the best maximum, then this 4-class solution is not useful and you should proceed with 3 classes. A 4-class solution like that can happen for instance if you let the variances differ across classes, but also in other cases. The one person is like an outlier.