I’m using Mplus to conduct LCA with 3 categorical indicators (2 with 3 options and 1 with 2 options) and 5 continuous indicators. I am able to run a 2 and 4 class solution with no errors; however, when I attempt LCA with 3, 5, etc. classes I get the following error: 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 0.205D-20. PROBLEM INVOLVING PARAMETER 34.
I’ve used tech1 to examine the parameter (an have an issues with C#2 parameter) and have changed start values, and yet this problem persist. Would you have any suggests on how I can ‘fix’ this issues?
Additionally, what are your thoughts about having more classes than variables in the model?
lisa Car posted on Tuesday, April 16, 2019 - 6:13 pm
I am getting a similar response when running a 3 class LPA with 15 indicators. I have tried increasing starts to 100 20 without any change. The model terminates normally. Do you have any suggestions for how to rectify this?
We need to see your full output to answer this - send to Support along with your license number.
lisa Car posted on Monday, November 04, 2019 - 8:53 am
I am getting a similar error message running a 3 or 4 class LPA with 18 indicators. Increasing start values to 200 40 does not help. I also seem to be having another issue in that one of the parameter estimates in all 3 or 4 classes is negative and there are no negative values in the data. I am using -99 for missing and it is properly coded in the dataset for all variables and in Mplus.