I am working on a multilevel mediation model and I get this warning:
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 0.105D-10. PROBLEM INVOLVING THE FOLLOWING PARAMETER: Parameter 15, %BETWEEN%: MAV
MAV = the Y variable in the mediation and parameter 15 is the residual variance of MAV in the between-level PSI matrix. The unstandardized residual variance is 0.014 with SE of 0.035, which is small, but does not seem to be too small to work with.
When I fix the residual variance of MAV to zero, the warning disappears, so it fixes the problem, but I have the feeling this is not the right solution.
1. Can you tell me if there is a better way to solve this issue?
2. Could you help me to understand the cause of this message? (Maybe by pointing me in the direction of literature about what is going on?)
3. The error message says that model parameters ‘may not be trusted’, which means that maybe they CAN be trusted. Is there a way to know when parameters can be trusted and when they can not?