|
Message/Author |
|
|
I want to regress a continuous predictor on a categorical variable (4 categories or three dummy coded predictors). I have missing data on the dependent but obviously not on the IVs. The N = 368. My purpose is to examine group differences and to use maximum likelihood to address missing data ( focusing specifically on one of the groups as the reference). I have specified the variances in the model and it produces interpretable results. But I keep getting a non-positive definite message. THE MODEL ESTIMATION TERMINATED NORMALLY 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 is a very simple model. The results make sense and my interest is in the resulting estimates. Do I need to address this error? If so, how can I improve the model to accommodate the error? Thanks very much. |
|
|
Perhaps you are including your dummies in the model in the sense of having their parameters estimated. In which case the message can be ignored. |
|
|
Yes, I believe that is it. Thank you, Bengt. |
|
Back to top |
|
|