|
Message/Author |
|
|
Hi, This is about full rank single-level models in Mplus. I placed it here because if it is possible to estimate full rank single-level models in Mplus, I wanted next to look at full rank multi-level models. Following is the model code I tried. T_Dum is a dummy code; C_Dum is 1 - T_DUM; T_pre is T_Dum*efpre where efpre is a pretest score; C_pre is C_Dum*efpre. I used ML. Starting values come from proc mixed code estimating the same model using the no intercept option and ML. model: efpost on T_Dum*86.521 T_pre*.2494 C_Dum*30.8764 C_pre*.7090; efpost*48.199; [efpost@0]; The Mplus estimates are not equal to the starting values and neither the Ho LL nor the H1 LL is equal to values obtained using model: efpost on T_Dum T_pre efpre; So it seems that I have either formulated the full rank model improperly or it is not possible to estimate full rank models in Mplus. Can you tell me which, if either explanation is correct and if I have formulated the model improperly can you tell me how to fix my code. Thanks, Jamie |
|
|
Jamie Send the two Mplus runs and data to support@statmodel.com and we will find the problem. The two runs should have the same H0 LL. |
|
|
To conclude, Mplus can indeed handle the model by using the Variance option of the Analysis command to do ridging needed for the singular covariance matrix for the covariates. |
|
Back to top |
|
|