Message/Author |
|
Wendong Li posted on Wednesday, April 29, 2009 - 7:28 pm
|
|
|
Hi, I came across a problem using Mplus 4 and I am wondering whether some expert could give me some inputs. The problem occurs when I am running multiple group SEM on a dataset with quite a number of missing values. I do use the MISSING = command in the VARIABLE section. But the results tell that Mplus only processed analyses based on full information group pairs--it did not use the information which is available for only one group. Mplus 5 seems to be able to handle this problem. But does that mean I should buy a new version? Thanks a lot in advance! Wendong |
|
|
I know of no changes related to missing values between Version 4 and 5 other than that the default in Version 4 was listwise deletion and in Version 5 it is TYPE=MISSING. I recommend always using the most recent version of Mplus as it is constantly being improved. |
|
Wendong Li posted on Thursday, April 30, 2009 - 8:07 pm
|
|
|
Thank you so much for the reply. I added a new line of command in the ANALYSIS section and it works now. I will think about ordering a new version of Mplus... |
|
Wendong Li posted on Thursday, April 30, 2009 - 8:09 pm
|
|
|
Oops, the new line of command is exactly TYPE=MISSING;. |
|
|
Hi, I'm running a model in which I am including 3 auxiliary variables (m) for missing values. However, it's also a mediation model that I am testing, thus, I want to use the bootstrap option. It does not seem to be possible. What is the recommendation in such a context? Should I simply model the effects (pathways: on statements) between the auxiliary variables and the variables with most missing values directly into the model? Thank you Simon C. |
|
|
Yes, you would have to model it. We describe the auxiliary missing modeling on pages 194-200 of our Topic 4 Short course video and handout. |
|
|
Thanks! From my understanding I need to model a saturated correlates model, is that it? Simon |
|
|
Right. It's laid out there on slide 198. |
|
|
My model: 4 factors & 5 controls (included their means due to missing). Is this syntax OK for a saturated correlate model with 3 aux. var: typeofjob_5 T1_EtSan T1_TIPI_Nevro? TIME BY T1_Concil_1 T1_Concil_2 T1_Concil_3; PMH BY T1_SMP_1 T1_SMP_2 T1_SMP_3; PWBW BY T1_BEPT_2 T1_BEPT_7 T1_BEPT_12; MIND BY T1_mind_1 T1_mind_2 T1_mind_3; T1_SMP_8 WITH T1_SMP_6 ; T1_SMP_8 WITH T1_SMP_7; PWBW PMH ON MIND; MIND ON TIME ; PWBW WITH PMH; PWBW PMH ON TIME; PWBW PMH ON Sexe Age SitEcono T1_PleinPartiel lengthcurrentwork; [Sexe Age SitEcono T1_PleinPartiel lengthcurrentwork]; typeofjob_5 WITH T1_EtSan T1_Concil_1 T1_Concil_2 T1_Concil_3 T1_SMP_1 T1_SMP_2 T1_SMP_3 T1_BEPT_2 T1_BEPT_7 T1_BEPT_12 T1_mind_1 T1_mind_2 T1_mind_3 Sexe Age SitEcono T1_PleinPartiel lengthcurrentwork; T1_EtSan WITH T1_TIPI_Nevro T1_Concil_1 T1_Concil_2 T1_Concil_3 T1_SMP_1 T1_SMP_2 T1_SMP_3 T1_BEPT_2 T1_BEPT_7 T1_BEPT_12 T1_mind_1 T1_mind_2 T1_mind_3 Sexe Age SitEcono T1_PleinPartiel lengthcurrentwork; T1_TIPI_Nevro WITH typeofjob_5 T1_Concil_1 T1_Concil_2 T1_Concil_3 T1_SMP_1 T1_SMP_2 T1_SMP_3 T1_BEPT_2 T1_BEPT_7 T1_BEPT_12 T1_mind_1 T1_mind_2 T1_mind_3 Sexe Age SitEcono T1_PleinPartiel lengthcurrentwork; |
|
|
I think you have to add T1_SMP_8 T1_SMP_6 T1_SMP_7 WITH typeofjob_5 T1_EtSan T1_TIPI_Nevro; The best way to verify is to check the difference between the number of parameters between the two models (the model with the Aux variables and the model without the Aux variables) If P is the number of variables in the model, I think P in your model is 20, and M is the number of Aux variables, M=3, the total number of added parameters should be P*M (covariances) + M*(M+1)/2 variance covariance + M (means) =69 Here is another reference you might find useful http://www.statmodel.com/download/AuxM2.pdf |
|
|
Thanks so much! Simon |
|
Back to top |