If you need to use TYPE=IMPUTATION, you would need to plot outside of Mplus. The chi-square is computed in a special way for TYPE=IMPUTATION. It sounds like something about your data make this computation impossible.
I have a question about a LGM I'm trying to run with MI-data. The LGM is based on data from three waves and the variables are continuous. The model also includes a binary outcome that is regressed on the intercept and the slope. I also want to include four additional covariates to the model which causes some trouble since this reduces the sample size by approximately half. I have therefore imputed data for my four covariates (but not for the other variables in the model).
In the first step of the analysis I run the model without including the covariates, which should produce identical results regardless if the analysis is based on one dataset or on 100. When I specify the outcome as continuous the results are the same, but when I specify it as categorical the the effects of I and S differs completely (changes from positive and significant to a minimal negative effect and a p-value of .99). Since the outcome is binary I want present ORs for the effects of I and S. I have used MLR as estimator in both analyses
Do you have any explanation for why this happens and any suggestions how to resolve the problem.
I don't understand what you are doing. Can you send the two outputs - continuous versus categorical and your license number to email@example.com.
Lina Homman posted on Friday, August 02, 2013 - 7:37 am
Dear Dr Muthen and Muthen,
I am having a problem with my plots when running a LCGA. I am running a LCGA with 3 time points, outcome variables are continuous. I add covariates and direct effects from covariates to outcome variables. The model runs fine and the plots look good, but reduces my sample size. I decided to use MI for the covariates (I have tried to add them into the analysis but this does not work). The analysis runs fine but the plots do not. The issues with the plots is that the estimated means looks normal but all the sample means are consistently zero. Do you have any idea where this is going wrong? Many thanks
You would need the Base plus Mixture or Base plus Combination Add-On.
Lina Homman posted on Friday, August 02, 2013 - 9:45 am
great, many thanks
Lina Homman posted on Friday, August 02, 2013 - 9:53 am
Linda, I tested the Demo version 7.11 but it does still not work. The output looks fine it is just the plots which do not, the sample mean. The estimated mean is what I expect it to be. Do you have any other ideas on how to deal with this? I cannot send the data files I am afraid due to data protection and confidentiality of the data.
DMello posted on Sunday, November 16, 2014 - 11:47 am
After following the guidance of GCM using multiple imputation, my model ran successfully. However, the N was less than it should be if the data was imputed.
In other words, were the parameters estimated using the imputed datasets?
Here is my code:
TITLE: ... DATA: ... VARIABLE: NAMES = a1ct1-a1ct6 mincome mbod mlotr; USEVARIABLES = a1ct1-a1ct6 mincome mbod mlotr mincomeMbod mincomeMlotr mbodMlotr mincomeMbodMlotr; MISSING = ALL (999); DATA IMPUTATION: IMPUTE = a1ct1-a1ct6 mincome mbod mlotr; NDATASETS = 20; SAVE = A1cImp*.dat; ANALYSIS: ESTIMATOR = ML; DEFINE: mincomeMbod = mincome*mbod; mincomeMlotr = mincome*mlotr; mbodMlotr = mbod*mlotr; mincomeMbodMlotr = mincome*mbod*mlotr; MODEL: i s | a1ct1@0 a1ct2@1 a1ct3@2 a1ct4@3 a1ct5@4 a1ct6@5; i s ON mincome mbod mlotr; i s ON mincomeMbod; i s ON mincomeMlotr; i s ON mbodMlotr; i s ON mincomeMbodMlotr; OUTPUT: TECH1 TECH8;