

Reference group in GMM with X 

Message/Author 

Anonymous posted on Friday, July 16, 2004  10:39 am



I have a 3 class GMM with continuous y's (4 time points) and 3 X's (and 100 random starts). When I test X1 and X2 the last class is the normative class, as expected, and the coefficients for C1 and C2 are relative to the normative class. However, when I test X3 (alone, without X1 and X2) the reference group changes so that the largest class (normative) is the first class is thus compared to the last class (which is now a nonnormative class)the class distributions stay essentially the same. I would like to report the results relative to the normative class, how might I change this in the command language? Thanks in advance. 

bmuthen posted on Friday, July 16, 2004  10:55 am



You have to give some key starting values to make a certain class come last. For instance, take the estimates for the normative class in the solution where this comes last and use them as starting values for the last class in the new run. 

Nassim Tabri posted on Tuesday, November 20, 2018  12:36 pm



Hi Mplus team, I am estimating a dual process conditional LCGA and want to change the reference class for each single process. My syntax is below: %OVERALL% cu with cy; I1 S1  Dep_T1@0 Dep_T2@1 Dep_T3@2 Dep_T4@3; I2 S2  ANX_T1@0 ANX_T2@1 ANX_T3@2 ANX_T4@3; I1s1@0; i2s2@0; CU CY on SCPT1 PSPT1; I1 S1 I2 S2 on SCPT1 PSPT1; MODEL cu: %CU#1% [I1  S1]; %CU#2% [I1  S1]; MODEL cy: %CY#1% [I2  S2]; %CY#2% [I2  S2]; %CY#3% [I2  S2]; %CY#4% [I2  S2]; I want to change the reference class for CU from class 3 to class 2. As well, I would like to change the reference class for CY from class 5 to class 3. I used the starting values approach (i.e., included estimates of intercept and slope factors for the desired reference class in the last class of each process) that has been suggested on the Mplus discussion board, but it did not work. Would appreciate your help with figuring out how to change the reference classes. Is there a syntax I am missing so that “ALTERNATIVE PARAMETERIZATIONS FOR THE CATEGORICAL LATENT VARIABLE REGRESSION” can be estimated? Thank you, Nassim 


It should word fine if you include all intercept and slope growth factor means in all classes. If you like, you can send your 2 outputs to Support along with your license number. 

Back to top 

