Message/Author 

Ahn,Taeyong posted on Wednesday, February 13, 2019  7:45 am



hi mplus team i want to know syntax of fixed effect model in gmm. i want to constrain covariate xmiss(unobserved variable) that influences c. what command should i fill in the next syntax. thanks. VARIABLE: names are ID SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND A1 A2 A3 B1 B2 B3 C1 C2 C3 D1 D2 D3 Autonomy Environmental Growth Relations Purpose SelfAcceptance Rate_life work_situation relationship_children marriage Life_Satis N ; usevar = SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND C1 C2 C3 N ; classes = c(2); nominal = N; missing = all(999); ANALYSIS: type = mixture missing; starts = 0; Model: %OVERALL% i s C1@0 C2@1 C3@2; i s on SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND; c on SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND ; %c#1% [N#1@0.158]; %c#2% [N#1@6.396]; OUTPUT: sampstat standardized tech1 TECH7 TECH8 ; PLOT: SERIES = C1C3 (s); TYPE = PLOT3; 


I don't know what you mean by "constrain covariate xmiss(unobserved variable) that influences c" Also, I don't see the xmiss variable you refer to. 

Ahn,Taeyong posted on Saturday, February 16, 2019  4:18 am



Sorry.I would like to know the net effect of the explanatory variables (covariance, Xs) on the latent variable (C), except for the timeinvariant effect (individualvariant effect). Can I get it by entering an additional command in the following syntax? If so, which command should I use? usevar = SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND C1 C2 C3 N ; classes = c(2); nominal = N; missing = all(999); ANALYSIS: type = mixture missing; starts = 0; Model: %OVERALL% i s C1@0 C2@1 C3@2; i s on SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND; c on SEX GENDER_ROLE PARENT_AFFEC FAMSUP worksup FS_GIVING WORK_DEMAND ; %c#1% [N#1@0.158]; %c#2% [N#1@6.396]; OUTPUT: sampstat standardized tech1 TECH7 TECH8 ; PLOT: SERIES = C1C3 (s); TYPE = PLOT3; 


I still don't understand. You say: " the net effect of the explanatory variables (covariance, Xs) on the latent variable (C), except for the timeinvariant effect (individualvariant effect)." This leads me to the question: Which of your explanatory variables are not timeinvariant? 

Ahn,Taeyong posted on Sunday, February 17, 2019  6:04 pm



I am analyzing the general growth mixture model with longitudinal data. I treated all covariates as timeinvariant variables in my GMM and performed the multinomial logistic regression of latent class on the covariates. But I have heard that covariates can have individualspecific effects and this is unobservable covariate effects. And I understand that net effects of independent variable (covariates / explanatory variable) can be obtained while unobservable individual (exogenous variable) effects are constrained. To do this, I first need to confirm whether the individualspecific effects model is a fixed effect model or a random effect model. For this, it is said that it is necessary to compare the fitness of models that allow correlation between independent variable and unobservable individual effects and those that do not. Is this my idea right? And in my case, is it necessary to perform the fixed / random effect constraint process discussed above and can it be done? And if it is necessary and possible, what I want to know is the specific mplus syntax. (I used the 3step approach in my analysis.) I am very embarrassed because I can not tell what I want. Do not you still understand me? Then I am really sorry. But if you understand me, please give me some help. thank. 


I think you are talking about 2 different things. There is the so called FE versus RE topic of longitudinal data where the random intercept (an unobserved variable) of the RE approach should be correlated with timevarying covariates. In growth mixture modeling you have random intercept and a random slope describing growth. And it doesn't sound like you have timevarying covariates. 

Ahn,Taeyong posted on Monday, February 18, 2019  6:49 pm



Are you saying that there are no random effects of covariates because I don't have timevarying covariates in my model. But I think that random intercept and random slope are correlated with timeinvarient covariate too, because some of timeinvarient covariate may have individualspecific effect as individualvarient covariate. Is this idea right? On the other hand, Are you saying that there is no fixed effect of covariate in GMM because we have random intercept and random slope in GMM? Then, is it possible to say that we can take fixed effect of covariate from LCGA? I am very sorry if you do not understand what I mean or if you feel my idea is nonsence. Thank. 

Ahn,Taeyong posted on Tuesday, February 19, 2019  4:52 pm



I've read the Fixed effect regression models written by Allison(2009), And I've got the answer what I want. Thanks. 


That's a good source for the FERE discussion. 

Back to top 