Message/Author 


Hi, I am working with a LCA model with two latent variables each with two categories. The model looks good and the categories makes a lot of sense and look like what I had expected. I wanted to use auxiliary variables in 3step process to check their effect on latent classes. However 3 step does not seem to work with the models with more than one latent variable. When I run a LCA model with one variable and 4 categories instead of having two variables with two categories I get different classes which don't look as good as two variable categories. So is there a way to constrain LCA model with one latent variable (4 categories) in a way that it will have the same categories as the model with two variables? In this case I would have the categories I want and still use 3step for multinomial regression. Thank you very much! 


See the LTA example in Web Note 15. 


Hi, I encounter a problem that did not occur when I modelled only one categorical latent variable. Now that I model a two categorical latent variable multiple group LCA, I always get the following error message: *** ERROR in MODEL command Unknown class label in MODEL : %CCYCLE#1.CC#1% However, CYCLE has been introduced in the names command and further: knownclass = CCycle (Cycle = 1 Cycle = 2); classes = CCycle(2) CC(4) CS(4); In my model command, I state: %OVERALL% CC CS ON CCycle; %CCycle#1.CC#1% [P333a$1] (1); [P333b$1] (2); [P333c$1] (3); [P333d$1] (4); [P333e$1] (5); […] This worked with only one categorical latent variable, and I had assumed I could introduce similar restrictions in a multiple group LCA with two latent variables. If not so, how can I fix the conditional response probabilities (and class sizes) to be the same in both (known) groups? Thank you! 


Please send output to support@statmodel.com 

PS posted on Monday, February 26, 2018  11:42 am



Good Afternoon, I am running an LCA wherein I would like to 1) use covariates to predict the classes and 2) use the classes to predict a distal outcome. Is this possible with DCAT? I am having trouble determining how to specify the covariates in the DCAT context. 


Use the manual approach described in our 2 web notes on 3step analyses #15 and #21. 

PS posted on Monday, May 07, 2018  10:43 am



Hello, I'm working on a model where latent classes AND some covariates predict DVs. I get: "ERROR The following MODEL statements are ignored:Statements in the OVERALL class: DV1 ON CLASS#1 DV1 ON CLASS#2 DV1 ON CLASS#3 ...DV2 ON CLASS#3" The step 3 syntax is below; where ind=step1 LCA indicator; cov=covariate; dv=dependent variable; n=my attempt at including the nominal variable Names = ind1 ind2 ind3 ind4 ind5 weight ID cov1 cov2 cov3 cov4 dv1 dv2 CPROB1 CPROB2 CPROB3 n !this was previously called ˇ°classˇ±, but I named it ˇ°nˇ±; VARSTRAT VARPSU; usevariables = cov1 cov2 cov3 cov4 dv1 dv2 n; classes = Class(3); nominal = n; WEIGHT = weight; Stratification = VARSTRAT; cluster = VARPSU; Analysis: type = COMPLEX MIXTURE; algorithm = integration; !added based on a prior error; Model: %OVERALL% Dv1 on cov1 cov2 cov3 class#1class#3; Dv2 on cov1 cov2 cov3 class#1class#3; %class#1% [n#1@1.783567498]; [n#2@3.171085161]; %class#2% [n#1@4.166665224]; [n#2@1.907530721]; %class#3% [n#1@2.852238888]; [n#2@2.119086373]; 

PS posted on Monday, May 07, 2018  10:52 am



In the message above, I forgot to mention that the DV is binary. 


You don't regress on the latent class variable. The effect of the latent class variable on the DV is seen in the change of means/thresholds over classes. 

PS posted on Tuesday, May 08, 2018  10:06 am



Thank you kindly for the response. Can you suggest any sources with annotated output or articles? I'm not sure how to interpret these, where to look in the output? I'm specifically not sure how to differentiate the effect of the classes and covariates on the DV. 


Have a look at NylundGibson, K., Grimm, R., Quirk, M., & Furlong, M. (2014): A latent transition mixture model using the threestep specification. Structural Equation Modeling: A Multidisciplinary Journal, 21, 439454. 

Back to top 