Message/Author 


Dear forum members, At the moment I am working on a Latent Profile Analysis, to make teacher profiles using continuous indicator variables. Because I am not working with categorical indicator variables, the output seems different than the LCA examples I found.I would like to know how to get the probabilities per indicator, now I only end up with the probabilities per individual (using cprob). Also in the plots, I have difficulties getting the probabilities per indicator. So, I wonder about the following: 1. how can I get the class probabilities per indicator? 2. how can I get plots based on that information? (might be also final cluster distance?) Right know, I feel I have not enough information to evaluate the classes conceptually, since I mostly have (standardized) model fit output and model estimates. My syntax is below in the next message. Thank you. 


Syntax: VARIABLE: names = id inst numsese adher opinion ratio edu pdmoney concobj percfut feasib manag workrel newwork percknow learnenv; auxiliary = id; usevariables = opinion manag workrel newwork percknow concobj percfut feasib learnenv; missing = all(999); classes = c(3); ANALYSIS: type = mixture; optseed = 351807; PLOT: type = plot3; !plots the conditional response probabilities in each class series = opinion(1) manag(2) workrel(3) newwork(4) percknow(5) concobj(6) percfut(7) feasib(8) learnenv(9); MODEL: %overall% %C#1% [opinion manag workrel newwork percknow concobj percfut feasib learnenv]; %C#2% [opinion manag workrel newwork percknow concobj percfut feasib learnenv]; %C#3% [opinion manag workrel newwork percknow concobj percfut feasib learnenv]; OUTPUT: stdyx SAVEDATA: file = "U:\SPELL\spell_3classes.dat"; save = cprobabilities; !probabilities and estimated class membership info !to a new external data file 


With continuous indicators, you focus on the means of those indicators (they don't have probabilities). These means can be plotted. We ask that postings be limited to one window. For longer messages, send to Support. 


I kindly thank you, Bengt, for the fast and helpful response. Excuse me for the syntax post. And of course, that makes sense with continuous indicators. Now I managed to plot them and request the right data with the plot option (Estimated means and observed individual values). These are the data of the plot, is this the information one would use to evaluate the classes conceptually? 1.00000 3.72315 3.25523 4.29975 2.00000 3.98789 3.44173 4.30157 3.00000 3.60931 3.71376 4.10772 4.00000 3.70370 3.57393 4.08713 5.00000 3.95627 3.99979 4.38125 6.00000 3.52472 3.54657 4.10125 7.00000 3.93443 4.89603 4.89500 8.00000 3.10861 3.18370 3.64267 9.00000 3.85497 3.88279 4.41883 (Of course the plot itself helps with this, but I will not be able to post it here) 


Yes. Those are the means for the 9 variables in the 3 classes. 


Thank you for your advice. 


Hello again, I want to account for the nested structure of the data, and therefore conduct a multilevel LPA. In Makikangas et al. 2018 they do this by setting the starting values according to the singlelevel LPA and random starting values as zero, to ensure the Level 1 profiles remain the same. If I understand well, you would use the means and variances of the single LPA and use them in your model specifications for the MPLA. Yet, I seem to get several warnings ('One or more individuallevel variables have no variation within a cluster for the following clusters.' and 'The means of variables on the between and within levels exist only on the between level). Thus, I was wondering if this is the right way to perform a MPLA. ANALYSIS: type = mixture twolevel; starts = 0; MODEL: %between% %overall% C#1; %within% %C#1% [opinion*3.855 manag*3.723 workrel*3.988 newwork*3.609 percknow*3.704 concobj*3.934 percfut*3.525 feasib*3.109 learnenv*3.956]; opinion*0.479 manag*0.387 workrel*0.288 newwork*0.241 percknow*0.204 concobj*0.052 percfut*0.260 feasib*0.252 learnenv*0.555; ... similar for the other profiles... 


I would not use singlelevel LPA estimates for twolevel LPA  that is, I think random starts are needed. We need to see your full output to answer this  send to Support along with your license number. 


Thank you for the response, I will send my full output. 

Back to top 