Message/Author 

Sanjoy posted on Sunday, May 01, 2005  6:12 pm



Dear Professors can we do "Cluster analysis" in MPlus while my variables are categorical I need to sort them by "Identification variable" as well ...I mean on my output sheet I have to have those "identification variable" arranged in their corresponding cluster we can do these things in SAS (using Proc Cluster, or proc fastclus) ... but I think those process are good mainly for continuous variable … are NOT they say in SAS code, this is what I have written PROC CLUSTER DATA=Attitude METHOD=SINGLE; ID Respondent; VAR r1 r2 r3; PROC TREE; run; we can use single/complete/average under “method” specification ... however, the problem of categorical or rather nonnormality won't be handled by this "option specification" thanks and regards 

bmuthen posted on Monday, May 02, 2005  7:13 pm



You can do cluster analysis using the latent class analysis that Mplus provides. For a discussion of this type of cluster analysis, see the HagenaarsMcCutcheon LCA book in the Mplus web site references. This can handle all the kinds of observed outcome types that Mplus allows. 

Sanjoy posted on Monday, May 02, 2005  8:02 pm



Thank you Prof. ... let me have a look at their book ... in between I'm going through our Mplus User's guide chapter 7 for LCA ...I'm still looking for the command/code so that in the result section we can have our Identifying variable in specified cluster/ class, since that is what I need first to establish my hypotheses(as an exploratory analysis only) ... can u help me please I have 240 observation ... my reasonable guess there should be 3, at most 4 classes ... Now, my output should show in which class these 240 observations belong to ... we are doing the clustering on Six indicator variables, and they are categorical thanks and regards 


See the SAVEDATA command and the SAVE = CPROBABILITIES option. 

Sanjoy posted on Tuesday, May 03, 2005  3:27 pm



Oh Madam!Thank u so much ... this is exactly what I was looking for ... Mplus is really cool :) ... thank u all once again 

Lois Downey posted on Friday, November 17, 2006  11:58 am



I ran 15 unclustered multiple regression analyses with an ordered categorical outcome, looking at 15 predictors of interest, with each analysis adjusted for a set of 9 potential confounders. (In most of the analyses, the sample size was over 200.) We then wanted to compare these analyses to a parallel set of analyses with the same observations, but clustered within 10 hospitals. I was expecting to see the standard error for the predictor of interest in each analysis to increase in the clustered analysis. In fact, however, in 11 of the 15 analyses, the standard error dropped. Should I be suspicious of the results? 


10 cluster units is not sufficient to estimate the SEs well. Our simulations indicate that at least 20, and preferrably many more are needed. 


I have a single continuous variable that represents country gender development index and a nominal variable that has the names of the country associated with the score in the continuous variable. If I am running a mixture model, how do I get the names of each country from the nominal variable in the model results and graph? 


I'm not sure what your question is but data in Mplus must be numeric. It cannot contains names. A variable not part of the analysis cannot be incorporated into a graph. 

Harpreet posted on Monday, May 09, 2016  12:55 pm



Dr. Muthens, I have a sample of 72 individuals who rated 12 items on a survey using a likert scale rating 1 'not at all' to 5 'a great deal'. I want to use the 12 items to conduct a cluster analysis. Is my sample size sufficient to run a cluster analysis in Mplus? Is there a minimum sample size required for conducting a cluster analysis? Thank you! 


I think you refer to mixture modeling. I think you can get 2 classes if they are clearly different. 

Back to top 