Anonymous posted on Tuesday, January 31, 2006 - 8:27 am
I am really quite new to the whole mplus-thing, so please forgive me naivité.
I want to specify some kind of 'factor mixture model' or 'latent profile analysis' and have some questions concerning syntax-writing. I have seven continous variables which I assume to have specific means, variances and covariances in two (or more) unobserved groups. Because by default Mplus (Version 3.0) assumes a) variances of observed to be equal, b) covariances of observed variables to be zero in these classes, I need to override these defaults. So my syntax goes as follows (cases are weighted):
TITLE: factor-mixture-model / latent profile analysis, englisch; DATA: file is englisch.dat; format is 1f10 8f4; VARIABLE: names are idstud ekenntn bedemut bedevat eunter umgange werte eltinte sampwgt; usevariables are ekenntn bedemut bedevat eunter umgange werte eltinte sampwgt; classes = c (2); weight is sampwgt; ANALYSIS: type = mixture; MODEL: %Overall% ekenntn with bedemut bedevat eunter umgange werte eltinte; bedemut with bedevat eunter umgange werte eltinte; bedevat with eunter umgange werte eltinte; eunter with umgange werte eltinte; umgange with werte eltinte; werte with eltinte; ekenntn bedemut bedevat eunter umgange werte eltinte; %c#1% ekenntn bedemut bedevat eunter umgange werte eltinte; ekenntn with bedemut bedevat eunter umgange werte eltinte; bedemut with bedevat eunter umgange werte eltinte; bedevat with eunter umgange werte eltinte; eunter with umgange werte eltinte; umgange with werte eltinte; werte with eltinte; %c#2% ekenntn bedemut bedevat eunter umgange werte eltinte; ekenntn with bedemut bedevat eunter umgange werte eltinte; bedemut with bedevat eunter umgange werte eltinte; bedevat with eunter umgange werte eltinte; eunter with umgange werte eltinte; umgange with werte eltinte; werte with eltinte; OUTPUT: tech1;
Question 1: Do you think, my syntax matches my assumptions mentioned above? Especially: Do I have to free variances and covariances in the %Overall% part of the Model-statement? To me, that seems a reasonable thing to do, since I expect my variables to covary in the whole sample.
Question 2: However, the estimation of the model above won't terminate. What Mplus (Version 3.0) tells me is:
THE LOGLIKELIHOOD DECREASED IN THE LAST EM ITERATION. CHANGE YOUR MODEL AND/OR STARTING VALUES.
THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ERROR IN THE COMPUTATION. CHANGE YOUR MODEL AND/OR STARTING VALUES.
I already tried starting values for the intercepts, as you suggest in your example 7.10 but that leads to another error message concerning the ESTIMATED OVARIANCE MATRIX FOR THE Y VARIABLES, which IN CLASS 2 COULD NOT BE INVERTED. Did I get something wrong? I don't think, my model is so complex after all, but fitting it just won't work...
Because the Mplus defaults can vary depending on the type of model being estimated, I suggest looking at your results to see if you are estimating the parameters that you intend. Even non-converged results will list the parameters that are being estimated.
I wonder if you were successful with the Mplus default model. If you have success with this, then you can relax the defaults but not all at the same time. I also suggest using Version 3.13 which is the current version. There have been many changes since Version 3. If you continue to have these problems which a data dependnet, send your input, data, output, and license number to firstname.lastname@example.org.
Vlad posted on Wednesday, February 24, 2010 - 4:19 am
Hello, When I run the FMA in the output file I have f and c_f, among other variables. As far I understood, f corresponds to the latent variable. Then, What about c_f? Also, I have the 2nd stage in my work. In this stage what values f/c_f should be used?
f are factors scores mixed over classes. c_f are factor scores based on most likely class membership. Regarding which to use, that depends on your research question. It is also related to measurement invariance. The following paper discusses the topic of measurement invariance for different FMA models:
Clark, S.L., Muthén, B., Kaprio, J., D’Onofrio, B.M., Viken, R., Rose, R.J., Smalley, S. L. (2009). Models and strategies for factor mixture analysis: Two examples concerning the structure underlying psychological disorders.
Vlad posted on Wednesday, February 24, 2010 - 10:18 am
You are calling the class variable "class", not c.
Samuel posted on Monday, February 04, 2019 - 3:01 pm
Dear Drs. Muthen, I am performing the 3-step procedure with a 5-class LPA and, for now, auxiliary covariates. My analysis ran properly, but it included a warning:
"TESTS OF CATEGORICAL LATENT VARIABLE MULTINOMIAL LOGISTIC REGRESSIONS USING THE 3-STEP PROCEDURE
WARNING: LISTWISE DELETION IS APPLIED TO THE AUXILIARY VARIABLES IN THE ANALYSIS. TO AVOID LISTWISE DELETION, DATA IMPUTATION CAN BE USED FOR THE AUXILIARY VARIABLES FOLLOWED BY ANALYSIS WITH TYPE=IMPUTATION."
I attempted to add "TYPE = IMPUTATION" to my input syntax, first under "ANALYSIS" and then under "VARIABLE," both of which resulted in error messages. What is the correct syntax for data imputation for the auxiliary variables?