Stat_love posted on Monday, November 14, 2005 - 12:28 pm
After fitting the model, my result have this kinds of message in output.
(1) ONE OR MORE PARAMETERS WERE FIXED TO AVOID SINGULARITY OF THE INFORMATION MATRIX. THE SINGULARITY IS MOST LIKELY BECAUSE THE MODEL IS NOT IDENTIFIED, OR BECAUSE OF EMPTY CELLS IN THE JOINT DISTRIBUTION OF THE CATEGORICAL VARIABLES IN THE MODEL. THE FOLLOWING PARAMETERS WERE FIXED:4
--> this mean that some parameteres are not meanningless isn't it? so I didn't use the any-result after showing this message. But not sure, could you give me some comment?
(2) In output, Tests of model fit part didn't give the p-value but, Chi-square test of model fit...part give two p-value like pearson chi-squre and Likelihood ratio chi-squre. Some output give different result. one is significant the other is no-significant. How to decise test of model fit result?
(3)In model results, they give estimates and S.E. est./S.E., how to get the P-value? just calcuate ourself? or some option?
p.s. I have a M-plus 'User's Guide' But, if I am a new user in Mplus, I have lots of Questions about output, error message, and interpretation...etc. Could you give me some comment? good book or something.?
bmuthen posted on Monday, November 14, 2005 - 11:58 pm
(1) you probably have a large threshold that is fixed. Such a large thresholds means that the probability of item endorsement is essentially 1. This helps interpretation and is not a problem.
(2) The Pearson and LR chi-squares are for the frequency table. They are only trustworthy if the table has no small or zero cell frequencies and they typically disagree when neither can be trusted. With many items, many cells are zero. In such cases you should instead go by other means of testing described in the literature (see e.g. Muthen 2004 in the Kaplan handbook on the Mplus web site) and in our courses (BIC, loglikelihood differences, etc).
(3) yes calculate yourself. I simply note if I have significance or not.
We have not yet written the Mplus books that we have in mind. Until then you can watch the web videos or come to our courses.
John posted on Sunday, December 11, 2005 - 12:07 pm
I am new to LCA multilevel analysis. My question is:
how can one reduce the number of parameters in a multilevel LCA?
Having read some papers and the mplus manual, I tried to estimate a LCA with 5 dichotomous indicators for 3000 individuals in 20 Groups. For 2, 3 and four classes, I get a normal output with the AIC /BIC needed for model comparison. However, when trying to estimate a LCA with 5 classes, I get the following message:
THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS -0.459D-16. PROBLEM INVOLVING PARAMETER 16.
THE NONIDENTIFICATION IS MOST LIKELY DUE TO HAVING MORE PARAMETERS THAN THE NUMBER OF CLUSTERS. REDUCE THE NUMBER OF PARAMETERS.
You could use a mixture regression model as shown in Examples 7.1 and 7.2 except your dependent variable would be specified as categorical rather than continuous or count.
Min Jung Kim posted on Thursday, February 07, 2008 - 7:44 pm
I am running a mixture model using continuous outcome variables repeatedly measured for 4 times. I don't know whether I need to fix intercepts (or means) of these outcome variables to zero. It seems that the default is to freely estimate intercepts of the outcome variables but to constrain those parameters to be equal across classes. Do I understand correctly? Could you clarify for me when to fix and when to release these intercepts. Also, what does mean fixing intercepts to zero? FYI, I copied my model command below. Thank you very much for your help in advance.
There are two ways a growth model can be parameterized. One is to fix the intercepts to zero and estimate the means of the growth factors. The other is to hold the intercepts equal, fix the intercept growth factor mean to zero, and estimate the other growth factor means. There is a table in Chapter 16 that shows the Mplus defaults for the special | growth model language.