Message/Author 

Anonymous posted on Tuesday, April 02, 2002  12:07 pm



What should be reported about the results! 


There are several examples of mixture applications at www.statmodel.com under Examples, Mixture Applications. Seeing how results are reported in these papers may be helpful. 

Monica posted on Wednesday, April 03, 2002  10:24 am



Reporting results. For a latent growth mixture model are the estimated factor mean test statistics a "t" or a "z"; for the multinomial logistic regression results it appears from Muthen & Muthen 2000 that the test statistic is a "t"Yes?? Thanks. Monica 


The test statistic is a z statistic. Actually, the cutoffs for z and t are the same for samples of any size, so this shouldn't be an issue. 

Anonymous posted on Sunday, September 29, 2002  3:51 pm



I'm constructing a mixture model in Mplus and would like to know how to interpret the "Condition Number for the Information Matrix" Mplus provides (under the heading "QUALITY OF NUMERICAL RESULTS"). 

bmuthen posted on Sunday, September 29, 2002  5:38 pm



A very small condition number, say of a magnitude less than 1 to the power of 10 or 12, is an indication that the model is not identified. The condition number in Mplus is defined as the ratio between the smallest and largest eigenvalue of an estimator of the ML information matrix. 

Yifu Chen posted on Thursday, March 20, 2003  11:56 am



Hi, Dr. Muthen, I am running a latent class model with covariates. There are 7 binary indicators and four classes are found. I also find a covariate (say, x) has significant relationship with one of binary indicators (say, u4). I read the user guide, the example 25.10 and it says that "for each class, the probability of u4 varies as a function of x". Suppose the coefficient (x predicts u4) I find is .429. I don't know how to interpret this coefficient in plain language? Could you provide more explanation about the interpretation of this relationship? Thanks a lot! 


This coefficent is a logistic regression coefficient. This means that when x increases one unit, the logit for u4 goes down .429 conditional on class membership. The important interpretation is that if this value is significant, it means that the probability of the u4 item given class is not constant for the values of x. Therefore, you don't have measurement invariance. 

dvl posted on Friday, November 16, 2012  1:44 am



Dear, How should I interpret the next results? QUALITY OF NUMERICAL RESULTS Condition Number for the Information Matrix 0.659E05 (ratio of smallest to largest eigenvalue) QUALITY OF NUMERICAL RESULTS Condition Number for the Information Matrix 0.182E04 (ratio of smallest to largest eigenvalue) 


If the condition number is very small, E10, it says that matrix is nearly singular and the model may not be identified. 


Dear Dr. Muthen, I have read the preceding messages and have a related question. If "QUALITY OF NUMERICAL RESULTS" turns out to be smaller than E10, is there any way to address this? In my analysis with categorical data, the value was 0.528E17. I tried different number of classes, but values were always low. Here is how I run the analysis. Data was obtained with 44 item, 5point scale, from 888 respondents.  TITLE: LCA DATA: FILE = '44itemN888.dat'; VARIABLE: NAMES = u1u44; CATEGORICAL = u1u44; CLASSES = C(3); ANALYSIS: TYPE IS MIXTURE;  Thanks very much for your advice, 


We would have to see the output to judge this. 

Back to top 