| Message/Author |
|
|
|
|
dear all,
I am calculating GMM with panel data on deviant behaviors, and I would like to specify a zero-class where the non-offenders should be placed. How can I specify that class in the MODEL part of the input?
...or better, if I want to put some constrains on class c#2, for instance [I-Q@0] or I-Q@0, how can I be sure that in the next run this class will be the same as in the previous run?
e.g. --> In the first run I identified class c#2 as the non-offenders class. Then I set the means and variances of the random effects for this class equal to zero and run the program again. However in the second run c#2 was not anymore the non-offenders class.
thanks a lot in advance,
luca |
|
|
|
|
I don't know what type of outcome you have. A zero class is specified as follows if you have count-inflated outcomes and use ZIP and a quadratic growth model for both the counts and the zero-inflation:
%c#3%
! zero class throughout
[i-q@0];
[ii-qi@0];
i@0;
[cage10-cage30@-15];
[cage10#1-cage30#1@-15];
This is the setup used in Kreuter-Muthen (2007).
If you use the right constraints in class 2, it will be the zero class. |
|
|
| Anonymous posted on Friday, June 22, 2007 - 7:49 am
|
|
|
First, I have been trying to run a ZIP model with no covariates (as baseline) with a response that is a "summed count". I know that this is not consistent with the Mplus setup. I have been using examples from the manual on-line material by Acock and Kreuter. There are differences in the code structures. A definitive example would be helpful. For example, does the data two part command do more than data manipulation? My runs have generated errors ranging from the nonzero derivative of the loglikelihood to statements about positive definite matrices, singularities, specific codes and parameters of issue, etc. Please indicate the best code structure. Then I can grapple with other issues.
Second, if the issues are at least in part due to my data or specification, I wonder about the following. Since my summed counts are not really counts, is there any way to model them as Poisson anyway, since the cross-sectional distributions at each wave seem to conform to Poisson. Or, if I were to transform the nonzero counts with a log transform, is there a way to make Mplus model the Vij (where Yij ne 0) response as Poisson, even though the data would be noninteger?
Third, perhaps a different model? The last post referring to Kreuter-Muthen 2007 may be an option, please provide a full eference or link. Running this model as a mixture may give me more flexibility in terms of the distributional form of the response. Please comment. |
|
|
|
|
Check the heading zero-inflated Poisson in the user's guide. There are several examples to look at. DATA TWOPART is not statistical. We do not have two-part modeling for count variables.
Please send your input, data, output, and license number to support@statmodel.com so I can see what your problems are. |
|
|
| Anonymous posted on Tuesday, June 26, 2007 - 8:28 am
|
|
|
| I would appreciate a more complete reply to the above questions. I am unable to share my model at this time. |
|
|
|
|
| It seems like you can model the sum of several count variables as Poisson because a sum of Poisson variables is Poisson distributed. The Kreuter Muthen paper is on the website under Papers. If you want the Mplus inputs for this paper, please send me your email. |
|
|
| Jungeun Lee posted on Thursday, November 15, 2007 - 1:32 pm
|
|
|
Hi,
I am trying to represent a class that is at zero throughout the study time points in growth mixture models with count variables. Based on Bengt's June 20, 2007 - 6:13 pm post above, I set up the mplus input like ;
%OVERALL%
i s | alacr96r@0 alacr99r@1 alacr02r@2 ALACR05r@3;
i@0;
s@0;
%c#1%
[i-s@0];
[alacr96r-ALACR05r@-15];
The model ran fine for two and three class models. But the program gave me a error for the four class model. The warningg is;
ONE OR MORE PARAMETERS WERE FIXED TO AVOID SINGULARITY OF THE
INFORMATION MATRIX. THE SINGULARITY IS MOST LIKELY DUE TO THE
MODEL IS NOT IDENTIFIED, OR DUE TO A LARGE OR A SMALL PARAMETER
ON THE LOGIT SCALE. THE FOLLOWING PARAMETERS WERE FIXED:
7 4
Please help me understand why I have this kind of error and what I can do about it... |
|
|
|
|
| You need to send your input, data, output, and license number to support@statmodel.com. |
|
|
|
|
| Hi! I am looking at the three different ways of handling zeros in count data in GMM(ignoring them, using ZIP modeling, and specifying a zero class) and wondering when each would be the most appropriate. Thanks for the help! |
|
|
|
|
See the following Web Talk where this is discussed:
Regression with a count dependent variable: The negative binomial model. |
|
|
| Back to top |
