Zero-Truncated Poisson PreviousNext
Mplus Discussion > Multilevel Data/Complex Sample >
 Scott Baldwin posted on Wednesday, February 22, 2012 - 3:48 pm
I am interested in a fitting a model with a count outcome using a zero-truncated poisson model.

The following discussion board postings seem to indicate that zero-truncated poisson is available:

- (see the Oct 16, 2011 posting by Bengt Muthen)
- (see the Feb 17, 2011 posting by Bengt Muthen as well as the July 23, 2011)

However, I've looked through the user manual and it doesn't seem to be an option. The truncated negative-binomial seems to the only option available for a zero-truncated count model. Is that correct? Have I overlooked something?

I'm using Mplus 6.12. Thanks for your help.

 Linda K. Muthen posted on Wednesday, February 22, 2012 - 4:35 pm
I agree that the only zero-truncated model is the zero-truncated negative binomial model.
 Bengt O. Muthen posted on Wednesday, February 22, 2012 - 4:49 pm
You can fix the alpha dispersion parameter of negbin to a small value to approximate Poisson. Like 0.01.
 Daniel Lee posted on Thursday, March 12, 2020 - 7:48 am
Hi Dr. Muthen,

I ran a zero-truncated negative binomial model. I am trying to get incident rate ratios (IRRs). Is it OK to run the following model constraint command to obtain these rates (by exponentiating the coefficients):

model constraint:
new(b1_exp b2_exp b3_exp);
b1_exp = exp(a1);
b2_exp = exp(a2);
b3_exp = exp(a3);
 Bengt O. Muthen posted on Thursday, March 12, 2020 - 2:37 pm
Yes, that's how you do it.
 Daniel Lee posted on Friday, March 13, 2020 - 10:59 am
I own a copy of the regression and mediation analysis using Mplus book. While there is some mention of the zero-truncated negative binomial model, it is within the context of a two-regression hurdle model.

Does the analysis become a zero-truncated negative binomial regression, if in table 6.5 (page 270), I remove the following code from Model command (while keeping "naffairs ON kids-yrsmarr6"):

naffairs#1 ON kids-yrsmarr6;
 Bengt O. Muthen posted on Saturday, March 14, 2020 - 2:38 pm
No, you should instead look at Table 6.7 in our book and the text on page 272 that goes with it.
 Daniel Lee posted on Saturday, March 14, 2020 - 7:38 pm
Hi Dr. Muthen,

In page 272, a hurdle model was estimated that specifies one process for zero counts and another process for positive counts. In my data, the value zero cannot occur. Hence, I thought that the following syntax would be correct:

Usevariable: Names = y x1 x2;
Usevar = y x1 x2;
Count = y(NBH);

Model: y ON x1 x2;

Analysis: Estimator=MLR;

Relative to Table 6.7, I removed the following code (y#1 on x1 x2) because this specifies a logistic regression for the binary outcome being greater than zero.

Please let me know if I am interpreting/comprehending something incorrectly in the book. Again, I very much appreciate your help.

Thank you.
 Bengt O. Muthen posted on Sunday, March 15, 2020 - 4:37 pm
You should use the NBT option and say:

Usevariable: Names = y x1 x2;
Usevar = y x1 x2;
Count = y(NBT);

Model: y ON x1 x2;

Analysis: Estimator=MLR;
Back to top
Add Your Message Here
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Options: Enable HTML code in message
Automatically activate URLs in message