

Clustered Data with a Count Outcome 

Message/Author 


Dear Drs. Muthen, I may be confronted with clustered data that are count data, and so need the analysis to handle clustering and to treat the count variable appropriately, for example, with Poisson model of the dependent variable. This would fall under the generalized linear model for clustered data. I would greatly appreciate it if you could tell me whether Mplus can handle clustered data with a count outcome, using an appropriate count model like Poisson. Thank you so much in advanced. Sincerely, ellen 


Yes, this can be handled in Mplus as either Type=Complex or Type=Twolevel. In the latter case, Level 2 considers a random intercept for the count outcome. 

Dave posted on Monday, June 23, 2014  4:55 pm



Dear Drs. Muthen, I have clustered data that are proportion data. Thus I need an analysis to handle clustering and the proportion variable appropriately. From earlier posts about offsets (e.g. http://www.statmodel.com/discussion/messages/23/781.html) I note that I can convert the proportion to a count variable and use an offset. From this post http://www.statmodel.com/discussion/messages/12/13569.html?1375213296 I note that Mplus can handle multilevel count data. I am wondering: 1  Is it possible to use an offset in a multilevel model in Mplus? 2  If the answer to 1 is yes, can you point me to any sample syntax? 3  Why would a modeler choose to use an offset as opposed to a data transformation (e.g., arcsine) to handle a proportion dv? Thank you in advance for any insight you have on these questions. Dave 


12. Just use the offset in the level1 (Within) part of the model. You still have a random intercept that is modeled on level2 (Between). 3. I am not familiar with the literature on this, but as a general rule it seems a good idea to build your model on the components of your data rather than summaries. 

Back to top 

