Message/Author 

Yu Ye posted on Friday, February 25, 2005  2:55 pm



I am wondering if Zeroinflated Poisson regression can be modelled in multilevel structure. I can't find the example in Mplus manual (Chapter 9, version 3). I am working on a project to predict the number of prior emergency room visits in which patients are nested in ERs. In particular I would like to know if two random intercept and two slopes would be modelled for both the counts and probability as in the normal ZIF regression. Thanks a lot in advance. Yu 


No, this is not available in the current version of Mplus. Only continuous and categorical outcomes can be modeled with multilevel models. If you look up ESTIMATOR in the Mplus User's Guide, you will find a summary of which types of analyses are available for different outcome types and which estimators are available for these combinations. You may find this tabel useful. 


I am trying to do a similar analysis as Yu Ye describes in the beginning of this thread where a zeroinflated Poisson regression or zeroinflated negative binomial multilevel model seems appropriate. In my case, the dependent variable is a measure of drug use, and thus, it seems very similar to the twopart modeling being done in MPlus with longitudinal data since there are many zeros. However, my data is nested, crosssectional. Can MPlus do this type of analysis? Thank you! Linda S. 


Yes, 2level ZIP regression can be done in Mplus using ML. Both the intercept and the slopes can be random. 2level, twopart regression modeling is also available. 


Hi Linda and Bengt, I'm trying to fit a multilevel ZIP model with a random intercept. Everything seems sensible with the exception of no intercept variance estimate. I'm not sure what I've missed in the syntax. A simplified specification, starting with the variable statement, follows. Thanks for your suggestions. Leslie VARIABLE: NAMES ARE some variables; USEVARIABLES ARE some variables ; COUNT IS bully (i); MISSING IS .; WITHIN with in variables; BETWEEN btwn vars ; CLUSTER schlid; BWEIGHT = schwgt; BWTSCALE = sample; WEIGHT = stuwgt; WTSCALE = cluster; STRATIFICATION = idstrate; SUBPOPULATION = (IDCNTRY EQ 51); ANALYSIS: TYPE = TWOLEVEL COMPLEX; MODEL: %WITHIN% bully ON ssupp1 pv1 age male slfest ageimm lang tearel1 s_pinv1 satch ses chimm fgnbor; bully#1 ON; %BETWEEN% bully on schsize supp urban dispol1 atch pinv mssupp1 mpv1 mage mmale mslfest mlang mses mtearel1 ms_pinv1 msatch ; bully#1 ON; 


Add bully; to the between part of the MODEL command. 


Hi Linda, Thanks for your response. Adding this variable to the between part results in the following: *** ERROR One or more betweenlevel variables have variation within a cluster for one or more clusters. Check your data and format statement. Between Cluster ID with variation in this variable Variable (only one cluster ID will be listed) This seems to make intuitive sense since the variable really does have within cluster variation given its role as a withingroup variable. Thanks for further insights. Leslie 


Don't put bully on the BETWEEN list. Add it to the between part of the MODEL command: %BETWEEN% bully on schsize supp urban dispol1 atch pinv mssupp1 mpv1 mage mmale mslfest mlang mses mtearel1 ms_pinv1 msatch ; bully#1 ON; bully; 


Thanks  that's clear. 


Hi, I am answering the reviewers comments related to my paper. I did structure equation modeling using Mplus. The reviewer asking about zero inflated poission regression and also suggest do to multilevel analysis. I did not know how to do it. I also read your instruction and try to to it in mplus, but I failed. I am looking for tutor who can help me. I am from Vietnam. I hope you can suggest me anybody who are very good at these things. I am looking forward to communicate through skype in English (only) and through email. I hope you can introduce me one person and also let me know how much do I have to pay. Many thanks and regards, Diep. 


You may want to ask on SEMNET. 


Hi, I'm running a zero inflated poisson regression and I'm testing an interaction between 2 continuous variables. The interaction shows to be significant improving significantly the fit. Now, what's the best way to followup this interaction on Mplus? Thanks a lot in advance 


You can plot interactions and get confidence intervals as shown for UG ex 3.18. See also our Mediation page. Your DV would be different but the principle is the same. 

Back to top 