

Generalized linear mixed modeling (GL... 

Message/Author 

Oliver posted on Friday, April 22, 2016  8:15 am



Hi everyone, My understanding is that Mplus can handle dichotomous outcomes as part of longitudinal multilevel models/analyses. In SPSS and the broader stats literature, this is typically termed Generalized Linear Mixed Modeling (GLMM). I am wondering how this analysis is termed based on the Mplus terminology ? When I look at the Mplus user guide that includes all the syntaxes, it seems that analysis 9.13 (Twolevel growth model for a categorical outcome/threelevel analysis) is pretty similar to GLMM. Am I right ? In my study, a total of 100 participants each provided ratings of anxiety (continous; 010 scale) and reports about the presence/absence of a chest pain episode (dichotomous; yesno), once a day for 14 consecutive days. In terms of data structure, Level 1 units (i.e, daily reports of anxiety and chest pain) are thus nested within participants (Level 2 units). In terms of research question, I am simply interested in examining the Level 1 association between anxiety and chest pain. That is, whether anxiety is associated with an increased likelihood of having chest pain episodes. I will be primarily interested in the fixed effect, but of course will have a look at potential random effects. Considering the information above, should I use the Mplus syntax provided in example 9.13 of the user guide ? Thanks in advance for letting me know ! O. 

Jon Heron posted on Friday, April 22, 2016  8:25 am



Hi Oliver SEMstyle models work with data in "wide" format so repeated measures are multivariate with a single row per participant. As a consequence a standard "latent growth model" is only one level as opposed to the two levels you'd have with multilevel modelling. Ultimately your model is I think going to be similar to 6.13 but I would get there via examples 6.1 and 6.4 You might find it useful as a starting point to estimate a simple model in both disciplines  for instance a latent growth model for your anxiety measure would be readily replicable either in mlwin or Stata (xtmixed) and give nearidentical results. best, Jon 

Oliver posted on Wednesday, April 27, 2016  8:25 am



Hi Jon ! Thank you for your response. I need to clarify something with regards to the way datasets are organized in Mplus. My understanding is that it is possible to run a typical multilevel models using Mplus using the "vertical/stacked" format. The data do not necessarily need to be organized in the "wide/horizontal" format. For instance, I've prepared a "vertical" Mplus dataset in which I have two distinct continous Level1 variables (i.e., Pain and mood) nested within participants/IDs (i.e., Level 2). I'm interested in examining if pain (predictor) is associated with mood (outcome). Let's say that I'm only interested in the fixed effect. In SPSS, this analysis can be easily conducted with the MIXED procedure. When I looked at the Mplus user guide, it seems that my analysis could be conducted using the syntax of model 9.1 described on page 238 of the Mplus userguide, which is named: "Twolevel regression analysis for a continuous dependent variable with a random intercept". **See my second message for the next part of my post. 

Oliver posted on Wednesday, April 27, 2016  8:25 am



I've tried the syntax below provided in the Mplus guide, and it works well. The output/findings are quite similar to SPSS, which makes sense. As you can see, I've removed the "between" part of the model to focus on the withinperson effects. DATA: FILE = Datab2.dat; VARIABLE: NAMES = ID L1Pain L1Mood; WITHIN = L1Pain; CLUSTER = ID; DEFINE: CENTER L1Pain (GRANDMEAN); ANALYSIS: TYPE = TWOLEVEL; MODEL: %WITHIN% L1Mood ON L1Pain; The syntax makes sense, right ? That would indicate that it's possible to run a typical multilevel analysis using the "vertical/stacked" format.. Thanks in advance for letting me know whether I understand correctly. O. 


You can use wide or long format in Mplus. 

Back to top 

