|
|
Poisson longitudinal multilevel modeling |
|
Message/Author |
|
|
Hi, I am trying to run a Poisson longitudinal multilevel model using Mplus. I know how to do it in HLM, and I am trying to compare both programs to make sure that I did everything correctly. However, I am running into some issues: 1. HLM gives unit specific estimates and population average estimates. However, Mplus only provides one and based on a random intercept model I am assuming it is unit-specific estimates, is this correct? 2. When I introduce random slope on time, I get quite a different variance terms and the fixed effect for the intercept is vastly different from HLM. I understand that the estimators can play a role in the difference. I ran a full PQL in HLM and ML in Mplus. Had all the estimates been approximately close to one on another I would proceed with Mplus and used MLR since I want to use FIML for my missing data. Here is the code for the random slope on time model for reference: missing = all(-9999); cluster is CASEID_1979; count is ACT_WMIN (p); within is t; usevariables are t ACT_WMIN; Analysis: estimator is ML; type is twolevel; !random model: %within% s | ACT_WMIN on t; %between% ACT_WMIN; s with ACT_WMIN; |
|
|
It sounds like HLM does not do ML. First, make sure your model has the same number of parameters in both software so they are comparable (also check sample size). I don't know what you mean by unit-specific estimates. The model you specified has parameters which are estimated. It also has random effects (intercept and slope) which can be estimated for each subject using FSCORES. |
|
Back to top |
|
|