'REML for multilevel model comparison/evaluation
With the very same data, I want to compare the two multilevel models. (these two models differs in both fixed and random effects)
so, I am planning simulation study. I will generate the data set, and fit this two models and compare this two model. For example, I have two proposed model and want to compare these
Model1:
$$y_{ij} = \gamma_{00} +\gamma_{01}mathscore_{j} + \gamma_{10}SEX_{ij} +
+ u_{0j}+e_{ij}$$
Model2:
$$y_{ij} = \gamma_{00} +\gamma_{01}mathscore_{j} + \gamma_{10}SEX_{ij} + \gamma_{30}SEX*mathscore_{ij}
+ u_{0j}+e_{ij}$$
the different thing between this model is whether it contains interaction term or not
it seems like when the fixed effect differs, researchers should not use REML.
However, it is using same data. (simulate the data and fit the both model)
in that case is REML still forbidden?
if I am interested in the each parameter estimates, not overall model fit, is REML allowed to use in this case?
Sources
This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.
Source: Stack Overflow
| Solution | Source |
|---|