Hello Everyone,
I am trying to fit a Multilevel Item Response Model (Rasch for now) with random effects using gsem. I know GLLAMM will let me do it too, but it takes over 24 hours for the model to fit, and I honestly am tired of fighting my office computer to keep it from shutting down while the model is fitting. The issue I am having is that I get the model fit with no problem.
Where I am running into issues is obtaining post-model estimates to help me interpret the additional predictors (since it is a logit model). I would like to obtain the exp(b) or even the odds ratio. Anything that is accepted within my area of study.
The example I am using is from Sulis and Toland's (2017) Journal of Early Adolescence because I cannot post my study data (it is protected).
The code I use to fit the model:
gsem (MATHSTU MATHCLASS[class]->(m1-m25)@1) (gender-> MATHSTU, ) (avg_pedc-> MATHSTU, ), logit
To pull out the latent constructs:
predict sf_rasch cf_racsch, latent(MATHSTU MATHCLASS[class])
Normally with a basic model I might use something like predict ___, pr (e.g., predict ps_rach, pr) to get the exp(b) but I have 3 levels (class, students, questions) and it does not seem right that I cannot pull out the exp(b) for each level. Honestly, I might be barking up the wrong tree on this one as well. I understand the model in theory but applied not so much.
Any help would be greatly appreciated.
I am trying to fit a Multilevel Item Response Model (Rasch for now) with random effects using gsem. I know GLLAMM will let me do it too, but it takes over 24 hours for the model to fit, and I honestly am tired of fighting my office computer to keep it from shutting down while the model is fitting. The issue I am having is that I get the model fit with no problem.
Where I am running into issues is obtaining post-model estimates to help me interpret the additional predictors (since it is a logit model). I would like to obtain the exp(b) or even the odds ratio. Anything that is accepted within my area of study.
The example I am using is from Sulis and Toland's (2017) Journal of Early Adolescence because I cannot post my study data (it is protected).
The code I use to fit the model:
gsem (MATHSTU MATHCLASS[class]->(m1-m25)@1) (gender-> MATHSTU, ) (avg_pedc-> MATHSTU, ), logit
To pull out the latent constructs:
predict sf_rasch cf_racsch, latent(MATHSTU MATHCLASS[class])
Normally with a basic model I might use something like predict ___, pr (e.g., predict ps_rach, pr) to get the exp(b) but I have 3 levels (class, students, questions) and it does not seem right that I cannot pull out the exp(b) for each level. Honestly, I might be barking up the wrong tree on this one as well. I understand the model in theory but applied not so much.
Any help would be greatly appreciated.
Comment