Model Building and Interpretation of Ordinal Multilevel Random Effects Models with Exogeneity and Endogeneity

摘要

We focus on multilevel random effects models for ordered response such as occur in educational achievement research. In model development changes in parameter values are difficult to compare because of implicit rescaling of parameters in the linear predictor. We combine a heuristic method to handle this with proposals for instrumental variable estimation when regressors are endogenous. Simulated and real educational data are used to evaluate these proposals.