ABSTRACT
This chapter discusses estimation and hypothesis testing for two-level hierarchical linear models (2LHLM). A more general approach is to formulate growth as a 2LHLM where population parameters, individual effects and within-subject variation are defined at the level-1 of model development, and between-subject variation is modeled at the level-2. Estimation of model parameters using the expectation-maximization algorithm has been improved upon by Longford which incorporates Fisher scoring. P. J. Everson and C. N. Morris develop a procedure for making inferences about the parameters of the two-level normal hierarchical linear model (HLM) which outperforms restricted maximum likelihood estimator and Gibbs sampling procedures. The last HLM example considered is a standard split-plot repeated measures design with constant, linear, quadratic, and cubic polynomials fit to the within-subjects effect.
