A comparison of generalized linear mixed model procedures with estimating equations for variance and covariance parameter estimation in longitudinal studies and group randomized trials

Response data in longitudinal studies and group randomized trials are gathe red on units that belong to clusters, within which data are usually positiv ely correlated. Therefore, estimates and confidence intervals for intraclas s correlation or variance components are helpful when designing a longitudi nal study or group randomized trial. Data simulated from both study designs are used to investigate the estimation of variance and covariance paramete rs from the following procedures: for continuous outcomes, restricted maxim um likelihood (REML) and estimating equations (EE); for binary outcomes, re stricted pseudo-likelihood (REPL) and estimating equations (EE). We evaluat e these procedures to see which provide valid and precise estimates as well as correct standard errors for the intraclass correlation coefficient or v ariance components. REML seems the better choice for estimating terms relat ed to correlation for models with normal outcomes, especially in group rand omized trial situations. Results for REML and EE are mixed when outcomes ar e continuous and non-normal. With binary outcomes neither REPL nor EE provi des satisfactory estimation or inference in longitudinal study situations, while REPL is preferable for group randomized trials. Copyright (C) 2001 Jo hn Wiley & Sons, Ltd.