Abstract: The two-stage random-effects model (Harville (1977), Laird and Ware (19- 82)) offers a powerful and flexible tool for the analysis of longitudinal data. This method assumes individual response may be modeled as the sum of an overall population effect, a random individual deviation, and random error. When individuals are nested within fam ilies or companies, this source of variability should also be considered. Building a mo del for hearing loss for minimally noise-exposed workers with data compiled from multip le sources motivated extending the two-stage model to include a nested random effe ct. Computational methods to compute population effects, random effects, and varian ce components in this more general setting using the EM algorithm are given.
Key words and phrases: Random effects models, EM algorithm, restricted maximum likelihood, mixed models, hierarchical models.