Likelihood-based analyses of longitudinal twin and family data: experiences with pedigree-based approaches

Substantial progress has been made recently in analyses of longitudinal twin and family data, principally for two reasons. The first is the continuing development of more refined models for describing longitudinal data; the second is the widespread availability of analytic methods (e.g. LISREL) with which to implement these models. Computational restrictions have limited likelihood-based analyses of longitudinal genetic data to analyses of covariance matrices or mean squares; however, advances in computer technology now make it feasible to conduct likelihood analyses of longitudinal pedigree data. We consider potential advantages of using pedigree-based methods. Our initial experiences with the application of these methods to simulated twin data, using the FISHER (Lange, K.L., et al., Genet. Epidemiol. 5:471, 1988) quantitative genetics package, are discussed, with particular attention to practical details such as running times on several computers. Preliminary results of the pedigree-based analyses, including robust estimation methods, convincingly demonstrate the failure of methods assuming the multivariate normal distribution for simulated twin data provided by Carey (this issue).