Ascertainment and Goodness of Fit of Variance
Component Models for Pedigree Data
Boehnke M, Lange K
Proggress in Clinical and Biological Research, 147:173-192 (1984)
Abstract
The multivariate normal parameterization of the polygenic model (Lange et al., 1976)
provides a great deal of flexibility for analyzing quantitative data on pedigrees. The
likelihood approach employed ensures statistical efficiency and allows for hypothesis
testing using the likelihood ratio criterion. The parameterization also facilitates
ascertainment correction and goodness-of-fit testing (Spence et al., 1977; Ott, 1979;
Hopper and Mathews, 1982; Boehnke, 1983). We reviewed these results and then
described a simulation study undertaken to determine their utility when applied to data.
Pedigree data were generated under polygenic and mixed models and sampled either
randomly or via probands. We found that the variance components of the model were
accurately estimated for random sampling, but less so for ascertained data analyzed by
conditioning on probands. Goodness-of-fit tests employing test statistics corresponding
to individual phenotypes and entire pedigrees were conservative, but pedigree tests did
demonstrate reasonable power to reject a variety of mixed model alternatives. In
addition, we found that the pedigree test statistics could be used to enrich a sample of
pedigrees for those pedigrees segregating at a major locus, providing an objective
criterion for choosing pedigrees to be included in a linkage analysis.
Variance component linkage analysis