Some empirical properties of an all-relative-pairs linkage test

Olson and Wijsman [Genet Epidemiol 10:87-102, 1993] recently proposed a single test of linkage which combines information from different types of relative pairs in a pedigree. Relative-pair-type-specific regression parameters that relate the squared pair trait difference to the estimated number of marker genes shared identical by descent between the pair are estimated using generalized estimating equation methodology, then combined to give a single linkage test statistic. Questions remain concerning the small sample and robustness properties of this test statistic; these questions are addressed in the present paper using simulation. The test is substantially anti-conservative for samples with fewer than about ten families and are approximately valid for samples larger than about 15 families. In addition, the test appears robust in the presence of trait genotype by environment interaction, trait family-specific errors, a second major trait locus, and trait dominance. Surprisingly, the sibpair test was more powerful than the all-relative-pairs test for dominant traits with high heritability. Finally, adjusting for the presence of a marker known to be linked to one trait locus only marginally improves the power for detecting a second trait locus.