An improved multipoint sib-pair analysis of quantitative traits.

Kruglyak and Lander (1995) recently published a multipoint sib-pair procedure based on the expected distribution of zero, one and two marker alleles shared identical by descent (ibd) and the method of maximum-likelihood (ML). Their approach uses phenotypic sib-pair differences, which ignores the bivariate structure of sib-pair data. Their simulations suggested that their method was more powerful than the regression method of Haseman and Elston (1972). We show through computation and simulation that their approach can be made more powerful still if the bivariate nature of sib-pair data is acknowledged. In addition, methods based on the average number of shared alleles that also employ bivariate ML procedures (Nance and Neale, 1989; Xu and Atchley, 1995) are more powerful than the approach they recommend and very similar to true ML using the distribution of ibd. The simple ML approach using the average number of shared alleles that we recommend seems to offer both optimal power and flexibility.