Although CHM provides a robust test when haplotypic association is present in a region, we are also concerned with an unbiased estimation of effects for particular haplotypes. Shibata et al. (54) suggested a parametric method that provides estimates of haplotype effects as well as haplotype-specific P-values based on a likelihood ratio test. A particular model (diplotypic, additive, recessive, dominant or overdominant) needs to be specified prior to the analysis. A haplotype would be flagged as an ‘associated’ if the significance persists across all methods. When the same haplotype is tested by the different methods, there is no need for a strict correction for multiple testing, because the same hypothesis is being tested. In this case, a conservative overall P-value for a particular haplotype is given by the maximum of the P-values (diplotypic, dominant, additive and CHM). In fact, this gives the upper bound for Simes’s adjustment/combination test (75), which is applicable under positive dependence among P-values, such as expected here. Haplotype frequencies were estimated using a program described by Shibata et al. (54,55), which implements the estimation via the EM algorithm.
