haplotypic frequency, this sum is directly estimated by counting from the unphased data. Thus, the HWE assumption is not required, and furthermore, the uncertainty in estimation of the composite frequencies is confined to just the sampling variation and does not involve an error related to the missing haplotype phase. This composite frequency (QAB = [PAB+PA/B]/2) also forms the basis for the LD inference by the composite disequilibrium approach (51). Instead of comparing the haplotype frequencies themselves, as dh = PAB (case) − PAB (control), the comparison in the CHM is based on the composite difference, dc = QAB (case) − QAB (control). Both differences are similarly expressed in terms of the population frequencies (fi) and the susceptibilities (gi) for the 10 possible di-locus genotypes (AB/AB, … , AB/ab, Ab/aB, … , ab/ab). Denote the population prevalence by g. Then the population frequency differences of interest are
