3. Transmission Disequilibrium Test Parameters

The transmission disequilibrium test [Spielman & Ewens 1996] can be performed using module paptctdt specifying a marker as the trait to be analyzed. The frequency module selected from the papfq roster must assume Hardy-Weinberg equilibrium and compute genotype frequencies from allele frequencies. This module requires category-specific autosomal GA, which in turn requires that a dichotomous OV for the disease of interest has been assigned to the Category Status node. Parameters of the model include transmission probabilities to both affected and unaffected offspring; the transmission to affected offspring may be estimated while fixing at 0.5 the transmission to unaffected offspring. The transmission of any number of marker alleles may be estimated simultaneously up to one fewer than the total number of alleles. The transmission parameters for different alleles may be equated, thereby reducing the number of degrees of freedom, either by setting an equality constraint in the parameter table so that they maximize together, or leaving multiple alleles unmaximized. The results obtained when equating alleles may differ slightly from the results obtained upon reassigning alleles due to different genotype inference for ungenotyped individuals. The transmission to unaffected offspring can also be estimated to test for distorted segregation of the marker; however, note that all unstudied pedigree member are classified in the unaffected category along with all truly unaffected pedigree members.

The transmission parameter and chi-square statistic may be overestimated when unstudied parents have few offspring. Parental genotype inference assumes the values of the transmission parameters and can differ substantially depending on the values. One way to test whether this is occurring is to set the transmission parameters to 0.5, is to simulate marker genotypes for individuals with missing values, then maximize over the transmission parameters. Using this approach, parental genotypes are inferred assuming transmission of 0.5 rather than the value used in the likelihood calculation.