A linkage test in complex disease is well motivated under either of two conditions: if a prior segregation analysis has given evidence for a major gene, or if there is a candidate locus that may influence liability. Although methods have been developed for pairs of affected relatives without specifying the parameters of segregation analysis, they extract only a small part of the linkage information (especially for markers that are not highly polymorphic) and give no valid estimate of map distance. Therefore most linkage analysis of complex disease is (and should be) based on prior segregation analysis, whose possible errors must be considered.
A major gene has a megaphenic effect (i.e., the displacement on the liability scale is large relative to the phenotypic standard deviation; therefore penetrance is high), but the frequency is often so low that the variance on the liability scale may be small. Microphenic effects are not resolvable by segregation analysis, but may be detected through linkage or association with a candidate locus. In segregation analysis a major gene may be simulated by inadequate allowance for ascertainment bias, by neglect of either parameter of multifactorial inheritance in children and sibs, by inappropriate definition of a liability indicator, by underestimation of affection risk in liability classes, by failure of the normality assumption for a quantitative trait, or by incorrect modeling of the relation between affection and a quantitative trait. Besides the above, misrepresentation of the genotype-phenotype relation or failure to include mutation may distort genetic parameters. In rejecting the null hypothesis there is always a logical disjunction: either a type I error, or incorrect specification of the multifactorial model, or a major gene. Tests of goodness of fit among sampling frames and even tests on transmission probabilities can detect some errors, but their power is never so high that they can be