Figure 1 shows how the effective additive parameter for a few models varies depending on the allele frequency. Here we have assumed an equal number of cases and controls in the sample; varying this ratio gives qualitatively similar results and is therefore a less important factor than the allele frequency (data not shown). One way to understand the results is think of them as similar to a weighted average of the disease risks at each genotype. When the allele frequency is at one extreme, only two of the three possible genotypes will be represented in the sample, and the model fit will be based mainly on the difference in risk between these two. Thus, for both the dominant and recessive models the limiting values are either zero effect, when the two equal-risk genotypes predominate, or a large effect, when the two genotypes differ in risk. In the later case, the effect is double (on the log scale) that of the multiplicative model which has the same homozygous RR as the original dominant/recessive model.
