One consideration in the analyses of GWAS data is which statistical test or model to use for the initial genome-wide scan. Since we expect to detect SNPs that are affected to a greater or lesser extent by distortion, a sensible default choice is the trend test, which is well-powered for multiplicative effects. It also has the benefit of being more robust to genotyping error than, for example, the general two degree of freedom test [Ahn et al., 2007]. We note that others have also made similar recommendations [Cantor et al., 2010; Iles, 2008]. Nevertheless, the trend test can be usefully complemented by the general test [Wellcome Trust Case Control Consortium, 2007], or other approaches for investigating nonmultiplicative models, such as the deviation test. The corresponding advice for Bayesian analyses is to place most of the prior weight on multiplicative models, and spread the rest out more widely [Stephens and Balding, 2009].
