association is difficult to interpret at the level of individuals. Some care is necessary when testing the associations of rare variants. In this case, the frequency of some genotypes may be too small (<5) for the χ2 test to be valid and the Fisher exact test should be used. A limitation of these different tests for categorical traits is that they cannot include the effects of covariates. Logistic regression can be used as an alternative by modeling the odds for association in the logarithmic scale using regression. The regression equation can include the genetic effect but also covariates and can be extended to include multiple interacting SNPs as well as gene-environment interactions. Additionally, these common statistical methods rely on large sample approximation, and therefore care is needed when testing the association of rare variants.
