The derived expressions involve which is the coefficient of determination on the observed scale. Lee et al have argued that, for a genetic predictor, R 2 on the liability scale is more interpretable for binary traits as it is invariant to the population prevalence and sampling ratio [25]. An approximate transformation to the liability scale is obtained by transforming the genetic effects using equation 13 and rescaling the trait variance from the binomial variance on the observed scale to the unit variance on the liability scale. Therefore,
