However, it is not always straightforward to estimate variance components. A variance component is only meaningful when measures are expressed on a scale of at least interval level. Moreover, many statistical methods require the phenotype to be normally distributed. Many phenotypes are not expressed in clearly defined units and are at best ordinal in character (e.g., conservatism, extraversion). Some traits have even only a nominal character (e.g., psychiatric disorders). There are several ways of dealing with such nominal data. One possibility is to focus on concordance rates and compute recurrence risk ratios (Risch 1990, 2001). Alternatively, one might assume a latent continuous trait with a threshold above which individuals are affected and estimate the heritability on that latent trait (Lynch and Walsh 1998; Falconer 1965; Crittenden 1961). This method can also be used with ordinal data.
