3 Fitting Genetic Models to Binary Data

It is very important to realize that binary or ordinal data do not preclude model-fitting. A large number of applications, from item analysis (e.g., Neale, et al, 1986; Kendler et al., 1987) to psychiatric or physical illness (e.g., Kendler et al, 1992b,c) do not have measures on a quantitative scale but are limited to discontinuous forms of assessment. In Chapter 2 we discussed how ordinal data from twins could be summarized as contingency tables from which polychoric correlations and their asymptotic variances could be computed. Fitting models to this type of summary statistic or directly to the contingency table data themselves involves a number of additional considerations, which we illustrate here with data on major depressive disorder. Although details of the sample and measures used have been provided in several published articles (Kendler et al., 1991a,b; 1992a), we briefly reiterate the methods to emphasize some of the practical issues involved with an interview study of twins.