Multiple regression analysis of twin data.

A multiple regression model for the analysis o f twin data is described in which a cotwin's score is predicted from a proband's score and the coefficient o f relationship (R = 1.0 and 0.5 for identical and fraternal twin pairs, respectively). This model is especially appropriate for the analysis o f data on twins in which one member o f each pair has been selected because o f a deviant score, e.g., low reading performance. When the model is fitted to such data, the partial regression o f the cotwin's score on the coefficient o f relationship provides a powerful test o f the extent to which the difference between the mean for probands and that for the unselected population is heritable, i.e., a test for genetic etiology. By fitting an augmented model containing an interaction term to either selected or unselected data sets, direct estimates o f heritability and the proportion o f variance due to shared environmental influences can also be obtained (subject, o f course, to the usual assumptions underlying twin analyses, e.g., a linear polygenic model, little or n~o assortative mating, and equal shared environmental influences for identical and fraternal twins).