not the same for MZ and DZ twin families). In RFGLS, rather than estimate the residual variance-covariance matrix for each family type in each SNP regression, we estimate it only once for each family type based on the model with covariates but no SNP effect. The residual variance-covariance matrix in the analysis of individual SNP effects is then fixed at the estimates from the model without any SNP effects. As we show in Li et al. (2011), because any SNP effect is likely to be very small, fixing the variance-covariance matrix in this way has no effect on the test statistics but greatly increases the speed of the analysis.
