The IRT models were similar to those used in Wray et al. [2008] but were adapted to accommodate dizygotic as well as MZ pairs. The data were entered as twin pairs to account for their interdependence. IRT models were analyzed in the WinBUGS (Bayesian Inference Using the Gibbs Sampler) program [Lunn et al., 2000]. After a burn-in phase of 4,000 iterations, the characterization of the posterior distribution for the model parameters was based on 8,000 iterations from two independent Markov chains. The independent variables were those used in the ordinal regressions.
