Statistical interaction is dependent on the scale of measurement and may not equate to biological interaction [Rothman et al., 1980] so that G × E can exist on the dichotomous disease scale even when none is present on the latent scale of disease risk [Eaves, 2006]. Item response theory (IRT) in combination with Markov Chain Monte Carlo (MCMC) estimation provides a flexible and efficient framework for modeling the underlying continuous liability to disease for behavioral phenotypes based on responses to multiple binary items in an interview framework [Reise and Waller, 2009]. Hence, IRT models the underlying scale, thereby controlling for interactions that are artefacts of scale [Kang and Waller, 2005], adding understanding to the interactions detected on the disease scale. In an earlier article, we used IRT analyses of 824 monozygotic (MZ) twin pairs to investigate the relationship between SLE and 5HTTLPR under the hypothesis that within pair variance would be greater for MZ twins homozygous for the S allele compared to those homozygous for the L allele if the interaction were real [Wray et al., 2008]. This novel design
