In contrast, both Purcell and Koenen (2005) and Turkheimer et al. (2005) explained that it is impossible to test for environmental mediation using a twin design when the environmental variable in question is obligatory-shared, or necessarily the same for twins reared together (e.g. socioeconomic status). Also, Purcell and Koenen’s (2005) simulation results showed that adding a measured ‘environmental’ variable to a DeFries-Fulker regression model or a univariate variance components model leads to type I errors at an unacceptably high rate when gene-environment correlation is present. Therefore, the conclusions that the associations between physical maltreatment and antisocial behavior (Jaffee et al. 2004b) and domestic violence and children’s problems (Jaffee et al. 2002) are environmentally mediated may not be warranted.
