9 Conclusions: Genetic Analyses of BMI Data

The analyses of Australian BMI data which we have presented indicate a significant and substantial contribution of genetic factors to variation in BMI, consistent with other twin studies referred to at the beginning of Section 6.2. In the young cohort like-sex pairs, we find significant evidence for genetic dominance (or other genetic nonadditivity), in addition to additive genetic effects, but in the older cohort non-additive genetic effects are non-significant. Further analyses are needed to determine whether genetic and environmental parameters are significantly different across cohorts, or indeed between males and females (see Chapter 9).

We have discovered unexpected mean differences between zygosity groups (in the young cohort), and between twins whose cotwin refused to participate in the 1981 survey, and twins from concordant-participant pairs. It is possible that these differences reflect only outlier effects caused by a handful of observations. In this case, if we recode BMI as an ordinal variable, we might expect to find no significant differences in the proportions of twins falling into each category. Alternatively, it is possible that there is an overall shift in the distribution of BMI, in which case we must be concerned about the undersampling of obese individuals. If the latter finding were confirmed, further work would be needed to explore the degree to which genetic and environmental parameters might be biased (cf. Lykken et al., 1987; Neale et al., 1989a; Neale and Eaves, 1992).