The Influence of Twin Pair Permutation on Likelihood-Based-Estimates of Genetic Variance That Require Ordering of Twin-Pairs

Abstract

Results from analyses of twin data that use models assuming a bivariate distribution of twin values will change when twins within pairs are reordered. We examined the effect of twin pair ordering on additive genetic variance estimates and hypothesis tests, from a bivariate normal model, both via simulation and through examination of real twin data. The simulations generated twin data for varying sample sizes and amounts of additive genetic and common environmental variance. The real data sets had sample sizes of 60 or less per zygosity. The results indicate that for moderate or large size studies, the effects of twin pair ordering are unlikely to greatly change the results of the data analysis; but for small studies the results can be sensitive to twin pair ordering. We therefore suggest that methods, not sensitive to within twin-pair differences be compared to the results obtained from twin-pair ordering. Methods not influenced by twin-pair ordering include least squares methods or covariance matrices approaches as described by Carey (2005, Behav Genet 35:667-670) or Guo and Wang (2002, Behav Genet 32:37-49).

On symmetric semiparametric two‐sample problem

“…This issue was discussed again by Carey (2005) and Byrne et al . (2007); the latter paper found that the maximum likelihood estimates of the additive genetic effect are sensitive to the within twin‐pair orderings especially for small sample sizes.…”

“…It was pointed out explicitly by Guo and Wang (2002) that a change in the within twin-pair orders may affect the estimation results substantially. This issue was discussed again by Carey (2005) and Byrne et al (2007); the latter paper found that the maximum likelihood estimates of the additive genetic effect are sensitive to the within twin-pair orderings especially for small sample sizes.…”