ACDE model and estimability Why can’t we estimate (co)variances due to A, C, D and E simultaneously in a standard twin design?
Covariances: MZ cov(y i1,y i2 |MZ) = cov(MZ) = A 2 + D 2 + C 2
Covariance: DZ cov(y i1,y i2 |DZ) = cov(DZ) =½ A 2 + ¼ D 2 + C 2
Functions of covariances 2cov(DZ) – cov(MZ) = C 2 - ½ D 2 2(cov(MZ) – cov(DZ))= A / 2 D 2
Linear model y ij = +b i +w ij y 2 = b 2 + w 2 y, b and w are random variables t = b 2 / y 2 –intra-class correlation = fraction of total variance that is attributable to differences among pairs
Data: “Sufficient statistics” (= Sums of Squares / Mean Squares) MZ –variation between pairs (= covariance) –variation within pairs (= residual) DZ –variation between pairs (covariance) –variation within pairs (residual) 4 summary statistics, so why can’t we estimate all four underlying components?
Causal components Between pairsWithin pairs A 2 + D 2 + C 2 E 2 DZ½ A 2 + ¼ D 2 + C 2 ½ A / 4 D 2 + E 2 Difference ½ A / 4 D 2 ½ A / 4 D 2 Different combinations of values of A 2 and D 2 will give the same observed difference in between and within MZ and DZ (co)variance: confounding (dependency), can only estimate 3 components
In terms of (co)variances “Observed”Expected MZ var A 2 + D 2 + C 2 + E 2 MZ cov A 2 + D 2 + C 2 DZ var A 2 + D 2 + C 2 + E 2 DZ cov½ A 2 + ¼ D 2 + C 2 MZ & DZ variance have the same expectation. Left with two equations and three unknowns
Assumption D 2 = 0 : the ACE model Between pairsWithin pairs A 2 + C 2 E 2 DZ½ A 2 + C 2 ½ A 2 + E 2 4 Mean Squares, 3 unknowns –Maximum likelihood estimation (e.g., Mx)