A
demonstration of the Additivity of
Variance
In the first block, two genetic loci A and B contribute to a combined
trait A+B. The
contributions of the two loci are (1) approximately equal
(means are similar) and (2) independent (A & B are not correlated).
Then, the phenotypic value
of the combined trait is simply the sum of the two
contributing loci, and the variance of (A+B) is the sum of the
variances of A & B separately. That is,
variance is additive.
In the second block,
variance is
not sensitive to moderate differences in the relative contribution of loci. C contributes
only 25% of the value
of the combined
trait. Means and variances remain essentially additive
(expected 1048 vs observed 1056). This is in part
because locus A with a greater mean than locus B
contributes more to the variance.
In the third block,
variance is much more sensitive to non-independence
of the contribution of loci. D simply adds 10
units to A: the variances of A and D are therefore identical.
The variance of the combined trait is much greater than
the sum of the contributing variances, because the two
contributing loci are perfectly correlated.
In the fourth block,
trait E has equal contributions from the transformed
values of A & B in C & D
in the second and third blocks. Because in D the A+10
contribution is still perfectly correlated with A, the
variance of E is substantially greater
than that of A + B
Additivity of variance is a
crucial assumption of many biological experiments and
analyses, especially Analysis
of Variance (ANOVA)
and estimation of Heritability.
Non-additivity contributes to interaction variance,
where a combined trait value cannot be predicted from
measurements of the component traits.