There is absolutely no reason to use the following for a number of reasons:
> The text book formula for h^2 is hard to apply for more general
> situations. I prefer to calculate h^2 from prediction error variances
> (pev) of BLUPS.
> pev = sigma^2_g * ( 1 - h^2)
Note that if you rearrange the this formula is becomes:
C=sigma^2_g*(1-h^2)/sigma^2_e = h^2 (if sigma^2_g+sigma^2_e=sigma^2_p)
(as pev=C sigma^2_e, where C is the diagonal element of the inverse of the coefficient matrix for a breeding value).
Henderson (1975 Biometrics) showed that PEV=var(a-estimated a) = sigma^2_a * (1 - reliability) under the animal model and known variance components (or ratios). None of this applies when estimating variance components for a variety of reasons. Not to mention reliability is not always heritability.
In order to obtain genetic parameters you need to to fit genetic components in the model as in White et al. (1999 JDS) - I don't know of any others as yet but interested in them if there are.
The definition of heritability remains the same regardless of the model. However, how it is interpreted is an other story - probably the proportion of genetic variance associated with the knots (which may relate to the measurement days or not).
Fred Dagg on the Socratic Paradox:
"The argument says in essence that you can't learn things you don't
already know, and given the widely accepted view that there's some
difficulty to be encountered in trying to learn something you do
already know, I'm afraid it's beginning to look as if the whole
business of learning is largely overated and should probably be left alone"
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml