For weighted analyses, whether (DH)GL(M)M or not, the base assumption is
that the weights are the inverse of the true variance, and therefore do not
require a scaling (heterogenity/dispersion) factor . In line with ASReml 4.3,
a scaling factor can be requested now via the RESIDUAL statement.
(It was previously set using the !DISPERSION qualifier.) So,
RESIDUAL id(units) #implies the dispersion factor is 1.
RESIDUAL idv(units) #requests the dispersion factor be estimated from the
#weighted residual variance of the (working) variable.
RESIDUAL idv(units !GF) #sets the dispersion factor at .
Traditionally, in a GLM model, is based on the deviance heterogeneity
which is reported by Echidna, but for non normal data, is not the same as
(though often close to) the value ‘weighted residual variance’.
A bug in calculating the ‘weighted residual variance’ has been fixed.
A bug parsing us(2).nrm(Lamb) has been fixed.
Remember, if something does not work as you expect, do not hesitate
to seek help (support@EchidnaMMS.org). I am sure there are bugs yet
to be discovered.