Changes to the model line
Introduction
half
is used in the linear model to include a column with a coefficient of 0.5.
It is primarily used to double the variance component of a model term as in
half.sire and half.dam.
Weighted analysis
!WT
w is a qualifier to the response variable.
Its
interpretation and specification have changed to rationalize the specification of
scaling factors in
weights
and
dispersion factors in GLMM.
Enhancements on Factor Analytic Models
ASReml 4.3 has the structures fak(), facvk(), xfak(), rrk(), and rrdk(), which are different parameterizations of the factor analytic model in
which Σ is modelled as Σ=ΓΓT + Ψ where Γ(ω×k) is a matrix of loadings on the covariance scale and Ψ is a diagonal vector of specific variances.The stru
It is not unusual for users to have trouble comprehending and fitting extended factor analytic models, especially with more than two factors. Two relevant structures are presented below.
rrk(rr for reduced rank) This is formally just an alternate specifcation of xfak() but with the specific variances (Ψ) all set to zero, which means that if you supply initial values, you need to supply initial values for the specific variances as zero.
rrdk (rr for reduced rank and d for diagonal) This is equivalent to xfak(), but fits it as two independent model terms: an rrk() term and a diag() term. This form of the mixed model equations runs faster with large relationship matrices. For example
rrk(trial).grm(entry) + diag(trial).grm(entry)
will run faster than
xfak(trial).grm(entry)
The structure rrdk() is new to ASReml 4.3. One advantage of the rrdk() structure is that it runs faster than xfak() when interacting with a GRM matrix. We also denote rrdk() as a RRD (random regression + diagonal) model.
Back to What's New
Return to index