On Thu, 8 Oct 1998, Chungui Qiao wrote:
> Hi there,
>
> I understand that if you want to fit a random spline row, you must fit a
> fixed linear row in the model as well. So what is the difference between the
> following two models:
yes thats right
We are considering changing this but at the moment spl() only fits the
random component of a natural cubic spline. That is if there are p knot
points (ie p rows) then spl() sets a random factor with p-2 levels. Thus
p must be greater than 3. If you are interested I can send you the final
version of the splines paper that we are presenting at Capetown in
December, as a read paper. This explains things quite well.
>
> Model (1) yield ~ lin(row) genotype !r spl(lin(row))
> Model (2) yield ~ lin(row) genotype !r spl(row)
>
> I found in some of the wheat variety trials I analyzed, fitting Model (1)
> caused a problem "Insufficient points for spl(lin(row))", while fitting
> Model (2) worked well. Is this due to the lack of degree of freedom for
> Model (1) and how is the degree of freedom estimated ? Could we just simply
> use Model (2) in this case?
I am not entirely sure why spl(lin(row)) presents problems. ASREML
should just examine the argument to spl() and form the matrix Z from the
set of unique values in the argument
hope this helps
brian cullis
> > Thanks for your
consideration. >
>
> Chungui
> Chungui Qiao
> School of Land and Food,
> Uni of QLD
> St Lucia 4072, QLD
> Ph: (07)33652859
> Fax:(07)33651177
>
>
...............................................................................
Brian Cullis Tel: 02 6938 1855
NSW Agriculture Fax: 02 6938 1809
Wagga Agricultural Institute email: brian.cullis@agric.nsw.gov.au
Pine Gully Rd
WAGGA WAGGA NSW 2650