Re: ASREML problem
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: ASREML problem

> Arthur,
> I ran an analysis with ASREML on August 25 (version 20 Aug 1998). Now this job
> fitted  a number of random effects which were fitted using default settings.
> The job converged within the default of 10 iterations (actually took 7
> a few of the random effects were fixed at boundaries, but I thought that this
was not
> a problem.
> I now need the ___.sln file for this job, but I have accidentally deleted it.
> But when I rerun the job now (so I can obtain a __.sln file so that I can
retrieve sire
> solutions) the same run does not converge within the expected 10
iterations. I can set maxit at 11 and it will converge, but it gives me
different values for the variance
> components. The version of ASREML I am now running is dated 30 Sep 1998.
> Is my original analysis incorrect or is there a problem with the most recent

Dear Sue,
The change is entirely related to the fact that you have 3 variance components
which ASREML is not permitting to go zero (because you have not included a !GU
qualifier and the default is !GP).

If you compare the LogL values from the earlier run, I expect they will
not be as high at the end as from the most recent run.

The problem is first one of detecting when a parameter is pushing against
the boundary. and then fixing it so that other parameters are not biased.

I do not consider it safe to fix the parameter the first time the
update would make it negative because then it may want to bounce back
and would not be able.  However, after several updates would make it
negative, I fix it at the current small positive value.
Then other parameters are conditioned on the fixed value rather
than being updated assuming the value was negative.

As this behaviour might interest other users
I will send it to the list.

I will use you job to look again at what is happening just in case
further refinement is needed.


Arthur Gilmour PhD                    email:
Senior Research Scientist (Biometrics)                 fax: <61> 2 6391 3899
NSW Agriculture                                             <61> 2 6391 3922
Orange Agricultural Institute               telephone work: <61> 2 6391 3815
Forest Rd, ORANGE, 2800, AUSTRALIA                    home: <61> 2 6362 0046
ASREML is currently free by anonymous ftp from pub/aar on
    or point your web browser at

To join the asreml discussion list, send the message

The address for messages to the list is

                        <> <> <> <> <> <> <>
"Ye are not your own for ye are bought with a price: therefore glorify God
    in your body, and in your spirit, which are God's" 1 Corinthians 6:20