[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

*To*: Arthur Gilmour <gilmoua@ornsun.agric.nsw.gov.au>*Subject*: RE:Unreplicated analysis*From*: "B. Cullis" <cullisb@wagsun.agric.nsw.gov.au>*Date*: Wed, 15 Jul 1998 09:31:18 +1000 (EST)*cc*: s185152@student.uq.edu.au, asreml@ram.chiswick.anprod.csiro.au*In-Reply-To*: <199807132217.IAA27161@ornsun.agric.nsw.gov.au>*Sender*: owner-asreml@chiswick.anprod.csiro.au

In reply to the query re error variance estimation. I agree with Arthur. I have done thousands of simulations, see Cullis et al (1989), Cullis et al (1992) to prove that REML estimation is effectively unbiassed for estimation of genetic and error variance parameters in the unreplicated trial setting. There is no difference between these trials and replicated trials as far as estimation is concerned. I dont fully agree with ARthur's model below with checks as fixed effects, though this is a minor point. On Tue, 14 Jul 1998, Arthur Gilmour wrote: > Dear Vince, > You wrote. > > Assume we have an unreplicated trial with repeated checks. Now, > > for the analysis of such a trial i believe that the estimate of 'error' > > comes (or should) come from an ANOVA on the check plots. However, I have a > > suspicion that the error estimate is from an ANOVA on the checks and > > genotypes. > > > Your suspicions are wrong. You can verify this from > doing the following series of analyses (without spatial > adjustement so you can see exactly what is happening). > > 1. yeild ~ Check !r geno > with all the data and Check having one level which > is assocoated with testlines, other levels assoc with > check varieties, geno being the test lines (coded zero > for check plots) > > 2. yeild ~ Check > dropping all the test-line data. OR > yeild ~ Check geno > treating test lines as fixed effects and having all the data. > > > These two analyses should give exactly the same residual variance > [unless the geno variance goes to zero boundary in which > case you will need to allow it to go negative !GU] to get > the same error variance. > > If replication is only in the Check lines and test lines are fitted, > the only information on the error variance comes from the check lines. > > Now, consider the case where all genotypes are treated as random. > Then things will not work out so neatly becasue the variance of > the check lines is not likely to be the same as the variance of the test > lines. If checks were equireplicated and had their own variance, > things would work out but otherwise, the residual would change slightly. yes thats fine Given that the linear model is 'correct' ie yield ~ mu ! varieties, where errors are modelled by ar1 x ar1, then there is NO ISSUE with bias! If the model is incorrect then there are problems. > > When we go to a spatial analysis, again, all the information on error > correlation comes from the check plots. This is why we recommend having a few > of the check plots additional to the regular grid spacing - so as to > get same lag 1 and lag 2 information as well as the lag5 information. > > > Date: Mon, 13 Jul 1998 17:12:10 +1000 (GMT+1000) > > From: Vincenzo Matassa <s185152@student.uq.edu.au> > > To: asreml@ram.chiswick.anprod.csiro.au > > Subject: RE:Unreplicated analysis > > MIME-Version: 1.0 > > > > Dear All > > Assume we have an unreplicated trial with repeated checks. Now, > > for the analysis of such a trial i believe that the estimate of 'error' > > comes (or should) come from an ANOVA on the check plots. However, I have a > > suspicion that the error estimate is from an ANOVA on the checks and > > genotypes. > > > > The problem with the latter approach is that typically in unreplicated > > trials check plots will usually have a smaller variance than the random > > genotypes. Hence if we run an ANOVA on checks and geno's in order to > > estimate the error component for our trial it may be poorly (over or > > underestimated) estimated . > > > > Is there a way in ASREML to just get the estimate of the error variance > > solely from an ANOVA on the checks when we fit a spatial model to our > > data Yield ~ mu c(checks) !r genotypes ? > > > > If you fit genotypes as fixed effects, then that will remove any > contribution they might make. > > > Many thanks for your help. > > > > Kind regards > > > > Vince > > > > Vince Matassa > > Department Of Agriculture > > BIOMETRICS SECTION > > University Of Queensland > > Brisbane 4072 > > Australia > > > > > > > <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> > Arthur Gilmour PhD email: Arthur.Gilmour@agric.nsw.gov.au > 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 ftp.res.bbsrc.ac.uk > Point your web browser at ftp://ftp.res.bbsrc.ac.uk/pub/aar/ > in the IACR-Rothamsted information system http://www.res.bbsrc.ac.uk/ > > To join the asreml discussion list, send the message > subscribe > to asreml-request@chiswick.anprod.CSIRO.au > > The address for messages to the list is asreml@chiswick.anprod.CSIRO.au > > <> <> <> <> <> <> <> > "The men marveled, saying 'Who can this be, > that even the winds and the sea obey Him?'" Matthew 8:27 > <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> > > > > ............................................................................... 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

**References**:**RE:Unreplicated analysis***From:*Arthur Gilmour <gilmoua@ornsun.agric.nsw.gov.au>

- Prev by Date:
**orthogonal polynomials** - Next by Date:
**Unable to form R inverse component** - Prev by thread:
**RE:Unreplicated analysis** - Next by thread:
**orthogonal polynomials** - Index(es):