Re: Boundary problems

I would interpret your result as when you specified keeping structures positive definite (!GP) the components tending to violate this will be kept at a boundary (B) preventing, for example, negative variances. Does the model suit estimation of everything specified?

Tore Ericsson
Skogforsk, Sävar

-----Original Message-----
From: ASReml users discussion group [mailto:ASREML-L_at_AGRIC.NSW.GOV.AU] On Behalf Of Tuomas Leinonen
Sent: Wednesday, April 02, 2008 4:14 PM
To: ASREML-L_at_AGRIC.NSW.GOV.AU
Subject: Boundary problems

Dear all,

I have a data set of shape variables from 84 half sib families produced by
crossing each sire with two dams, with each family raised in two replicate
blocks. I keep on getting problems with univariate (and hence of course
multivariate) analyses, especially with the variance components associated
with maternal environmental (dam) effects. ASREML gives a warning that the
dam variance component is fixed at a boundary. Does anyone have suggestions
on why this should happen and how could the problem be fixed? Is there a
problem with the model I am using or is it more likely to be due to the
data. The variables are continuous and normally distributed. I will attach
the model and a typical result file below.

Thank you in advance,

Tuomas Leinonen

I have been using the following model. (I added the MAXIT and EXTRA
qualifiers, because for some of the variables, the LogL did not converge
with less iterations.)

ASREML DAA in big cross

  id !P
  sire !A
  dam !A
  block !A
  size
  shape1
  shape2
  shape3
  shape4
  shape5
  shape6

Pedigree.ped !ALPHA
BodyData.asd !SKIP 1 !EXTRA 100 !MAXITER 700 !AISING !dopart 2

###########################################################

!part 1 # single trait model

pcsize ~ mu block ucsize !r id dam

#########################################################

!part 2 # Covariance between traits Va only

rw1 csize ~ Trait block !r Trait.id Trait.dam

1 2 2
0
Trait 0 US !GP
0.5484E-04
0.01 0.1266

Trait.id 2
Trait 0 US !GP
0.1821E-03
0.01 0.7649E-01
id

Trait.dam 2
Trait 0 US !GP
0.8117E-05
0.01 0.1331E-01
dam

#########################################################

The asr file typically looks like this:

ASReml 2.00a [01 Jul 2006] ASREML DAA in big cross
     Build: v [27 Feb 2007] 32 bit
 01 Apr 2008 11:30:19.665 125.00 Mbyte Windows bigcrossPmulti
 Licensed to: hy
 ***********************************************************
 * SYNTAX change: A/B now means A A.B *
 * *
 * Contact support_at_asreml.co.uk for licensing and support *
 ***************************************************** ARG *
 Folder: C:\Documents and Settings\asennus.000\My
Documents\Tuomas\ASRemlAnalyses130308
   id !P
   sire !A
   dam !A
   block !A
 A-inverse retrieved from ainverse.bin
 PEDIGREE [Pedigree.ped ] has 2276 identities, 6657 Non zero elements
 QUALIFIERS: !SKIP 1 !EXTRA 100 !MAXITER 700 !AISING
 QUALIFIER: !DOPART 2 is active
 Reading BodyData.asd FREE FORMAT skipping 1 lines

 Bivariate analysis of shape1 and shape2
 Using 2147 records of 2147 read

  Model term Size #miss #zero MinNon0 Mean
MaxNon0
   1 id !P 2276 0 0 6.000 1142. 2276.
   2 sire 42 0 0 1 21.6833 42
   3 dam 84 0 0 1 42.8668 84
   4 block 2 0 0 1 1.5207 2
   5 size 0 0 4.170 5.792 7.333
   6 shape1 0 0-0.3741 -0.3532 -
0.3336
   7 shape2 0-0.2138E-01 0.1928E-02 0.2125E-01
   8 shape3 0 0-0.4081E-01 0.1677E-10
0.4334E-01
   9 shape4 0 0-0.4864E-01-0.6800E-10
0.2897E-01
  10 shape5 0 0-0.2663E-01 0.2329E-10
0.2061E-01
  11 shape6 0 0-0.2324E-01-0.9502E-10
0.3225E-01

   2147 identity
      2 UnStructure 0.0000 0.0000 0.0000
    4294 records assumed pre-sorted 2 within 2147
      2 UnStructure 0.0000 0.0000 0.0000
   2276 Ainverse
 Structure for Trait.id has 4552 levels defined
      2 UnStructure 0.0000 0.0000 0.0000
     84 identity
 Structure for Trait.dam has 168 levels defined
 Forming 4726 equations: 6 dense.
 Initial updates will be shrunk by factor 0.038
 Notice: LogL values are reported relative to a base of 10000.000
 Notice: 2 singularities detected in design matrix.
   1 LogL= 9008.88 S2= 1.0000 4290 df
   2 LogL= 9089.98 S2= 1.0000 4290 df
   3 LogL= 9441.45 S2= 1.0000 4290 df
   4 LogL= 9957.93 S2= 1.0000 4290 df
   5 LogL= 10317.9 S2= 1.0000 4290 df : 2 components
constrained
   6 LogL= 10415.3 S2= 1.0000 4290 df : 2 components
constrained
   7 LogL= 10452.4 S2= 1.0000 4290 df
   8 LogL= 10467.7 S2= 1.0000 4290 df
   9 LogL= 10470.1 S2= 1.0000 4290 df
  10 LogL= 10470.1 S2= 1.0000 4290 df
  11 LogL= 10470.1 S2= 1.0000 4290 df
  12 LogL= 10470.1 S2= 1.0000 4290 df
  13 LogL= 10470.1 S2= 1.0000 4290 df
  14 LogL= 10470.1 S2= 1.0000 4290 df
  15 LogL= 10470.1 S2= 1.0000 4290 df
  16 LogL= 10470.1 S2= 1.0000 4290 df
  17 LogL= 10470.1 S2= 1.0000 4290 df
  18 LogL= 10470.1 S2= 1.0000 4290 df
  19 LogL= 10470.1 S2= 1.0000 4290 df
  20 LogL= 10470.1 S2= 1.0000 4290 df
  21 LogL= 10470.1 S2= 1.0000 4290 df

 Source Model terms Gamma Component Comp/SE % C
 Residual UnStructured 1 1 0.267061E-04 0.267061E-04 14.44 0 P
 Residual UnStructured 2 1 0.205932E-06 0.205932E-06 0.22 0 P
 Residual UnStructured 2 2 0.133494E-04 0.133494E-04 14.33 0 P
 Trait.id UnStructured 1 1 0.156594E-04 0.156594E-04 5.32 0 P
 Trait.id UnStructured 2 1 -0.340451E-06 -0.340451E-06 -0.23 0 P
 Trait.id UnStructured 2 2 0.790775E-05 0.790775E-05 5.32 0 P
 Trait.dam UnStructured 1 1 0.605679E-09 0.605679E-09 0.00 0 B
 Trait.dam UnStructured 2 1 0.725274E-09 0.725274E-09 0.00 0 B
 Trait.dam UnStructured 2 2 0.421248E-09 0.421248E-09 0.00 0 B
 Warning: Code B - fixed at a boundary (!GP) F - fixed by user
               ? - liable to change from P to B P - positive definite
               C - Constrained by user (!VCC) U - unbounded
               S - Singular Information matrix
 S means there is no information in the data for this parameter.
 Very small components with Comp/SE ratios of zero sometimes indicate poor
           scaling. Consider rescaling the design matrix in such cases.
 Covariance/Variance/Correlation Matrix UnStructured Residual
  0.2671E-04 0.1091E-01
  0.2059E-06 0.1335E-04
 Covariance/Variance/Correlation Matrix UnStructured Trait.id
  0.1566E-04 -0.3059E-01
 -0.3405E-06 0.7908E-05
 Covariance/Variance/Correlation Matrix UnStructured Trait.dam
  0.6057E-09 1.436
  0.7253E-09 0.4212E-09

 Analysis of Variance NumDF F_inc
 142 Trait 2 398396.90
 143 Trait.block 2 16.99

                     Estimate Standard Error T-value T-prev
 143 Trait.block
                    2 0.744107E-03 0.256887E-03 2.90
                    4 -0.919168E-03 0.181836E-03 -5.05
 142 Trait
                    1 -0.353225 0.418373E-03 -844.28
                    2 -0.797485E-02 0.297075E-03 -26.84 665.30
 145 Trait.dam 168 effects fitted
 144 Trait.id 4552 effects fitted ( 6 are
zero)
 Finished: 01 Apr 2008 11:30:29.680 LogL Converged

--
Tuomas Leinonen
Ecological Genetics Research Unit
Department of Biological and Environmental Sciences
PO Box 65 (Biocenter 3, Viikinkaari 1)
FIN-00014 University of Helsinki
Finland
email: tuomas.leinonen_at_helsinki.fi
tel. +358(0)9 19157801
cell. +358(0)50 5409278