Further problem with prediction
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Further problem with prediction



I would be grateful for feedback on the ASREML command file and output
embedded below.   The objective is to determine the relationship between
field establishment (plants/m2) and grain yield (t/ha) in three varieties of
lupin.   

I'd like to know whether the output indicates that the model I am fitting is
appropriate.   In particular:

i.  I put two starting values after the codeword AR with the intention of
getting first- and second-lag autocorrelations.   Is this the correct method?

ii.  If so, does the output indicate that the inclusion of second-lag
autocorrelation gives a significantly better fit?   I am assuming that it
does because the Compnt/StndErr values for both AR terms are quite large
(>3).   

iii.   Why are the gamma and component values for these two terms so
similar?   Is this just coincidence?

iv.   I don't want to fit any column effects.   So is the codeword IDEN
okay?   How should I decide whether the identify matrix should be scaled?

v.  Is it okay to fit a linear effect of the variate ESTAB (and its
interaction with Variety), to account for any overall trend, before
including the spline in the model?

vi.  Should the spline (and its interaction with Variety) be put in the
random effects model?   If so, will the variance explained by the spline be
included in the error against which the linear trend is tested? 

vii.  Am I right in thinking that it makes no difference to the fitted value
of a data point whether a term (e.g. spl(ESTAB)) is fitted as a fixed or a
random effect?

The desired outcome is a curve relating yield to establishment in each
variety, and it is therefore necessary to obtain predicted values.   I would
like to avoid the complexities of setting up a .pin file, and the ASREML
manual suggests adding some "missing data" and fitting them with "missing
values" (p 86, version of 2 October 1998).   However, it seems to me that
this approach won't work for a spatial analysis, as each observation must be
allocated to a row and column.   I had hoped that the row and column values
would not influence the fitted value (just as the variogram is not
influenced by whether AR terms are fitted), but it turns out that they do
have an effect.   Any suggestions?

__________________

96wh14.as

Restricted branching lupin - density trials
 Variety   3 !A
 TDENS     6
 Rep       5
 PLOT     90
 COL       2
 ROW     101
 THA
 ESTAB

C:\DOCS\Dracup_M\96WH14\96WH14.dat !skip 1 !maxit 20

THA ~ mu mv c(Variety) ESTAB c(Variety).ESTAB,
 !r Rep spl(ESTAB) c(Variety).spl(ESTAB) 
1 2
ROW ROW AR .1 .1
COL COL IDEN
______________________
96wh14.asr

  ASREML [30 Sep 1998]  Restricted branching lupin - density trials

 10/29/98 12:09:43.22       8.00 Mbyte    c:\DOCS\Dracup_M\96WH14\96wh14.as
 QUALIFIERS: !skip 1                                                     
 Reading C:\DOCS\Dracup_M\96WH14\96WH14.dat  FREE FORMAT skipping            1
 lines
 Univariate analysis of THA                 
 Using      108 records [of     108 read from     108 lines of
C:\DOCS\Dracup_M\96W]
  Model term      Size Type    COL   Minimum    Mean      Maximum   #zero #miss
   1 Variety         3 Factor    1      1     2.0000          3         0    18
   2 TDENS           6 Factor    2     25    71.6667        125         0    18
  WARNING - More levels in factor than expected
   3 Rep             5 Factor    3      1     3.0000          5         0    18
   4 PLOT           90 Factor    4      1    45.5000         90         0    18
   5 COL             2 Factor    5      1     1.5000          2         0     0
   6 ROW            54 Factor    6      1    27.5000         54         0     0
   7 THA             1 Variate   7 0.7530      1.014      1.428         0    20
   8 ESTAB           1 Covariat  8  19.60      71.19      137.6         0    18
   9 mu              1 Constant Term
  10 mv_estimates   20 Missing value
  11 c(Variety)      2 Factor    1      1     2.0000          3         0    18
  12 c(Variety).E    2 Interaction 11 c(Vari:    2    8 ESTAB         :    1
  13 spl(ESTAB)     47 Spline    8   19.60     71.19      137.6         0    18
  14 c(Variety).s   94 Interaction 11 c(Vari:    2   13 spl(ESTAB)    :   47
    54  AR=AutoR    0.10    0.10
     2  identity
 Forming          173 equations:           27 dense
 Initial updates will be shrunk by factor    0.548
 LogL= 80.6926     S2= 0.11703E-01     82 df  0.10000    0.10000    0.10000    
   1.0000    0.10000    0.10000    
 LogL= 104.813     S2= 0.11911E-01     82 df  0.16662    0.10000E-010.10000E-01
   1.0000    0.21445    0.27837    
 LogL= 116.546     S2= 0.13897E-01     82 df  0.30356    0.10000E-020.10000E-02
   1.0000    0.30181    0.35295    
 LogL= 121.136     S2= 0.15483E-01     82 df  0.40011    0.10000E-030.10000E-03
   1.0000    0.35163    0.34838    
 LogL= 122.124     S2= 0.14720E-01     82 df  0.44461    0.58089E-040.10000E-04
   1.0000    0.35465    0.31151    
 LogL= 122.722     S2= 0.18048E-01     82 df  0.42452    0.49216E-040.10000E-06
   1.0000    0.34659    0.41749    
 LogL= 122.719     S2= 0.20917E-01     82 df  0.36671    0.45774E-040.10000E-06
   1.0000    0.40451    0.39840    
 Final parameter values                       0.56309    0.65929E-040.10000E-06
   1.0000    0.37353    0.37798    

   Source               Model  terms     Gamma     Component    Compnt/StndErr
  Rep                       5      5  0.563094      0.117781E-01      1.77 P  
  spl(ESTAB)               47     47  0.659291E-04  0.137902E-05      0.75 P  
  c(Variety).spl(ESTAB     94     94  0.100000E-06  0.209167E-08      1.86 B  
   Variance               108     82   1.00000      0.209167E-01      1.86 P  
   Residual           AR=AutoR    54  0.373534      0.373534          3.20 U  
   Residual           AR=AutoR    54  0.377980      0.377980          3.20 U  
 WARNING: Code B - fixed at a boundary (!GP)
               C - Constrained by user (!CON)
               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
 Fitted Spline  49 (X) for spl(ESTAB)          
    19.600    21.800    24.400    27.050    28.200    29.800    32.800    37.400
    41.200    42.900    44.000    45.200    46.600    48.400    50.350    52.400
    54.200    56.000    57.700    59.000    60.600    63.350    64.733    69.900
    71.400    72.800    76.000    79.500    80.800    82.400    85.000    87.267
    89.400    91.600    93.400    95.600    98.600   100.000   103.800   105.800
   111.000   115.267   116.900   118.800   121.000   122.700   128.400   130.800
   137.600
 Fitted Spline  49 (Y) for spl(ESTAB)            1
  0.028825  0.025985  0.022633  0.019228  0.017760  0.015734  0.011984  0.006362
  0.001891 -0.000024 -0.001228 -0.002504 -0.003938 -0.005686 -0.007444 -0.009122
 -0.010445 -0.011620 -0.012588 -0.013231 -0.013901 -0.014735 -0.015003 -0.015187
 -0.015038 -0.014829 -0.014125 -0.013057 -0.012588 -0.011960 -0.010830 -0.009747
 -0.008657 -0.007466 -0.006443 -0.005138 -0.003267 -0.002361  0.000183  0.001561
  0.005231  0.008293  0.009465  0.010819  0.012369  0.013552  0.017448  0.019085
  0.023756
 ',                                              ,    0.03
   '',,                                        ,'     0.02
       '                                  ,,,''       0.01
        ',,                            ,,'            0.00
           ''',,                  ,,,''               0.00
                ''',,,,,,,,,,,''''                   -0.01
                                                     -0.02
                                                     -0.03
 Fitted Spline  49 (Y) for c(Variety).spl(ESTAB  1
 -0.74E-06 -0.58E-06 -0.40E-06 -0.23E-06 -0.16E-06 -0.58E-07  0.14E-06  0.50E-06
  0.89E-06  0.11E-05  0.13E-05  0.15E-05  0.17E-05  0.20E-05  0.24E-05  0.29E-05
  0.32E-05  0.35E-05  0.38E-05  0.39E-05  0.39E-05  0.37E-05  0.35E-05  0.20E-05
  0.14E-05  0.73E-06 -0.97E-06 -0.30E-05 -0.37E-05 -0.45E-05 -0.57E-05 -0.65E-05
 -0.70E-05 -0.74E-05 -0.76E-05 -0.76E-05 -0.72E-05 -0.69E-05 -0.59E-05 -0.51E-05
 -0.26E-05 -0.24E-06  0.73E-06  0.19E-05  0.32E-05  0.42E-05  0.76E-05  0.91E-05
  0.13E-04
 '''''''''''''''''''''''''''''''''''''''''''''''''    0.00
                                                      0.00
                                                      0.00
                                                     -0.01
                                                     -0.01
                                                     -0.01
                                                     -0.01
                                                     -0.01
 Fitted Spline  49 (Y) for c(Variety).spl(ESTAB  2
 -0.61E-05 -0.52E-05 -0.41E-05 -0.31E-05 -0.26E-05 -0.20E-05 -0.93E-06  0.40E-06
  0.11E-05  0.13E-05  0.14E-05  0.16E-05  0.17E-05  0.17E-05  0.18E-05  0.17E-05
  0.16E-05  0.15E-05  0.13E-05  0.12E-05  0.94E-06  0.53E-06  0.33E-06 -0.14E-06
 -0.18E-06 -0.17E-06  0.13E-07  0.44E-06  0.65E-06  0.93E-06  0.14E-05  0.19E-05
  0.22E-05  0.26E-05  0.28E-05  0.29E-05  0.30E-05  0.30E-05  0.27E-05  0.24E-05
  0.13E-05  0.12E-06 -0.38E-06 -0.10E-05 -0.18E-05 -0.24E-05 -0.46E-05 -0.55E-05
 -0.82E-05
 '''''''''''''''''''''''''''''''''''''''''''''''''    0.00
                                                      0.00
                                                      0.00
                                                     -0.01
                                                     -0.01
                                                     -0.01
                                                     -0.01
                                                     -0.01
                     Solution       Standard Error    T-value     T-prev
  12 c(Variety).ESTAB         2        3.46        3.46             [DF
F_inc F_all]
                    1   0.582442E-03   0.385741E-03      1.51
                    2  -0.101474E-02   0.388952E-03     -2.61     -2.41
   8 ESTAB                    1       23.84       24.87             [DF
F_inc F_all]
                    3  -0.196641E-02   0.394323E-03     -4.99
  11 c(Variety)               2        0.37        2.08  0.5136E-01 [DF F_i
F_a SED]
 MERRIT                -0.305525E-01   0.302952E-01     -1.01
 85SO46-37              0.607833E-01   0.298523E-01      2.04      1.73
  10 mv_estimates            20       13.49       14.01             [DF
F_inc F_all]
                    6  -0.882628       0.916597E-01     -9.63
                    7  -0.647879       0.941482E-01     -6.88      1.92
                    8   -1.07879       0.116429         -9.27     -2.83
                    9   -1.11726       0.122686         -9.11     -0.34
                   10   -1.10920       0.133636         -8.30      0.07
                   11   -1.12127       0.138845         -8.08     -0.10
                   12   -1.12293       0.143739         -7.81     -0.01
                   13   -1.12842       0.147031         -7.67     -0.05
                   14   -1.13130       0.149708         -7.56     -0.02
                   15   -1.13465       0.151712         -7.48     -0.03
                   16   -1.13715       0.153298         -7.42     -0.02
                   17   -1.13950       0.154532         -7.37     -0.02
                   18   -1.14145       0.155512         -7.34     -0.02
                   19   -1.14317       0.156290         -7.31     -0.01
                   20   -1.14464       0.156913         -7.29     -0.01
                   21   -1.14592       0.157416         -7.28     -0.01
                   22   -1.14703       0.157823         -7.27     -0.01
                   23   -1.14799       0.158154         -7.26     -0.01
                   24   -1.14882       0.158426         -7.25     -0.01
                   25   -1.14953       0.158650         -7.25     -0.01
   9 mu                       1      157.88      287.57             [DF
F_inc F_all]
                   26    1.15412       0.680586E-01     16.96
   3 Rep                      5 effects fitted
  13 spl(ESTAB)              47 effects fitted
  14 c(Variety).spl(ESTAB    94 effects fitted
 Finished: 12:10:59.39                LogL Converged

_____________________________________________________________________
N.W. Galwey,
Faculty of Agriculture,
University of Western Australia,
Nedlands, WA 6709, Australia.

Tel.: +61 9 380 1959 (direct line)
      +61 9 380 2554 (switchboard)
Fax:  +61 9 380 1108