Re: missing value
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: missing value




Dear ASREMLers,

Action on missing values depends on where they occur.

Missing values in independent variables and factors are treated as
zeros. 

The manual says you may delete the whole record by specifying !D0
but I just tested it and it doesn't.  I will rectify that
but in the meanwhile you will have to manually delete records with
missing values in independent variables if this is what you want
to happen.  An alternative strategy which is sometimes appropriate
for covariates is to CENTRE the covariate - which is equivalent
to imputing the MEAN value to the missing value.


Concerning the dependent variable,  ASREML drops records with
missing values unless the  'mv'  term is included in the model
or the analysis is multivariate.

> 
> A question about missing values. 
> I thought asreml would recognize the star (*) as a missing value,
> However, in the following simple example, the program has taken it as a
> valid zero-value
> This in spite of the fact that one missing is given in the variable list.
> 
> 
> ------data--------
> 
> 12 73
> 14 89
> 16 95
> 13 76
> * 97
> 
> ------as file------
> 
> test data to checking missing
>   x
>   y
> test.dat
> y ~ mu + x
> 
> ----output---------
>  ASREML [15 Feb 1999]  test data to checking missing 
> 
>  Wed May 12 10:14:14 1999 128.00 Mbyte    test
>   Reading test.dat  FREE FORMAT skipping  0 lines
>  Univariate analysis of y
>  Using        5 records [of       5 read from       5 lines of test.dat
>    ]
>   Model term      Size Type    COL   Minimum    Mean      Maximum   #zero
> #miss
>    1 x               1 Covariat  1  12.00      13.75      16.00         0
>   1
>    2 y               1 Variate   2  73.00      86.00      97.00         0
>   0
>    3 mu              1 Constant Term
>   Forming  3 equations:  3 dense
>  Initial updates will be shrunk by factor    0.548
>  LogL=-12.2458     S2=  139.17          3 df   1.0000
>  Final parameter values                        1.0000
> 
>    Source               Model  terms     Gamma     Component    Compnt/StndErr
>    Variance                 5      3   1.00000       139.167          1.22 P
>                      Solution       Standard Error    T-value     T-prev
>    1 x                        1        0.45        0.45 [DF F_inc F_all]
>                     1  -0.625000       0.932626         -0.67
>    3 mu                       1      265.72       64.82 [DF F_inc F_all]
>                     2    92.8750        11.5359          8.05
>  Finished: Wed May 12 10:14:14 1999   LogL Converged
>  
> 
> 
> 2 more:
> 
> I was originally wondering whether records with missing covariates of 
> fixed effect codes are discarded, I suppose they are.

NOT unless you deliberately discard them

> 
> How is the log likelihood calculated? Is it not
> logl=-.5*N*log(2*pi)-.5*log(det(V))-.5*(e'*inv(V)*e)
> but this gives me  logl =  -18.4339
> 

  The formula in Chapt 2 says
  -0.5(log | X'HX | + log|H| + e log S2 + yPy/S2)
  
  This is parameterized differently from yours but some terms are equivalent.
  
  
  In your example H=I and X'X = 765 55
                                 55  5           from .asl file
                                 
  So LogL  is  -0.5 ( 6.6846 + 0 + 3 x 4.9357 + 3 ) = 12.2458
  
  My expression omits the pi term but includes a term in X
  
  
  regards
  
  Arthur
  
                                 
> julius
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> Julius van der Werf				e-mail:jvanderw@metz.une.edu.au
> Sr. Lecturer in Animal Genetics		www:   
http://metz.une.edu.au/~jvanderw/
> Division of Animal Science				phone: +61-2-6773-2092
> School of Natural Resources and Rural Science		fax  : +61-2-6773-3275
> University of New England				home : +61-2-6775-2442
> Armidale, NSW 2350, AUSTRALIA	
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> --
> Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml


<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
Arthur Gilmour PhD                    mailto: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 still free by anonymous ftp from pub/aar on ftp.res.bbsrc.ac.uk
    or point your web browser at ftp://ftp.res.bbsrc.ac.uk/pub/aar/ 

To join the asreml discussion list, send the message  
     subscribe
mailto:asreml-request@chiswick.anprod.CSIRO.au

To send messages to the list, mailto:asreml@chiswick.anprod.CSIRO.au

Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml

                        <> <> <> <> <> <> <>
"The Holy Scriptures are able to make you wise for salvation
   through faith which is in Christ Jesus.
 All Scripture is given by inspiration of God, and is profitable for
   doctrine, for reproof, for correction, for instruction in righteousness,
 That the man of God may be complete,
   thoroughly equipped for every good work."   II Timothy 3:15-17.
<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>


--
Asreml mailinglist archive: http://www.chiswick.anprod.csiro.au/lists/asreml