Page 330 - Linear Models for the Prediction of Animal Breeding Values 3rd Edition
P. 330

Appendix D: Methods for


         Obtaining Approximate Reliability

         for Genetic Evaluations





         D.1   Computing Approximate Reliabilities for an Animal Model

         Presented below is a method published by Meyer (1989) for obtaining approximate
         values of repeatability or reliability for genetic evaluations from an animal model and
         has been used to estimate reliabilities in the national dairy evaluation system in
         Canada. The reliability for each animal is derived from the corresponding diagonal
         element in the MME, adjusting for selected off-diagonal coefficients. For instance, the
         section of the coefficient matrix (C) pertaining to an animal i with parents f and m and
         with a record in a subclass h of a major fixed effect as HYS could be represented as:

            ⎡  ii c  -a  -a   1⎤
            ⎢                  ⎥
            ⎢ -a    ff c  0.5a  0 ⎥
            ⎢           c mm  0⎥
            ⎢ -a  0.5a         ⎥
            ⎣  1    0     0  n h⎦

         where n  is the number of records in subclass h of the major fixed effect and a =
                h
             2
          2
                                                                   −1
         s /s . If this were the complete coefficient matrix for this animal, C  and hence true
          e  a
                                                                             ii
         reliability could be obtained using partition matrix results. Thus the coefficient c  can
         be calculated as the reciprocal of the ith diagonal element of C after absorbing all
         other rows and columns. For animal i:
                                                  1
                            2
                                                    2
             ii
            c  = (c  − 1/n  − a (c  + c   − a)/(c c   −  a )) −1
                  ii   h      ff  mm       ff mm  4
         and for parent f:
                          1
             ff
                                  2
            c  = (c  − Q − ( a − Q) /(c   − Q)) −1
                  ff      2         mm
         with:
                  2
            Q = a (c  − 1/n ) −1
                    ii    h
         Exchange m for f for parent m.
            However, if there are other off-diagonals for animal i, the above equations will
         yield approximations of the diagonal elements of C and hence reliability. Based on
         the above principle of forming and inverting the submatrix of the MME for each
         animal, Meyer outlined three steps for calculating approximate r , which were similar
                                                                2
                   2
         to the true r  from her simulation study. These steps are:
         1. Diagonal elements (D) of animals with records are adjusted for the effect of the
         major fixed effects such as HYS. Thus:
            D  = D  − 1/n
              1i   0i    h
          314            © R.A. Mrode 2014. Linear Models for the Prediction of Animal Breeding Values,
                                                                3rd Edition (R.A. Mrode)
   325   326   327   328   329   330   331   332   333   334   335