Page 275 - Linear Models for the Prediction of Animal Breeding Values 3rd Edition
P. 275

¶L/¶s  = (0.5){8.8753 − (−12.5000) − 18.0733} = 1.6510
                 2
                 e
                            −1
                                4
                 2
            ¶L/¶s  = (0.5){a ′ A a/s  − q/s  + trace[C A ]s /s }
                                                          4
                                                   −1
                                                       2
                                       2
                                                22
                 a              a      a               e  a
                 2
            ¶L/¶s  = (0.5){6.3461 − 40.0000 + 36.1466} = 1.2464
                 a
        and:
                2
            A(¶L /¶s ) = −(0.5){(y − Xb − Za) ′ P(y − Xb − Za)/s  }
                                                         4
                    4
                    e                                    e
                2
                    4
            A(¶L /¶s ) = −(0.5)16.5346 = −8.2673
                    e
            A(¶L /¶s ) = −(0.5){a ′ Z ′ PZa}/s  4
                    4
                2
                    a                  a
                    4
                2
            A(¶L /¶s ) = −(0.5)9.1163 = −4.5582
                    a
                                                     2
                                                       4
                    2
                       2
                2
            A(¶L /¶s ¶s ) = −(0.5){a ′ Z ′ P(y − Xb − Za)}/(s  s   )
                    a  e                             e  a
                    2
                2
                       2
            A(¶L /¶s ¶s ) = −(0.5)11.3070 = −5.6535
                    a  e
        and:
                  ⎡8.2673 5.6535 ⎤            ⎡  0.7967 − 0.9882⎤
            Ainf =  ⎢            ⎥   so Ainf  −1  = ⎢           ⎥
                  ⎣ 5.6535 4.5582 ⎦           ⎣ −00.9882  1.4450 ⎦
            Using Eqn 15.4 and replacing Einf by Ainf:
                                                        ⎤ ⎡
           q =  q + Ainf  −1 ( L∂ /  ∂ q = )  ⎡0.4 ⎤ ⎥  + ⎢ ⎡  0.7967  −0.9882 1.6510⎤ ⎥ =  ⎡ ⎢ 0.4⎤ ⎥ +  ⎡ ⎢ 0.0838⎤ ⎥
                                 ⎢
                                                        ⎥ ⎢
            n
                                                      0
                                                        ⎦ ⎣
                                 ⎣ 0.2 ⎦  ⎣ −0.9882  1.4450 1.2464 ⎦  ⎣ 0.2 ⎦  ⎣ 0.1695 ⎦
                                                                    2
                               2
                                      2
                                                    2
        so that new estimates of s  and s  are 0.4838 (kg ) and 0.3695 (kg ), respectively.
                                e     a
            Table 15.3 gives six successive iterates and log-likelihood for this data.
            In the last iteration:
                   ⎡  2 4436  −3 2532.  ⎤
                      .
            Ainf =  ⎢ ⎢             ⎥ ⎥
                −1
                               .
                   ⎢ ⎣ −3 2532.  5 3481 ⎥ ⎦
        so that the estimate of s  is 0.4835 with standard error  2.4436 = 1.563 and the esti-
                             2
                             e
                 2
        mate of s  is 0.5514 with standard error  5.3481 = 2.313.
                 a
                                   2
                                             2
            By contrast, if estimates of s  = 0.4 and s  = 0.2 are used in conjunction with Eqns 15.5
                                   e         a
                             2
                                                                    2
                                                                                2
                                                     2
        and 15.6 then: (n − p)s  = (y – Xb − Za) ′ (y) so 3s  = 1.9277 so s  = 0.6426 (kg )
                             e                       e              e
                                                                2
        and qs  = a ′ A a + trace[C A ]s  so 8s  = 0.2538 + 1.4458 so s  = 0.2125 (kg ) with
                                           2
                                                                            2
                                 −1
                              22
                    −1
                                     2
               2
               a                     e     a                    a
        L = −2.3852. After 1000 iterations, the algorithm gives s  = 0.4842 (kg ) and s  =
                                                                               2
                                                                       2
                                                           2
                                                           e                   a
                  2
        0.5504 (kg ) with L = −2.1817, showing that this algorithm is slower to converge.
                   Table 15.3. Estimates of s  and s  and L.
                                              2
                                        2
                                        e     a
                                                      2
                                  2
                                                  2
                                      2
                   Iterate       s  (kg )        s  (kg )          L
                                  e               a
                   1             0.4000          0.2000          −2.3852
                   2             0.4838          0.3695          −2.2021
                   3             0.4910          0.5126          −2.1821
                   4             0.4839          0.5500          −2.1817
                   5             0.4835          0.5514          −2.1817
                   6             0.4835          0.5514          −2.1817
        Estimation of Genetic Parameters                                     259
   270   271   272   273   274   275   276   277   278   279   280