Page 121 - Linear Models for the Prediction of Animal Breeding Values 3rd Edition
P. 121

Analysis with FA model

                                                                             *
        The main differences with fitting an FA model with no specific effects is that Z  for
        an animal in Eqn 6.11 is of order n by n and the last row of Eqn 6.11 is omitted.
                  *
        Note that Z is now a product of the eigenvectors of G and the square root of a diago-
                                                      *
        nal matrix of eigenvalues (see Section 6.4.2). Thus Z  is:
                      ⎛  2.2974 3.2100  −0.1259  −2.0983⎞
                      ⎜  −1.7761 5.8683  −0.1348  −1.5333 ⎟
                  Z = ⎜                                  ⎟ ⎟  and for animal i:
                   *
                   i
                      ⎜  0..2453 2.0018  4.4348    1.1272⎟
                      ⎜ ⎝  0.5878 4.3506 − 1.7658  3.0977⎠ ⎟
                      ⎛  0.372  −0.226  −0.042  0.054⎞
                      ⎜  −0.226  1.236  −0.141  −00.246 ⎟
              ′
                −1
             *
            ZR Z   *  = ⎜                            ⎟ ⎟
              i    i
                      ⎜ − 0.042 − 0.141  0.410  0.019⎟
                      ⎜ ⎝  0.054 − 0.246  0.019  0.537⎠ ⎟
        Setting the MME follows the usual rules and the MME has 40 equations for the
        example but with only 388 non-zero elements. The low number of non-zero elements
                                              −1
                                                                             2
        is due to the fact that only n elements of A  are contributed compared with n  for
        the MBLUP. Solving the equations gives the following solutions.
        Solutions for calf sex effects
             WWG      PWG      MSC      BFAT

        M    4.352    6.795    9.412    0.231
        F    3.487    5.959    7.095    0.535
        Animal solutions

                   Untransformed solutions              Transformed solutions a
             WWG       PWG      BFAT     MSC     WWG       PWG      MSC      BFAT
        1    −0.011    0.035    0.063   −0.008    0.095    0.227    0.340    0.010
        2    −0.003   −0.005    0.066    0.028   −0.089   −0.073    0.313   −0.050
        3     0.010   −0.020    0.010    0.021   −0.086   −0.169    0.031   −0.032
        4     0.000    0.002   −0.177   −0.067    0.168    0.136   −0.855    0.113
        5     0.015   −0.062   −0.099    0.019   −0.191   −0.407   −0.539   −0.029
        6    −0.022    0.061    0.267    0.045    0.017    0.290    1.350   −0.082
        7     0.003   −0.069   −0.153    0.005   −0.208   −0.399   −0.813   −0.015
        8    −0.007    0.050    0.285    0.060   −0.017    0.178    1.431   −0.101
        a Transformed solutions = vectors of solutions multiplied by Z*

        6.4.2  Principal component analysis

        Analysis with full PC model

        The application of a full PC model with no rank reduction is similar to the FA analysis
                                                                      *
        except that Z  is now a matrix of eigenvectors of order n by n and Z ′R Z  + (I ⊗ A )
                                                                               −1
                                                                 *
                   *
                                                                   –1
                                                                          m
        Methods to Reduce the Dimension of Multivariate Models               105
   116   117   118   119   120   121   122   123   124   125   126