Table 5 (con't)
*** No RANDOM model specified RESIDUAL term only will be used
INPUT READ
 36 EXPERIMENTAL UNITS
FIXED DF 17


REML RESIDUAL MAXIMUM LIKELIHOOD COMPONENTS OF VARIANCE
 ITERATION NO 2

Analysis of Design 1 with unstructured treatments
 ITERATIONS HAVE CONVERGED

 MEAN EFFES Best Linear Unbtased Estimates

MEAN EFFECTS (B.L.U.E.'S)Jof Treat

 A B C D E F G from
 15.917 16.361 16.750 16.972 18.083 18.306 18.583 directive
 'MEAN
 H I J K L MARGIN EFFECT'
 19.028 19.917 20.306 20.694 21.083 18.500

 STANDARD ERROR OF DIFFERENCES BETWEEN PAIRS
 AVERAGE 0.385 MAXIMUM 0.404 MINIMUM 0.361


 CANONICAL DECOMPOSITION T.INV INF MX.T TO BE DIAGONAL

 DIAGONALISED INVERSE INF MATRIX (D)
 SIGMA
 SQUARED Output
 SIGMA SQUARED 0.004 from
 directive
 'CANONICAL'
 Coefficient matrix (T)

 SIGMA
 SQUARED ERROR mean square ERROR degrees
 1.000 offreedom

 TC = T. (Components_fromg revio s_iteration) and 2*TC*TC/D i.e
 Approximate Stratum Variances and effective d


 SIGMA SQUARED 0.196 19.000


 NOTE: In this example, the MINIMUM standard error
 of differences between pairs (0361) is the
 standard error of differences between pairs
 of treatments concurring once, as for instance
 treatments A and B. Whereas the MAXIMUM
 (0.404) is the standard error of differences
 between pairs of treatments not concurring
 at all, asfor instance A andE.