Info

0.46

76

1

\

71

0.52

1

86

0.48

0.61

1

73

0.46

0.50

0.61

1

A85

(1

\

0.54

1

0.54

0.54

1

0.54

0.54

0.54

1

B68

(1

\

0.39

1

0.39

0.39

1

0.39

0.39

0.39

1

C76

(1

\

0.63

1

0.63

0.63

1

0.63

0.63

0.63

1

A85

(1

\

0.58

1

0.48

0.58

1

0.50

0.48

0.58

1

B68

(1

\

0.42

1

0.33

0.42

1

0.42

0.33

0.42

1

C76

(1

\

0.69

1

0.61

0.69

1

0.46

0.61

0.69

7467.4

7489.3

7458.6

7462.3

7459.5

7448.0

On the basis of the fact that Model 5 indicates that separate covariances for each treatment group may be necessary and that Model 3 suggests a Toeplitz pattern, Model 6 incorporating both these features was tested. Models 3 and 5 are nested within Model 6 and Model 6 shows significant improvements over them both, x2 = 26.59(p = 0.0008) and x62 = 23.57(p = 0.0006).

Thus, we have statistically justified the use of a fairly complex covariance pattern. This is likely to be partly because the trial is relatively large, so the covari-ance parameters are estimated with a reasonable accuracy. Model 6 has given us statistical evidence that the treatment groups have different variances. Also, the Toeplitz structures indicate that correlations between repeated measurements are highest for treatment C and lowest for treatment B. These differences are likely to be, in some way, due to the different actions of the treatments. In smaller trials, however, it is often not possible statistically to justify any pattern more complex than the compound symmetry or first-order autoregressive. This is not necessarily because a more complex pattern does not exist, but because there is insufficient information to determine it.

Was this article helpful?

0 0

Post a comment