Significance tests are illustrated for Model 5. Tests of treatment effects were carried out using Wald F tests with Satterthwaite's approximation to the denominator DF (by using the DDFM=KR option). An F2 67 value of 3.68 was obtained for the composite test of treatment equality. This gave a significant p-value of 0.03. Wald t tests were used to perform pairwise treatment comparisons:
A - B t55 = -1.04, p = 0.30, A - C t82 = 1.99, p = 0.08, B - C t75 = 2.71, p = 0.008.
Thus, cold feet are significantly less likely on treatment C than on treatment B. Treatment A is intermediate in its effect. The t statistic for baseline cold feet was 6.5 7 with 272 DF. This was highly significant (p < 0.0001) indicating that fitting baseline has greatly increased the sensitivity of the analysis.
Was this article helpful?