Contrast Results

This table displays results for each contrast. Polynomial contrasts were specified. Thus, the first contrast tests the linear component of each effect, the second contrast tests the quadratic component, and the third contrast tests the cubic component. The significance values for most of the contrasts are greater than 0.10, so you can conclude these components of the effects are not significant. The significance value for the cubic component of WEEK*PROMO*MARKETID is between 0.10 and 0.05, small enough to be interesting but not significant. However, you can fairly safely ignore this contrast, because of the complexity involved. It is difficult to believe that the cubic component of a three-way interaction might have an effect when none of the two-way effects or the other components of the three-way effect are significant. The linear component of WEEK is more plausible as an effect. Its significance value is also between 0.10 and 0.05, but it is also the most basic component of the simplest within-subjects effect.

This result runs somewhat counter to the results of the univariate and multivariate tests for WEEK, which indicated that the within-subjects factor has no effect. Thus, you are in doubt as to whether WEEK should be included in the model.

Next