Question: Two researchers examine the results from a 26 factorial arrangement with one replication. Researcher A decides in advance to use the residuals from a model with main effects and 2-factor interactions to estimate variability, assuming all 3-factor, 4-factor, 5-factor, and 6-factor interaction terms are zero. Researcher B is willing to assume that the 5-factor and 6-factor interaction terms are zero, but uses a backward stepwise procedure to eliminate 3-factor and 4-factor interactions if they fail to meet the p = :05 criterion for significance. Researcher B finds that all 3-factor and 4-factor interactions can be eliminated and therefore ends up with the same model structure as Researcher A.
(a) Can the same inferences be drawn from these two approaches?
(b) Should the two report their results the same way?
(c) Is one approach preferred?