Tukey Test for Additivity If we believe interaction is a problem, this is a possible way to test it without using up all our df. One additional term is added to the model ( θ ) , replacing the ( αβ )ij with the product θα iβ j : µij = µ + α i + β j + θα iβ j We use one degree of freedom to estimate θ, leaving two left to estimate error (in the terminals example). Of course, this only tests for interaction of the specified form, but it may be better than nothing. Using SAS for Tukey’s test for additivity: Terminal Example. Find µ̂ (grand mean) proc glm data=terminals; model usage=; output out=overall p=muhat; proc print data=overall;run; Obs 1 2 3 4 5 6 7 8 usage 16.5 11.8 12.3 16.6 21.4 17.3 16.9 21.0 location 1 2 3 4 1 2 3 4 time Midterm Midterm Midterm Midterm Final Final Final Final muhat 16.725 16.725 16.725 16.725 16.725 16.725 16.725 16.725 Find µ̂ A (treatment means) proc glm data=terminals; class location; model usage=location; output out=meanA p=muhatA; proc print data=meanA;run; Obs 1 2 3 4 5 6 7 8 usage 16.5 11.8 12.3 16.6 21.4 17.3 16.9 21.0 location 1 2 3 4 1 2 3 4 time Midterm Midterm Midterm Midterm Final Final Final Final A 18.95 14.55 14.60 18.80 18.95 14.55 14.60 18.80 Find µ̂ B (treatment means) proc glm data=terminals; class time; model usage=time; output out=meanB p=muhatB; proc print data=meanB;run; Obs 1 2 3 4 5 6 7 8 usage 16.5 11.8 12.3 16.6 21.4 17.3 16.9 21.0 location 1 2 3 4 1 2 3 4 time Midterm Midterm Midterm Midterm Final Final Final Final muhat B 14.30 14.30 14.30 14.30 19.15 19.15 19.15 19.15 Combine and compute data estimates; merge overall meanA meanB; alpha = muhatA - muhat; beta = muhatB - muhat; atimesb = alpha*beta; proc print data=estimates; var location time alpha beta atimesb; run; Obs 1 2 3 4 5 6 7 8 location 1 2 3 4 1 2 3 4 time Midterm Midterm Midterm Midterm Final Final Final Final alpha 2.225 -2.175 -2.125 2.075 2.225 -2.175 -2.125 2.075 beta -2.425 -2.425 -2.425 -2.425 2.425 2.425 2.425 2.425 atimesb -5.39562 5.27437 5.15312 -5.03188 5.39563 -5.27437 -5.15312 5.03188 Run GLM proc glm data=estimates; class location time; model usage=location time atimesb; run; Source location time atimesb Error Corrected Total DF 3 1 1 2 7 Sum of Squares 37.00500000 47.04500000 0.07855763 0.26644237 84.39500000 Mean Square 12.33500000 47.04500000 0.07855763 0.13322119 F Value 92.59 353.13 0.59 Pr > F 0.0107 0.0028 0.5228 The test for atimesb is testing H0: θ = 0, which is not rejected. According to this, the interaction is not significant. Compare to original analysis (Additive model): Source location time Error Corrected Total DF 3 1 3 7 Squares 37.00500000 47.04500000 0.34500000 84.39500000 Mean Square 12.33500000 47.04500000 0.11500000 F Value 107.26 409.09 Pr > F 0.0015 0.0003 Notice the p-values on the main effect tests changed, because we used up an extra error df when conducting Tukey’s additivity test.