Levi Waldron
2008-Apr-28 21:07 UTC
[R] restricting pairwise comparisons of interaction effects
I'm interested in restricting the pairwise comparisons of interaction effects in a multi-way factorial ANOVA, because I find comparisons of interactions between all different variables different to interpret. For example (supposing a p<0.10 cutoff just to be able to use this example):> summary(fm1 <- aov(breaks ~ wool*tension, data = warpbreaks))Df Sum Sq Mean Sq F value Pr(>F) wool 1 450.7 450.7 3.7653 0.0582130 . tension 2 2034.3 1017.1 8.4980 0.0006926 *** wool:tension 2 1002.8 501.4 4.1891 0.0210442 * Residuals 48 5745.1 119.7 --- Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1> TukeyHSD(fm1,conf.level=0.90)Tukey multiple comparisons of means 90% family-wise confidence level Fit: aov(formula = breaks ~ wool * tension, data = warpbreaks) $wool diff lwr upr p adj B-A -5.777778 -10.77183 -0.7837284 0.058213 $tension diff lwr upr p adj M-L -10.000000 -17.66710 -2.332900 0.0228554 H-L -14.722222 -22.38932 -7.055122 0.0005595 H-M -4.722222 -12.38932 2.944878 0.4049442 $`wool:tension` diff lwr upr p adj B:L-A:L -16.3333333 -30.112566 -2.554101 0.0302143 A:M-A:L -20.5555556 -34.334788 -6.776323 0.0029580 B:M-A:L -15.7777778 -29.557010 -1.998545 0.0398172 A:H-A:L -20.0000000 -33.779233 -6.220767 0.0040955 B:H-A:L -25.7777778 -39.557010 -11.998545 0.0001136 A:M-B:L -4.2222222 -18.001455 9.557010 0.9626541 B:M-B:L 0.5555556 -13.223677 14.334788 0.9999978 A:H-B:L -3.6666667 -17.445899 10.112566 0.9797123 B:H-B:L -9.4444444 -23.223677 4.334788 0.4560950 B:M-A:M 4.7777778 -9.001455 18.557010 0.9377205 A:H-A:M 0.5555556 -13.223677 14.334788 0.9999978 B:H-A:M -5.2222222 -19.001455 8.557010 0.9114780 A:H-B:M -4.2222222 -18.001455 9.557010 0.9626541 B:H-B:M -10.0000000 -23.779233 3.779233 0.3918767 B:H-A:H -5.7777778 -19.557010 8.001455 0.8705572 It would seem to make sense (and please correct me if I'm wrong) to restrict the pairwise comparisons of wool:tension to terms like B:L-A:L, and A:M-A:L, and not calculate or try to interpret differences like B:M-A:L. How can I accomplish this (note that there are actually 5 factors in the experiment I'm analyzing)?