I am wondering how to customize a pairwise comparisons plot of a factorial
ANOVA, without doing a lot of manual manipulation of a TukeyHSD object. The
customizations I'd like are:
1. The aov used log-transformed response data, but I'd like to plot the
intervals on their original, untransformed scales
2. Plot all the main and interaction effects together, rather than in a
separate window for each.
3. Omit certain comparisons, specifically interactions without some common
level. ie plot A1:B1 - A1:B2, but not A1:B2 - A3:B4.
For example (I know not a valid one since the factor wool isn't significant,
and it doesn't need to be log-transformed):
> data(warpbreaks)
> summary(fm1 <- aov(log(breaks) ~ wool*tension, data = warpbreaks))
Df Sum Sq Mean Sq F value Pr(>F)
wool 1 0.3125 0.3125 2.2344 0.141511
tension 2 2.1762 1.0881 7.7792 0.001185 **
wool:tension 2 0.9131 0.4566 3.2642 0.046863 *
Residuals 48 6.7138 0.1399
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
The general features of plot(fm1) are good but I'd like to plot all the
comparisons together in one plot, on the original scale, skipping the
comparisons B:M-A:L, B:H-A:L, A:M-B:L, A:H-B:L, B:H-A:M, A:H-B:M while
keeping the rest of the interaction comparisons. Advice on any of these, if
not all, would be much appreciated.
[[alternative HTML version deleted]]
