Dear R community,
I'm trying to understand this behavior of TukeyHSD. My goal is to obtain
defensible, labelled multiple comparisons of an interaction term.
Firstly, if I plot the TukeyHSD from the model that calculates its own
interactions, then the y-axis labels appear to be reflected on their median
when compared to the text output of the TukeyHSD statement. The labels are
integers.
Secondly, if I provide an interaction term for the model, to try to coerce
TukeyHSD to label the comparisons, then the multiple comparison outcome is
quite different, as is the output from coefficients(). It must be using a
different parameterization, because the anova statements that summarize the
model are identical.
However, if two different parameterizations give rise to two different sets of
multiple comparisons, how ought we choose between them?
The following snippet illustrates.
======================================================================
data(warpbreaks)
warpbreaks$WT <- interaction(warpbreaks$wool, warpbreaks$tension)
summary(fm1 <- aov(breaks ~ wool * tension, data = warpbreaks))
summary(fm2 <- aov(breaks ~ wool + tension + WT, data = warpbreaks))
summary(fm1) # Identical
summary(fm2) # Identical
coefficients(fm1) # Different
coefficients(fm2) # Different
TukeyHSD(fm1, "wool:tension")
# (1) My first concern is that the y-axis labels seem to be reflected
# in their median.
plot(TukeyHSD(fm1, "wool:tension")) # Labels are upside down?
# (2) My second concern is that the estimates and intervals are
# different from fm2
TukeyHSD(fm2, "WT")
======================================================================
I would appreciate any advice.
Andrew
--
Andrew Robinson Ph: 208 885 7115
Department of Forest Resources Fa: 208 885 6226
University of Idaho E : andrewr at uidaho.edu
PO Box 441133 W : http://www.uidaho.edu/~andrewr
Moscow ID 83843 Or: http://www.biometrics.uidaho.edu
No statement above necessarily represents my employer's opinion.
