Interpreting contrasts for a two-way interaction in the presence of a
significant
three-way interaction is dangerous. They might not be interpretable.
I would start by examining the interaction2wt plot (in the HH package)
interaction2wt(activity ~ pH + I + f, data=yourdataframe)
Look at the vulcan example in ?interaction2wt
You might want to look at the set of two-way interactions, conditioned on
the value
of the third factor, for example similar to
require(HH)
interaction2wt(wear ~ filler + raw,
data=vulcan,
sub="all values of pretreat",
ylim=c(50,600))
vulcan.pretreat <- split(vulcan, vulcan$pretreat)
lapply(1:length(vulcan.pretreat),
function(i) {
interaction2wt(wear ~ filler + raw,
data=vulcan.pretreat[[i]],
sub=paste("pretreat =", i),
ylim=c(50,600))
})
On Fri, Apr 6, 2012 at 5:01 AM, Jinsong Zhao <jsz...@yeah.net> wrote:
> On 2012-04-05 10:49, Richard M. Heiberger wrote:
>
>> Here is your example. The table you displayed in gigawiz ignored the
>> two-way factor structure
>> and interpreted the data as a single factor with 6 levels. I created
>> the interaction of
>> a and b to get that behavior.
>> ## your example, with data stored in a data.frame
>> tmp <- data.frame(x=c(76, 84, 78, 80, 82, 70, 62, 72,
>> 71, 69, 72, 74, 66, 74, 68, 66,
>> 69, 72, 72, 78, 74, 71, 73, 67,
>> 86, 67, 72, 85, 87, 74, 83, 86,
>> 66, 68, 70, 76, 78, 76, 69, 74,
>> 72, 72, 76, 69, 69, 82, 79, 81),
>> a=factor(rep(c("A1", "A2"), each = 24)),
>> b=factor(rep(c("B1", "B2", "B3"), each=8, times=2)))
>> x.aov <- aov(x ~ a*b, data=tmp)
>> summary(x.aov)
>> ## your request
>> require(multcomp)
>> tmp$ab <- with(tmp, interaction(a, b))
>> xi.aov <- aov(x ~ ab, data=tmp)
>> summary(xi.aov)
>> xi.glht <- glht(xi.aov, linfct=mcp(ab="Tukey"))
>> confint(xi.glht)
>>
>> ## graphs
>> ## boxplots
>> require(lattice)
>> bwplot(x ~ ab, data=tmp)
>> ## interaction plot
>> ## install.packages("HH") ## if you don't have HH yet
>> require(HH)
>> interaction2wt(x ~ a*b, data=tmp)
>>
>>
> Thank you very much for the demonstration.
>
> There is still a small difference between the results of glht() and the
> the table displayed in gigawiz. I try my best to figure out, but fail...
>
> By the way, I have a similar question. I built a ANOVA model:
>
> activity ~ pH * I * f
>
>
> Df Sum Sq Mean Sq F value Pr(>F)
> pH 1 1330 1330 59.752 2.15e-10 ***
> I 1 137 137 6.131 0.0163 *
> f 6 23054 3842 172.585 < 2e-16 ***
> pH:I 1 152 152 6.809 0.0116 *
> pH:f 6 274 46 2.049 0.0741 .
> I:f 6 5015 836 37.544 < 2e-16 ***
> pH:I:f 6 849 142 6.356 3.82e-05 ***
> Residuals 56 1247 22
>
> ---
> Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.
>
> Now, how can I do a multi-comparison on `pH:I'?
>
> Do I need to do separate ANOVA for each `pH' or `I', just as that in
> demo("MMC.WoodEnergy", "HH")? And then do multi-comparison on `I' or `pH'
> in each separate ANOVA?
>
> Thanks again.
>
> Regards,
> Jinsong
>
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
