R help - Aug 2006 - best way to calculate per-parameter differences in across-subject means

Hello,

I have some data in a data.frame where for each of a number of
subjects, I have scores for all of a number of symptoms.

Subjects are subdivided in a number of groups, which have unequal sizes.

I'd like to plot between-group differences in the scores on the
various symptoms. Ideally, that would be in a form as would be
produced by
> bwplot( Score~Symptom )
but I'm not sure one can say anything about the distribution of
differences when the sample sizes differ as much as they do.

So I will start plotting the per-group differences in the per-symptom
mean scores.

Is there a better way (rather than using loops) to get a table of
those per-symptom means, something like
> with( subset(dat, group==1, drop=FALSE), Score~Symptom )
Thanks in advance,
Ren? Bertin

Ren? J.V. Bertin <rjvbertin <at> gmail.com> writes:
> I have some data in a data.frame where for each of a number of
> subjects, I have scores for all of a number of symptoms.
> 
> Subjects are subdivided in a number of groups, which have unequal sizes.
> 
> I'd like to plot between-group differences in the scores on the
> various symptoms. Ideally, that would be in a form as would be
> produced by
....

Maybe it's a bit more than you want, but possibly you are happy with it: see
the
example under TukeyHSD.

summary(fm1 <- aov(breaks ~ wool + tension, data = warpbreaks))
TukeyHSD(fm1, "tension", ordered = TRUE)
plot(TukeyHSD(fm1, "tension"))


Dieter

Thanks, I'll look at that.

In the meantime, the code below is what I came up with myself. It does
what I want

# SelectCases(dat,crit) == subset(dat, crit, drop=FALSE)

SENSICK.AvScores<- function( dat=SENSICK.items.tr )
{
	n<-nlevels(dat$Symptom)
	data.frame(
		Patient=c( rep(1,n), rep(0, 3*n)),
		WasSick=c( rep(1,2*n), rep(NA,2*n)),
		StrictSick=c( rep(NA,2*n), rep(-1,n), rep(1,n)),
		Symptom=rep(levels(dat$Symptom),4),
		AvScore=c(
			with( SelectCases(dat, 'Patient==1 & WasSick==1'), tapply(Score,
Symptom , mean) ),
			with( SelectCases(dat, 'Patient==0 & WasSick==1'), tapply(Score,
Symptom , mean) ),
			with( SelectCases(dat, 'Patient==0 & StrictSick==-1'),
tapply(Score, Symptom , mean) ),
			with( SelectCases(dat, 'Patient==0 & StrictSick==1'),
tapply(Score,
Symptom , mean) )
		)
	)
}

AvScores<-SENSICK.AvScores

with( AvScores, (barchart( AvScore[Patient==1] - AvScore[Patient==0 &
WasSick==1]) ~ Symptom, scales=list( rot=c(45,0)) )

------ Dieter wrote:
Maybe it's a bit more than you want, but possibly you are happy with it: see
the
example under TukeyHSD.

summary(fm1 <- aov(breaks ~ wool + tension, data = warpbreaks))
TukeyHSD(fm1, "tension", ordered = TRUE)
plot(TukeyHSD(fm1, "tension"))


Dieter