[R] plotting results from tapply

[Marc Schwartz]

On Fri, 2007-01-26 at 11:50 -0500, Michael Rennie wrote:
> Hi, there
> 
> I'm trying to plot what is returned from a call to tapply, and can't figure 
> out how to do it. My guess is that it has something to do with the 
> inclusion of row names when you ask for the values you're interested in, 
> but if anyone has any ideas on how to get it to work, that would be 
> stellar.  Here's some example code:
> 
> y1<-rnorm(40, 2)
> x1<-rep(1:2, each=20)
> x2<-rep(1:2, each=10 times=2)
> 
> ex.dat<-data.frame(cbind(y1,x1,x2))
> 
> ex.dat$x1<-factor(ex.dat$x1, labels=c("A", "B"))
> ex.dat$x2<-factor(ex.dat$x2, labels=c("C", "D"))
> 
> attach(ex.dat)
> 
> xbar<-tapply(ex.dat$y1, ex.dat[,-1], mean)
> xbar
> 
> #values I'd like to plot:
> row.names(xbar) #levels of x1
> xbar[,1] #mean response of y1 for group C (x2) across x1
> 
> #plot mean response y1 for group C against x1 (i.e., using x2 as a grouping 
> factor).
> plot(row.names(xbar), xbar[,1], ylim=range[y1])
> 
> #This is where things break down. The error message states that I need 
> "finite xlim values" but I haven't assigned anything to xlim. If I just 
> plot the data:
> 
> plot(x1, y1)
> 
> #This works fine.
> 
> #And, I can get this to work:
> 
> stripchart(xbar[1,]~row.names(xbar), vert=T)
> 
> #However, I'd like to then add the values for the second set of means 
> (i.e., mean values for group D against x1, or (xbar[,2])) to the plot.
> #I tried following up the previous command with:
> 
> points(row.names(xbar), xbar[,2])
> 
> #But that returns an error as well (NAs introduced by coercion).
> 
> 
> 
> Any suggestions?
> 
> Cheers,
> 
> Mike
> 
> PS- some of you might suggest for me to use interaction.plot, since this is 
> essentially what I'm building here. True, but I can't get error bars using 
> interaction.plot. I'm therefore trying to build my own version where I can 
> specify the inclusion of error bars. Presumably the interaction.plot has 
> figured out how to do what I'm attempting, so I have some faith that I am 
> on the right track....

Michael,

The problem is that when you are using the rownames for 'xbar', they are
a character vector:

> str(rownames(xbar))
 chr [1:2] "A" "B

When you attempt to use the same values from 'ex.dat$x1', they are
factors, which are being coerced to their numeric equivalents, so that
they can work as x axis values.

> str(ex.dat$x1)
 Factor w/ 2 levels "A","B": 1 1 1 1 1 1 1 1 1 1 ...

> as.numeric(ex.dat$x1)
 [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2
[35] 2 2 2 2 2 2



I might note using the following:

par(mfcol = c(1, 3))

with(ex.dat, plot(1:2, xbar[, 1], ylim = range(y1),
                  type = "b", col = "red",
                  lty = c("dashed", "solid"),
                  xaxt = "n", xlab = "x1",
                  ylab = "mean of y1"))

with(ex.dat, points(1:2, xbar[, 2], col = "blue",
                    type = "b"))

axis(1, at = 1:2, labels = c("A", "B"))


matplot(1:2, xbar, col = c("red", "blue"),
        pch = 21, type = "b", ylim = range(y1),
        lty = c("dashed", "solid"),
        xaxt = "n", xlab = "x1",
        ylab = "mean of y1")

axis(1, at = 1:2, labels = c("A", "B"))


with(ex.dat, interaction.plot(x1, x2, response = y1, 
                              type = "b", pch = 21,
                              col = c("red", "blue"),
                              ylim = range(ex.dat$y1),
                              legend = FALSE))



You get basically the same 3 plots, with the principal difference in
interaction.plot() being a different x axis range.

Using interaction.plot(), you can get the proper basic plot and then add
the CI's if you wish, since you know the x axis values of the mean
estimates, which is 1:NumberOfGroups (as in a boxplot).

interaction.plot() actually uses matplot() internally.

HTH,

Marc Schwartz