[R] interaction plot with SE bar

[Gerrit Eichner]

Hi, Kristi,

I have adapted the interaction plot example of function panel.average() of 
the lattice package and modified the code of panel.average() to a new 
function panel.loc_and_scale(). See below. (There might be a much simpler 
solution, though.)

Remark on the side: I can't resist to point you to Tatsuki Koyama's very 
instructive explanation on why plots of SE bars are not quite 
recommendable: 
http://biostat.mc.vanderbilt.edu/wiki/pub/Main/TatsukiRcode/Poster3.pdf


Nevertheless, here's the code (note that it also allows to graph other 
location +/- scale bars if you provide fun1 and fun2 with respective 
functions):

  Regards  --  Gerrit


panel.loc_and_scale <- function(x, y, fun1 = mean, fun2 = sd,
                                  horizontal = TRUE,
                                  lwd = add.line$lwd, lty = add.line$lty,
                                  col, col.line = add.line$col, ...) {
     x <- as.numeric(x)
     y <- as.numeric(y)
     add.line <- trellis.par.get( "add.line")
     if (!missing(col)) {
         if (missing(col.line))
             col.line <- col
     }
     if (horizontal) {
         vals <- unique(sort(y))
         yy <- seq_along(vals)
         xx1 <- xx2 <- numeric(length(yy))
         for (i in yy) {
          xx1[i] <- fun1(x[y == vals[i]])
          xx2[i] <- fun2(x[y == vals[i]])
          }
         panel.arrows( x0 = xx1 - xx2, y0 = vals[yy],
                       x1 = xx1 + xx2, y1 = vals[yy],
                       code = 3, angle = 90, length = 0.1,
                       col = col.line, lty = lty, lwd = lwd)
         panel.points( x = xx1, y = vals[yy], col = col, pch = 18)

     }
     else {
         vals <- unique(sort(x))
         xx <- seq_along(vals)
         yy1 <- yy2 <- numeric(length(xx))
         for (i in xx) {
          yy1[i] <- fun1(y[x == vals[i]])
          yy2[i] <- fun2(y[x == vals[i]])
          }
         panel.arrows( x0 = vals[xx], y0 = yy1 - yy2,
                       x1 = vals[xx], y1 = yy1 + yy2,
                       code = 3, angle = 90, length = 0.1,
                       col = col.line, lty = lty, lwd = lwd)
         panel.points( x = vals[xx], y = yy1, col = col, pch = 18)
     }
}


stripplot(yield ~ site, barley, groups = year,  # fun1 = median, fun2 = IQR,
        panel = function(x, y, groups, subscripts, ...) {
            panel.grid(h = -1, v = 0)
            panel.stripplot(x, y, ..., jitter.data = TRUE,
                            groups = groups, subscripts = subscripts)
            panel.superpose(x, y, ..., panel.groups = panel.loc_and_scale,
                            groups = groups, subscripts = subscripts)
            panel.superpose(x, y, ..., panel.groups = panel.average,
                            groups = groups, subscripts = subscripts)
        },
        auto.key = list(points = FALSE, lines = TRUE, columns = 2))



> Hi R user,
> I am just wondering how I can add Standard error bar in the interaction 
> plot. I used the following code but I don't know how i can edit this to 
> put a started error's bar on the mean.
> Would you give me some hints? or do other packages provide the 
> information about plotting the SE bar for interaction.plot?
>
> require(graphics)
> with(OrchardSprays, {
>  interaction.plot(treatment, rowpos, decrease)
>  interaction.plot(rowpos, treatment, decrease, cex.axis = 0.8)
>  ## order the rows by their mean effect
>  rowpos <- factor(rowpos,
>                   levels = sort.list(tapply(decrease, rowpos, mean)))
>  interaction.plot(rowpos, treatment, decrease, col = 2:9, lty = 1)
> })
>
> thanks