Make dotplot scale y axis as for histogram

Question

We are using dotplots in a classroom setting to introduce the histogram, because the binning concept is confusing to many students. So we start with the dotplot which is similar but more intuitive:

x <- rnorm(100)
qplot(x, geom = "bar")
qplot(x, geom = "dotplot", method="histodot")

Because students do this on their own data, the code needs to work without manual fiddling. However the geom_dotplot seems to use different scaling defaults than geom_bar. The y axis does not adjust with the data, but seems to depend only on the size of the dots. For example:

x <- runif(1000)
qplot(x, geom = "bar")
qplot(x, geom = "dotplot", method="histodot")

How can I make geom_dotplot with stat_histodot scale the y axis exactly as it would do for the histogram, either by using smaller or overlapping dots?

Answer 1

I came up with the following workaround that shrinks the binwidth until things fit on the page:

# This function calculates a default binwidth that will work better
# for the dotplot with large n than the ggplot2 default.
calculate_smart_binwidth <- function(x, aspect_ratio = 2/3){
  x <- as.numeric(x)
  nbins <- max(30, round(sqrt(length(x)) / aspect_ratio))
  range <- range(x, na.rm = TRUE, finite = TRUE)
  if(diff(range) == 0) return(NULL)
  repeat {
    message("trying nbins: ", nbins)
    binwidth <- diff(range)/nbins;
    highest_bin <- max(ggplot2:::bin(x, binwidth = binwidth)$count);
    if(highest_bin < aspect_ratio * nbins) return(binwidth)
    nbins <- ceiling(nbins * 1.03);
  }
}

Examples:

x <- runif(1e4)
qplot(x, geom="dotplot", binwidth=calculate_smart_binwidth(x))

x <- rnorm(1e4)
qplot(x, geom="dotplot", binwidth=calculate_smart_binwidth(x))