Boxplots widths too narrow if x == by

[grantmcdermott]

library(tinyplot)

tinyplot(
  Petal.Length ~ Species, iris,
  type = "boxplot"
)

tinyplot(
  Petal.Length ~ Species | Species, iris,
  type = "boxplot",
  legend = list("topleft", bty = "o")
)

The bottom boxplots should be the same width as the top ones. But they're too narrow.

I think I know why this is happening (related to a factor coercion) and should hopefully be a simple fix.

[grantmcdermott]

Oops, turning off the legend triggers a resizing (and positioning) error of its own.

tinyplot(
  Petal.Length ~ Species | Species, iris,
  type = "boxplot",
  legend = FALSE
)

I believe that this is a separate bug, albeit also limited to x==by case. I'll try to address it at the same time.

[zeileis]

Note also that you try to detect x == by only based on their levels which is not general enough. As a simple example, consider a coordination game of two players, where you want to know whether the outcome of the experiment depends on the combination of the gender of player1 and player2. To visualize the boxplots of the four groups you can use outcome ~ player1 | player2. Thus x != by but levels(x) == levels(by).

When legend = FALSE this case is handled correctly but by default it is not:

d <- data.frame(
  outcome = sin(1:40 - 3),
  player1 = factor(rep(c("female", "male"), each = 20)),
  player2 = factor(rep(rep(c("female", "male"), each = 10), 2))
)
tinyplot(outcome ~ player1 | player2, data = d)
tinyplot(outcome ~ player1 | player2, data = d, legend = FALSE)

[grantmcdermott]

Good catch.

Hmmm. I'll probably have to move this logic higher up in the code and do something like x_by = identical(x, by).

[zeileis]

Oh, and playing around with this some more: We currently cannot set any of the boxplot.default arguments, can we? For example, range, varwidth, width, notch, etc. seem to be ignored at the moment.

iris2 <- iris[c(1:30, 61:100, 101:150), ]
plot(Petal.Length ~ Species, data = iris2, varwidth = TRUE, range = 0.5, notch = TRUE)
tinyplot(Petal.Length ~ Species, data = iris2, varwidth = TRUE, range = 0.5, notch = TRUE)

But maybe this should be a separate issue?

I just noticed this because I wanted to check whether I can set the width argument of boxplot.default to address the original issue. But this fails because tinyplot has its own width argument which does something completely different (height/width of the plot).

[grantmcdermott]

Oh, and playing around with this some more: We currently cannot set any of the boxplot.default arguments, can we? For example, range, varwidth, width, notch, etc. seem to be ignored at the moment.

It's a bit hidden, but boxwex, staplewex, and outwex at least are supported through .... In this case, adjusting boxwex would have had a similar effect.

tinyplot(Petal.Length ~ Species, data = iris2, boxwex = .75)

I didn't include support for notch because I personally don't like it (see also the warnings about weird behaviour in ?boxplot) But varwidth and range are oversights; I'll add those. (I can add notch too if you would like to advocate for it.)

[zeileis]

I don't like notch either but I would still support it for compatibility with boxplot. And then maybe outline could be supported as well (even though I have never used it myself)?

[grantmcdermott]

Cool. I'll also add support for boxwidth as more direct argument alias that gets passed down to boxplot(..., width = boxwidth) internally (and so doesn't create a clash with our primary tinyplot(..., width) argument).