Allow stacked bars for data containing negative y-values in `type_barplot()`

[634523]

When plotting a stacked bar chart with negative values, tinyplot doesn't stack them the way ggplot2 does for example. Currently for data that contains negative y-values the following warning is shown:

Warning message:
In settings$type_data(settings, ...) :
  'beside' must be TRUE if there are negative 'y' values

and the bars are put beside. For some data it would make sense to show the bars stacked even if some values are negative (e.g. balances like below):

rm(list = ls())

library(tinyplot)
library(worldbank)
library(ggplot2)

df <- wb_data(
  indicator = c("BN.GSR.GNFS.CD", "BN.GSR.FCTY.CD", "BN.TRF.CURR.CD"),
  country = "US"
)

# tinyplot
plt(
  value ~ date | indicator_id,
  data = df,
  type = "barplot"
)

# ggplot2
ggplot(df, aes(fill=indicator_id, y=value, x=date)) + 
  geom_bar(position="stack", stat="identity")

[zeileis]

Hmm, if the signs of the value were always the same within each indicator_id group, then I can see the value of this plot. If the signs change (as in your example), then I find the plot somewhat hard to interpret. This is brought out more clearly when we focus on the first and on the last years:

I think it would be reasonably easy to implement the stacked height as the absolute sum of all heights and then center by the sum of the negative heights. But the reordering of the categories is not completely obvious in the code. Let's see what Grant @grantmcdermott says when he's back from vacation.

[grantmcdermott]

TBH, I'd defer to your judgement here @zeileis!

@634523 not to dismiss your proposal, but I'm also struggling to think of a case where these make good sense from a visual/interpretability sense. As Achim points out, if there is a reversal of sign for different series, then the ordering switches and is hard to reason about IMO.

However, note that you kind of get what you want by providing a (zero) offset:

plt(
  value ~ date | indicator_id,
  data = df,
  type = "barplot", offset = 0,
  lighten = FALSE, fill = 0.3,  # optional: transparent fill to see overlaps
  theme = "clean"               # optional: theme
)
lines(tapply(df$value, df$date, sum), lwd = 2, col = "gray20")  # optional: highlight cumulative amount

^{Created on 2026-07-27 with reprex v2.1.1}

Alternatively, maybe this another case for a stacked area plot (#632)?

[634523]

@grantmcdermott @zeileis
I agree that it's not the most intuitive chart, but many institutions use it nonetheless. For my example, I chose the current account because it's a very common way of presenting it there (e.g. BEA, ONS (Figure 5), ECB).

Thanks for the hint of using the offset = 0 argument, this is indeed already resulting in what I wanted.

[zeileis]

Thanks for the links to the examples @634523

Grant @grantmcdermott, given that this seems to be commonly used in practice, I think we should try to support it, even if it is a bit hard to read. I'm not sure, though, how to best implement this because we can have these changes of the position of certain bar segments.