stacked (relative) bar chart: silly stacking order

[salim-b]

If I want to plot a stacked bar chart of mixed positive and negative values, say to recreate a likert plot, the order given in the relevant factor fed to plot_ly's name parameter is not displayed in the chart as one would expect!

Below you find a reprex which does

1. create some sample data

2. plot a bar chart without stacking -> order is fine

3. plot a bar chart with stacking (barmode = "relative") -> category very good comes before not so good while I would expect the opposite (same order as in barmode = "normal")!

As a workaround I could reorder the factor levels before plotting, but then the legend entry order is messed up. Since there's no way to fully customize the order of legend entries, reordering factor levels seems a dead end.

Now this problem might not be specifically R-related, but present in the underlying plotly.js library, I can't tell. Therefore I thought it'd make most sense to report it here and maybe you can bring it up upstream if necessary :)

---

Brief description of the problem

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union

# create example dataset
data_example <- data_frame(question = c("q1", "q1", "q1", "q1",
                                        "q2", "q2", "q2", "q2",
                                        "q3", "q3", "q3", "q3",
                                        "q4", "q4", "q4", "q4"),
                           category = factor(x = c(0, 1, 2, 3,
                                                   0, 1, 2, 3,
                                                   0, 1, 2, 3,
                                                   0, 1, 2, 3),
                                             labels = c("very bad",
                                                        "not so bad",
                                                        "not so good",
                                                        "very good"),
                                             ordered = TRUE),
                           percentage = c(-12, -22, 45, 21,
                                          -33, -7, 41, 19,
                                          -50, -27, 13, 10,
                                          -1, -11, 56, 32))

# print example dataset
data_example
#> # A tibble: 16 x 3
#>    question category    percentage
#>    <chr>    <ord>            <dbl>
#>  1 q1       very bad           -12
#>  2 q1       not so bad         -22
#>  3 q1       not so good         45
#>  4 q1       very good           21
#>  5 q2       very bad           -33
#>  6 q2       not so bad          -7
#>  7 q2       not so good         41
#>  8 q2       very good           19
#>  9 q3       very bad           -50
#> 10 q3       not so bad         -27
#> 11 q3       not so good         13
#> 12 q3       very good           10
#> 13 q4       very bad            -1
#> 14 q4       not so bad         -11
#> 15 q4       not so good         56
#> 16 q4       very good           32

# plot bar chart without stacking
data_example %>%
    plotly::plot_ly(type = "bar",
                    orientation = "h",
                    name = ~category,
                    x = ~percentage,
                    y = ~question) %>%
    plotly::layout(barmode = "normal",
                   xaxis = list(title = NA,
                                ticksuffix = " %"),
                   yaxis = list(title = NA))

# plot bar chart with stacking (relative mode)
data_example %>%
    plotly::plot_ly(type = "bar",
                    orientation = "h",
                    name = ~category,
                    x = ~percentage,
                    y = ~question) %>%
    plotly::layout(barmode = "relative",
                   xaxis = list(title = NA,
                                ticksuffix = " %"),
                   yaxis = list(title = NA))

^{Created on 2018-11-03 by the reprex package (v0.2.1)}

[cpsievert]

I think layout(., legend = list(traceorder = "reversed")) does what you want? If not, please be more specific about what ordering you expect.

[salim-b]

I think layout(., legend = list(traceorder = "reversed")) does what you want?

Nope, not at all. The order in the legend is fine (as long as I don't reorder the factor levels) but at the same time the order in the stacked bar chart is not!

If not, please be more specific about what ordering you expect.

On the right side of the zero line the blue area (very good) comes before the orange area (not so good) while I would expect it the other way round (as per factor level ordering). I have to admit that in the end, the whole plot is a bit of a hack (negative values) to recreate a likert plot like this one. Maybe I should have labelled the different levels very good > rather good > rather bad > very bad, so that the ordering would be more intuitive – sorry for that.

But do you agree that the order changes from barmode = normal (orange comes before blue) to barmode = stacked (orange comes after blue)? That's my main point: This shouldn't be the case!

[cpsievert]

But do you agree that the order changes from barmode = normal (orange comes before blue) to barmode = stacked (orange comes after blue)?

No, not if you take the approach that the first levels should start from the origin, which I believe the way most visualization tools interpret it.

[salim-b]

No, not if you take the approach that the first levels should start from the origin (...)

What do you mean by "origin"? The zero line? Then I have to disagree because very good is a "pole" level (the highest one) and is placed right next to the zero line while the other "pole" level very bad (the lowest one) is not placed right next to the zero line... I can't see any sense in that.

[YsoSirius]

The ordering is correct if you include type="category" to the xaxis options, but then the layout changes.

[salim-b]

The ordering is correct if you include type="category" to the xaxis options, but then the layout changes.

Well...

The order of the categories is fine indeed, but everything else isn't much more than "modern art" 🤣