Plotting Likert-scale data with ggplot2, tidyverse, and lemon.

The following graph was presented in Avisen.dk on March 22nd 2018.

Hvor enig eller uenig er du i følgende udsagn:

“Vi har som samfund råd til at give de offentligt ansatte en mærkbar lønforhøjelse under forhandlingerne om nye overenskomster?”

Kilde: Wilke for Avisen.dk (marts 2018) og Analyse Danmark for Avisen.dk (oktober 2017). Note: I undersøgelsen foretaget af Wilke deltog 1.014 respondenter, mens 2.680 deltog i undersøgelsen for Analyse Danmark.

This translates to:

As a society, we can afford to pay public servants a considerable raise during the negotiations of the collective (union) agreements??

The article's topic was to compare the shift in the proportion of agreeing participants between October '17 and March '18. I thought the given visualisation could be approved upon; not only showing an increasing proportion of agree participants, but also that the increase was drawn from every other item.

I started by using divergent stacked bars, laid horizontally, as described in e.g. Data Revelations (and other posts I cannot recall). The issue with this, is the proportion of 'Don't know's (“Ved ikke”). Should they be excluded? Do they bundle up with the neutrals (“Hverken-eller”)? In this instance, I opted to put the 'Don't know's aside, but not excluded. The comparison was also partly to compare the agreeable vs. the disagreeable. The neutrals where grouped with the disagreeable; this puts the divergence in the centre of the plot. Keeping the same colours as the source, the result is as below:

Using less space, I would argue it is easier to digest.

To improve it further, the horizontal y-axis should have a split where the scale restarts between black and dark grey, and dark grey and grey.

library(dplyr)
library(tibble)
library(tidyr)
library(lemon)
 
df <- tribble(
  ~stat, ~Marts_2018, ~October_2017,
  'enig', 66, 50,
  'hverken-eller', 14, 18,
  'uenig', 15, 20,
  'ved-ikke', 5, 12)
 
 
(p <- df %>% gather(year, val, Marts_2018, October_2017) %>%
    mutate(ymin=case_when(.$stat == 'uenig' ~ 18,
                          .$stat == 'ved-ikke' ~ 18 + 20,
                          TRUE ~ 0),
           year=c(October_2017='Oktober 2017', Marts_2018='Marts 2018')[year]) %>%
    mutate(y = case_when(.$stat == 'enig' ~ - .$val, 
                         TRUE ~ .$val + .$ymin)) %>%
    mutate(stat=factor(stat, levels=c('enig','hverken-eller','uenig','ved-ikke'))) %>%
    ggplot(aes(x=year, xend=year, y=y, yend=ymin, colour=stat)) + geom_segment(size=10) +
    coord_flex_flip() +
    scale_y_continuous('Procent', position='top', breaks=c(-60,-45,-30,-15,0,15,18,33,38,53), labels=c(60,45,30,15,0,15,0,15,0,15)) +
    scale_colour_manual(values=c(enig=rgb(235, 96, 29, maxColorValue = 255), 'hverken-eller'=gray(0.15), uenig=grey(0.5), 'ved-ikke'=grey(0.8)),
                        labels=c(enig='Helt eller delvist enig', 'hverken-eller'='Hverken-eller', uenig='Helt eller delvist uenig', 'ved-ikke'='Ved ikke')) +
    theme(panel.grid.minor.x = element_blank(), panel.grid.major.y = element_blank(), axis.title.y=element_blank(), 
          axis.ticks =  element_blank(),
          legend.title=element_blank(), legend.position='bottom',
          panel.border = element_blank(), axis.line.x = element_line()) +
    labs(caption='Wilke for Avisen.dk (marts 2018, n=1.014) og Analyse Danmark (oktober 2017, n=2.680)\nhttps://www.avisen.dk/danskerne-i-ny-maaling-der-er-raad-til-loenstigninge_490703.aspx')
)