How to plot a rating scale in R

Cleaning the data

In order to solve your problem, we have to transform your data, in order to convert it into tidy format.

Observation

There are few particular problems with your original dataset:

• Data are in a wide format, i.e. most of the columns from your data frame, can be represented by 3 variables;
• Names of the variables are not self-explanatory. Names are in upper case which, by itself, does not hold any useful information, they are not readable and not good for typing/writing.
• There is additional information we can extract from the variable names: Party and Feelings toward the Party. First one is an abbreviation ('dem' or 'rep') second one is the numerically encoded feeling towards the political party. However the order of numbers encoding the feeling does not reflect natural order of emotions from the disgust up to joy;
• Variable PARTYID_Strength is numerically encoded Political Party [self-]Identification it also does not reflect natural order from strongest democrats through independent towards strongest republicans;

Plan

1. Convert data from wide into long format using all variables starting with TRAIT, and leaving PARTYID_Strength variable unchanged;
2. Extract useful information from the TRAIT... variables (Political Party, Feelings Toward the Party);
3. Convert all numerically encoded variables into the factors with reasonably ordered levels;
4. Give all variables meaningful names;
5. Summarize the data;

Transformations

We need to create several lookup tables, which will simplify the workflow.

Affiliation lookup table:

aff_lookup <- c(
  'Strong Democrat',
  'Not very strong Democrat',
  'Strong Republican',
  'Not very strong Republican',
  'Independent',
  'Independent-Democrat',
  'Independent-Republican',
  'Other'
)

We can further order aff_lookup by this vector:

aff_order = c(1, 2, 6, 5, 7, 4, 3, 8)

Emotions/Feelings lookup table:

emo_lookup <- c(
  'Delighted',    
  'Angry',
  'Happy',
  'Annoyed',
  'Joy',
  'Hateful',
  'Relaxed',
  'Disgusted'
)

And we can order emo_lookup by this vector:

emo_order <- emo_order <- c(8, 6, 2, 4, 7, 3, 1, 5)

Political party lookup table:

party_lookup <- c(
  dem = 'National Democratic Party',
  rep = 'National Republican Party'
)

Finally, with all helper variables, we can transform our data into desirable form.

library(tidyverse)

dat %<>%
  rename_all(tolower) %>%
  pivot_longer(
    cols          = starts_with('trait'),
    names_to      = c('party', 'emotion'),
    names_pattern = 'trait(dem|rep)(\\d)',
    values_to     = 'score'
  ) %>%
  mutate(
    party = factor(party_lookup[party]),
    affiliation = factor(
      aff_lookup[partyid_strength], 
      levels = aff_lookup[aff_order]
      ),
    emotion = factor(
      emo_lookup[as.numeric(emotion)], 
      levels = emo_lookup[emo_order]
      )
  ) %>%
  group_by(party, emotion, affiliation) %>%
  summarise(score = median(score)) %>%
  ungroup()

head(dat)

## A tibble: 6 x 4
#  party                     emotion   affiliation                score
#  <fct>                     <fct>     <fct>                      <dbl>
#1 National Democratic Party Disgusted Strong Democrat                1
#2 National Democratic Party Disgusted Not very strong Democrat       2
#3 National Democratic Party Disgusted Independent-Democrat           2
#4 National Democratic Party Disgusted Independent                    3
#5 National Democratic Party Disgusted Independent-Republican         3
#6 National Democratic Party Disgusted Not very strong Republican     5

Plot the data

Plan

Now we can plot the data, as two separate plots for Democrats and Republicans with Affiliation (Political Party Identification) on X-axis and Emotions (Feelings) on Y-axis.

Each Emotion/Affilation point is going to be represented as a bar with the height of the bar representing the Score.

We can also add color encoding to our plot. From my point of view, encoding Emotions/Feelings with a color gradient from red (Disgust) to green (Joy) could help as to gather the internal structure of our data.

Plot

dat %>%
  ggplot(
    aes(
      x      = affiliation, 
      y      = as.numeric(emotion) +  (score / max(score) * .95) / 2, 
      height = (score / max(score) * .95), 
      width  = .95,
      fill   = emotion,
      label  = score
      )
    ) +
  geom_tile(show.legend = FALSE) +
  geom_text(size = 3.5, color = 'gray25', alpha = .75) +
  facet_wrap(~ party, scales = 'free') +
  scale_fill_brewer(palette = 'RdYlGn') +
  scale_y_continuous(breaks = sort(emo_order), labels = emo_lookup[emo_order]) +
  labs(x = 'Affiliations', y = 'Emotions') +
  ggthemes::theme_tufte() +
  theme(
    axis.text.x  = element_text(angle = 45, hjust = 1),
    axis.ticks.x = element_blank(),
    axis.text.y  = element_text(hjust = 0, vjust = -0.025),
    axis.ticks.y = element_blank()
  )

Which gives as following figure:

Explanation

There is a trick with this plot: it looks like a series of barplots, bot it is not real barplots (by the fact, not functionally).

What I do:

The core of this solution is the use of geom_tile() for each data point. It is just a rectangle (square by default) with geometrical center of mass determined by the given coordinates (Affilation, Emotion).

Both Affilation and Emotion are factors, not numerics. And it is OK for Affiliation, because we want only to position our tile according to the Affiliation it represents.

It is more complicated with Emotion, because we want to position each tile according to the Emotion it represents, but also we want to encode Score by the height of the tile.

To define the height of the tile we use height parameter within the aes(). We want our tile height to be less or equall to one (with 0.05 offset) so the tiles between let say Angry and Annoyed do not overlap. That's why we use (score / max(score) * .95 for the height parameter.

We also need to give different y-coordinates for each tile, so the center of the tile is placed not on the imaginary line representing each emotion, but half-height up. So when tile is drawn, it's center (on y-axis) is placed half-height up from the "base line" and the tile extends half-height up and down, creating a fake barplot. That's what the following line of code does as.numeric(emotion) + (score / max(score) * .95) / 2.

We also give a tile a fixed width of .95 by width = .95, file the tile with Red-Yellow-Green gradient and lable each tile with the relevant Score.

The rest are just decorations. However, note how we relable the Y-axis. Because, as it defined in aes() it is continuous scale, but we want to make it fake discrete axis we use this row:

scale_y_continuous(breaks = sort(emo_order), labels = emo_lookup[emo_order])

Here we just use our emo_order to say that we want breaks for integers from 1 to 8, and after that we label this breaks with feelings from ordered emo_lookup table.