Practical: Styling

In this practical, you will visualize the development of wealth in Basel’s quarters across time and learn how to make your visualization prettier.

0 - Preliminaries

1. If you haven’t done so already, load the tidyverse and the taxation.csv data.

library(tidyverse)
basel <- read_csv('1_data/taxation.csv')

1 - Plot base

1. Begin by creating a plot showing, as lines (geom_line()), the relationship of wealth_median (y) and years (x) for each of the quarters (col).

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX

2 - Themes

1. Make the plot prettier by adding a theme_*(). I like theme_minimal(), but you can go for a different one if you like, e.g., theme_light().

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX

3 - Theme

1. Now use theme() to fix the legend. I bet on your screen the legend is also taking up more space than the actual plot. Move the legend to the bottom using legend.position = "bottom".

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX)

2. That’s better. Now you may find that the legend is not fitting entirely into the plotting area. If the case, use legend.title = element_blank() and legend.text = element_text(size = 7) to make it fit as I’ve done in the presentation.

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX)

4 - Labs

1. Ok, the plot itself is pretty decent already. What’s missing, is appropriate annotation. Use labs() to add titles for the axes, a plot title and subtitle, and a caption.

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX")

5 - Colors

1. Add scale_color_viridis_d to change the color set to the popular viridis set. Note the d at the end. This stands for discrete as opposed to continuous, a distinction made for most scaling functions. discrete should be used whenever you are scaling a categorical variable and continuous whenever you are scaling a continuous variable. Here we are scaling a categorical variable: quarter.

basel %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX") + 
  XX

6 - Factors

1. To change the order of categorical variables, one must rely on factors. The two data wrangling lines added below make sure that the factor levels in quarter are ordered according to wealth_median in year 2001. The as_factor function btw is part of the forcats package, which exists exactly to solve these kinds of problems.

basel %>% 
  arrange(year, wealth_median) %>% 
  mutate(quarter = as_factor(quarter)) %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX") + 
  XX

7 - Aggregation

1. Let’s see whether the richer quarters saw a larger increase in wealth than the poorer quarters. To this end, use the dplyr code below to create a new grouping variable that codes whether a quarter belongs to the richer or poorer quarters in 2001.

basel <- basel %>% 
  arrange(year, wealth_median) %>% 
  group_by(quarter) %>%
  mutate(wealth_2001 = first(wealth_median)) %>% 
  ungroup() %>% 
  mutate(rich = wealth_2001 > median(wealth_2001))

2. Next, use some more dplyr to calculate mean and standard deviations of medians as a function of the newly-created variable rich and year.

averages <- basel %>% 
  group_by(XX, XX) %>%
  summarize(wealth_median_mean = mean(XX),
            wealth_median_sd = sd(XX))

3. OK, now, use the code from above to create a plot showing lines and points for the average wealth median of rich and poor quarters.

averages %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX") + 
  XX

4. It seems the rich quarters have indeed grown faster in wealth than poor quarters. Add some error bars geom_errorbar reflecting confidence intervals. To this end divide the standard deviation by sqrt(10) and then multiply by qt(.975, 10).

averages %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  XX +
  XX + 
  geom_errorbar(mapping = aes(ymin = XX - XX/XX*XX,
                              ymax = XX - XX/XX*XX)) +
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX") + 
  XX

5. Looks like there is a credible difference, which is not so surprising when you think about it. We effectively selected the groups based on the dependent variable. Nevertheless, the results underpin that the relative wealth of Basel’s quarters has been very constant, with rich quarters staying rich (and getting richer) and poor quarters staying poor. If you want a slightly prettier version of the plot, try geom_ribbon() instead or in addition to geom_errorbar(). Since geom_ribbon() plots an area, it is better placed before the points and lines. Furthermore, for the same reason, you have control the color using scale_fill_*.

averages %>% 
  ggplot(aes(x = XX,
             y = XX,
             col = XX)) + 
  geom_ribbon(mapping = aes(ymin = XX - XX/XX*XX,
                            ymax = XX - XX/XX*XX),
              alpha = .5) +
  XX +
  XX + 
  XX(XX = XX,
     XX = XX,
     XX = XX) +
  XX(XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX",
     XX = "XX") + 
  XX + scale_fill_viridis_d()