Examples of aligning against a common baseline, using Cleveland-style dot plots

align_common_baseline.R

# Response to a post at Storytelling with Data:
# \url{http://www.storytellingwithdata.com/blog/orytellingwithdata.com/2015/07/align-against-common-baseline.html}
# Demonstrates
#  * Cleveland-style dot plots (improvement over pie and bar charts)
#  * Sorting categorical data by a numerical variable with more than one grouping variable
#  * Highlighting differences between groups graphically

library(ggplot2)
library(scales)

# Create the base data
df <- data.frame(group = rep(c("African-American","White"), each = 6),
                 cause = rep(c("Homicide","Intoxication","Accident","Unknown","Suicide","Natural causes"), times = 2),
                 value = c(3.4, 0.8, 0.4, 0.4, 0.3, 0.3, 1.8, 0.3, 0.1, 0.1, 0.5, 0.2))

temp_df <- df[df$group == "African-American",]
levels_order <- temp_df[order(temp_df$value, decreasing = FALSE),]$cause
df$cause <- factor(df$cause, levels = levels_order)

# Create a dotplot version of the improved bar chart
ggplot(df) +
  geom_point(aes(x = cause, y = value, colour = group, shape = group)) +
  scale_shape_manual(values = c(16, 4)) +
  ylab("Deaths per 100,000 arrests by race in the U.S., 2003 - 09") +
  ggtitle("Being Arrested is Deadlier for African-Americans") +
  coord_flip() +
  theme_minimal() +
  theme(axis.title.y = element_blank(), 
        legend.background = element_rect(fill = alpha("white", 0.75), colour = "white"), 
        legend.position = c(1,0), 
        legend.justification = c(1,0))
#ggsave(filename = "~/Dropbox/Analyses/Align_Against_Common_Baseline/common_baseline_dot_plot.png", width = 6, height = 3, units = "in")

# Since the story is about the difference, create a new data frame with percent difference
# using (a - b) / b. This gives equal weight when a > b and when b > a (as opposed to,
# e.g., a / b, which crams all b > a between 0 and 1 while all a > b range > 1...inf.
df_diff <- data.frame(cause = df$cause[1:6],
                      value = (df$value[1:6] - df$value[7:12])/df$value[7:12])
levels_order <- df[order(df_diff$value, decreasing = FALSE),]$cause
df_diff$cause <- factor(df_diff$cause, levels = levels_order)

# Plot the differences, again using a Clevelend-style dot plot.
# Some work is needed on the colour gradient to ensure negative differences stand out from positive ones--the
# transition from green to red is not abrupt enough.
ggplot(df_diff) +
  geom_point(aes(x = cause, y = value, colour = value)) +
  scale_y_continuous(labels = percent) +
  scale_color_gradient2(high = muted("red"), mid = "grey50", low = muted("green"), guide = "none") +
  geom_hline(yintercept = 0, colour = "grey50") +
  ggtitle("Being Arrested is Deadlier for African Americans") +
  ylab("Percent increase in deaths per arrest for African- vs. White-Americans") +
  coord_flip() +
  theme_minimal() +
  theme(axis.title.y = element_blank())
#ggsave(filename = "~/Dropbox/Analyses/Align_Against_Common_Baseline/common_baseline_diff_plot.png", width = 6, height = 3, units = "in")