The Example Dataset: The Counties of North Carolina

When I sat down to write this post I considered several different datasets before eventually deciding to use the North Carolina Sudden Infant Death Syndrome (SIDS) dataset (nc). I’ll admit that may seem like an odd choice for an example, as the data are related neither to the focus of my career (urban planning) nor to the place I live (Seattle, Washington). I basically have no connection to these data, so why use them in a post about creating graphics?

I want to introduce nc for three reasons:

1. Easy Access: It’s easy to access because it comes bundled in the simple features (sf) package.¹
2. Relatable Geometries: The county boundaries are polygon geometry objects, which are one of the more common types of spatial objects and are easy for viewers to understand.
3. Reprex-ible: The dataset is small and simple enough to be used in a reprex², which means that it can be easily adapted to illustrate a problem and used get help from others.

But given my lack of public health expertise I doubt I’ll be able to produce an informative graphic with the SIDS data contained in the dataset.³ Instead, I’ll search for more accessible data (which can be joined to nc’s spatial geometries) in my go-to open data portal:

Prompted by the Wikipedia search bar autocomplete, I find that the “List of counties in North Carolina” page contains a table with several columns that might be interesting to visualize in a plot:

With the spatial data from nc and the descriptive data from the Wikipedia table, I’m ready to begin the process of creating a data visualization graphic.

Loading the data

We’ll use functions from several packages during this post, so the first step is to load the packages:

library(ggrepel)
library(patchwork)  
library(rvest)
library(janitor)  
library(sf)
library(tidyverse) 

As I said before, the nc dataset gets installed as one of the example datasets in the sf package. The data is stored in the package’s gpkg/ directory⁴, so we’ll begin by loading it with read_sf() and taking a look at the data with glimpse(), my preview function of choice⁵:

nc <- read_sf(system.file("gpkg/nc.gpkg",package = "sf")) 

glimpse(nc)
## Observations: 100
## Variables: 15
## $ AREA      <dbl> 0.114, 0.061, 0.143, 0.070, 0.153, 0.097, 0.062, 0.0...
## $ PERIMETER <dbl> 1.442, 1.231, 1.630, 2.968, 2.206, 1.670, 1.547, 1.2...
## $ CNTY_     <dbl> 1825, 1827, 1828, 1831, 1832, 1833, 1834, 1835, 1836...
## $ CNTY_ID   <dbl> 1825, 1827, 1828, 1831, 1832, 1833, 1834, 1835, 1836...
## $ NAME      <chr> "Ashe", "Alleghany", "Surry", "Currituck", "Northamp...
## $ FIPS      <chr> "37009", "37005", "37171", "37053", "37131", "37091"...
## $ FIPSNO    <dbl> 37009, 37005, 37171, 37053, 37131, 37091, 37029, 370...
## $ CRESS_ID  <int> 5, 3, 86, 27, 66, 46, 15, 37, 93, 85, 17, 79, 39, 73...
## $ BIR74     <dbl> 1091, 487, 3188, 508, 1421, 1452, 286, 420, 968, 161...
## $ SID74     <dbl> 1, 0, 5, 1, 9, 7, 0, 0, 4, 1, 2, 16, 4, 4, 4, 18, 3,...
## $ NWBIR74   <dbl> 10, 10, 208, 123, 1066, 954, 115, 254, 748, 160, 550...
## $ BIR79     <dbl> 1364, 542, 3616, 830, 1606, 1838, 350, 594, 1190, 20...
## $ SID79     <dbl> 0, 3, 6, 2, 3, 5, 2, 2, 2, 5, 2, 5, 4, 4, 6, 17, 4, ...
## $ NWBIR79   <dbl> 19, 12, 260, 145, 1197, 1237, 139, 371, 844, 176, 59...
## $ geom      <MULTIPOLYGON [°]> MULTIPOLYGON (((-81.47276 3..., MULTIP...

After staring at the glimpse() output for a bit, I’ll hazard a guess that the variables can be divided into four categories:

My plan is to join the rows from the Wikipedia table to the nc rows using the county name as the join key, so I’ll probably only keep the NAME and geom columns and the rest can be dropped.

Now I’ll scrape the Wikiepedia page using the rvest package and prepare the column names for joining:

url <- "https://en.wikipedia.org/wiki/List_of_counties_in_North_Carolina"

nc_county_table <- read_html(url) %>% 
  html_node("table.wikitable") %>% 
  html_table(header = TRUE) %>% 
  as_tibble %>% 
  clean_names(case = "screaming_snake") %>% # remove unwanted characters from names and applies a specific case
  rename_all(~ str_remove_all(.x, "_\\d+")) %>% 
  mutate(NAME = str_remove(COUNTY," County")) %>% # transform COUNTY to match nc$NAME
  select(NAME, FIPS_CODE:MAP) 

glimpse(nc_county_table)
## Observations: 100
## Variables: 9
## $ NAME                <chr> "Alamance", "Alexander", "Alleghany", "Ans...
## $ FIPS_CODE           <int> 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23,...
## $ COUNTY_SEAT         <chr> "Graham", "Taylorsville", "Sparta", "Wades...
## $ CREATED             <int> 1849, 1847, 1859, 1750, 1799, 1911, 1712, ...
## $ ORIGIN              <chr> "Orange County", "Caldwell County, Iredell...
## $ ETYMOLOGY           <chr> "The Battle of Alamance which was derived ...
## $ POPULATION_ESTIMATE <chr> "7005153291000000000<U+2660>153,291", "7004370870...
## $ AREA                <chr> "7002435000000000000<U+2660>435 sq mi\n(700311270...
## $ MAP                 <lgl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA...

The two datasets can now be joined:

nc_ready <- nc %>% 
  select(NAME) %>% 
  full_join(nc_county_table, by = "NAME") %>% 
  rename(YEAR = CREATED) %>% # YEAR is a clearer column name
  st_sf

glimpse(nc_ready)
## Observations: 100
## Variables: 10
## $ NAME                <chr> "Ashe", "Alleghany", "Surry", "Currituck",...
## $ FIPS_CODE           <int> 9, 5, 171, 53, 131, 91, 29, 73, 185, 169, ...
## $ COUNTY_SEAT         <chr> "Jefferson", "Sparta", "Dobson", "Currituc...
## $ YEAR                <int> 1799, 1859, 1771, 1668, 1741, 1759, 1777, ...
## $ ORIGIN              <chr> "Wilkes County", "Ashe County", "Rowan Cou...
## $ ETYMOLOGY           <chr> "Samuel Ashe (1725–1813), a Revolutionary ...
## $ POPULATION_ESTIMATE <chr> "7004271430000000000<U+2660>27,143", "70041105200...
## $ AREA                <chr> "7002427000000000000<U+2660>427 sq mi\n(700311060...
## $ MAP                 <lgl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA...
## $ geom                <MULTIPOLYGON [°]> MULTIPOLYGON (((-81.47276 3....

A couple columns appear to have encoding issues (POPULATION_ESTIMATE, AREA, and MAP), but otherwise the data look as I expected them to.

Brainstorming

Now I’ll begin probing the data for plot ideas. The COUNTY_SEAT column catches my attention first (cities!), but I’m then drawn to YEAR which seems to be the year that each county was formed. Visualizations of temporal data are appealing because they naturally lend themselves to narrative creation. The years in this column tell the story of which counties are old and which are new (relatively speaking), as well as the general pattern of county formation

Let’s have a look at the distribution of the years when the counties were created:

nc_ready %>% 
  st_set_geometry(NULL) %>% 
  ggplot(data = ., aes(x = YEAR)) +
  geom_histogram() + 
  theme_min_blogdown() + # a theme customized to match the aesthetics of this blog
  theme(axis.title = element_blank())

The histogram above suggests a major increase around 1775 and another smaller one near 1850. Increasing the bins parameter might clarify exactly when these counties were added:

years <- max(nc_ready$YEAR) - min(nc_ready$YEAR) + 1

nc_ready %>% 
  st_set_geometry(NULL) %>% 
  ggplot(data = ., aes(x = YEAR)) +
  geom_histogram(bins = years) + 
  scale_y_continuous(breaks = 0:10, minor_breaks = NULL) +
  theme_min_blogdown() + 
  theme(axis.title = element_blank()) 

While not an attractive chart, it clarifies the difference between the spike in counties around 1776 and the small but steady additions seen in the mid-1800s.

This period of rapid growth seems worth exploring as a topic for a data visualization.⁶ While the histogram above clearly illustrates the number of counties added each year, I’m curious if there is a visually-richer way to tell this story.

I like the linear x-axis because it is a familiar way to representing time. The y-axis, however, feels like an opportunity for improvement. I’ll see what this pattern of county creation looks like when represented as a cumulative count rather than an individual one:

nc_year <- nc_ready %>% 
  st_set_geometry(NULL) %>% 
  group_by(YEAR) %>% 
  summarise(NAME = list(NAME) %>% map_chr(str_c, collapse = ", "),
            COUNT = sum(n())
            ) %>% 
  mutate(COUNT_CUMULATIVE = cumsum(COUNT)) %>% 
  arrange(YEAR)

print(nc_year, n=5)
## # A tibble: 55 x 4
##    YEAR NAME                                      COUNT COUNT_CUMULATIVE
##   <int> <chr>                                     <int>            <int>
## 1  1668 Currituck, Pasquotank, Perquimans, Chowan     4                4
## 2  1705 Craven                                        1                5
## 3  1712 Beaufort, Hyde                                2                7
## 4  1722 Bertie, Carteret                              2                9
## 5  1729 Tyrrell, New Hanover                          2               11
## # ... with 50 more rows

ggplot(data = nc_year, aes(x = YEAR, y = COUNT_CUMULATIVE)) +
  geom_col() +
  theme_min_blogdown() +
  theme(axis.title = element_blank())

This is a definite improvement! The lines act like a rug plot and a bar chart at the same time, drawing attention to the period of dramatic county creation between 1776 - 1779.

Refining the Draft

I’d like to take this draft and run with it, but before I start tweaking the ggplot2 parameters I’ll do some quick stetches on paper to figure out what the final product should look like. I’ve found this step helpful because it provides me with a clear end goal which reduces the amount of time spent messing around with minor changes.

Here’s my sketch of what I’ll try to create:

The idea is to create a combination plot with the county boundaries placed above the county creation history bar chart. I’ll use a label to emphasize the spike in 1779 and distinguish those counties from the rest by giving them a different fill color. The American Revolutionary War will be illustrated with a simple gray rectangle behind the county data.

With this goal in mind, I’ll now implement it using the ggplot2 and patchwork packages. Rather than showing all the iterations I’ll go through, I’ll just list the methods and functions that make it work:

• Polygon Plot:
  • Highlighting the counties created in 1779 could be achieved by mapping a variable to the fill aesthetic, but in this case it’s easier to just add a second geom_sf() layer to the plot which only contains the 1779 county geometries.
  • The datum lines are hidden using coord_sf(datum = NA)⁷
• Step Plot:
  • Augmenting a timeseries plot with a “callout” bar is a common tactic used by economic and financial data analysts.⁸ ggplot’s layering system makes it relatively straightforward to create this type plot by combining geom_step() and geom_col().
  • The label for the year with the most counties is created using group_by(YEAR) and summarize(), followed by an if_else() that replaces all labels other than the year 1779 with "".
  • The hollow circle point highlighting the 1779 step is created by taking advantage of R’s point shape options⁹ and ggplot’s scale_shape_identity() function.¹⁰
  • The label for the Revolutionary War is created by adding an additional plot that only contains the geom_text() layer and is combined with the rest of the step plot using the patchwork package. This is a little heavy-handed but it gets the job done.
  • The y-axis ticks/labels do not match the panel grid. Normally the ticks and labels are show at the major breaks, but in this plot I want the ticks/labels to show only the min and max while the panel grid will have typical set of major and minor breaks. The way I achieve this is pretty hacky: set major breaks at the min and max, hide the grid panel, and add geom_hline() and geom_vline() layers that mimic a grid panel but do not correspond with the major breaks.
  • I experimented with adding a fill color to the area beneath the geom_step() line but decided against it because it didn’t add any information.¹¹
• Rug plot:
  • While geom_rug exists, I don’t like the way the lines clutter the x-axis, so I create a similar effect using geom_col. To do this I add a column to the data (COUNT_RUG) which has a value of 4 for every year that a county was added.
  • In order to better distinguish the years were many counties were added, I map the COUNT variable to geom_col( alpha = COUNT) and set the scale with scale_alpha(range = c(0.2, 0.8)).¹²
• Custom themes:
  • In order to match the typography of this blog, I create custom ggplot2 themes and install the fonts using the extrafont package.
  • The polygon plot should have anything except the polygons, so I create a version of theme_void which I call theme_void_blogdown(). The filled step plot needs axes, labels, and ticks, so a verison of theme_minimal is created (theme_min_blogdown()) and used to style this plot.¹³
• Multi-Plot Layout:
  • The patchwork package makes it trivial to combine multiple plots - highly recommended!
  • I arrange the plots with plot_layout() and fine-tune the sizes using the heights and widths parameters
  • Labels (title, subtitle, caption) are added using plot_annotation()

Banging Out the Code

The following code chunks show the steps needed to transform the data and create the composite data graphic. Note: for those interested in running the code on their own computer I recommend copying the entire script from the gist.

 max_year <- nc_ready %>% 
  count(YEAR, sort = TRUE) %>% 
  slice(1) %>% 
  pluck("YEAR") 

max_year_header <- str_c(max_year,":", sep = "")

nc_year <- nc_ready %>% 
  group_by(YEAR) %>% 
  summarise(COUNTY_SEAT = list(COUNTY_SEAT),
            NAME = list(NAME),
            COUNT = sum(n()),
            COUNT_RUG = as.integer(COUNT > 0) * 4
            ) %>% 
  mutate(MAX_LGL = COUNT == as.integer(max(COUNT)),
         CUMULATIVE = cumsum(COUNT), 
         LABEL = map_chr(NAME, str_c, collapse = "\n"),
         LABEL = if_else(MAX_LGL, str_c(max_year_header, LABEL, sep = "\n"), ""),
         SHAPE = if_else(MAX_LGL, 1, 32 ) %>% as.integer()
         )   
# Check out the result:
print(st_set_geometry(nc_year, NULL),n=3) 
## # A tibble: 55 x 9
##    YEAR COUNTY_SEAT NAME    COUNT COUNT_RUG MAX_LGL CUMULATIVE LABEL SHAPE
## * <int> <list>      <list>  <int>     <dbl> <lgl>        <int> <chr> <int>
## 1  1668 <chr [4]>   <chr [~     4        4. FALSE            4 ""       32
## 2  1705 <chr [1]>   <chr [~     1        4. FALSE            5 ""       32
## 3  1712 <chr [2]>   <chr [~     2        4. FALSE            7 ""       32
## # ... with 52 more rows

war <- tibble(
  WAR_NAME = "Revolutionary War",
  START = as.integer(1775),
  MID = as.integer(1779),
  END = as.integer(1783)
)

# Setup

title <- "1779: The Year of Many Counties"

subtitle <- "In the middle of the American Revolutionary War, North Carolina created nine new counties\nin a single year - more than any other year in the state's history."

caption <- "Source: Wikipedia <https://en.wikipedia.org/wiki/List_of_counties_in_North_Carolina>"

black_15pct <- adjustcolor( "black", alpha.f = 0.15)

black_15pct_no_opacity <- rgb(218,218,218, max = 255)

highlight_red <- "#f03838"
 
labels <- c("First 4 Counties", "All 100 Counties")

x_intercepts_major <- seq(1700,1900, by = 50)

x_intercepts_minor <- seq(1675,1875, by = 50)

y_intercepts_major <- seq(50,100, by = 50)

y_intercepts_minor <- seq(25,75, by = 50)
 
year_limits_min <- min(nc_year$YEAR)

year_limits_max <- max(nc_year$YEAR)

# Step plot
gg_step <- ggplot(data = nc_year, aes(x = YEAR, y = CUMULATIVE, label = LABEL)) + 
  geom_vline(xintercept = x_intercepts_major, alpha = .25, size = .25) +
  geom_vline(xintercept = x_intercepts_minor, alpha = .15, size = .15) +
  geom_hline(yintercept = y_intercepts_major, alpha = .25, size = .25) +
  geom_hline(yintercept = y_intercepts_minor, alpha = .15, size = .15) + 
  geom_col(mapping = aes(x = YEAR, y = COUNT_RUG, alpha = COUNT)) + 
  scale_alpha(range = c(0.2, 0.8)) +
  geom_rect(inherit.aes = FALSE,
                    data = war, 
                    mapping = aes(xmin = START, xmax = END, ymin = 0, ymax = Inf), 
                    fill = black_15pct) + 
  geom_step(size = 1) +   
  scale_shape_identity() +   
  geom_point(data = nc_year, aes(x = YEAR, y = CUMULATIVE, shape = SHAPE), fill = "white", size = 3, color = highlight_red) +
  geom_label_repel(min.segment.length = .1, segment.colour = "black",point.padding = 1.5,nudge_x = -1, nudge_y = 1, fill = "white", color = "black", size = 2.5,label.size = NA, box.padding = 0.15, family = "Work Sans", fontface = "plain", angle = 0, lineheight = 1.15) +
  theme_min_blogdown() + 
  scale_y_continuous(breaks = c(4,100), labels = labels) + 
  theme(legend.position = "none",
        panel.grid = element_blank(), 
        axis.title = element_blank(), 
        axis.ticks = element_line(size = .5),
        axis.line = element_line(size = .5)) 

# Step plot war label

gg_label <- ggplot(data = war, aes(x = MID, y = 0, label = WAR_NAME)) +
  scale_x_continuous(limits = c(year_limits_min, year_limits_max)) +
  geom_text(size = 2.5,nudge_x = 0,nudge_y = 0, family = "Work Sans", fontface = "plain",
                                      color = "black",
                                      hjust = 0.5, vjust = 0.5, angle = 0, lineheight = 0.9) +
  theme_void() 



# Map plot
gg_map <- ggplot() + 
  geom_sf(data = filter(nc_ready, YEAR != max_year), fill = black_15pct_no_opacity, color = "black", size = .1) + 
  geom_sf(data = filter(nc_ready, YEAR == max_year), fill = "black", color = black_15pct_no_opacity, size = .25) +
  coord_sf(datum = NA) + 
  theme_void_blogdown() 

# Combined plot
gg_plot <-  gg_map + 
  {gg_label + gg_step + plot_layout(ncol = 1, heights = c(1,20))} + 
  plot_layout(ncol = 1, heights = c(1,1.15)) + # composing plots like this is made possible by the patchwork pacakge 
  plot_annotation(title, subtitle, caption,
                  theme = theme_min_blogdown()
        )  

Final Result

After all of that, the plot is finally complete! Let’s take a look:

gg_plot

Conclusion

As I said at the top, polished plots are a ton of work!

When it comes to data visualization, you need to be able to take control of nearly every detail of the image. Packages like ggplot2 and patchwork are powerful tools, but that doesn’t means it’s easy to create a graphic exactly as you envision it. You have to come up with creative workarounds within the limitations of the tools, and to do that you need to become good at searching the documentation and community resources for the help.

I hope that this description of my approach serves as a useful reference to other R users and people who create data visualizations.

DIY & Feedback

If you’d like to reproduce this example on your own computer please download the gist.

Questions, feedback, and suggested improvements are always welcome!

Resources

• Data Visualization by Claus O. Wilke
• ggplot2 References
• patchwork Github repo README
• ggrepel Github repo README