Prerequisites

If you are not experienced with R, it is strongly advised that you read-up on and more importantly test out R and RStudio before attempting analyse road crash data with R. See the stats19-training-setup vignette at https://docs.ropensci.org/stats19/articles/stats19-training-setup.html for guidance on getting started with R, RStudio and installing R packages.

The completing the course requires that the following packages, which can be installed with install.packages(), can be loaded as follows:

library(pct)      # access travel data from DfT-funded PCT project 
library(sf)       # spatial vector data classes
library(stats19)  # get stats19 data
library(stplanr)  # transport planning tools
library(tidyverse)# packages for 'data science'
library(tmap)     # interactive maps

You should type, run and ensure you understand each line of code in this document.

#> Warning in utils::citation(..., lib.loc = lib.loc): no date field in DESCRIPTION
#> file of package 'stats19'
#> Warning in utils::citation(..., lib.loc = lib.loc): could not determine year for
#> 'stats19' from package DESCRIPTION file

Projects and scripts

Projects are a way to organise related work together. Each project has its own folder and Rproj file. Advice: always working from projects will make your life easier! Start a new project with:

File > New Project You can choose to create a new directory (folder) or associate a project with an existing directory. Make a new project called stats1-course and save it in a sensible place on your computer. Notice that stats1-course now appears in the top right of RStudio.

Scripts are the files where R code is stored. Keeping your code in sensibly named, well organised and reproducible scripts will make your life easier: you could simply type all our code into the console, but that require retyping commands each time you run it. Instead, code that you want to keep and share should be saved script files, plain text files that have the .R extension.

Make a new script with Flie > New File > Rscript or Ctrl+Shift+N

Save the script and give it a sensible name like stats19-lesson-1.R with File > Save, the save button on the toolbar, or Ctrl+S.

Pro tip: You can also create new R scripts by typing and running this command in the R console:

file.edit("stats19-lesson-1.R")

Keeping scripts and other files associated with a project in a single folder per project (in an RStudio project) will help you find things you need and develop an efficient workflow.

Writing and running code

Let’s start with some basic R operations. Write this code into your new stats19-lesson-1.R R script and execute the result line-by-line by pressing Ctrl+Enter

x = 1:5
y = c(0, 1, 3, 9, 18)
plot(x, y)

This code creates two objects, both are vectors of 5 elements, and then plots them (bonus: check their length using the length() function). Save the script by pressing Ctrl+S.

There are several ways to run code within a script and it is worth becoming familiar with each. Try running the code you saved in the previous section using each of these methods:

1. Place the cursor in different places on each line of code and press Ctrl+Enter to run that line of code.
2. Highlight a block of code or part of a line of code and press Ctrl+Enter to run the highlighted code.
3. Press Ctrl+Shift+Enter to run all the code in a script.
4. Press the Run button on the toolbar to run all the code in a script.
5. Use the function source() to run all the code in a script e.g. source("stats19-lesson-1.R")

Pro tip: Try jumping between the console and the source editor by pressing Ctl+1 and Ctl+2.

Viewing Objects

Create new objects by typing and running the following code chunk in a new script, e.g. called objects.R.

vehicle_type = c("car", "bus", "tank")
casualty_type = c("pedestrian", "cyclist", "cat")
casualty_age = seq(from = 20, to = 60, by = 20)
set.seed(1)
dark = sample(x = c(TRUE, FALSE), size = 3, replace = TRUE)
small_matrix = matrix(1:24, nrow = 12)
crashes = data.frame(vehicle_type, casualty_type, casualty_age, dark)

We can view the objects in a range of ways:

1. Type the name of the object into the console, e.g. crashes and small_matrix, and run that code. Scroll up to see the numbers that didn’t fit on the screen.
2. Use the head() function to view just the first 6 rows e.g. head(small_matrix)
3. Bonus: use the n argument in the previous function call to show only the first 2 rows of small_matrix
4. Click on the crashes object in the environment tab to View it in a spreadsheet.
5. Run the command View(vehicle_type). What just happened?

We can also get an overview of an object using a range of functions, including summary(), class(), typeof(), dim(), and length().

You can, for example, view a summary of the casualty_age variable by running the following line of code:

summary(casualty_age)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
#>      20      30      40      40      50      60

Exercise try these functions on each of the objects, what results do they give?

Bonus: Find out the class of the column vehicle_type in the data frame crashes with the command class(crashes$vehicle_type). Why has it changed? Create a new object called crashes_char that keeps the class of the character vectors intact by using the function tibble::tibble() (see tibble.tidyverse.org and Section 4 for details).

Autocompletion

RStudio can help you write code by autocompleting it. RStudio will look for similar objects and functions after typing the first three letters of a name.

When there is more than one option you can select from the list using the mouse or arrow keys. Within a function, you can get a list of arguments by pressing Tab.

Getting help

Every function in R has a help page. You can view the help using ? for example ?sum. Many packages also contain vignettes, these are long form help documents containing examples and guides. vignette() will show a list of all the vignettes available, or you can show a specific vignette for example vignette(topic = "sf1", package = "sf").

Commenting Code

It is good practice to use comments in your code to explain what it does. You can comment code using #

For example:

# Create vector objects (a whole line comment)
x = 1:5 # a seqence of consecutive integers (inline comment)
y = c(0, 1, 3, 9, 18.1) 

You can comment/uncomment a whole block of text by selecting it and using Ctrl+Shift+C.

Pro tip: You can add a comment section using Ctrl + Shift + R

The global environment

The Environment tab shows all the objects in your environment, this includes datasets, parameters, and any functions you have created. By default, new objects appear in the Global Environment but you can see other environments with the drop-down menu. For example, each package has its own environment.

Sometimes you wish to remove things from your environment, perhaps because you no longer need them or things are getting cluttered.

You can remove an object with the rm() function e.g. rm(x) or rm(x, y) or you can clear your whole environment with the broom button on the Environment Tab.

1. Remove the object x that was created in a previous section.
2. What happens when you try to print the x by entering it into the console?
3. Try running the following commands in order: save.image(); rm(list = ls()); load(".RData"). What happened?
4. How big (how many bytes) is the .RData file in your project’s folder?
5. Tidy up by removing the .Rdata file with file.remove(".Rdata").

Debugging Code

All the code shown so far is reproducible. To test RStudio’s debugging features, let’s write some code that fails, as illustrated in the figure below.

1. What is the problem with the code shown in the figure?
2. Create other types of error in the code you have run (e.g. no symetrical brackets and other typos)
3. Does RStudio pick up on the errors? And what happens when you try to run buggy code?

Always address debugging prompts to ensure your code is reproducible

Saving R objects

We have already seen that you can save R scripts. You can also save individual R objects in the RDS format.

saveRDS(crashes, "crashes.Rds")

We can also read back in our data.

crashes2 = readRDS("crashes.Rds")
identical(crashes, crashes2)
#> [1] TRUE

R also supports many other formats, including CSV files, which can be created and imported with the functions readr::read_csv() and readr::write_csv() (see also the readr package).

readr::write_csv(crashes, "crashes.csv")
crashes3 = readr::read_csv("crashes.csv")
identical(crashes3, crashes) 

Notice that crashes3 and crashes are not identical, what has changed? Hint: read the help page associated with ?readr::write_csv.

Subsetting by index or name

Subsetting returns part of an R object. It can be done by providing numbers representing the positions of the elements we want (e.g. the 2^nd element) or with a logical vector, with values associated with TRUE returned. Two dimension object such as matrices and data frames can be subset by rows and columns. Subsetting in base R is done with square brackets [] after the name of an object. Run the following commands to practice subsetting.

casualty_age[2:3] # second and third casualty_age
crashes[c(1, 2), ] # first and second row of crashes
crashes$vehicle_type # returns just one column
crashes[, c("casualty_type", "casualty_age")] # first and third columns

1. Use the $ operator to print the dark column of crashes.
2. Subset the crashes with the [,] syntax so that only the first and third columns of crashes are returned.
3. Return the 2^nd row and the 3^rd column of the crashes dataset.
4. Return the 2^nd row and the columns 2:3 of the crashes dataset.
5. Bonus: what is the class() of the objects created by each of the previous exercises?

Subsetting by values

It is also possible to subset objects by the values of their elements. This works because the [ operator accepts logical vectors returned by queries such as ‘is it less than 3?’ (x < 3 in R) and ‘was it light?’ (crashes$dark == FALSE), as demonstrated below:

x[c(TRUE, FALSE, TRUE, FALSE, TRUE)] # 1st, 3rd, and 5th element in x
x[x == 5] # only when x == 5 (notice the use of double equals)
x[x < 3] # less than 3
x[x < 3] = 0 # assign specific elements
casualty_age[casualty_age %% 6 == 0] # just the ages that are a multiple of 6
crashes[crashes$dark == FALSE, ]

1. Subset the casualty_age object using the inequality (<) so that only elements less than 50 are returned.
2. Subset the crashes data frame so that only tanks are returned using the == operator.
3. Bonus: assign the age of all tanks to 61.

Dealing with NAs and recoding

R objects can have a value of NA. This is how R represents missing data.

z = c(4, 5, NA, 7)

NA values are common in real-world data but can cause trouble, for example

sum(z) # result is NA

Some functions can be told to ignore NA values.

sum(z, na.rm = TRUE) # result is equal to 4 + 5 + 7

You can find NAs using the is.na() function, and then remove them

is.na(z)
z_nona = z[!is.na(z)] # note the use of the not operator !
sum(z)

If you remove records with NAs be warned: the average of a value excluding NAs may not be representative.

Changing class

Sometimes you may want to change the class of an object. This is called class coercion, and can be done with functions such as as.logical(), as.numeric() and as.matrix().

1. Coerce the vehicle_type column of crashes to the class character.
2. Coerce the crashes object into a matrix. What happened to the values?
3. Bonus: What is the difference between the output of summary() on character and factor variables?

Recoding values

Often it is useful to ‘recode’ values. In the raw STATS19 files, for example, -1 means NA. There are many ways to recode values in R, the simplest and most mature of which is the use of factors, as shown below:

z = c(1, 2, -1, 1, 3)
l = c(NA, "a", "b", "c") # labels in ascending order
z_factor = factor(z, labels = l)
z_charcter = as.character(z_factor)
z_charcter
#> [1] "a" "b" NA  "a" "c"

1. Recode z to Slight, Serious and Fatal for 1:3 respectively.
2. Bonus: read the help file at ?dplyr::case_when and try to recode the values using this function.

Now you are ready to use R

Bonus: reproduce the following plot

eyes = c(2.3, 4, 3.7, 4)
eyes = matrix(eyes, ncol = 2, byrow = T)
mouth = c(2, 2, 2.5, 1.3, 3, 1, 3.5, 1.3, 4, 2)
mouth = matrix(mouth, ncol = 2, byrow = T)
plot(eyes, type = "p", main = "RRR!", cex = 2, xlim = c(1, 5), ylim = c(0, 5))
lines(mouth, type = "l", col = "red")

What are packages?

R has over 15,000 packages (effectively plugins for base R), extending it in almost every direction of statistics and computing. Packages provide additional functions, data and documentation. They are very often written by subject-matter experts and therefore tend to fit well with the workflow of the analyst in that particular specialism. There are two main stages to using a package: installing it and loading it. A third stage is updating it, this is also important.

Install new packages from The Comprehensive R Archive Network with the command install.packages() (or remotes::install_github() to install from GitHub). Update packages with the command update.package() or in Tools > Check for Package Updates in RStudio. You only need to install a package once.

install.packages("sf")
# remotes::install_github("r-spatial/sf")

Installed packages are loaded with the command library(). Usually, the package will load silently. In some cases the package will provide a message, as illustrated below.

library(sf)
#> Linking to GEOS 3.9.1, GDAL 3.3.2, PROJ 7.2.1

To use a function in a package without first loading the package, use double colons, as shown below (this calls the tibble() function from the tibble package).

crashes_tibble = tibble::tibble(
  vehicle_type,
  casualty_type,
  casualty_age,
  dark
)

1. Take a look in the Packages tab in the Files pane in RStudio (bottom right by default).
2. What version of the stats19 package is installed on your computer?
3. Run the command update.packages(). What happens? Why?

ggplot2

Let’s take a look at a particular package. ggplot2 is a generic plotting package that is part of the ‘tidyverse’ meta-package, which is an “opinionated collection of R packages designed for data science.” All packages in the tidyverse “share an underlying design philosophy, grammar, and data structures.” ggplot2 is flexible, popular, and has dozens of add-on packages which build on it, such as gganimate. To plot non-spatial data, it works as follows (see figure below, left for result):

library(ggplot2)
ggplot(crashes) + geom_point(aes(x = casualty_type, y = casualty_age))

Note that the + operator adds layers onto one another.

1. Install a package that build on ggplot2 that begins with with gg. Hint: enter install.packages(gg) and hit Tab when your cursor is between the g and the ).
2. Open a help page in the newly installed package with the ?package_name::function() syntax.
3. Attach the package.
4. Bonus: try using functionality from the new ‘gg’ package building on the example above to create plots like those shown below (hint: the right plot below uses the economist theme from the ggthemes package, try other themes).

dplyr and pipes

Another useful package in the tidyverse is dplyr. It provides functions for manipulating data frames and using the pipe operator %>%. The pipe puts the output of one command into the first argument of the next, as shown below (note the results are the same):

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
class(crashes)       
#> [1] "data.frame"
crashes %>% class()
#> [1] "data.frame"

Useful dplyr functions are demonstrated below.

crashes %>%
  filter(casualty_age > 50) # filter rows
crashes %>%
  select(casualty_type) # select just one column
crashes %>%
  group_by(dark) %>% 
  summarise(mean_age = mean(casualty_age))

1. Use dplyr to filter row in which casualty_age is less than 18, and then 28.
2. Use the arrange function to sort the crashes object in descending order of age (hint: see the ?arrange help page).
3. Read the help page of dplyr::mutate(). What does the function do?
4. Use the mutate function to create a new variable, birth_year, in the crashes data.frame which is defined as the current year minus their age.
5. Bonus: Use the %>% operator to filter the output from the previous exercise so that only observations with birth_year after 1969 are returned.

sf objects

All road crashes happen somewhere and, in the UK at least, all collisions recorded by the police are given geographic coordinates, something that can help prioritise interventions to save lives by intervening in and around ‘crash hotspots.’ R has strong geographic data capabilities, with the sf package provides a generic class for spatial vector data: points, lines and polygons, are represented in sf objects as a special ‘geometry column,’ typically called ‘geom’ or ‘geometry,’ extending the data frame class we’ve already seen in crashes.

Create an sf data frame called crashes_sf as follows:

library(sf) # load the sf package for working with spatial data
crashes_sf = crashes # create copy of crashes dataset
crashes_sf$longitude = c(-1.3, -1.2, -1.1)
crashes_sf$latitude = c(50.7, 50.7, 50.68)
crashes_sf = st_as_sf(crashes_sf, coords = c("longitude", "latitude"), crs = 4326)
# plot(crashes_sf[1:4]) # basic plot
# mapview::mapview(crashes_sf) # for interactive map

1. Plot only the geometry column of crashes_sf (hint: the solution may contain $geometry). If the result is like the figure below, congratulations, it worked!).
2. Plot crashes_sf, only showing the age variable.
3. Plot the 2^nd and 3^rd crashes, showing which happened in the dark.
4. Bonus: How far are the points apart (hint: sf functions begin with st_)?
5. Bonus: Near which settlement did the tank runover the cat?

Reading and writing spatial data

You can read and write spatial data with read_sf() and write_sf(), as shown below (see ?read_sf).

write_sf(zones, "zones.geojson") # save geojson file
write_sf(zones, "zmapinfo", driver = "MapInfo file")
read_sf("zmapinfo") # read in mapinfo file

See Chapter 6 of Geocomputation with R for further information.

sf polygons

Note: the code beyond this point is not evaluated in the vignette:

knitr::opts_chunk$set(eval = FALSE)

sf objects can also represent administrative zones. This is illustrated below with reference to zones, a spatial object representing the Isle of Wight, that we will download using the pct package (note: the [1:9] appended to the function selects only the first 9 columns).

zones = pct::get_pct_zones("isle-of-wight")[1:9]

1. What is the class of the zones object?
2. What are its column names?
3. Print its first 2 rows and columns 6:8 (the result is below).

Spatial subsetting and sf plotting

Like index and value subsetting, spatial subsetting can be done with the [ notation. Subset the zones that contain features in crashes_sf as follows:

zones_containing_crashes = zones[crashes_sf, ]

To plot a new layer on top of an existing sf plot, use the add = TRUE argument. Remember to plot only the geometry column of objects to avoid multiple maps. Colours can be set with the col argument.

1. Plot the geometry of the zones, with the zones containing crashes overlaid on top in red.
2. Plot the zone containing the 2^nd crash in blue.
3. Bonus: plot all zones that intersect with a zone containing crashes, with the actual crash points plotted in black.

Geographic joins

Geographic joins involve assigning values from one object to a new column in another, based on the geographic relationship between them. With sf objects it works as follows:

zones_joined = st_join(zones[1], crashes_sf)

1. Plot the casualty_age variable of the new zones_joined object (see the figure below to verify the result).
2. How many zones are returned in the previous command?
3. Select only the geo_code column from the zones and the dark column from crashes_sf and use the left = FALSE argument to return only zones in which crashes occured. Plot the result.

See Chapter 4 of Geocomputation with R (Lovelace, Nowosad, and Muenchow 2019) for further information on geographic joins.

CRSs

Get and set Coordinate Reference Systems (CRSs) with the command st_crs(). Transform CRSs with the command st_transform(), as demonstrated in the code chunk below, which converts the ‘lon/lat’ geographic CRS of crashes_sf into the projected CRS of the British National Grid:

crashes_osgb = st_transform(crashes_sf, 27700)

1. Try to subset the zones with the crashes_osgb. What does the error message say?
2. Create zones_osgb by transforming the zones object.
3. Bonus: use st_crs() to find out the units measurement of the British National Grid?

For more information on CRSs see Chapter 6 of Geocompuation with R.

Buffers

Buffers are polygons surrounding geometries of a (usually) fixed distance. Currently buffer operations in R only work on objects with projected CRSs.

1. Find out and read the help page of sf’s buffer function.
2. Create an object called crashes_1km_buffer representing the area within 1 km of the crashes.
3. Bonus: try creating buffers on the geographic version of the crashes_sf object. What happens?

Attribute operations on sf objects

Because sf objects are data.frames, we can do non-spatial operations on them. Try the following attribute operations on the zones data.

# load example dataset if it doesn't already exist
zones = pct::get_pct_zones("isle-of-wight")
sel = zones$all > 3000  # create a subsetting object
zones_large = zones[sel, ] # subset areas with a popualtion over 100,000
zones_2 = zones[zones$geo_name == "Isle of Wight 002",] # subset based on 'equality' query
zones_first_and_third_column = zones[c(1, 3)]
zones_just_all = zones["all"]

1. Practice subsetting techniques you have learned on the sf data.frame object zones:
   1. Create an object called zones_small which contains only regions with less than 3000 people in the all column
   2. Create a selection object called sel_high_car which is TRUE for regions with above median numbers of people who travel by car and FALSE otherwise
   3. Create an object called zones_foot which contains only the foot attribute from zones
   4. Bonus: plot zones_foot using the function plot to show where walking is a popular mode of travel to work
   5. Bonus: bulding on your answers to previous questions, use filter() from the dplyr package to subset small regions where car use is high.
2. Bonus: What is the population density of each region (hint: you may need to use the functions st_area(), as.numeric() and use the ‘all’ column)?
3. Bonus: Which zone has the highest percentage of people who cycle?

Matching roads to crashes

I think you forgot something here. For example we could introduce st_nearest_feature? Or counting using st_within and st_buffer.