R primer (reference)

The same material as the browser primers, for RStudio

Use this page if the browser primers will not run on your network, or if you prefer to work in RStudio.

Everything taught in the browser primers is taught here, including the 2×2 table from Primer 7. The difference is that you run the code in your own R, so you also have to load packages yourself.

NoteRun this first

Open RStudio, then run these 2 lines in the Console.

library(dplyr)
library(ggplot2)

In the browser primers these packages are loaded for you. In RStudio they are not. Every script you write from now on starts with its library() lines.

Objects

<- puts the value on its right into the name on its left.

n_screened <- 2400
district   <- "Kolar"

n_screened
district

Nothing prints when you assign. Typing the name on its own line prints it.

Functions

A function takes arguments in brackets and returns a value.

sqrt(144)                     # 12
round(3.14159, 2)             # 3.14
round(x = 3.14159, digits = 2)  # the same, with the arguments named

Vectors

c() combines values of the same type.

cases <- c(14, 22, 31, 27, 19, 11)

length(cases)   # 6
sum(cases)      # 124
mean(cases)     # 20.66667
range(cases)    # 11 31

Arithmetic applies to the whole vector at once. There is no loop to write.

cases      <- c(14, 22, 31, 27, 19, 11)
population <- c(12000, 12000, 12500, 12500, 13000, 13000)

cases / population * 1000

Types

counts   <- c(14, 22, 31)                    # numeric
district <- c("Kolar", "Tumkur", "Hassan")   # character
exposed  <- c(TRUE, FALSE, TRUE)             # logical

class(counts)

Mixing types converts everything to the most forgiving one, usually character. That is how a column of numbers becomes text.

Comparison and counting

A comparison returns one TRUE or FALSE per element. sum() then counts, and mean() gives a proportion.

age <- c(34, 51, 22, 67, 45, 19, 58, 40)

age >= 45          # TRUE/FALSE for each
sum(age >= 45)     # 4
mean(age >= 45)    # 0.5

mean() of a comparison is prevalence in miniature.

Missing values

sbp <- c(128, 142, NA, 136, 151, NA, 119)

mean(sbp)                 # NA
mean(sbp, na.rm = TRUE)   # 135.2
sum(is.na(sbp))           # 2

na.rm = TRUE changes the question from “the mean of these 7 people” to “the mean of the 5 we measured”. Ask why the values are missing before you drop them.

Tibbles

A tibble is several equal-length vectors side by side: one row per observation, one column per variable.

library(dplyr)

participants <- tibble(
  id       = c("P01", "P02", "P03", "P04", "P05"),
  district = c("Kolar", "Kolar", "Tumkur", "Hassan", "Tumkur"),
  age      = c(34, 51, 22, 67, 45),
  smoker   = c(TRUE, FALSE, FALSE, TRUE, TRUE),
  sbp      = c(128, 142, 118, 156, 134)
)

participants

Look at any new dataset with these before you analyse it.

nrow(participants)
names(participants)
str(participants)
summary(participants)

$ pulls a column back out as a vector.

participants$age
mean(participants$sbp)
mean(participants$smoker)          # proportion of smokers
sum(participants$sbp >= 140)       # how many have raised BP

table() counts a categorical column. It hides NA unless you ask.

table(participants$district)
table(participants$district, useNA = "ifany")

A dataset to practise on

The examples below use survey. Build it by running this once.

library(dplyr)

set.seed(2026)

survey <- tibble(
  id           = sprintf("S%03d", 1:200),
  district     = sample(c("Kolar", "Tumkur", "Hassan"), 200, replace = TRUE),
  residence    = sample(c("Urban", "Rural"), 200, replace = TRUE),
  sex          = sample(c("Female", "Male"), 200, replace = TRUE),
  age          = round(runif(200, 18, 80))
) |>
  mutate(
    sbp          = round(rnorm(200, 104 + 0.38 * age, 14)),
    dbp          = round(rnorm(200, 68 + 0.14 * age, 9)),
    hypertension = sbp >= 140 | dbp >= 90
  )

glimpse(survey)

These numbers are made up. They exist so the code below has something to run on.

Packages

2 separate steps, and people mix them up.

Step Command How often
Install install.packages("dplyr") once, ever
Load library(dplyr) every session

install.packages() belongs in the Console, never in a script.

The native pipe

|> puts the value on its left into the first argument of the function on its right.

# Nested: read inside out
round(sqrt(mean(c(4, 9, 16, 25, 36))), 2)

# Piped: read left to right
c(4, 9, 16, 25, 36) |>
  mean() |>
  sqrt() |>
  round(2)

Turn on the native pipe shortcut in Tools → Global Options → Code → Use native pipe operator. The shortcut is Ctrl+Shift+M, or Cmd+Shift+M on macOS.

You will see %>% in older code. It behaves almost the same. We use |> because it is part of R itself.

The 5 dplyr verbs

Verb Does what
filter() keeps rows matching a condition
select() keeps columns you name
mutate() adds or changes a column
group_by() splits into groups
summarise() collapses each group to one row 
library(dplyr)

survey |> filter(residence == "Urban")
survey |> filter(residence == "Urban", age >= 50)
survey |> filter(district %in% c("Kolar", "Hassan"))

survey |> select(id, age, sbp)
survey |> select(-dbp)

survey |> mutate(age_group = if_else(age >= 50, "50+", "under 50"))

survey |>
  mutate(age_band = case_when(
    age < 40 ~ "under 40",
    age < 60 ~ "40-59",
    .default = "60+"
  ))

filter() drops rows where the condition is NA, silently. Count your rows before and after.

survey |>
  group_by(district) |>
  summarise(
    n            = n(),
    mean_age     = round(mean(age), 1),
    n_hypertense = sum(hypertension),
    prevalence   = round(mean(hypertension), 3)
  )

Put n() in every summarise() you write. A mean over 3 people and one over 300 look the same on the page without it.

Plots

3 pieces: the data, the aes() mapping, and a geom_.

library(ggplot2)

ggplot(survey, aes(x = age, y = sbp)) +
  geom_point()

Layers join with +, not |>.

Inside aes() means the property comes from a column. Outside aes() means it is a fixed setting for the whole layer.

ggplot(survey, aes(x = age, y = sbp, colour = residence)) +
  geom_point(size = 2, alpha = 0.6)
ggplot(survey, aes(x = sbp)) +
  geom_histogram(binwidth = 10, colour = "white")

ggplot(survey, aes(x = district, y = sbp)) +
  geom_boxplot()

ggplot(survey, aes(x = district, fill = hypertension)) +
  geom_bar(position = "dodge")

Label every plot. A title that states the finding beats one that names the axes.

ggplot(survey, aes(x = age, y = sbp, colour = residence)) +
  geom_point(size = 2, alpha = 0.6) +
  labs(
    title    = "Systolic blood pressure rises with age in both urban and rural adults",
    x        = "Age (years)",
    y        = "Systolic BP (mmHg)",
    colour   = "Residence",
    caption  = "Simulated data for teaching."
  ) +
  theme_minimal()

Epidemiological measures from a 2×2

The 4 cells are called a, b, cc and d. The third is cc because c() is a function.

Disease No disease
Exposed a b
Unexposed cc d 
a  <- 90
b  <- 910
cc <- 30
d  <- 970

risk_exposed   <- a / (a + b)          # 0.09
risk_unexposed <- cc / (cc + d)        # 0.03

risk_exposed / risk_unexposed          # risk ratio   3.00
risk_exposed - risk_unexposed          # risk difference 0.06
1 / abs(risk_exposed - risk_unexposed) # NNH  16.7 (harmful exposure)
(a * d) / (b * cc)                     # odds ratio  3.20

On the course we use epitools, which does this and gives confidence intervals as well.

library(epitools)

tab <- matrix(c(d, cc, b, a), nrow = 2, byrow = TRUE)
riskratio(tab, method = "wald")

epitools wants the unexposed row first and the non-case column first. Read the table it prints back before you trust the number.

Where to go next

Back to top