Logistics

• When: Thursday June 11, start of class (9:35 a.m.)
• How long: 75 minutes
• Points: 100 total, 8 parts (A-H)
• Closed book

You may assume tidyverse and palmerpenguins are loaded. You’ll see these datasets: penguins, mpg, diamonds, msleep.

Format breakdown:

Study advice

The exam tests whether you can read, write, and reason about code. Some study suggestions:

• For multiple choice, know why each answer is correct or incorrect, not just the letter
• For predict-the-output, run the expressions yourself and understand the rule
• For code writing, practice writing from scratch instead of just reading
• For debugging think about “what rule does this code violate?”

Assignment and objects

The standard assignment operator is <-. = works too, but <- is the course convention and the idiomatic R style.

x <- 7.3 # preferred
x

[1] 7.3

x = 8.1 # works, not preferred
x

[1] 8.1

1.2 -> x # also works, really not preferred
x

[1] 1.2

== is a comparison, not assignment; it tests whether two things are equal and returns TRUE or FALSE.

x == 1.2     # is x equal to 1.2?

[1] TRUE

x == pi    # is x equal to pi (~3.14)?

[1] FALSE

install.packages() vs library()

Data frames

A data frame is a two-dimensional table:

• Each row is one case/observation
• Each column is one variable
• Columns can each hold a different type (numeric, character, logical, factor, etc.)
• Within a column, every value is the same type

This is what distinguishes a data frame from a matrix (all values must be the same type) and from a vector (one-dimensional).

glimpse(penguins)  # see the structure: rows, columns, types

Rows: 344
Columns: 8
$ species           <fct> Adelie, Adelie, Adelie, Adelie, Adelie, Adelie, Adel…
$ island            <fct> Torgersen, Torgersen, Torgersen, Torgersen, Torgerse…
$ bill_length_mm    <dbl> 39.1, 39.5, 40.3, NA, 36.7, 39.3, 38.9, 39.2, 34.1, …
$ bill_depth_mm     <dbl> 18.7, 17.4, 18.0, NA, 19.3, 20.6, 17.8, 19.6, 18.1, …
$ flipper_length_mm <int> 181, 186, 195, NA, 193, 190, 181, 195, 193, 190, 186…
$ body_mass_g       <int> 3750, 3800, 3250, NA, 3450, 3650, 3625, 4675, 3475, …
$ sex               <fct> male, female, female, NA, female, male, female, male…
$ year              <int> 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007, 2007…

Named vs positional arguments

Functions take arguments. You can pass them positionally (by order) or by name.

round(3.14159, 2)              # positional: x first, digits second

[1] 3.14

round(x = 3.14159, digits = 2) # named

[1] 3.14

round(digits = 2, x = 3.14159) # named; shows that order doesn't matter

[1] 3.14

Named arguments are safer and more readable, but both styles are correct R.

Arithmetic and vectors

R operates element-wise on vectors of the same length:

c(1, 2, 3) + c(10, 20, 30)

[1] 11 22 33

c(2, 4, 6) * 3

[1]  6 12 18

Key operators to know:

17 %% 5  # remainder when 17 is divided by 5

[1] 2

17 %/% 5 # how many times does 5 go into 17?

[1] 3

Practice: predict the output

Write your answer before clicking.

(a) sqrt(81)

(b) c(3, 6, 9) - c(1, 2, 3)

(c) 10 %% 3

(d) seq(0, 12, by = 3)

(e) length(c("alpha", "beta", "gamma", "delta"))

(f) mean(c(5, 10, 15, 20), na.rm = TRUE)

NA and na.rm

NA in R means “not available”, i.e., a missing value. Almost any calculation with NA returns NA:

mean(c(2, 4, NA, 8))               # NA

[1] NA

mean(c(2, 4, NA, 8), na.rm = TRUE) # drop NAs first

[1] 4.666667

Any function that summarizes values across a vector, such as mean(), sum(), min(), max(), sd(), accepts na.rm = TRUE.

Mapping vs setting aesthetics

# color varies by species (mapping):
ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm)) +
  geom_point(aes(color = species))

# every point is red (setting):
ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm)) +
  geom_point(color = "red")

Glyphs

In the grammar of graphics, a glyph is the basic mark that represents one case.

• In a scatterplot: one dot per row
• In a bar chart: one bar segment per group
• In a line plot: one point connected by a line segment

The glyph is what geom_* functions draw. Each row of data produces one glyph (or contributes to one group glyph).

Choosing the right geom

Zooming: coord_cartesian() vs xlim()

Both zoom in on a plot, but the difference matters when you have fitted curves:

• xlim(0, 50) drops data outside the range before any computation, which changes what a smoother sees
• coord_cartesian(xlim = c(0, 50)) only adjusts the viewport. Here, all data is still used (you’re just zooming in).

# Safer zoom since it doesn't affect statistics:
ggplot(mpg, aes(x = displ, y = hwy)) +
  geom_point() +
  geom_smooth() +
  coord_cartesian(xlim = c(1.5, 4))

Position adjustments

With geom_bar(aes(x = drv, fill = class)):

"fill" is best when you want to compare proportions across groups.

Facets

Faceting splits the plot into panels, one per level of a variable.

facet_wrap(~ class)    # one variable; auto-wraps into a grid
facet_grid(drv ~ year) # drv in ROWS, year in COLUMNS

Key rules for facet_grid(row_var ~ col_var):

• Left of ~ \(\rightarrow\) rows
• Right of ~ \(\rightarrow\) columns
• Number of panels = levels(row_var) \(\times\) levels(col_var)
• By default all panels share the same axis scales

# 3 drv values × 2 year values = 6 panels
ggplot(mpg, aes(x = hwy)) +
  geom_histogram(bins = 15) +
  facet_grid(drv ~ year)

Practice: ggplot debugging

What is wrong with this code? There are two separate problems.

ggplot(diamonds, aes(x = carat, y = price))
  geom_point()

ggplot(diamonds, aes(x = carat, y = price)) +
  geom_point()

Practice: skewed data

A scatterplot of msleep with bodywt on the x-axis and sleep_total on the y-axis shows almost all points crammed into the bottom-left corner, with one or two animals far to the right.

(a) Why does this happen?

(b) Write one line of code to fix it:

+ scale_x_log10()

The six core verbs

group_by() + summarise() always go together.

Reduction functions

A reduction (or summary) function takes a whole column (vector) and returns one value:

mean(c(10, 20, 30))
sum(c(10, 20, 30))
n() # only works inside summarise(): counts rows in the group

Compare to a transformation function — it returns one value per row:

c(10, 20, 30) * 2  # three inputs, three outputs

[1] 20 40 60

log(c(10, 20, 30)) # still three outputs

[1] 2.302585 2.995732 3.401197

Use reductions in summarise(). Use transformations in mutate().

Pipeline order matters

The pipe (|>) feeds the output of one step into the next. Order determines what each step sees.

# CORRECT: filter rows first, then summarise
penguins |>
  filter(body_mass_g > 4000) |>
  summarise(avg = mean(body_mass_g))

# WRONG: summarise removes the rows — nothing left to filter
penguins |>
  summarise(avg = mean(body_mass_g)) |>
  filter(body_mass_g > 4000) # body_mass_g no longer exists!

Reading code: anatomy of a pipeline

mpg |>
  filter(class == "compact") |>
  group_by(manufacturer) |>
  summarise(avg_hwy = mean(hwy))

Be able to identify each piece:

• Data table: mpg
• Functions: filter, group_by, summarise, mean, == (operator)
• Variables (columns): class, manufacturer, hwy, and the new avg_hwy
• Constants: "compact" (a character string literal)
• Verb that keeps only some rows: filter()

Practice: write a pipeline

Write a pipeline on penguins that:

1. Keeps only rows where sex is not NA
2. Groups by species
3. Reports the count (n) and mean flipper length (avg_flipper) per species
4. Sorts from longest to shortest average flipper

Try it before looking.

penguins |>
  filter(!is.na(sex)) |>
  group_by(species) |>
  summarise(n = n(), avg_flipper = mean(flipper_length_mm, na.rm = TRUE)) |>
  arrange(desc(avg_flipper))

# A tibble: 3 × 3
  species       n avg_flipper
  <fct>     <int>       <dbl>
1 Gentoo      119        217.
2 Chinstrap    68        196.
3 Adelie      146        190.

Practice: pipes and nested calls

Nested \(\rightarrow\) pipe: Rewrite with |>, one step per line.

head(arrange(filter(mpg, year == 2008), desc(hwy)), 3)

mpg |>
  filter(year == 2008) |>
  arrange(desc(hwy)) |>
  head(3)

Pipe \(\rightarrow\) nested: Rewrite without the pipe.

penguins |>
  filter(!is.na(bill_length_mm)) |>
  summarise(avg_bill = mean(bill_length_mm))

summarise(filter(penguins, !is.na(bill_length_mm)), avg_bill = mean(bill_length_mm))

The over-grouping pitfall

penguins |>
  group_by(species, island, sex) |>
  summarise(avg_mass = mean(body_mass_g, na.rm = TRUE), .groups = "drop") |>
  arrange(desc(avg_mass)) |>
  head(3)

# A tibble: 3 × 4
  species island sex    avg_mass
  <fct>   <fct>  <fct>     <dbl>
1 Gentoo  Biscoe male      5485.
2 Gentoo  Biscoe female    4680.
3 Gentoo  Biscoe <NA>      4588.

The top row has the highest average, but how many penguins is it based on?

Wide vs narrow (long)

The same data can be stored in two shapes:

Wide: one row per student, one column per exam:

student   midterm   final
Ava            88      91
Ben            72      85
Cy             95      90

Narrow (long): one row per measurement:

student   exam       score
Ava       midterm       88
Ava       final         91
Ben       midterm       72
Ben       final         85
Cy        midterm       95
Cy        final         90

Wide has 3 rows and 3 columns. Narrow has 6 rows and 3 columns. They contain the same data, just in a different shape.

pivot_longer() and pivot_wider()

# Wide to narrow (long)
wide_scores |>
  pivot_longer(
    cols = c(midterm, final), # which columns to stack
    names_to  = "exam",       # new column holding the old names
    values_to = "score"       # new column holding the values
  )

# Narrow to wide
narrow_scores |>
  pivot_wider(
    names_from  = exam, # column whose values become new column names
    values_from = score # column whose values fill the new cells
  )

Memory trick: pivot_longer() makes the table taller (more rows). pivot_wider() makes it wider (more columns). They are exact inverses.

Practice: identify the pivot

(a) You have one row per country and columns pop_2000, pop_2010, pop_2020. You want to plot population over time. Should you pivot_longer() or pivot_wider()?

(b) You have a long table with one row per (student, subject) pair. You want to compute each student’s math score minus their reading score. Should you pivot_longer() or pivot_wider()?

ggplot template

ggplot(DATA, aes(x = XVAR, y = YVAR)) +
  geom_SOMETHING(aes(MAPPED_AESTHETICS), FIXED_AESTHETICS)

Mapped (inside aes()): changes row by row based on a column Fixed (outside aes()): same for every point/bar/line

Practice: Write a scatterplot of penguins with bill_length_mm on x, bill_depth_mm on y, points colored by species, and a fixed size of 2.

ggplot(penguins, aes(x = bill_length_mm, y = bill_depth_mm)) +
  geom_point(aes(color = species), size = 2)

Quick-reference: things to know cold

R basics

• <- is assignment; == is equality test
• install.packages() once; library() every session
• Data frame: 2D, columns can have different types
• na.rm = TRUE to ignore NAs in summary functions

ggplot

• aes() maps data; constants go outside aes()
• geom_histogram for distribution, geom_point for scatter, geom_bar for counts
• coord_cartesian() to zoom without dropping data
• facet_grid(row ~ col): rows left of ~, columns right
• position = "fill" for proportions

dplyr

• filter rows, mutate columns, summarise collapses with group_by
• filter() must come before summarise() if the condition is on individual rows
• Always add n = n() to summaries so you know the group size
• arrange(desc(...)) for descending order

Tidy data

• pivot_longer(): wide to narrow (more rows)
• pivot_wider(): narrow to wide (more columns)