NZLS Workshop Day

Session 2: Introduction to R and RStudio

Joshua Wilson Black

joshua.black@canterbury.ac.nz

Te Kāhui Roro Reo | New Zealand Institute of Language, Brain and Behaviour

Te Whare Wānanga o Waitaha | University of Canterbury

2024-12-11

Overview

Overview

1. Motivations (Why am I doing this?)
2. What is R? What is RStudio?
3. A realistic(ish) example
4. How do I get help?

This session is a basic introduction to R and RStudio for linguistics researchers.

Lots of people go into linguistics without much of a background in computing or statistics. So I’ll talk a bit about why we’re doing this at all.

Then we’ll look at R and RStudio, get a little bit of a sense of what is infront of us and do a few simple things in an R script.

We’ll then look at a realistic example of a bit of linguistic data analysis together.

Then, in order to make you a little more independent, I’ll talk about ways in which you can get help when you don’t know what to do.

We won’t be looking in any detail at any statistical models. This would be too much for an hour and a half!

Motivations

I never asked for this!

Many people who sign up for linguistics don’t expect that they’re going to end up having to learn how to code.

Perhaps you avoided this kind of thing like the plague at school.

So I’ll say a little something about why this is important.

Reproducibility

• Important for you
  • Review comes back after six months, you get to it in two months, but now you have to regenerate your results from scratch and you can’t figure out how.
  • Another researcher wants to collaborate and extend your idea, but you can’t retrace your steps. Opportunity lost!
• Important for the rest of us
  • Provides a reliable record for other researchers to retrace your steps.
  • Protection against (some!) scientific fraud.
• We need clear step-by-step instructions from data collection to final analysis.
• i.e. a computer programme!

Research is reproducible if, given the same data, we can generate the same results (including both statistical models and any plots which might appear in papers). This is different from replicability, which is also important, but isn’t entirely up to you! Replicability is the ability to generate consistent results with different data.

This is important for you, because it helps to avoid this kind of case:

It’s also important for the rest of us. We want to be able to retrace your steps and to be protected against certain kinds of scientific fraud (they’re out there).

But what does this look like: basically — a computer programme!

Spreadsheet hell

• Data analysis in a spreadsheet programme (e.g. Excel) is not step-by-step.
• Spreadsheets allow direct manipulation of data (e.g. here)
• Automatic conversion: (sometimes) good for business not for scientific research.
  • “Scientists rename human genes to stop Microsoft Excel from misreading them as dates” (link)
• Many more horror stories are collected here

Perhaps you’ve made peace with the fact that you have to be an amateur statistician. But why do you have to be an amateur coder as well?

The dream

• Preregistration of planned analysis.
• Open data (as much as ethical).
• A plain text file which contains all steps to get from the open data to your results in a commonly used programming language.
• Sharing of other analytic methods which were tried.
• Human-readable commentary alongside the code.
• An example: https://osf.io/m8nkh/

What is R? What is RStudio?

R is a programming language, created by two statisticians at the University of Auckland and with a focus on data analysis.

RStudio is a piece of software for carrying out data analysis using R. It makes it easier to use R interactively, to view data in a convenient way, and to see any visualisations etc which R can be made to produce.

Ctrl + Shift + N

You’ll see a lot of terminology out there. But here’s what the bits of the RStudio window are called according to the official documentation.

The keyboard shortcut is to open a new R script.

Avoiding bad habits

• Change default settings (Tools > Global Options).

RStudio projects

• Open an RStudio project (File > New project)
• Give it a name like nzls_workshop and store it in your documents folder.
• A RStudio project is a directory (folder) which contains everything you need for a data analysis project.
• Use projects!

It may be useful to show this on the big screen.

First steps in R

• Go to the console pane.
• You can enter code after the >
• Let’s start simple: type ‘2 + 2’ and press enter.

🥳🥳🥳🥳🥳

Help! I’m bored!

https://nzilbb.github.io/statistics_workshops/

• This is for total beginners.
• If you get bored, you are welcome to work through anything at this link
• Ask me in the break if you have any questions about it.

R scripts

• The console can be very useful for small tasks.
• An R script allows us to keep a step-by-step record.
• Open a new R script (Ctrl + Shift + N, or File > New File > R Script)

More maths

1 * 2
2 ^ 3
2 / 4
2 %% 4
2 %/% 4
"kea" + "tūī"

• Run the line you are currently on by pressing ‘Ctrl + Enter’
• Output will appear in the console pane.
• Final statement produces an error

Help! I just see a ‘+’!

• If you enter an incomplete statement, R will wait for you to complete it in the console pane.

5 *

• Two options:
  1. Complete the statement in the console.
  2. Enter “Ctrl + C” to escape.

Comments

• Anything after a # will be ignored by R.

# Add 4 to 3
3 + 4

• Use this to add ‘comments’ for human readers.

Functions

toupper("this is an example sentence")
?toupper
paste("tūī", "kea", sep = "-")
rnorm(n = 100, mean = 0, sd = 1)
rbinom(1, 1, 0.5)

• Functions take input(s) and produce an output.
• The function’s name is followed by the input(s) inside brakets.
• The inputs are called arguments
• ? before function name produces help (in output pane)

Logic

"tūī" == "kea"
1 == 2
1 == "kea"
"tūī" != "kea" 
"tūī" != "kea" & 1 != 2 # "And"
"tūī" == "kea" | 1 != 2 # "Or"
1 >= 4

• Statements can be TRUE or FALSE
• Combine multiple statements with ‘and’ (‘&’) or ‘or’ (|)
• Useful for filtering data

Vectors

• Data isn’t just one or two numbers…
• Vectors allow us to combine many observations.

c("I", "went", "to", "the", "shop")
c(1, 2, 3, 4, 5)
c(TRUE, TRUE, FALSE, FALSE, TRUE)
mean(c(1, 2, 3, 4, 5))
paste0(c(1, 2, 3), ". ", c("I", "went", "to"))

• The c stands for “combine”.
• All the entries in a vector must be of the same type.
• A vector counts as a single argument to a function.

Variables

• We want to reuse the same data multiple times
• Variables are the solution.
• Variables associate a name with an object using ‘<-’

age <- c(2, 4, 2, 3)

• Look in the environment pane
• Your variable is there now.

age + 1 # What does this do?

Data frames

• How do we associate multiple data points from the same individual?
• Data frames
  • Rows are ‘observations’
  • Columns are ‘variables’ 🫤🫤🫤🫤🫤

toddlers <- data.frame(
  age = c(2, 4, 2, 3),
  happiness = c(5, 4, 5, 1),
  name = c("Basil", "Glenys", "Tahu", "Gronk")
)

• NB: Statements can go across multiple lines.

Accessing data

• View(toddlers): a spreadsheet-style view of data.
• Square brackets give access to portions of the data.

age[2]
age[3:4]
toddlers[1,3]
toddlers[2,]
toddlers[,3]

• Why two numbers for data frames?

Accessing data with names

• Use ‘$’ to access columns in a data frame.

toddlers$name
toddlers$happiness

Accessing data with vectors

age[c(2, 4)] # the second and fourth entry in age.
age[age > 2] # filter using a logical statement
toddlers[, c(2, 3)]
toddlers[c(1, 2), ]
toddlers[toddlers$happiness < 2, ]

• Use of logical statements lets us filter data.
• Logical statements produce ‘logical vectors’.

toddlers$happiness < 2

[1] FALSE FALSE FALSE  TRUE

There’s a package for that!

R packages

• Most problems can be solved with a package.
• Packages contain R code and data for solving specific problems.

install.packages("tidyverse") # only run this once
library(tidyverse)

• We install packages with install.packages().
• Packages are loaded with library().
• Sometimes they take a while to install!

The tidyverse

• The tidyverse is a popular collection of packages.
• It includes, e.g.:
  • dplyr: functions for data filtering and transformation.
  • ggplot2: a popular package for data visualisation.
  • readr: functions for reading and writing data.
  • stringr: functions for manipulating strings.
• Code written with tidyverse packages has a different style. (…better)

Pause

• We’ve covered a lot!
  • Getting around RStudio
  • R as a calculator
  • Functions
  • Logical statements
  • Vectors and variables
  • Data frames
  • Accessing and filtering data
  • R packages
• It’ll take a while to remember this material.

😮‍💨

A realistic(ish) example

Get some data

• We’ve borrowing data from (Mattingley et al. 2024).
• Download the file here: https://osf.io/ucx8n

• Put the data file inside your project folder.
• Start a new script (name it protolexicon.R)
  • Ctrl + Shift + N

Data inspection (text)

library(tidyverse)
library(here)

# read in data
wellform <- read_tsv(here('wellFormTask2021.tsv'))

# examine data
View(wellform)
summary(wellform)

• read_tsv() loads ‘tab separated values’ (simlar to a .csv file)
• summary() tells us about some of the variables.

Data inspection (visualise)

• hist() is a built-in function for making histograms.

# Look at distribution of reaction times
hist(wellform$reactionTime)

Data inspection (visualise)

# Look at shorter reaction times
hist(wellform$reactionTime[wellform$reactionTime < 20])

Filter

# Filter out long reaction times.
# base R style
wellform_filtered <- wellform[wellform$reactionTime < 5, ]

# tidyverse style
wellform_filtered <- wellform |> 
  filter(reactionTime < 5)

Chaining functions

# Chaining functions together with pipes.
wellform_filtered <- wellform |> 
  filter(reactionTime < 5) |> 
  select(
    workerId, stimulus, word, 
    enteredResponse, reactionTime, score.shortv
  ) |> 
  rename(
    participant = workerId,
    response = enteredResponse,
    reaction_time = reactionTime,
    phonotactic_score = score.shortv
  )

• Pipes send output of a function to the next function.
• Commonly used in tidyverse style.

Visualise again

# visualise again using ggplot
wellform_filtered |> 
  ggplot(
    aes(
      x = factor(response),
      y = phonotactic_score
    )
  ) +
  geom_boxplot()

• More on visualisation later in the day.

Visualise again

Model

• Let’s fit a model using the built-in lm() function.

# model
wellform_fit <- lm(
  response ~ phonotactic_score + reaction_time,
  data = wellform_filtered
)

summary(wellform_fit)

Model

Call:
lm(formula = response ~ phonotactic_score + reaction_time, data = wellform_filtered)

Residuals:
    Min      1Q  Median      3Q     Max 
-3.0752 -0.8669  0.1387  1.0105  2.8295 

Coefficients:
                  Estimate Std. Error t value Pr(>|t|)    
(Intercept)        6.16334    0.11274   54.67   <2e-16 ***
phonotactic_score  3.53412    0.12287   28.76   <2e-16 ***
reaction_time      0.04016    0.02039    1.97   0.0489 *  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 1.217 on 4327 degrees of freedom
Multiple R-squared:  0.1605,    Adjusted R-squared:  0.1602 
F-statistic: 413.8 on 2 and 4327 DF,  p-value: < 2.2e-16

Plot predictions

# Get predictions from model.

# Step 1. Decide what we want predictions for. In this case,
# the full range of phonotactic scores at the mean value for
# reaction time.
to_predict <- data.frame(
  phonotactic_score = seq(-1.2, -0.6, by = 0.01),
  reaction_time = mean(wellform_filtered$reaction_time)
)

# Step 2: Get predictions using the `predict()` function.
model_predictions <- predict(
  wellform_fit, 
  newdata = to_predict,
  se.fit = TRUE
)

# Step 3: add predictions and 95% confidence intervals to the `to_predict` data
# frame.
to_predict$prediction <- model_predictions$fit
to_predict$upper <- model_predictions$fit + 1.96 * model_predictions$se
to_predict$lower <- model_predictions$fit - 1.96 * model_predictions$se

# Step 4: visualise again
to_predict |> 
  ggplot(
    aes(
      x = phonotactic_score,
      y = prediction,
      ymin = lower,
      ymax = upper
    )
  ) +
  geom_ribbon(alpha = 0.4) +
  geom_line()

Plot predictions

Pause again

• We have:
  • loaded some real data,
  • inspected it,
  • visualised it,
  • filtered it,
  • fit a model, and
  • plotted model predictions.
• Every decision is recorded in a script.
• This is reproducible

Getting help

Documentation

• For help with a function, add ? before its name in the console pane.
• Packages often come with ‘vignettes’
  • e.g. run vignette("dplyr", package="dplyr") in console.
• Many packages host their documentation online.
  • e.g. https://nzilbb.github.io/nzilbb_vowels/

Peers

• Usually there is a some expertise around you, ask other students and academics.
• There are many online tutorials (text and video).
• Check for public code from papers you are using in your research.

Resources

• Winter (2019): Statistics for Linguists: An Introduction Using R. New York: Routledge.
  • An excellent intro to states for linguists.
• R for Data Science: https://r4ds.had.co.nz/
  • A general introduction to data science using the tidyverse.
• https://swcarpentry.github.io/r-novice-gapminder/
  • Introduction to R and Rstudio for reproducible science.
• https://nzilbb.github.io/statistics_workshops
  • NZILBB statistics workshops (work in progress)

AI coding assistants

• Be very careful! Output always looks good.
• Often worse for R than for more popular languages.
• https://claude.ai works quite well.
• Standard problem solving tips apply:
  • Break down the problem into small steps.
  • Test each step yourself to make sure it’s doing what you think.
• You can also ask for explanations of code blocks from elsewhere.

Summary

Summary

1. Why are we doing this?
2. Basics of R and RStudio
3. A realistic(ish) example
4. How to get help.

References

Allaire, JJ, Yihui Xie, Christophe Dervieux, Jonathan McPherson, Javier Luraschi, Kevin Ushey, Aron Atkins, et al. 2024. rmarkdown: Dynamic Documents for r. https://github.com/rstudio/rmarkdown.

Mattingley, Wakayo, Forrest Panther, Simon Todd, Jeanette King, Jennifer Hay, and Peter J. Keegan. 2024. “Awakening the Proto-Lexicon: A Proto-Lexicon Gives Learning Advantages for Intentionally Learning a Language.” Language Learning 74 (3): 744–76. https://doi.org/10.1111/lang.12635.

Müller, Kirill. 2020. here: A Simpler Way to Find Your Files. https://CRAN.R-project.org/package=here.

R Core Team. 2024. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.

Wickham, Hadley, Mara Averick, Jennifer Bryan, Winston Chang, Lucy D’Agostino McGowan, Romain François, Garrett Grolemund, et al. 2019. “Welcome to the tidyverse.” Journal of Open Source Software 4 (43): 1686. https://doi.org/10.21105/joss.01686.

Winter, Bodo. 2019. Statistics for Linguists: An Introduction Using R. New York: Routledge. https://doi.org/10.4324/9781315165547.

Xie, Yihui, J. J. Allaire, and Garrett Grolemund. 2018. R Markdown: The Definitive Guide. Boca Raton, Florida: Chapman; Hall/CRC. https://bookdown.org/yihui/rmarkdown.

Xie, Yihui, Christophe Dervieux, and Emily Riederer. 2020. R Markdown Cookbook. Boca Raton, Florida: Chapman; Hall/CRC. https://bookdown.org/yihui/rmarkdown-cookbook.