Tsinghua Workshop on R

Workshop on R National Tsinghua University
2017/6/2
James Myers National Chung Cheng University, Graduate Institute of Linguistics

Main tutorial materials

Data file 1 (TXT)
Data file 2 (TXT)
Data file 3 (to be distributed in the tutorial)
R code (TXT)

Introduction

R is free software for doing statistical programming
The weird name:

C is a famous computer programming language
S is a not-free programming language for Statistics
R is a free version almost like S (R is before S in alphabet)
R is extremely popular in linguistics, because R is free (linguists are poor), powerful (linguistic data are complicated), and flexible (linguists also need programming, not just statistics).
Some textbooks:

Beginner:

Johnson, K. (2008). Quantitative methods in linguistics. Wiley.

Intermediate:

Gries, S. Th. (2013). Statistics for linguistics with R: A practical introduction (2nd edition). De Gruyter Mouton.

Levshina, N. (2015). How to do linguistics with R: Data exploration and statistical analysis. John Benjamins.

Advanced:

Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R. Cambridge University Press.

Free:

Myers, J. (2017). Yet another statistics for linguists book. National Chung Cheng University ms.
Getting R:

Go to R website: https://www.r-project.org/
Under "Download", click CRAN (Comprehensive R Archive Network)
Choose a "mirror" (copy) near you (e.g. http://cran.csie.ntu.edu.tw/ at National Taiwan University, Taipei)
Choose Download R for Windows, or whatever your computer is
Choose base and download it
Find the R installation file on your computer and follow the instructions that it shows you:

If you have a new version of Windows, it's probably 64 bit.
When R asks what language you want to use, that's just for the installation instructions. By default, it will translate the R menus into the language of your local system (i.e., Chinese). If you want R to have English menus in Chinese Windows (like me!), then during the installation, turn OFF "Message translations".

Some people like to run R inside another free program: RStudio (https://www.rstudio.com/). RStudio makes it easier to program, keep track of your variables, and update your graphs.
Using R:

To find the folder you're going to work with:

File menu, then Change dir...

To quit the program:

Click red "X" on main window, or use menu to choose File, then Exit

"Save workspace image?"

If you choose "Yes", all the commands you typed will be saved in your folder as .Rhistory and all the stuff created by your commands (variables, statistical models, etc) will be saved as .RData

To reload your old commands or stuff (respectively):

File, then Load History... or Load Workspace... (respectively)

To install a new package (set of special-purpose R functions):

Packages, then Install package(s)...

Talking to R:

R is a programming language, so you talk to it using commands and special symbols.

R talks back with text output (including statistical analyses) or graphs.

You can repeat an old command by hitting the up-arrow button (and scroll down again through your command history by hitting the down-arrow button).

If you can't remember how a function works, just type ? before the function name, and you'll get a help page.

I also have an occasionally updated file called StatsFunctions.pdf on this statistics resources page

Anyway, there are thousands of R users on the internet, so almost any R question you search for has already been asked: search wisely, and you will find what you need!

The rest of the tutorial

Download the data and handout / R code above, and follow along!

Practice exercises

These are all independent, so try whichever you want, in any order. However, they do get harder towards the end....

Programming

Use this R code to take George Washington's first inaugural address (from Yale University's Avalon Project), and convert it into this word-by-word version. Use ? to figure out how the functions work.
Use the function(), as.character(), and nchar() functions to create a new function that finds all words with N letters in a vector of words. Test it on some short word vectors, and on this word-by-word version of Washington's speech.

Graphs

This file has fake data showing children's ages in months and their accuracies on some language task. Use plot() to make a dot plot. Does there seem to be any correlation between the two variables? Add a linear best-fit line using the abline() and lm() functions.
Amy got 87 on the English test, and Bill got 43. Put these results into a vector using the c() function, add names using the names() function (works like colnames() in the handout), and then use the barplot() function to make a bar plot (using the ylab argument to add the y-axis label "Scores", the main argument to add the main title "Test results", and the col argument to make the first bar red and the second bar blue).

Frequencies

Use the table() function to count the word frequencies in this word-by-word version of George Washington's first inaugural address (from Yale University's Avalon Project).
Use hist(), plot()/density(), and/or qqnorm() to look at the distribution of the Washington speech word frequencies. Does this distribution look normal? Transform the frequencies using log() and plot them again. Does the log-transformed distribution look more normal?

Distributions and probability

This file contains lots of real reaction times in milliseconds (ms). If you select one of them at random, what is the probability that it will be below 500 ms?
Use the as.character() and nchar() functions to get the word lengths in the word-by-word version of George Washington's first inaugural address (from Yale University's Avalon Project). Use hist(), plot()/density(), and/or qqnorm() to check if the word lengths are normally distributed.
These are the real reaction times for two different participants in a psycholinguistic experiment. Use mean() to find out which participant was slower on average. Use sd() to find out which participant showed more variability in RT. Apply the scale() separately to each participant to convert their RTs to z scores. What happens to their means and standard deviations? Why?

t tests

Why would it be appropriate to run an unpaired t test assuming equal variance on these fake scores from two English classes where different teaching methods were used? Which teaching method is associated with significantly higher scores? Report the results in the standard reporting format.
Why would it be appropriate to run a paired t test on these fake mean reaction times for nouns vs. verbs in this file? Which type of word was responded to significantly faster? Report the results in the standard reporting format.

ANOVA

This file has fake data about the speaking rates (mean number of syllables per second) from monolingual speakers of Mandarin, Southern Min, and Hakka. Use the tapply() function to compute the mean speaking rates for each of the three languages. Use aov() and summary() to test the null hypothesis that all of these speaking rates are the same. Report the results in the standard reporting format.
This file has fake data from a phonetics experiment on two Venusian tones (X and Y), where the phonetic pitch (F0) was measured at the beginning and end of each syllable to see if the tones are level, rising, or falling. If X and Y differ in slope, why do we expect to see a statistical interaction between the factors Tone and Point? To find out if they do, use the aggregate() and colnames() functions to compute the mean F0 values by subjects (so each subject gives just four values: mean X-begin F0, mean X-end F0, mean Y-begin F0, mean Y-end F0). Why is it appropriate to run a repeated-measures ANOVA on these means? Use the summary() and aov() functions (incuding the Error() argument and the as.factor() function). Report the results in the standard reporting format.

Regression modeling

This file has fake data showing children's ages in months and their accuracies on some language task. Use the cor.test() function and/or the lm() and summary() functions to compute the p value for the correlation between these two variables.
This file contains real data on three lexical variables, namely age of acquisition (AoA), familiarity (Fam), and frequency (Freq) for a bunch of English words, plus fake data on one phonetic variable: duration (Dur). Use the lm() and summary() functions to find out which of the three lexical variables has/have significant effects on the phonetic variable.

Logistic regression

This file show fake results of a fake sociolinguistic study on phlugugooglification (a fake phonological process) in the fake Venusian language. Each row represents a different Venusian speaker. A 1 in the Phlug column indicates that this speaker shows phlugugooglification, and a 0 shows that s/he doesn't. The other columns represent social variables. Use the glm() and summary() functions to run a logistic regression that tests which, if any, of these social variables affect the probability of a speaker showing phlugugooglification.

Mixed-effects modeling

This file gives the fake results of a syntactic judgment experiment that the famous fake linguist Norm Chompsky ran on 10 participants and 20 minimal pairs of sentences. He hypothesized that one sentence in each pair is grammatical (A) and the other is ungrammatical (B), but he wanted to know if this contrast affects the judgments of ordinary people. Participants gave their judgments on a 10-point scale (1 = terrible, 10 = perfect). Thus the responses were grouped not just by participants (each person saw many sentences of type A and type B), but also by sentence pairs (each A-B pair). Based on this information, run a mixed-effects modeling analysis that predicts the judgment scores from the two sentence types, taking both types of random effects into account, using the lmer() function in the lme4 package (which first needs to be downloaded/installed, then opened with the library() function).

Other links