The Open Workspace

Only 36 percent

Open Science Collaboration, 2015

• 270 researchers set out to redo 100 published psychology studies
• Same methods, similar participants, same analyses
• Only 36 percent replicated at conventional thresholds
• Among those that did, effect sizes were on average half the original magnitude
• Published in Science, and one of the decade’s most cited findings

Not fraud. Friction.

• Methods sections did not contain enough detail to re-run the analysis
• Researchers had to email original authors for clarification
• Some authors had moved. Some had lost the data.
• Some had used software whose menus had since changed
• The work was never built to be retraceable in the first place

The point-and-click problem

• SPSS, Excel, and JMP work through menus and dialog boxes
• You click Analyze, then Compare Means, then One-Way ANOVA
• The result appears. The click history vanishes.
• To verify your work, a collaborator needs a prose description of every click
• Almost no one does this well, and almost no one wants to

Code as documentation

• A script is a description of every step, in a language a computer can re-run
• When you write your analysis as code, the code is the methods section
• Reinhart and Rogoff was caught because the spreadsheet was eventually shared
• It would have been caught faster if the analysis had been code from the start
• Reproducibility: the analysis can be re-run and produce the same result

Four tools, one ecology

Free, open, cross-platform

• VS Code: the editor, where prose, code, and notes all live in tabs
• R: the language that does the counting and the statistics
• Quarto: the publishing system that turns source into finished output
• Git and GitHub: version control and a home for the history
• Each produces a plain-text record of what you did

Why R, why Quarto

• R dominates social science: communication, sociology, political science
• Its packages for content analysis, reliability, and regression are the deepest
• Python is also excellent. The skills transfer both ways.
• Quarto takes one .qmd source and produces HTML, PDF, Word, or slides
• The source mixes prose with code, and the code runs when you render

Why Git and GitHub

• Git tracks every change to every file in a project
• GitHub hosts that history on the web, publicly by default
• You get a permanent audit trail and the ability to roll back to any state
• Students get the GitHub Student Developer Pack at no cost
• This is the standard workflow in research labs and industry data teams

Install order, part one

Order matters. Do these first:

1. R, from cran.r-project.org, installer defaults
2. VS Code, from code.visualstudio.com
3. The R extension for VS Code, by REditorSupport
4. Quarto, from quarto.org, install with defaults

R has no interface of its own. It is a backend other tools call.

Install order, part two

5. The Quarto extension for VS Code, for live preview of .qmd files
6. Git, from git-scm.com, default options are fine
7. A GitHub account, with a professional username you will link work to
8. The v2v package, install command on the V2V Hub

Click-by-click walkthroughs with screenshots live on the Hub.

The setup check

Create a file named setup-check.qmd and add one chunk:

library(v2v)
v2v::setup()

The validator confirms that R, Quarto, Git, and the v2v package are all present at acceptable versions. Success prints “V2V setup complete”. Anything else, and the Hub has a troubleshooting page.

First contact with the data

stream_log, first ten rows

The data moves

• This is a time series, not a snapshot. Four channels repeat across ten rows.
• Sodapoppin climbs 27,934, then 28,076, then 28,203
• That is 269 viewers gained in 122 seconds
• xQc lost 150 and then held steady. Forsen lost 120. Giantwaffle gained 3.
• Each number is a real moment: a clip going viral, a raid arriving

Three more things to notice

• The timestamps look like nonsense. 1542578200214 is Unix epoch time, milliseconds since January 1, 1970. Chapter 11 fixes it.
• There is a hidden newline inside sodapoppin’s title. Real data is messy in exactly this way.
• Titles are themselves data. “No Smoking in Chat” performs etiquette. Chapter 12 returns to title content.

The v2v package

The course companion ships three things:

• Two fixtures: twitch_chat_sample, roughly 35,000 messages, and twitch_streams_sample, roughly 32,000 rows across 50 channels
• Helper functions for loading data, codebook skeletons, reliability, plots
• The setup validator, v2v::setup(), which you can rerun any time

Load a fixture with data(twitch_chat_sample, package = "v2v").

Looking ahead

• Thursday is a lab. Bring a laptop with as much of the stack installed as you managed. We troubleshoot together.
• You will build your project repository and make your first commit
• Before then: finish the installs, or get far enough to know where you stuck
• Read Chapter 3 on ethics before Tuesday
• GitHub Profile is worth 25 points and is due at the end of Week 2
• Project Setup is worth 25 points and follows at the end of Week 3