Setup guide — do this before August 31

Everything here is one-time setup, so class time can go to actual work. Budget about an hour, plus a few days’ wait for the GitHub student-benefits verification (step 4), which is out of your hands — start that one today.

Getting stuck somewhere is normal — note where, move on, and bring it to the first session.

Checklist

1. ChatGPT subscription

The course workflow uses an AI agent that comes with a paid ChatGPT plan. Two routes — either is fine:

  • Route A (always works): personal ChatGPT Plus, $20/month. Go to chatgpt.com, sign in (create an account with any email), and upgrade to Plus from the account menu. A Plus plan should comfortably cover this course’s workload; you can cancel after the semester.
  • Route B: UMD ChatGPT Enterprise. UMD has an institutional ChatGPT agreement. Email UMD IT and ask whether graduate students can get a ChatGPT Enterprise/Edu seat that includes Codex access. If the answer is slow or unclear, don’t wait — use Route A.

Prefer Claude? That’s allowed — but at the $20 tier expect to hit usage limits with agent-style work; you’d likely need a heavier plan. The class demos use the ChatGPT/Codex stack.

2. VS Code + the Codex extension

VS Code is a free code editor; the Codex extension puts the AI agent in a panel next to your code.

  1. Download and install VS Code: code.visualstudio.com
  2. Open VS Code → click the Extensions icon in the left sidebar (four squares) → search “Codex” → install the extension published by OpenAI.
  3. Click the Codex icon that appears in the sidebar and sign in with your ChatGPT account from step 1.
  4. Reference docs if anything looks different: developers.openai.com/codex/ide

3. Git

Git tracks versions of your code. (Concepts come later — right now just install it.)

  • Mac: open Terminal, type git --version, press Enter. If it’s not installed, macOS will offer to install it — accept.
  • Windows: download from git-scm.com/downloads and install with default options.

Then set your identity (in Terminal / Git Bash), using the same email you’ll use for GitHub:

git config --global user.name "Your Name"
git config --global user.email "you@umd.edu"

4. GitHub account, student benefits, and your course repo

  1. Create an account at github.com using your UMD email (or add your UMD email to an existing account).
  2. Apply for the free GitHub Student Developer Pack: education.github.com/pack. Verification can take a few days — this is the one “start today” item. It includes GitHub Copilot free for verified students (AI autocomplete inside VS Code — different tool-fit from Codex; we’ll cover when to use which in the lecture).
  3. Create your own personal repository for the course: on GitHub click New repository → name it something like lsms-project → Private → check “Add a README” → Create. This repo is yours — it outlives the course, and your submission at semester’s end is simply its link.
  4. Share it with your instructor: on your repo’s page, Settings → Collaborators → Add people → her GitHub username (she’ll send it). A private repo is invisible to everyone else — the link you submit only works once she’s a collaborator.
  5. Connect it to your computer: in VS Code, open the Command Palette (Cmd/Ctrl+Shift+P) → type “Git: Clone” → paste your repo’s URL → choose a folder you’ll remember (e.g., Documents/lsms-project).

Want the concepts behind steps 3–5? Short reads: Terminal Basics and Git & GitHub Essentials. And you never have to type Git commands if you don’t want to: the Source Control panel is all buttons, GitHub Desktop is a friendly standalone app, and the agent will happily drive Git for you.

5. World Bank microdata account

Assignment 1 uses an LSMS survey, which you download through a free World Bank microdata account. Access is granted per survey: once you’ve picked a country-year, you request access to that survey’s files (a short form, approved instantly).

  1. Go to the LSMS catalog: microdata.worldbank.org/index.php/catalog/lsms
  2. Create an account (top right).
  3. When you’ve chosen a survey (a country-year with a full household module — details in Assignment 1), open its page, click Get Microdata, and request access to the files.

Background on the program: worldbank.org/en/programs/lsms

6. The five-minute test

You’re done when all four of these work:

  1. Open your cloned course repo folder in VS Code (File → Open Folder).
  2. Open the Codex panel and ask: “Create a file called hello.do that displays ‘setup works’.” Watch it propose the file; accept the change.
  3. Run the file in Stata (or your package of choice) — confirm it runs.
  4. Commit and push: in VS Code’s Source Control panel (left sidebar), type a message like test: setup works, click Commit, then Sync/Push. Refresh your repo page on GitHub — hello.do should be there.

If all four worked, that’s the entire semester’s workflow in miniature: agent writes → you review → it runs → you check the output → you commit. See you August 31!

Nice extras (optional, five minutes)

VS Code earns its keep beyond the agent. Worth installing while you’re in the Extensions marketplace:

  • LaTeX Workshop — compile your papers in the same window as your code (the agent can edit .tex files too)
  • A Stata syntax-highlighting extension — search “Stata,” pick the popular one
  • Markdown preview is built in — open any .md file and press Cmd/Ctrl+Shift+V

A note on Stata

Stata stays our default language (matching your econometrics sequence), and it plays nicely with this workflow. Two ways to run your code, and you’ll use both:

  • You run it: open the .do file in Stata as usual. Good when you’re exploring interactively.
  • The agent runs it: Stata has a batch mode (stata-mp -b do myfile.do from the terminal), which writes a .log file — so the agent can run your do-file, read the log, and fix its own errors.

Either way, the part that can’t be delegated is the same: you read the output and decide whether it’s right. (You may also use R, Python, or anything else — the workflow is identical, minus the batch-mode wrinkle.)