Overview

Author
Affiliation

Eduard Bukin

Distributional Impact of Policies. Fiscal Policy and Growth Department

What the self-study period is for:

How It Works

The self-study period runs for 2–3 days between the two sessions. You work independently through the structured case studies below and document your experience.

Share your experience in the Teams channel: ask questions, share screenshots of what worked or didn’t, and pick up tips from colleagues.

Exercises are designed to be progressively more complex, but feel free to jump around based on your interests and needs.

  • Example 0 is a quick test of your setup. If you haven’t completed it yet, start here. All you need are fulfilled minimal prerequisite requirements,

  • Example 1 is an introduction to the Positron + Copilot workflow using Stata. It’s a great place to start if you’re new to AI-assisted coding or want to get comfortable with the tools. It covers the basics AI tools such as: inline suggestions, chat mode, inline chat, and agent mode. Example 1 contains detailed instructions and suggested prompts to guide you through the process. Finally, there is an 18 min video walkthrough in case you want to follow along.

  • Example 2 is a live demonstration of using AI to clean and prepare for analysis Mexico’s ENIGH survey (2016 and 2018). Full details, suggested prompts, and instructions are in Example 2.

    The exercise follows a structured seven-step workflow:

    1. Set the overall objective — name the survey, years, target spec, and language up front
    2. Define inputs and outputs — specify unit of observation, file format, and modules
    3. Create a data dictionary — capture variable names, types, labels, and missings without sharing raw data
    4. Gather external metadata — fetch variable definitions, units, recall periods from official catalogs
    5. Verify understanding — produce a mapping table before writing any code
    6. Develop code iteratively — generate modular, DIME Wiki-compliant scripts one module at a time
    7. Verify with an independent review — audit the cleaning code in a separate AI session

    Its main objective is to demonstrate how versatile AI assistance can be in understanding data structure and metadata.

  • Example 3 is a final example with less hand-holding that gives you the opportunity to practice AI for data analysis tasks with bare minimum guidance. It contains an example of a data analysis from the past and asks you to reproduce, revise, and reuse it with the help of AI by adding an additional survey year. This is one of the most common tasks. Full details and instructions are in Example 3.

Supporting materials