Data Science and Analysis for Economists

Course goals

At the end of the course the student is able to:

• Distinguish predictive from causal problem framings, and formulate empirical questions as regression, classification, or unsupervised learning tasks;
• Construct reproducible data science projects in R using Git, renv, and Quarto, following professional conventions for project structure and version control;
• Explain the bias–variance tradeoff, common loss functions, and regularisation, and relate these to model complexity and overfitting;
• Implement tidymodels pipelines — including feature engineering recipes, cross-validation, and hyperparameter tuning — that are free of data leakage;
• Compare and select among predictive models (linear, penalised, tree-based) using appropriate evaluation metrics and resampling procedures;
• Apply unsupervised methods (PCA, clustering) to discover structure in data and interpret results in an economic context;
• Implement Post-Double Selection and Double/Debiased ML to estimate treatment effects under selection-on-observables assumptions;
• Critically evaluate ML-based empirical analyses for assumption validity, methodological choices, and reliability of reported estimates.

Content

This course equips students with a modern, practical toolkit for data science with structured data, using R and the tidymodels ecosystem. It opens with ML problem framing — translating economic questions into prediction tasks, choosing targets and evaluation metrics, and understanding the boundary between predictive and causal inference.

The core of the course covers supervised learning: its foundations (loss functions, the bias–variance tradeoff, regularisation) and their application to regression and classification tasks using linear models, penalised regression (LASSO, Ridge, Elastic Net), and tree-based methods including random forests. Students learn to build rigorous evaluation pipelines using cross-validation, with explicit attention to data leakage and its consequences. The course also introduces unsupervised learning (clustering/PCA) for pattern discovery and economic segmentation.

The course concludes with a module on Causal Machine Learning, showing how the supervised learning machinery developed throughout can be deployed responsibly for treatment-effect estimation under selection-on-observables assumptions, covering Post-Double Selection and Double/Debiased ML.

Equal emphasis is placed on the technical methods and on professional reproducible workflows: students build R projects using Git, renv, and Quarto — tools standard in modern empirical work.

Format
Two weekly meetings: one lecture and one tutorial. Attendance of both is mandatory. Lectures introduce theoretical concepts and modelling choices, with emphasis on intuition, pitfalls, and economic applications. Tutorials are hands-on and reproducibility-focused: students implement methods on their own computers, building a professional analysis pipeline step-by-step across the seven weeks. Each tutorial produces a rendered Quarto report committed to a version-controlled repository.

Assessment method

• Midterm exam. Closed-book, computer-based (Chromebook); Weeks 1–3.
  • Weight = 40%
  • Individual
• Endterm exam. Closed-book, computer-based (Chromebook); Weeks 4–7.
  • Weight = 45%
  • Individual
• Tutorial portfolio. Weekly reproducible R project submitted via GitHub.
  • Weight = 15%
  • Group

Intended Learning Outcomes
1b apply models for policy development/testing
2a render objectives, hypotheses & values
2b assess results, arguments & problems
2d select & account for research method
2e select & account for analysis methods
2g specific economic research skills
2h defend results in English
3a (jointly) solving academic problems from econ perspective
4a professional collaboration
6a independently track academic developments
7d has acquired the programming, data analysis and data interpretation skills necessary for this purpose.

In case online access is required for this course and you are unable (or do not want) to buy the access code, you are advised to contact the course coordinator for an alternative solution. Please note that access codes are not re-usable meaning that codes from second hand books do not work, as well as access codes from books with a different ISBN number. Separate or spare codes are usually not available.