Applied Econometrics: Mostly Harmless Big Data
Graduate applied econometrics: regression, matching, instrumental variables, differences-in-differences, regression discontinuity and big data methods.
Applied micro research depends on choosing an empirical strategy that fits the question. This class, taught by Joshua Angrist and Victor Chernozhukov, covers regression and matching, instrumental variables, differences-in-differences, regression discontinuity designs and standard errors, then adds high-dimensional data sets. It uses authentic questions that real data can answer, along with long and challenging problem sets.
What You Will Achieve
- Apply regression and matching to causal questions
- Interpret instrumental variables estimates when effects differ across people
- Use Lasso for prediction and inference with a large number of covariates
- Handle clustering and serial correlation in standard errors
Program Overview
Regression and Matching
The opening topics recap the core tool and how to compare treated and untreated groups.
- Regression recap
- Omitted variables bias
- Matching and the propensity score
Instrumental Variables and Designs
Most of the class develops the main strategies for causal research.
- Constant-effect models and 2SLS
- Local average treatment effects
- Differences-in-differences
- Regression-discontinuity designs
Inference and Big Data
Later topics address standard errors and large data.
- Clustering and the Moulton factor
- Prediction with a large number of covariates
- Lasso and Post-Lasso
- Large sample sizes and Hadoop
Course Completion
Job Outlook
The instructor says the audience is mostly headed for research in academia, government or NGOs, with a growing group going to technology companies.
Good for: PhD students, professional school students and research assistants doing applied empirical work who already have a first course in econometrics.
Keep in mind: Notes are selected, so some topics appear only in the syllabus, and there are no videos. The first problem set’s data file is 114MB and the book is not supplied.
Bottom line: Graduate empirical methods taught around real questions, a natural companion to Mostly Harmless Econometrics.
Frequently Asked Questions
How closely does it use Mostly Harmless Econometrics?
Mostly Harmless Econometrics by Angrist and Pischke supplies basic and review material for the first two-thirds, with a few core and frontier articles listed for each topic.
What does the big data section cover?
Prediction and valid confidence intervals with many covariates, using Lasso and Post-Lasso, and analysis with large sample sizes, including parallelization, Google Map Reduce and Apache Hadoop.
What is in the problem sets?
Three, some with accompanying files, such as a data set for Problem Set 1 and IK Software for Problem Set 2. Recitation notes are published too.



There are no reviews yet.