About the Course

Course Description

This is a foundational course on machine learning, with emphasis on applications in social science and economics. The course covers standard econometric theory and its empirical applications, as well as an introduction to statistical machine learning. The first part of the course reviews core econometric estimation approaches (e.g., multiple linear regression analysis), including hypothesis testing, heteroskedasticity, measurement error, and omitted variable bias. We will also discuss non-linearities in regression, limited dependent variable models, and fundamentals of regression-based prediction.

The second part of the course discusses selected machine learning methods that may be of interest to social science researchers and aspiring data science practitioners including regularized linear regression, logistic regression, decision trees and random forests, neural networks, and unsupervised methods like clustering and principal component analysis. We will then highlight recently developed methods at the intersection of machine learning and econometrics which improve upon traditional econometric methods in settings of heterogeneous treatment effect estimation in causal inference.

Course Objectives

By the end of this course, students will be able to do the following:

Discuss the objectives of traditional econometrics and statistical machine learning. We will relate and contrast the two branches of statistics and identify how they use similar tools for different purposes.
Apply common probability rules, derive and interpret simple derivatives and integrals, design and evaluate hypothesis tests, and interpret summary statistics.
Derive, code, and interpret multiple linear regression coefficient estimates, fit statistics, and standard errors, including the concepts of statistical significance and hypothesis testing.
Discuss and interpret out-of-sample fit statistics for linear models, including diagnosing overfitting.
Apply best resampling practices to real-world data: coding k-fold cross-validation algorithms, bootstrapping algorithms, and regularization. Discussing why and when these are useful/necessary.
Describe regularization in linear regression and contrast its' usage in Lasso vs. Ridge vs. Elastic Net regressions.
Construct and evaluate decision trees and random forests both mathematically and in code.
Visualize the common neural network architectures, explain the processes of backpropagation and gradient descent, identify which architecture is designed for which predictive task.
Derive and construct logistic regression and Support Vector Machine equations mathematically and in code. Interpret coefficient estimates and out-of-sample fit statistics.
Define unsupervised learning, identify its' use-cases, and appraise its' value in different settings. Construct and interpret principal components and K-means clusters.
Define and derive treatment effects, explain the assumptions which create unbiased and/or consistent estimates of them.
Define heterogeneity and explain its' implications for policy analysis.