Machine Learning Methods for Political Science

About

"Learn about what constitutes 'Machine Learning', the varying model types and how to construct and run different methods using social scientific data-sets.

Need to know
Students intending to take this course must have a basic understanding of statistical analysis, including (linear) regression. We will build extensively on these statistical concepts in the early components of this course. While we will occasionally consider mathematical formulae, knowledge of linear algebra is not required.

This course will also include guided coding exercises using R. To gain the most from this course, you should have some basic proficiency in R. For example, you should be able to manipulate vectors, write for-loops, and use conditionals (e.g. if-then statements).

Students should expect to spend 1-3 hours outside of the core teaching hours consolidating the material covered in these classes

In depth
Day 1: What is machine learning?

Day 2: Regularised methods and the bias variance trade-off

Day 3: Tree-based methods and hyperparameter tuning

Day 4: Neural networks and feature engineering

Day 5: Ensemble learning
How the course will work online
The course is structured into five live Zoom sessions, each lasting 3 hours. The first 1.5-2 hours will focus on the major theoretical components of each day’s topic. The remaining 1 hour will be spent walking through hands-on coding exercises, where you will apply the concepts and methods we discuss in the lecture to real-world data.

Prior to each session, the instructor will distribute the slides and R script for you to explore at your own pace. During the session, the instructor will go through the code and models with you. Additionally, we will take time to discuss the benefits and limitations of machine learning models, as a group.

After each session, there will be an exercise for you to complete, which will help consolidate your understanding of both the theoretical and practical topics we cover."

Course leader

Thomas Robinson is a methodologist and political scientist, whose research focuses on the application of machine learning (ML) methods within experimental research.

Target group

The course is designed for a demanding audience (researchers, professional analysts, advanced students).

Course aim

"Purpose of the Course
By the end of this course, you will:

-understand and implement fundamental concepts in the use of machine learning;
-distinguish between major machine learning model types (including regression-, tree- and network-based forms);
-construct and run machine learning methods on social scientific datasets;
-compare and critique the appropriateness of machine learning methods for various use-cases in political science (and the wider social sciences).
Overall, the course will equip you with advanced knowledge and skills that will help you develop convincing and important research designs, in political science, using advanced computational methods."

Fee info