Causal Inference: prediction, explanation, and intervention

Location & Time

Fall 2013
Mondays 3:00pm-5:30pm

What is causality and why is it useful? Causes are what allow us to predict what will happen in the future (that a stock price will rise based on a news report), explain why something happened in the past (what actually led to a patient's seizure), and intervene to produce particular outcomes (crafting political speeches to influence voter opinion). Whether you want to buy stocks, develop effective treatments, or manipulate elections, you need to know that you are not acting on a mere surrogate but rather the true culprit.

http://xkcd.com/552/

Overview

This course covers the practical tools needed for evaluating causal claims and making causal inferences. We will explore two primary questions -- 1) what is causality? 2) how can we find it? After covering the conceptual and theoretical underpinnings of causal inference, we discuss how causal inference is handled by different fields and how to responsibly test it in real-world cases.

Syllabus [pdf]

Prerequisites

None. The course is intended for advanced undergraduate and graduate students from computer science and other disciplines.

Evaluation

Discussion of the readings is an important part of the course and will count towards the final grade. There will be a final project (and presentation), which may be theoretical or experimental in nature (for example, applying causal inference methods to data, writing a critique of a methodology or study, proposing a new inference method).

Grades will be: 5% homework, 15% participation, 30% midterm exam, 50% final project.

Schedule and Lecture notes

See the syllabus for detailed list of readings for each week and due dates

Week 1 (8/26): Introduction to causal inference [slides]
Week 2 (9/9): Regularities, counterfactuals, and token causality [slides]
Week 3 (9/16): Probabilistic causality [slides]
Week 4 (9/23): Introduction to graphical models, probability review [slides]
Week 5 (9/30): Conditions for inference, Bayesian networks [slides]
Week 6 (10/7): Causality in time series: DBNs, logic-based methods [slides]
Week 7 (10/15): Causality in time series: Granger causality [slides]
Week 8 (10/21): Midterm
Week 9 (10/28): Mechanisms, interventions, RCTs [slides]
Week 10 (11/4): Bias, evidence, evaluation [slides]

Week 11 (11/11): Applications/recent work in causality
Week 12 (11/18): Journal club
Week 13 (11/25): Presentations
Weel 14 (12/2): Presentations


Note: If you'd like the original powerpoint files to use these slides in your (academic, noncommercial) presentations or teaching, email me at samantha.kleinberg@stevens.edu and I'd be happy to send them to you.