Causal Data Science with Directed Acyclic Graphs

Overview

This course offers an introduction into causal data science with directed acyclic graphs (DAG). DAGs combine mathematical graph theory with statistical probability concepts and provide a powerful approach to causal reasoning.

Originally developed in the computer science and artificial intelligence field, they recently gained increasing traction also in other scientific disciplines (such as machine learning, economics, finance, health sciences, and philosophy). DAGs allow to check the validity of causal statements based on intuitive graphical criteria, that do not require algebra.

In addition, they open the possibility to completely automatize the causal inference task with the help of special identification algorithms. As an encompassing framework for causal thinking, DAGs are becoming an essential tool for everyone interested in data science and machine learning.

Syllabus

• Introduction to Causal Data Science
Overview of Causal Inference
Importance of Causal Reasoning in Data Science
Historical Context and Applications of DAGs
• Fundamentals of Directed Acyclic Graphs (DAGs)
Basic Concepts in Graph Theory
Properties of DAGs
Graphical Representation of Variables and Causal Relationships
• Building and Interpreting DAGs
Constructing DAGs from Domain Knowledge
Common Patterns and Structures in DAGs
Identifying Causal Paths and Blocking Paths
• Causal Identification with DAGs
D-separation and Conditional Independence
Graphical Criteria for Causal Statements
Identifying Confounders and Adjusting for Bias
• Algorithms for Causal Inference
Overview of Causal Identification Algorithms
Introduction to Pearl’s Do-Calculus
Automatic Causal Inference from Data
• Applications of DAGs in Various Disciplines
Use Cases in Machine Learning
Causal Frameworks in Economics and Finance
Applications in Health Sciences and Epidemiology
• Practical Exercises with DAGs
Hands-on Construction and Analysis of DAGs
Applying DAGs to Real-world Problems
Software Tools for DAG Analysis (e.g., DAGitty, bnlearn)
• Advanced Topics in DAGs
Structural Causal Models (SCM)
Feedback and Cyclic Models
Recent Developments and Future Directions
• Summary and Review
Key Takeaways from the Course
Integration of DAGs with Other Causal Methods
Further Reading and Resources
• Final Project
Design and Analysis of a DAG-based Causal Study
Presentation and Discussion of Findings

---

Taught by

Paul Hünermund

---

Subjects

Data Science