Causal Inference from Observational Data

Berkeley Methods Workshop, 2015 

New methods of causal inference are transforming science. This course provides an in-depth survey of modern causal inference for social and health researchers. We introduce key tools from statistics, economics, and computer science to learn when and how researchers can draw causal conclusions from observational (non-experimental) data. We adopt the modern potential outcomes framework of causal inference and draw liberally on graphical notation and conventional structural equations to aid understanding.

Topics include:

• The art of formulating causal questions
• Working with directed acyclic graphs (DAGs)
• Identification analysis: When can data reveal causal effects?
• Avoiding bias: confounding, selection, and overcontrol 
• Propensity score analysis
• Regression-based causal inference
• Problems of effect heterogeneity
• Causal mediation analysis
• Time-varying treatments 
• Marginal structural models

We mingle theoretical exposition with extensive hands-on training to develop transferable skills. We analyze many real applications and discuss when causal inference is possible and when it is not. The goal is to empower social scientists and health researchers to apply causal inference with confidence.

Participants should have a good applied background in multiple regression. The course does not require calculus or matrix algebra, though neither will hurt. We will execute empirical examples in Stata 13. 

Fee: $1500 (USD)    Register at RegOnline

UC Berkeley Graduate Students:  A limited number of fellowships will be available to cover up to $1400 of workshop fees. Please complete the Fellowship Application  to be considered.