Statistical methods are discussed for inferring causal effects from data from randomized experiments or observational studies.
Examples will come from many disciplines: economics, education, other social sciences, epidemiology, and biomedical science. Specific examples include evaluations of job training programs, educational voucher schemes, medical treatments, smoking, and military service.
The primary textbook
is "Causal Inference for Statistics, Social, and BiomedicalSciences: An Introduction", by Guido W. Imbens and Donald B. Rubin, Cambridge University Press (2015)
Additional journal articles for discussion will also be made
available.
Learning Objectives
Students will develop expertise to assess the credibility of causal claims and the ability to apply the relevant statistical methods for causal analyses.
Prerequisites
Statistical Inference
Teaching Methods
Lectures, presentations of case studies.
Type of Assessment
Final score will be based on evaluation of home assignments, 1 mid-term exam, one final project with oral discussion.
Course program
Part I:
The Basic Framework
-A Brief History of the Potential
Outcome Approach
to Causal Inference
- A Taxonomy of Assignment Mechanisms
Part II:
Classical Randomized Experiments
- A Taxonomy of Classical Randomized Experiments - Fisher’s Exact P‐values
for Completely Randomized Experiments -
Neyman’s Repeated Sampling
Approach to Completely Randomized Experiments
Regression Methods for Completely
Randomized Experiments
- Model‐based Inference in Completely Randomized Experiments
- Stratified Randomized Experiments
- Paired Randomized Experiments
Part III
Regular Assignment Mechanisms
– Unconfounded Treatment Assignment
– Estimating the Propensity Score
– Assessing Overlap in Covariate Distributions
- Design in Observational Studies: Matching and Subclassification - Sensitivity Anlalysis
Part IV:
Irregular Assignment Mechanisms
Non‐Compliance and Instrumental Variables Analysis
- Principal Stratification
- Additional topics depending on interest and potentially including: Rgression Discontinuity Designs, Broken Randomized Experiments, Recent
Developments in Causal Inference