Missing Data - Theory and Methods
[Syllabus]
- Lectures notes [merged]
- Chapter 1. Introduction [slides; addendum]
- Chapter 2. The EM: Basic Theory and Practice [slides; addendum]
- Chapter 3. MLE in Regression Models with Missing Covariates [slides]
- Chapter 4. Advanced Topics of the EM [slides]
- Chapter 5. Multiple Imputation and Bayesian Analysis [slides]
- Chapter 6. The Estimating Equation Approach [slides]
- Chapter 7. A Review [slides]
- Problem sets
- R code
An illustration of non-identifiability with not-missing-at-random (NMAR) data.
Left column: full data distributions; right column: observed data distributions. A: missing at random; B and C: not missing at random. Despite varied full-data distributions (left column), when combined with appropriate missingness mechanisms, all lead to seemingly identical distributions of observed data (right column).