Missing Data Theory and Causal Effects

Course goals

Overview

Missing data seriously complicate statistical data analysis, and the means by which missing data are treated can substantially impact statistical results. Yet, missing data are not always a simple data quality issue. Indeed, many statistical inference and estimation problems can be reframed as missing data problems. Causal inference is a prime example. This course will introduce the fundamentals of missing data theory and causal modeling and explore the connections between these fields. Students will learn how to evaluate the extent of a missing data problem and treat missing data via (multiple) imputation. With these techniques in hand, students will be able to apply familiar statistical methods (e.g., regression, ANOVA) to incomplete data while minimizing the deleterious effects of the missing data. Students will also learn how to identify a causal effect and how to estimate that effect from data. With these skills, students will be able to critically evaluate any causal effects they may estimate in their own work as well as the strength of causal claims reported in the literature.

Learning Goals

Students who successfully complete this course will be able to:

1. Describe fundamental concepts in missing data theory and apply these concepts in hypothetical research scenarios
   1. Missing data mechanisms
   2. Ad hoc missing data treatments and their consequences
   3. The multiple imputation (MI) algorithm and the procedure for MI-based data analysis
   4. The strengths and weaknesses of MI relative to ad hoc missing data treatments
2. Describe fundamental concepts in causal inference and apply these concepts in hypothetical research scenarios
   1. The definition of a causal effect
   2. Counterfactuals and the potential outcomes framework
   3. Graphical methods for causal inference and the fundamental causal structures
   4. Empirical methods for identifying/isolating causal effects in data
3. Describe how missing data theory and causal inference are related
4. Use open-source software to apply basic methodological and statistical techniques for missing data analysis
5. Use open-source software to apply basic methodological and statistical techniques for causal inference
6. Use a markup language (e.g., Quarto) to document statistical analyses as reproducible reports

Relation between Assessment and Objective

1. Final exam:
   1. Demonstrate theoretical knowledge of the concepts underlying missing data analysis and causal inference (Learning Goals 1, 2, 3)
   2. Apply the above concepts to hypothetical research scenarios and interpret related statistical software output (Learning Goals 1, 2, 3)
2. Data analysis assignments:
   1. Analyze incomplete data and draw meaningful statistical inferences therefrom (Learning Goal 4)
   2. Identify and estimate causal effects in real data (Learning Goal 5)
   3. Use a markup language to document the above analyses as reproducible reports (Learning Goal 6)

In addition to these two types of summative assessment, weekly practical exercises provide formative assessment of Learning Goals 4, 5, and 6.

Content

During every lecture, we will treat a different theoretical aspect of missing data analysis or causal inference. Through weekly practical exercises (to be completed at home), you will connect the statistical theory to practice by using open-source software to apply the techniques from the lecture. You will also attend weekly workgroup meetings wherein you will work on your data analysis assignments with a group of your peers. 

Assumed knowledge: Basic theoretical understanding of statistical testing (e.g., t-test, ANOVA), linear regression, and correlation. Participants should have some experience using statistical software (e.g., R, JASP, SAS, STATA, SPSS) to conduct statistical tests and linear regression analyses.

Note: Students who cannot comply with the entrance requirements mentioned below will be asked to provide further information on their eligibility. The course coordinator will decide on their eligibility.