This course develops a unified inference toolkit for real-world data where the data-collection mechanism is nontrivial: survey sampling, missing data, and causal inference. A recurring theme is that valid inference requires explicit reasoning about the mechanism—sampling design, nonresponse/missingness, or treatment assignment—and that modern methods (IPW, doubly robust estimators, semiparametric efficiency, ML-assisted inference) can be understood within a common framework. To motivate these ideas, we take a historical perspective, tracing how classical problems and early debates in statistics led to today’s emphasis on valid inference, efficiency, and robustness, and using these milestones to cultivate statistical thinking.

Syllabus

Lecture 1

Lecture 2

HW1

Lecture 3