"Invitation to Causal Inference"
Leonard Schulman (Caltech)
Monday, March 4, 2019, 2:00 pm
EBU3B, Room 4258
The scientific method rests upon the careful collection analysis of empirical data. Much of this data is gathered for the purpose of potential intervention in medicine, public health, environmental regulation, social policy, etc. A decision-maker cannot take advantage of correlations and other structural characterizations of the data without knowing about causal relationships between variables.
Historically, causality has been teased apart from correlation through controlled experiments. However, we must often make do with passive observation---and then this challenge has no easy answer.
Remarkably, there are some situations in which statistically defensible causal inference is possible even in the absence of controlled experiments. This is the subject of the theory of structured causal models which has been developed mainly in the last three decades. I plan to briefly describe the framework and mention (a) some results, joint with Piyush Srivastava, about the robustness of this theory; (b) some ongoing work with students relating to inference of mixture models; (c) many open questions.