Hal Varian: Causal Inference, Econometrics, and Big Data

Lecture
https://www.youtube.com/watch?v=8xa-hSlJFo0

Lecture Notes
http://www.nasonline.org/programs/sackler-colloquia/documents/varian.pdf

• Train: a model on training data
• Test: a model on holdout data (placebo experiment)
• Treat: apply treatment to treatment group
• Compare: treated to predicted counterfactual (model and/or control group)

Machine Learning folks very good at training and testing data. And, identifying patterns and making predictions.

Social Scientists much better at looking at causal inferences.