2026 Notes

[nickeubank]

• After potential outcomes by hand, have people think about 2x2 grid. Answer: (1) what's observable, (2) what would no baseline differences be, and (3) what would ATT = ATU be?
• I would really love to come up with a full presentation on how hypothesis testing, MLE, and bayesian inference approaches all relate to decision theory. Maybe in an exercise?
  • I have tried this — in solutions_and_quizzes/250_bayesian_notautograded_yet I have my attempt. But I don't explicitly engage with Bernstein-von Mises theorem (Under regularity conditions, as n → ∞, the posterior distribution converges to a normal distribution centered at the MLE with variance equal to the inverse Fisher information. This is true regardless of the prior (as long as it's not too pathological and gives positive density near the true parameter)), I just bootstrap.