We provide two empirical examples in the paper:
- Estimating GE effects of cash-transfers when we cannot run country-wide experiments
- Site selection for microfinance interventions – relevant when we can only include a small fraction of villages in our experiment.

The solution balances (i) variance over the smallest achievable variance and (ii) a sensitivity component that captures bias rescaled by its smallest attainable value. (ii) only depends on parameters known ex-ante without requiring priors on the bias. It formalizes trade-off betw/ variance and bias.

A challenge is that observational studies may be biased in unknown ways. We study minmax regret without requiring knowledge of the bias. We optimize
- Which experiment to run (arms/sites)
- Sample size allocation
- How to combine experimental and observational evidence

The full order: www.whitehouse.gov/presidential...

Because this cam out of the blue and sudden: other, similar orders could come out of the blue and sudden.

So those on visas will surely consider international travel a potential risk, and opt for zoom calls over flying when possible