Thread by @anup_malani

I want to highlight this MR post by @ATabarrok. It has AFAIK a novel approach to trials that warrants serious discussion: release a drug & randomize access via lottery. Then compare outcome across ppl that won lottery and those who lost. Like the Oregon Health Ins Experiment.

https://twitter.com/anup_malani/status/1320387684179824642

If we do this, we may want to subsidize the price, so that demand for the lottery is high. This may also address ethical issues.

But there are some important issues with this approach. Not all insurmountable. Here is a first cut.

1/Selection. Ppl who are randomized to no access will get other treatment. Some will take nothing, others will take another conventional treatment if available.

But this is a problem with existing trials. I noted an old article on selection into RCTs.

sciencedirect.com/science/articl…

I'm not sure this is a big or feature. I agree w eg @BartHamilton1, T Philipson & others, who argue that attrition can be used to measure value from treatment. The broader point is that we want to compare to real world controls. Not the control in trials.

journals.uchicago.edu/doi/abs/10.108…

If you like the design issues here, consider this question: what is the optimal price for subsidizing entry into the lottery? It is an analogous question to how trial wages affect selection into trials, and a question the selection into trial paper above examines.

2/Deciding when to do a "population" lottery. An extreme version of Alex's suggestion is Richard Epstein's ideas about only testing for safety before approving drugs: we'd get data on efficacy from the population. Alex's idea uses randomization to ensure we have causal estimates.

But this is analytically tractible. Earlier release increases risk and reduces demand for the lottery. Later release involves less efficient clinical testing. Bread & butter for economists.

A related issue is choosing outcomes. Should we do population lotteries just for efficacy or both safety & efficacy? This too fits within the framework in the last tweet.

BTW, those worried about ethics here should note that most product markets, even many dangerous ones (including non-FDA regulated medical care) use the population testing approach. Drugs are the exception. Elsewhere handle risk via exclusively via ex post tort liability.