The number of test subjects isn't small, but the number that have something bad happen is small (they're not going to accept test subjects with obvious, serious

The number of test subjects isn't small, but the number that have something bad happen is small (they're not going to accept test subjects with obvious, serious problems because that makes the data noisier.) Reducing noise means increasing the sample size--to get clean data here you would need a truly huge test group. I wouldn't be surprised if you would have to use the whole nation as your test group--and what's the point??