A/B testing is what people talk about, but in the IT world there are even better variants. See http://stevehanov.ca/blog/index.php?id=132 for an explanation of "bandit testing", which is a more adaptive approach which works better when the intermediate outcomes are important.
It's a really useful technique. I use it on testing variations of charity donation appeals, when the length of time to get a purist statistically significant result can mean you lose out on significant funds. Converging more rapidly on the answer which is likely to turn out to correct has more benefit. I would expect the same to be true for medical trials.
A/B testing is what people talk about, but in the IT world there are even better variants. See http://stevehanov.ca/blog/index.php?id=132 for an explanation of "bandit testing", which is a more adaptive approach which works better when the intermediate outcomes are important.
Thanks for this, I didn't know about bandit testing so this is really interesting reading.
It's a really useful technique. I use it on testing variations of charity donation appeals, when the length of time to get a purist statistically significant result can mean you lose out on significant funds. Converging more rapidly on the answer which is likely to turn out to correct has more benefit. I would expect the same to be true for medical trials.