News for hackers. Stuff that matters.
Re:Causal Information Affects Decisions by gwern on Friday December 10, 2021 @02:49PM (#29514659)
It is more common than you realize. Turkers are all over the place in psychology, sociology, machine learning, etc. It's something of a methodological issue: Turkers are extremely convenient and cheap, but they are also not necessarily representative of any realistic population, and the data can be absolute garbage sometimes... but at other times, they're great and they get you results at 1/10th or better the cost, so you can't just write off Turk entirely. Too useful to not use a lot, too dangerous to use blithely.
Re:Causal Information Affects Decisions by dsizzle on Friday December 10, 2021 @09:29PM (#29517859)
Are they less representative than other common sources of subjects, like undergrads?
Re:Causal Information Affects Decisions by gwern on Saturday December 11, 2021 @06:43PM (#29525406)
That's difficult to say. Undergrads are at least less likely to be composed of a bunch of bots randomly-clicking on answers and adversarially bypassing your attention checks.