By Franklin Kramer
Crowdsourcing web service Have turkers complete HITs for small amounts of money (most being 1-25 cents) Can filter workers by specific criteria Typical worker earns $1.40 (Horton & Chilton) US Turkers are fairly representative of general population- much better than University participants (Ipeirotis) What is Mechanical Turk?
Original experiment (Robert Mille) 31/19 Coin Flips with Bonuses (same or different) 32/18, 23/27 Coin Flips: A further look (time) Die rolls (Suri et. al) Background
If people are going to lie, they will tend to choose the first result offered. People are more likely to give honest answers when dishonesty is more transparent in potential results given. Price will have little effect on results. People are less likely to lie when they are more aware of what is at stake. People will be less likely to lie when they feel representative of a group they belong to. Hypotheses to Test:
Lying is a very interesting social phenomena Benefit to lying minimal Only detectable on large scale MT is fairly representative Valuable to understand how to best give valid results in MT Why?