So, you did a survey on a self-selected group of people who think their time is worth less than a buck an hour, and asked them questions about behavioural economics?

Maybe I'm missing something.

The point is to get access to the utility function (versus perception of expected value) across a relatively constant set of people. I'm not sure how they value their own time is relevant.

First of all, it's a self-selected group of people-- not a very good sample. Secondly, and more interestingly, these are people who are self-selected precisely along a very specific axis, one regarding a non-normative notion of utility-- the very thing you are asking them about.

Very cool. I think there is a tremendous amount of neat hacks like this to be done with ever more available data.

To make it more than a 'fun experiment', you'd have to actually offer those kinds of terms. Actually, what might be interesting is to see the difference between what people say they would accept, and what they would choose if they were getting a real offer.

I think the problem is that there are other tasks that require a high task approval rate, so by rejecting people's answers (so they don't get a payout) they are otherwise adversely affected. Perhaps there is a way to programmatically bonus people.

Also I don't think there's an easy way to force a choice between taking different tasks.

Love it, but the real way to test this is to present the monetary choice itself. For example, they could pick to get payed 1 cent with 100% certainty or 25 cents with 4% certainty. (although I don't think mturk supports this natively) EDIT: right, exactly as been suggested by other commenters!

Sure, but at lower prices, you know they'll choose the higher payoff with higher risk.

The experiment's results aren't even vaguely surprising. It's that it was so easy and rapid to conduct.

I'm still thinking about how to phrase a question around "free"