Slashdot Mirror
Experimenting On Mechanical Turk
itwbennett writes "In a recent article, Dr. Markus Jakobsson, a Principal Scientist at PARC, offers some tips on effectively running human-subject research studies on Amazon's Mechanical Turk. '...[B]enefits [include] very low experiment costs, quick turn-around rates, and relatively simple approvals from human subjects boards. But you have to be careful to avoid bias and error.' says Dr. Jakobsson. For example, in many situations subjects may be biased just from knowing that they are participating in a study, or by knowing the goals of a study. To avoid this bias, you need to 'convey a different task to your subject than what you are observing — essentially deceive them — to see how they react when faced with the situation of interest. Consider a study of user reactions to phishing sites. You may, for example, say that you are studying the common reaction to online e-commerce sites, and ask them to rate how helpful various sites are, with a free-text input field where they can add other observations. You first show them three or four legitimate websites, asking them to rate and describe them; then you show them a phishing site and do the same. Will they tell you that this is a site run by fraudsters? If they do, they noticed signs of fraud without you prompting them.'" The author also gives tips on avoiding cheaters, and determining how much to pay and when.
For mechanical turkish delight.
Sent from my PDP-11
The title makes it sounds so exciting, like we're experimenting with our robotic Turks.
But the statistician within me is also fascinated with this. It always made me wonder, as the human mind can precondition itself. The study about whether or not prayer helps the sickly followed this mindset, and since the sick humans had no idea what the study was about, when they were told that people were praying for them they thought they were much sicker and actually recovered more slowly. If they hadn't told the sick humans, however, would there have been a large difference?
Math and the brain, it's amazing how they meet.
I often see websites where I feel there is a hidden agenda other than to make me happy enjoying the content to make the website owners some money. But people get savvy over time.. and it will start to get obvious. One thing is for sure.. you can only do this at most a few times before individuals start to figure it out, or you have get a new pool every time. Unless, you are looking for possible even more real world scenarios where there are a lot of websites that are frauds...
Like a city whose walls are broken down is a man who lacks self-control.
If I was in a study gauging the helpfulness of various websites and one I was shown a fake website, I would simply assume they ran out of, or could not have access too the real site and so were using a mockup.
If the mockup was particularly bad I might tell them, but otherwise I'd probably chuckle to myself and then just rate it as usual.
There is a huge margin of error with this type of thinking.
i make my mechanical Turk have sex with my mechanical JD.
Great article, however you should realize it's impossible to completely avoid bias. For example, he has restricted his population to MTurk users while generalizing to the population of web users. He also "weeded out" the lazy people for the convenience of his experiment - aren't there lazy people in the real world?
"Before God we are all equally wise - and equally foolish"
Albert Einstein
The subject basically says it all. If you're conducting a study, either for academia or industry, do yourself a favor and take a good research methods class. I can't count then number of promising studies I've rejected for publication because their methods were poor. While Dr. Jacobson makes some good points, most of them are pretty obvious to anyone who has taken a good class on creating experiments with humans -- for example, deception is a cornerstone of many human studies. Also, for you budding young scientists, make sure you get IRB approval before conducting your study. I reviewed a paper where the authors were clearly from a University and I had questions about the ethics of their methods so I asked for the IRB data for the study...whoops, they never got it. The paper was withdrawn by the authors shortly after.
My Slashdot account is old enough to drink...
When I was a kid, my parents received a free episode of some potential new comedy show to evaluate for them. We watched it, commercials and all. (Yeah, the commercials were a surprise.) At the end, unsurprisingly, the questions were actually about the commercials and only the last was about the show. It didn't fool us at all, of course.
"If you make people think they're thinking, they'll love you; But if you really make them think, they'll hate you." - DM
But I am a digital Turk you insensitive clod ! Allah cezanizi verir insallah ...
The ability to ignore extraneous data is basically what separates the good mind from the bad. If some procedure requires me to concentrate on X, and Y is not part of X, there's no way in hell I'm going to concern myself with Y unless it's threatening to shoot me in the face. Put at another way, I aim to be highly perceptive where I need to be, and in doing so I aim to filter out everything else. I conjecture that this is how many people who are focused in their work tend to act.
Applied to this example, when entering personal data on the web I *am* on the lookout for phishing attempts. When doing a supposedly controlled study, I'm not. Because I am focused, I would be likely to succeed in the real-world test but fail at the deceptive one.
(Actually, this isn't quite true, because when I am given a study to take part in I *always* add a "how might the tester be trying to deceive me?" factor to my response. But if your subject is always looking for ways you might be trying to deceive him, he's again not a valid subject. In conclusion, all controlled tests based on deception are invalid.)
Some funny shit right there. I also like the story about the third amendment rights group.
You are welcome on my lawn.
Thank you for saying this. IRBs exist for a reason and it's concerning to see an article like this essentially saying "Hey kids! Go do some human experimentation!"
This sig is not the Zahir. Lucky for you.
This is like psych 101, don't tell subjects what they are doing, or they'll just game you and your results will suck. Add internet. Same concept
Ze Atomic Device! It iz Ztolen!
Amazon Mechanical Turk, or MTurk for short, is a cloud computing platform that permits outsourcing of tasks to other users, using a built-in payment scheme to compensate users. People (often referred to as "Turkers") perform MTurk tasks, which are called Human Intelligence Tasks (HITs), and are paid just a few cents for completing them.
Although I'm still not sure what is mechanical or turkish about it. The Amazon part apparently refers to the fact that payment is made in way of credits to Amazon.com.
Damn_registrars has no butt-hole. Damn_registrars has no use for a butt-hole.
There've been enough comments about why you need to trick your participants, so I won't go into it, but the specific content of this study made me think.
I usually go to websites thinking that they are probably going to scam me, and don't let my guard down until they've demonstrated that they aren't. But in this situation, where I'm being directed to it as part of a survey on e-commerce sites, I'd probably show a lot less scepticism at the start, and would be less likely to flag it as if I had just stumbled onto it normally.
Ah, the things we have to take into account.
Is 1563649 a prime number?
You think a bunch of indians would care? As long as they get the 1 cent they will move on fast and type whatever.
Is it something like an Electric Monk?
Bow-ties are cool.
...for spotting the chess reference in the name.
Contrary to the article (at CMU at least) we _are_ required to run our Mturk studies past the IRB.
They all get rubber stamped 99% of the time, but it's still an important formality.