The Hunting of the Snark

by Joel Spolsky

on July 31, 2012

You can't fix what you can't measure, so the first thing we did as a part of our Summer of Love campaign was try to measure friendliness in an objective and repeatable way. We gathered 7000 comments from Stack Overflow and submitted them to Mechanical Turk. For each comment, we asked 20 people to rate the comment as Friendly, Unfriendly, or "Neutral/Unclear."

There are different ways of massaging the data, but I do want to give you a flavor for the kind of comments we're talking about when we're talking about unfriendly comments. Here is a snapshot of the complete results, showing comments where 95% of the reviewers rated a comment as unfriendly (warning, if there are any kids in the area, you may want to send them away):

> > > * Can you not google?! > > * Dean, don't be a f___ing lamer. You clearly don't have the slightest clue of what your ass… > > * Neurofluxation, haha f___ you should be f___ you kid > > * try to give answers. you baby kid shut up your mouth. this is forum to share problems not … > > * yea a__hole im saying send me the code > > * This is not correct, for many reasons, many of which were pointed out by plinth below. I d… > > * @Rich, It is clear (and always was clear) that you don;t care about what anybody else thin… > > * could you please stop reposting all your questions 4 times? > > * You know, I really dislike the attitude here that a question can only be asked once. > > * @cee: And rolling back is not intended to solve whatever gripes you may have with me. If y… > > * Now you're just proving your douchiness.  Editing your question to hide your true intent, … > > * man, you guys have no sense of humor.  I don't see how hundreds of idiotic and non-program… > > * NO. GOD. these comments are getting irritating. > > * Indeed. Although demonstrating a severe lack of ability to ask questions is a bad start. > > * I mean, really, WTF? > > * Don't vote down, the sooner I sabotage this the sooner we can tell the client f___ you and… > > * @TheTXI: That's exactly my point. If the person actually bothered to *google* the damn que… > > * u discuss all kind of questions here but when i ask a question and if u people are not abl… > > * @TStamper: If they haven't bothered to look through the FAQ, what makes you think they are… > > * It's amusing for a while, babby, but even the funniest jokes get tedious when they're done… > > * If you're so desperate to have your account removed why not just leave and not come back? … > > * He asks lot's of these troll questions > > * Seriously? WFT Dude? > > * Jesus! Start fixing your question. > > * Spencer, my tone? You sir are political correctness gone mad! > > * If I said your mouse sucked, are you gonna take me out to the parking lot and fight me?  i… > > * This isn't a programming question, it is a psychology question. It doesn't ask for an expl… > > * @mario why in your opinion should I not link to it? Because of the 5 pageviews the site ga… > > * Rec, you are not asking a question that can be answered in the form at the bottom of this … > > * @user336502, you're pushing your luck with cruddy questions (http://stackoverflow.com/ques… > > * Jebus, @AKA, did you even read your own question?  This is the worst piece of crap I've ev… > > * Jeez dont' people read web sites. What do you think Stackoverflow careers is for? This pla… > > * Hmya, how can 4 in 5 programmers be wrong?  Or 1 in 2?  We don't know how your brain works… > > * @M.H: Don't blame the language because you don't know how to use it. Don't blame the gun w… >

Of course, "friendliness" can be subjective. But when we're talking about making Stack Overflow a friendly place, we're not talking about being terse or even snippy -- we're talking about lighting a bag of dog poo on fire and throwing it at people.

Of 7000 comments submitted, there were 161 that were rated as "unfriendly" by 75% or more of the reviewers… that's about 2.3%. If you browse Stack Overflow for a few minutes, it's likely that you will come across one of these extremely unfriendly comments. Of those 2.3% extremely unfriendly comments, less than 1/5th have been deleted. Most of them are still on Stack Overflow right now.

The "friendliness" situation is much better, mainly because our reviewers tend to universally interpret thank you's as friendly.

> > > * +1 This is a good question, as this programming practice is even used in some big projects. (www.ogre3D.com for ex.) > > * @Visage Haha, thanks for that > > * Great analogy, @Guffa ! > > * Nice find, Chad Birch! > > * It was tagged javascript before Joel edited it. Clearly it is programming related :) > > * Thanks much for the feedback everyone > > * i wonder how did you manage to create the compoennts tag? ;) like your question +1 > > * Thanks for the suggestion ! > > * That's nice.  Let us know how it goes! > > * You're welcome... > > * hehe...that was fun! > > * Thanks, see my edit. > > * @Mike: Thank you. > > * Cool, will do that. > > * Okay, thanks all. > > * O_o That's very cool! +1 > > * @spudly - apologies, turns out the link I posted was a dud - sorry for wasting your time! > > * Love the question.  What you're after is domain knowledge which is exactly the type of information that a company guards because it's a barrier to entry to blokes in garages writing television clients :) > > * Cool. thank you! > > * Ha! I didn't even catch it in your post. :) > > * LOVING all the images that have been added :-) > > * thanks.. thats exactly what I was looking for :) > > * Thanks everyone > > * @meagar wow that looks awesome! I'm only about a several hour drive from there, I'll see if I can make it. Juggling for the win!!! :) >

Of 7000 comments submitted, 557 were rated as "friendly" by 75% of more of the reviewers.

I think this proves that the methodology is reasonably sound. I think everyone can agree that the Mechanical Turk reviewers, who were shown comments out of context and who probably did not know anything about our site (all they knew was that it was a "programmer's discussion forum"), did, nevertheless, produce results that seem to agree with how we, inside the community, would judge the comments. That gives me confidence that we have a reliable measure of friendliness that we can track.

There's a lot of other interesting stuff in the data, so here's an Excel workbook containing the raw data and friendliness ratings.