Wednesday, 30 April 2014

Liking vs Sharing Bias

Sometimes I come across posts on Twitter or Facebook that say "Like/Fav for Option A; Share/RT for Option B"! In fact, give me two seconds, an example won't be hard to find...
This system is flawed!

Am I the only person who gets actively annoyed whenever I see that sort of post? It seems to ignore the fact that if you SHARE something, more people will SEE it - completely screwing up your counting system!!

Okay, so maybe people don't care because "it's all in fun". But it worries me that some people may honestly be blind to the mathematics involved.

NOT ROCKET SCIENCE


For instance, pretend I'm running a poll on my favourite serial character - Para (Fave) or ParaB (ReTweet).  I have 448 followers, and we'll pretend they know what the heck I'm talking about. Notice that as long as people like Para (Fave), the only way anyone else is going to see this message is if they happen to search for one of the words I used.

But perhaps Audrey McLaren (a follower) prefers ParaB, so she ReTweets. And Audrey has 1,456 followers. There is now a MUCH larger audience! The single retweet means up to three times as many votes are possible (might be less - crossover friends). Simultaneously, as most humans tend to follow like-minded people, many of those new votes on her side would also be for ParaB, creating more ReTweets, which throw off the numbers even more. (Insert your favourite sports team there, if my analogy is breaking down.) But it's even worse than that.

In the unlikely event that I'm following someone, even though I disagree with most of their life/sports choices, the most I can do to disagree with their "ReTweet" is to "Fave", thereby cancelling it out. I cannot ReTweet it to my followers for them to "Fave" as well, because that would mean I'm voting for the other side! (I suppose I could do both, cancelling myself out, hoping a follower cancels out the ReTweet that brought it to me, but I doubt that this much thought is involved when you see these informal polls.)

In other words, once something starts getting ReTweeted, it gains perpetual motion, thanks to like-minded individuals... until that motion is negated by Favourites. Or more likely by Apathy. Which I think is really what you end up measuring.  To wit, I foresee two possible interpretations from every single one of these "polls":

1) The Favourites win. In other words, the post didn't get very far away from the source - or possibly it got so far away that the immense new population was able to trounce the people that brought said post to that wider audience. Conclusion: Everyone got riled up and shot the messenger(s).

2) The ReTweets win. In other words, the post got out there, and when most people saw it they went "meh", rather than voting against the people that brought said post to the wider audience. Conclusion: Everyone couldn't care less.

I suppose you could say you're measuring the popularity of the question itself, rather than the actual choices offered. (More faves = good question, more RTs = lame question.) Which might be clever marketing, but to me really serves no point as far as answering the question as posed.

OPEN QUESTION


So here's a thought. Is it actually possible to create an algorithm which adjusts to the constantly fluctuating population? Which can actually factor in percentages to know how many people honestly DID favour choice A or B over choice C? (Choice C being: Stop, Stop, Stop It, Stop.) I doubt it.

After all, non-response bias occurs when your survey results are influenced owing to the fact that the majority of responses were due to people with very strong opinions. (As opposed to a more typical person.) At best, I think that's what's happening. Meaning an algorithm which fixes the problem... would also fix a huge problem in regular data gathering.

But I might be way off here. Do you think I'm way off here? Comment on the post if you agree with me. Share it with others if you don't agree.

No comments:

Post a Comment