How to lie with Twitter polls
Category : rants
Twitter has a polling feature. It’s not hard to use, but for some reason many people don’t like using it. Maybe they access Twitter through an app that doesn’t support polls, or maybe they just aren’t very good at clicking the right buttons; it doesn’t matter. Instead of using the simple poll-creating tool they go with an alternative method that uses standard Twitter functions. In its way this is even simpler; you just ask a question in this format:
“What’s your favourite drink? Retweet for organic raw kale smoothies; favourite for beer.”
This sort of poll is pointless, and people need to stop doing it.
I’m actually astonished that I need to explain why it’s pointless, but apparently I do, so here’s a worked example. We are going to ask the question I posed in the first paragraph, and see how many people prefer kale smoothies to beer. To do this we’ll make some assumptions, just to keep the numbers tidy, but you’ll be able to see how it works in reality. Here are the assumptions:
1. The average person has ten followers on Twitter
2. For every nine sensible people who prefer beer, there’s one sandal-wearing loon who prefers kale smoothies
3. People who follow weirdo kale drinkers are three times as likely to be weirdo kale drinkers themselves
4. The average dumb question will be retweeted ten times before it dies of its own stupidity
So, based on these assumptions, let’s run the poll and see how it goes.
The genius who thought this was a sensible question to ask tweets it to his ten followers. Nine of them, not being insane, favourite the tweet. The other one, peering at the screen through a choking cloud of lentil farts, retweets it.
So far the score is as follows:
Beer: 9 votes
Unappetising green slime: 1 vote
So beer has a 90% rating so far.
Because some maniac retweeted it, another 10 people receive the poll. They all follow grass munchers so are three times as likely to be grass munchers themselves. Seven of them think, “Beer, of course,” and favourite it. The other three think “Yummy, liquidised weeds” and retweet. Let’s look at the running total:
Beer: 16 votes
Compost in a glass: 4 votes
Despite beer being as superior to blended garden waste as ever, its rating has suddenly dropped to 80%.
Thirty more people receive the poll; all of them follow bacon dodgers. Nevertheless 21 of them would rather have a glass of hoppy, refreshing bitter than a pint of revolting vegetable matter. They favourite the tweet. The other nine brush a handful of granola crumbs off their screen and retweet it. The scores continue to change:
Beer: 37 votes
Sheep vomit: 13 votes
Beer’s apparent popularity relative to semi-edible pond scum continues its decline, with a rating of 74%.
I won’t bore you with too much maths, so let’s skip to the tenth iteration, at which point the poll fizzles out. Here is the final score:
Beer: 68,896 votes. This is a lot of votes, because beer is good.
Ick. 29,524 votes. This is a lot fewer votes, because diluted mulch is revolting.
But let’s look at the percentages. Despite 90% of people actually preferring beer, gritty liquid fertiliser is now scoring 30% in the poll. In just ten generations it has trebled its apparent – but not its real – popularity. Why is that?
It’s fucking obvious, of course. Because of the way the poll is set up, people who favour one option will retweet it while those who favour the other option will not. And because we tend to follow people who have generally similar interests to ourselves, with every retweet the poll becomes more biased towards one answer. This is inevitable. There is nothing you can do about it. And these things can circulate on Twitter for days, becoming less accurate with every click.
Polls like this do not give accurate answers. They cannot give accurate answers. Stop creating them.