Hacker News new | ask | show | jobs
by geofft 1809 days ago
But those are all examples of data collection working fine, though, and people not listening to the data because their anecdotes went the wrong way.

FiveThirtyEight predicted a 28.6% chance of a Trump victory: https://projects.fivethirtyeight.com/2016-election-forecast/ This is a number far greater than zero - it is, for comparison, more likely than flipping two coins and them both coming up heads, or drawing a random card from a deck and having it be diamonds.

Opinion polling of the 2019 Conservative Party leadership election showed Boris Johnson favored to win: https://en.wikipedia.org/wiki/Opinion_polling_for_the_2019_C...

Opinion polling on Brexit showed that it clearly could have gone either way, with a narrow margin between the two options: https://en.wikipedia.org/wiki/Opinion_polling_for_the_United...

I'm less familiar with Australian politics, but assuming you mean the 2013 election, opinion polls showed the Liberals were clearly in the lead: https://en.wikipedia.org/wiki/Opinion_polling_for_the_2013_A...

If people look at data that says "28.6% chance of winning" and they choose to read it as "0% chance of winning," that's not a problem with the way anyone's measuring anything, that's a problem with people trying to treat data as if it's an anecdote.

3 comments

I agree it's a bad example re: the Bezos quote, but I also don't think it's an example of anecdotes leading people astray. IMO it's from the brand of bad pop sci that misinterprets perfectly good data while ranting about how amazing science is and how stupid anecdotes are.
> I'm less familiar with Australian politics, but assuming you mean the 2013 election

They definitely mean the 2019 election.

> FiveThirtyEight predicted a 28.6% chance of a Trump victory:

This is cherry-picking of the highest widely-known prediction by far, practically an outlier. And I don't think they do any actual polling, just analysis using others' polling data.