Hacker News Reader

Comment by yohann305 | Hacker News Reader

Conducting polls can be quite misleading. Remember the last US presidency election polls? They were totally inaccurate* Hedge funds could have lost a lot of money on this one.

*: https://www.nytimes.com/2017/05/31/upshot/a-2016-review-why-...

View on HN

The polls were not that bad: https://fivethirtyeight.com/features/the-polls-are-all-right.... It's just that the outcome of US presidential elections can be changed by relatively small shifts in a handful of swing states.

> relatively small shifts in a handful of swing states.

And if a poll doesn't reflect that reality then it's not a good poll.

Forecasting isn't betting. One forecast doesn't tell you anything. If a weather forecast that says there's a one in five chance of rain and it rains, that's not wrong. They did tell you there's a chance.

To show whether a forecaster is good at what they do, you need to track their results for multiple forecasts and see what their track record is.

They were pretty indicative though. Trump only outperformed the national poll by 1-2 points and 2-3 points in the swing states. The whole 'polls are garbage' story that came out of 2016 isn't really supported by the data if you look at it.

The error has been pretty consistent forever, it's just hard to do a poll and have it line up exactly with reality because a poll is a sample in time of the thoughts of a subset that may not match who shows up. All of those factors go into why polls miss; some populations are harder to reach than others, news can come out between the poll and election day or bad weather could mess up your turnout model by keeping a lot of people home.

https://fivethirtyeight.com/features/the-polls-are-all-right...

> The whole 'polls are garbage' story that came out of 2016 isn't really supported by the data if you look at it.

There were a lot of garbage poll-based predictions (the Clinton 90+% ones) based on assuming independence of polling errors. Predictions with claims of quantified likelihood based on polls are a fairly recent media phenomenon, and I think some of the legitimate reaction to how bad many of those were spilled over into polls themselves, because people don't understand that the polls weren't the source of the problem. (Which fed into existing political narratives against polling; propaganda definitely played a role but there was a real issue it interacted with.)

Yeah the use of the polls was the real failure in 2016. A narrative and conventional wisdom bias set in to the analysis in a lot of places that discounted close results in important states.

I don't remember seeing any polls asking 'who will win the national election?'. All the ones I saw attempted to answer the question 'who will individual people vote for in this state/nation?'.

Are you perhaps thinking of a model?

Except for the one poll that was accurate, but everyone was giving it shit because it was giving a result they didn't want to hear.

http://graphics.latimes.com/usc-presidential-poll-dashboard/

That poll had Trump winning the national popular vote by 3.2 points. He lost by 2.1 points. The poll missed by a total of 5.3 points.

The average poll missed by 4.8 points. This result wasn't much worse than average, but it definitely was below average.

I can see why someone might be confused by this - one might think these polls are predicting who will become president.

Maybe the polls were bad because the people conducting them were pushing to have sufficient difference between private and public polls in order to maximize their profits.

True, although exit polls such as here probably can be more reliable than asking days beforehand, especially to capture voter turnout.

Why the downvote? Isn't this comment bringing good critic?

No. The polls were fairly accurate; the problem was how people used the polls. This is also a topic that's been beaten to death over the last two years.

Fair point - the Ashcroft polls, funded by the UK's Conservative party, called the 2015 election as a tie, although it turned out to be a major victory by the Conservatives.

https://en.wikipedia.org/wiki/2015_United_Kingdom_general_el...

Reply on news.ycombinator.com