3 ms·

The polls were not that bad: https://fivethirtyeight.com/features/the-polls-are-all-right/ https://fivethirtyeight.com/features/the-polls-are-all-right.... It's

The polls were not that bad: https://fivethirtyeight.com/features/the-polls-are-all-right/ 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.

ng12 8y ago
> 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.
- rtkwe 8y ago
  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/ https://fivethirtyeight.com/features/the-polls-are-all-right...
  - dragonwriter 8y ago
    > 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.)
    - rtkwe 8y ago
      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.
- Armisael16 8y ago
  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?
- skybrian 8y ago
  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.

ng12 8y ago

> 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.

rtkwe 8y ago
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/ https://fivethirtyeight.com/features/the-polls-are-all-right...
- dragonwriter 8y ago
  > 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.)
  - rtkwe 8y ago
    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.
Armisael16 8y ago
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?
skybrian 8y ago
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.