3 ms·

The results of the 2016 election tell me that polling can be hilariously inaccurate and cannot account for things like the shy tory effect[1]. Everything else

The results of the 2016 election tell me that polling can be hilariously inaccurate and cannot account for things like the shy tory effect[1]. Everything else you're saying is spot on, though. [1]: https://en.wikipedia.org/wiki/Shy_Tory_Factor https://en.wikipedia.org/wiki/Shy_Tory_Factor

learc83 8y ago
The 2016 polling wasn't close to hilariously inaccurate. The average of the 13 final national polls had Clinton ahead by ~3 points, Clinton won the national vote by ~2. Some individual state polls were more wrong, but they were conducted less frequently and so many of them didn't fully factor in the latest news cycle developments.
- Karunamon 8y ago
  If those polls were measuring just raw voter preference rather than controlling for state (since the electoral college is what elects the President), we can add "intentionally dishonest" to the list of problems. "Who will win the popular vote" can be a datapoint since more often than not they line up, but it is absolutely less important overall. Even if we exclude mendacity and downgrade that to mere irresponsibility, terms like "98% chance of Hillary winning" were being thrown around left and right on election day. That speaks to something very, very wrong in how that data was being collected.
  - learc83 8y ago
    >That speaks to something very, very wrong in how that data was being collected The problem was mostly with how it was analyzed, not how it was collected. Some publications were making bad assumptions that state by state polling errors would end up being independent. 538, who didn't make that assumption, gave trump close to 1/3 chance of winning. Trump won by less than 80k votes in 3 states. Even the state polls weren't that far off--unlikely events happen.
dragonwriter 8y ago
> The results of the 2016 election tell me that polling can be hilariously inaccurate The polling was fairly accurate; given the nature of the electoral college and the fact that state-by-state deviations from polling results tend to be correlated rather than independent, and how close many of the state contests were, the winner was different than the expected winner (and this was very surprising to people who followed predictors who based their odds on the assumption that state variation was independent, less so those who followed those like 538 that pointed out that that was a mistake), but the polling was not particularly inaccurate.

learc83 8y ago

The 2016 polling wasn't close to hilariously inaccurate. The average of the 13 final national polls had Clinton ahead by ~3 points, Clinton won the national vote by ~2. Some individual state polls were more wrong, but they were conducted less frequently and so many of them didn't fully factor in the latest news cycle developments.

Karunamon 8y ago
If those polls were measuring just raw voter preference rather than controlling for state (since the electoral college is what elects the President), we can add "intentionally dishonest" to the list of problems. "Who will win the popular vote" can be a datapoint since more often than not they line up, but it is absolutely less important overall. Even if we exclude mendacity and downgrade that to mere irresponsibility, terms like "98% chance of Hillary winning" were being thrown around left and right on election day. That speaks to something very, very wrong in how that data was being collected.
- learc83 8y ago
  >That speaks to something very, very wrong in how that data was being collected The problem was mostly with how it was analyzed, not how it was collected. Some publications were making bad assumptions that state by state polling errors would end up being independent. 538, who didn't make that assumption, gave trump close to 1/3 chance of winning. Trump won by less than 80k votes in 3 states. Even the state polls weren't that far off--unlikely events happen.

dragonwriter 8y ago

> The results of the 2016 election tell me that polling can be hilariously inaccurate The polling was fairly accurate; given the nature of the electoral college and the fact that state-by-state deviations from polling results tend to be correlated rather than independent, and how close many of the state contests were, the winner was different than the expected winner (and this was very surprising to people who followed predictors who based their odds on the assumption that state variation was independent, less so those who followed those like 538 that pointed out that that was a mistake), but the polling was not particularly inaccurate.