If Nate Silver had thrown out the Bad Polls would he have been more Accurate?

[cthulu2016]

The answer is, of course, no.

Nate Silver could not have been much more accurate. (Particularly when you understand that his EV prediction is an average of thousands of simulated elections, not a prediction of a specific electoral map)

And to the tiny degree he was inaccurate it was not that he was too pro-Republican.

Without the "bad" polls, which I agree are bad polls, Nate Silver would have missed on the popular vote, and would have had North Carolina as leaning Obama. Ohio was probably a little bit closer than Nate Silver anticipated.

This is a lesson about poll aggregation that is slow to learn. You cannot throw out obviously bad polls unless you are 100% sure you are throwing out all of the bad polls, including the less obviously bad polls.

2001, the Walker recall, last night... all of these would have been missed if the obviously bad bad polls had been thrown out.


I don't seek to explain the effect here, merely to note it.

For whatever reason, in the world as it exists, including the "bad" polls, with adjustments such as Silver made for methodology (robocalling, etc.) yields better results than excluding them.

It is possible to manufacture so much bad polling that this would no longer be the case, but we are apparently not at that point.

For whatever reason.

[pnwmom]

1. That thought had occurred to me, too. Without the bad polls, it would appear Obama

had more support than he did in the final analysis.

It's kind of amazing, though, that all that messiness could end in such a clearcut result.

[cthulu2016]

2. It is counter-intuitive

but there were obviously some "bad" polls in our favor also, but probably not idealogically motivated.

PPP was the closest, and they do robocalls. Tough to figure.