Paul GrahamVerified account

And then another adjusted poll accounting for education

Keep in mind that’s pre-comey letter tho. Race tightened after that. Also - many pollsters have adjusted assumptions based on 2016 polling errors.

Key point: that 2016 poll occurred before the Comey letter

True, but n is larger and it’s closer to election this time. That extra column alone could be misleading.

something to keep in mind is that you'd rather be leading 49-43 than 46-39, because there's a lot more undecideds in the latter. At the end of the day, hillary did get 47% of the vote in Pennsylvania, so their poll wasn't too off in that regard

That Siena poll was two weeks before the election. Siena polls conducted closer to the election (in other states; that was the last in PA) showed a tight race. Late-deciding voters (and there were more than usual in 2016) went for Trump.

Good call! It does not appear that NYT / Siena polled PA post-Comey letter (10/28) so this was the latest. 2016 also had 15% undecided vs. the only 8% undecided in the final Biden-Trump poll. So I would say it seems like a more robust result this year but no sure thing.

In a number of ways it’s a bit unfair to call that “the same poll”, given timing, sampling methodology etc

Not a like to like comparison of the turnout model and weighting has changed. Which is very likely the case. Particularly around non college white voters

But that comparison is only valid if the 2016 poll sample leaned R. Then you’re isolating a flawed methodology and can extrapolate the polling error.