NH was finally called, for Clinton. I'm going to assume MI is gonna be called for Trump. If it's called for Clinton, just note that I got 1 more Likely state right.
- 100% accuracy on Solid states. That's, by my count, 31 states. (Note that this discludes one of Maine's and Nebraska's districts, which were tossups.)
- 80% accuracy on Likely states. There were 10 Likely states, and 8 went to who I expected.
- 83% accuracy on Lean states. There were 6 Lean states, and I got 5 right.
- Simply due to what I labeled "Tossup" states, no candidate should have won more than 3 of the 5. Since Trump took all of them, I count that as a 60% accuracy with Tossup states, with 2 wrong.
So, I got 5 wrong out of 53 predictions, for an overall accuracy of 94%.
Not bad, if I say so myself, even if I got the final results wrong.
North Dakota and Hawaii haven't finished their tallying yet for Senate, according to Decision Desk HQ, but Hawaii so far (10% in) looks strongly Dem (no surprise there) and ND is a strong Repubilcan state, so I'm going to assume from the get-go that they're going that way. If they don't, count that as two Solid states wrong.
- Out of the 24 Solid senate seats, all 24 were right.
- Out of the 3 Likely senate seats, all 3 were right.
- Out of the 4 Lean senate seats, I got 2 right. The two I got wrong were Pennsylvania and Wisconsin, which I expected to go Democratic.
- Due to the way the math works with tossup seats, I effectively got all 3 of them right.
This means I got 32 of the 34 seats effectively right, for a 94% accuracy. Also, I predicted a Republican lean for the Senate, and they won.
I'm not going to go seat-by-seat in the House, but I can say, looking at it broadly, I got it about right. If I wanted to be more accurate, I would've put a few tossup states leaning Republican, but I still got the number right, give or take. If everyone took their lean and likely seats, and then took no more than they're "supposed to" out of the tossup seats, then you'd get what's about called right now.
Lastly, Governorships. I got all 5 solid and, seemingly, 1 of the 2 likely seats right. I say seemingly because North Carolina is one of those that continues to be contested, and may go to recount. I also got only 2 out of the 4 lean seats correct. That's 75% accuracy. I must admit that this was one of the weakest areas, and I had to make some guesses. Not much polling was available for some of these.
And with that, I make the case for the polls, since people seem to be thrashing them now.
If polling was so messed up, how could I, and forecasters (of whom I based most of my predictions off of) get accuracies 94% for the Presidency and Senate, and the House about right for seat numbers? The governorships were odd, because I had to use polling data that was, at times, extremely sparse and low quality. If I spent more time, I probably could've done better on that.
Right now, according to the Associated Press, Clinton had 47.66% of the vote, and Trump 47.5%. Any more numbers brought up aren't likely to move this by a lot, as far as i can tell. RCP got the popular vote off, by their count, 3.1 points. Here's an interesting article by 538, on November 4th, that's very relevant.
Even at the end of a presidential campaign, polls donāt perfectly predict the final margin in the election. Sometimes the final polls are quite accurate. An average of national polls in the week before the 2008 election had Barack Obama winning by 7.6 percentage points. He won by 7.3 points. Sometimes, however, the polls miss by more. Four years ago, an average of survey results the week before the election had Obama winning by 1.2 percentage points. He actually beat Mitt Romney by 3.9 points.
If that 2.7-point error doesnāt sound like very much to you, well, itās very close to what Donald Trump needs to overtake Hillary Clinton in the popular vote. She leads by 3.3 points in our polls-only forecast.
My "optimistic" Trump map was, as I said, a "very possible one". It had him at 299 electoral votes. While the map didn't play out exactly how I expected, it did go mostly how I expected.
If we look at the polls that were outliers in Trump's favor, they aren't faring too much better. LATimes had Trump 3 points over Clinton at the end, about the margin that the average of polls had Clinton over Trump, according to RCP and 538. In the end, that one ended up being about as wrong as everyone else. (Remember, these are national polls, predicting the national vote, not the winner of the election.)
IBD/TIPP had Trump up 2% (1% if you're looking at the 2-way match up instead of 4-way, which appears to be marginally more accurate in the end). So overall, they were better, but still off by about 2%.
Rasmussen's final poll had Clinton up 2%, which was about as wrong as IBD/TIPP. Given that Rasmussen has generally just had a history of being several points red compared to other pollsters, I'd chalk this up to house effect, and not true accuracy. Regardless, they were off by about a standard polling error, give or take.
What I'm trying to show through this is that, when you bother to actually handle the polls properly, they actually didn't do bad. As 538 and I believed, a Trump presidency was very possible. They had Trump at about 30% chance of winning on election night, and that was with polls that suddenly really liked the number 4.
The polls were off, by notable margins. But if you had come to know them well, you'd have known that it's not a surprise for the polls to be off by enough for Trump to win. Next time, the pollsters are going to try to fix whatever they didn't get this time, so the polling error is lower. That's how it works.
I rest my case.
Locked
