As I said in the predictions thread, I believe that this, combined with underpolling of young minority voters and Clinton’s greatly superior GOTV operation, is going to make her beat her numbers rather than vice versa.
Three reasons - you forgot wishful thinking. Or forgot to name it, at least.
#1) I don’t really buy this one. Not exactly. It could be that potential Trump voters are less likely to participate in polls. . . I can kinda see that. No trust in the “establishment” likely translates into no trust in pollsters, which is also a recent Trump mantra.
#2) Isn’t whether a voter is considered likely based upon their response to a question within the actual poll? So, I’m not sure I buy this either. Do pollsters get their phone lists from past voter rolls? I don’t think so.
I think the main thing that throws off polling is enthusiasm. I think Trump has more enthusiasm from a smaller pool of voters, and Hillary is counting on Anti-Trump enthusiasm from a larger pool of voters. I think it’s going to be close.
Voter registrations can indeed lapse, depending on state law. For example, here is North Carolina’s rule:
Ohio, to name another swing state, has been in and out of federal court over the methods it uses to remove inactive voters.
It’s the Great Trumpkin, Charlie Brown!
Just going to point out that the polling systems used under-count millennial voters who by and large reject trump.
Millennials do not use land-line phones. Phone surveys over-represent our population as elderly, and that’s the only demo where he’s competitive, outside of white men earning 50k a year.
That’s why people who try to “unskew” polls suck so badly at reading the electorate. There’s a good 20 percent of the country you’re not even getting opinions from, who sure, have a spotty track record voting in non presidential years. But this is a presidential year. A normal Republican has a VERY hard time winning when people 18-30 show up in massive numbers. Donald Trump is a “Republican” that will cause half of his millennial moderate Republican supporters to not vote for him.
Silver also addressed your #1 by pointing out that there was no evidence of a shy Trump vote in the primaries.
This criticism is about 10 years out of date. Modern pollsters call cell phones, and use appropriate demographic weighting to account for the rest of the response problem. Moreover, there are lots of polling methods in the field now, including non-phone polls, and they do not show significant divergence. So there is no real reason to think Millenials (who are pretty unreliable voters anyway) are being undercounted.
The central fact that most of these theories miss is that there’s a ton of data out there now: live phone polls, robot phone polls, landline polls, cell phone polls, polls in apps, sophisticated online polls, early voting, registration data, exit polls, etc. The idea that one of these methods is systematically under-counting someone is fine, but the idea that they are all systematically under-counting is very tough to believe. And the proof is in the pudding. Polls in the US in the last three decades have been very accurate, and none of these purported problems are new with Trump.
There is also what I observed months ago, the polls during the primaries had some failures but it was mostly in states with caucuses.
Overall though they were very accurate in predicting a Trump victory after Trump found his reprehensible voice.
But what was important to me was that at the same time Trump was beating the other Republicans with his bile, the same pollsters that pointed at a trump victory also reported that Clinton remained ahead of Trump in the match-up polls.
http://elections.huffingtonpost.com/pollster/2016-general-election-trump-vs-clinton
They still do.
They undercounted somebody in 2012. Obama significantly outperformed his final polling numbers.
Me trick-or-treating tonight in my Trump mask, with Tic-Tacs at the ready in case I run into any beautiful women and don’t want to wait: x.com
Calling cell phones is still more expensive (because robo-dialers can’t be used; federal law requires that the calls be dialed manually), so no, not every pollster uses cell phones.
For example, this evening’s updates on 538 include a batch of polls from Auto Alliance / ESA / Pulse Opinion Research, which according to their website uses “a combination of land line and online interviews.”
National polls being off by 1-2, as in 2012, is perfectly plausible. Polls being off by 6, not so much in the US.
Of course. But many of them do. And, very importantly for your thesis, there is not a systematic difference. When you aggregate multiple methods, you get reliable results when there is a bunch of polling.
It was actually more than three points, not “1-2”:
In any case though, I don’t understand what your point was in saying they are unlikely to be off by 6. So what?
If you use a more comprehensive aggregator like 538 or HuffPollster, the error is much smaller. RCP leans GOP.
And the point is that there are many sources of polling error. So a race within a couple points is a toss up, while a race outside that isn’t.
You have lost the thread.
I’d be careful not to overly interpret or excessively extrapolate from that one experiment with one specific Florida dataset.
The impact in that one experiment of using voting history as a sole criteria (the T+1 result, one extreme) and excluding those who claim they are unlikely to vote only approach (the C+4 result, the other extreme) may be different in a state with a large Hispanic population that has many newly declaring a plan to vote, than in many other states.
Also choice of LV screen was not the only difference: the T+1 result also used a different and novel model to weight the demographics than did the others.
FWIW HuffPo has it currently at C+6.2 and NowCast C+4.4. RCP is C+3.2 (in the 4-way). PEC is C+3.5.
HuffPo’s final Obama v Romney was O+1.5, underestimating Obama’s win by 2.4%. 538’s final was O+2.5, missing by 1.4%. PEC was O+2.2, missing by 1.7%. RCP was O+0.7, missing by 3.2.
Minimally RCP’s rolling average is less … rigorous … than are the others. Easy to follow though!
Being off from the core rigorous aggregators’ final calls by more than 2.5% would be inconsistent with their past performance.
That said I do believe that the final aggregated results will be off by perhaps a bit more than they were in 2012 but I doubt highly they will be off Trumpward. I believe that the screens that using the intent to vote will overestimate Trump turnout in many parts of the country (including in Ohio and Iowa) and that the past voting screen will both underestimate Hispanic turnout and overestimate traditional GOP voters who say Trump but may decide to stay home this time. And of course there is the impact of the ground game that is simply hard to model.
IMHO the way Trump would exceed his poll numbers would be if Clinton voters became complacent. I think this last smokebomb, which should result in some tightening before election day, will shake them out of that. All we need is Trump to say one last horribly offensive ugly thing in the final day or two to completely commit people to making the time and not assuming that enough others will vote for Clinton that theirs does not matter
One reason it has always been hard to model is that in the past, one side or another may have had a more effective GOTV operation, but both sides always put plenty of money and volunteer hours into it. This time, that will not be true on the Republican side, and will be true in spades on the Democratic side, offering a great chance for political scientists to get a better idea of how much it matters.
My money is on it mattering more than most people, even political scientists, think. Trump will go down in history as “that idiot who ______” with all manner of choices for how to fill in that blank, but one of them will likely be “didn’t think he needed to bother with GOTV, except for holding rallies and urging people at them to vote”.
Well, the polls are actually skewed all the time, there’s just no way to know which way they are skewed.
Early voting data suggests that white and Latino turnout is up, black turnout down. That’s not really good for Clinton. Latino voters being motivated is nice and helps in some states, but black turnout is key in some extremely important states and it’s just not happening. Maybe they are waiting for election day. Even Latino turnout isn’t as amazing for Clinton as it could be, because while Trump is putting up historically awful numbers among Latinos, Clinton may still underperform because Gary Johnson is getting double digit support from Latino voters in polls.
Now I’m no expert, as you all know, but those tea leaves look problematic to me. They seem to promise a much closer election night than we even expect now. This could be 2000 all over again.
To me, that is only possible if Trump catches her, and in fact passes her, in the polls–because, as I keep saying, I guarantee he’s not going to outperform his polling. (If he does, I will have to eat as much crow as you apparently did in 2012.) He has been narrowing the gap the past couple weeks, yes. But if you notice, that has been from increases in his own number and not from decreases in hers. Since she’s been consistently up over the 49% line for more than three weeks now, that doesn’t really leave him much room to catch her unless her number actually starts dropping.
Actually, she’s at 45.6% in four way polling, 48% in 2-way. She is up 3, with 7% undecided or third party.
nate silver’s odds look pretty accurate to me, at 73% in polls plus, and the betting markets seem to have become more Clinton-skeptical: 71% on Predictit.