I think one partial explanation as to why people were shocked by the Trump election, given the pro-Clinton polls in the final weeks of polling of the race, is that we may have blown the significance of a “3 to 4 percent lead” out of proportion to begin with.
Ever since the razor-thin margin of victory in 2000, and also the tightest reelection victory by an incumbent president in 2004, American polling/media has entered this new norm whereby a lead of 3 to 4 percent in the polls (Hillary’s lead over Trump in the waning weeks) is considered a “big” lead, like how even relatively close election outcomes are nowadays termed “landslides” by some. In this new partisan, divided era, small polling leads are considered big leads. A “big lead” used to be something like 10%; now even 5% will qualify.
But a lead of 3-4% in the polls is not a big lead at all. For perspective, it’s akin to a room of 65 voters, in which Candidate A carries 33 voters while Candidate B gets 32 voters. It’s an incredibly *narrow *lead, mathematically. In almost any other democracy in the world, a 3-4% lead in the polls would not be considered “big” at all. In that hypothetical room of 65 voters, all it takes is for *one *voter to switch sides for that 3-4% lead for Candidate A to suddenly be flipped to a 3-4% lead for Candidate B.
Yes, the polls were undoubtedly off and there were many shy Trump voters who wouldn’t admit to pollsters to voting Trump. But even if the polls were accurate, and Hillary was leading by 3-4% in the final weeks, it is not far-fetched at all for the trailing candidate to close that gap on Election Day. Any lead of less than, say, 7-8% should be considered a small lead.
I would replace “Trump shy” with “Hillary shrug”. Lots of place in America where voting is a pain in the butt, and you know it before you leave the house. By an astonishing coincidence, those places are usually the ones that Dems might be expected to carry. Why, one might almost see the hand of Providence in such a coincidence!
At any rate, if the Hillary voter lives in “four hours standing outside” district, and is pretty sure she’s going to win, he might shrug it off.
When one poll shows a 3 or 4% lead, its not a big deal. When all of them show the same or similar lead, then the law of large numbers kicks in and it becomes more reliable unless they were all polling the same people the same way.
There was something wrong with the way pollsters were accounting for turnout and veracity. Trump supporters were more likely to lie about who they were going to vote for and Hillary supporters were more likely to lie about their likelihood of bothering to go vote.
I used to think that rural voting was a disadvantage for the GOP due to the need to drive a while to get to the polls, but you brought up the point that in urban Blue-leaning areas you have to stand in line longer, so I suppose it tends to cancel out, too.
Agreed. What made the polls so convincing was that they all showed her consistently in the lead. A single poll with a 4% lead is a good sign- but dozens of polls with a consistent 4% lead over *months *is a pretty good indication of who the public prefers. It’s pretty obvious that even *Trump *was surprised by the win.
I do think that the GOP’s voter suppression tactics had a pretty big influence on the election. If you think your candidate’s going to handily win, how enthusiastic will you be about the idea of waiting four hours in a line to vote?
It looks like by the time the count is over, her actual nation-wide lead will be close to 1.5 or 2 points. That’s not far at all from 3 or 4 points. What was significantly off were polls in WI, MI, and PA, and a few other states.
I think that this year was something of a special case.
Hillary is an indifferent campaigner, and part of her strategy for overcoming that is projecting (through surrogates, etc.) the inevitability of her success. It’s a reasonable strategy overall, but like other strategies it has weaknesses. It failed against the excitement of Obama in 2008, and it failed against the “don’t tell me what to do” crowd in 2016.
In fact, I wonder if Trump would have gotten FEWER votes if people thought he might actually win.
We have because the majority of polls in this last election were statistically insignificant by themselves but we talk about them like they aren’t Pew points to a rule of thumb of about twice the sampling margin of error being the margin of error for the difference. (There’s a better discussion for actually computing the difference MOE here.)
Aggregation can help smooth out sources of error when you have a lot of different data points with different methodologies. Like iiandyiiii points out, polling got the national difference generally right. They predicted a Clinton win and she won the popular vote. There tends to be a less poll data to aggregate for the individual state races though. Some of the other sources of error, like estimates for likely voters or attempting to correct demographic sampling issues, also get more problematic in the smaller pools that don’t mirror the overall US. Think of the blue collar labor Democrats that are over-represented in MI, OH, and PA compared to the nation and voted differently this cycle. It’s hard to use old data to correct in those smaller pools of voters when there’s a non-traditional candidate.
Most of the stories about polls in this cycle could have been “New Poll from ____.; still a dead heat.” Unsurprisingly, we ended up with a very close election.
As a qualifier, on this side of the puddle 3-4% margin in the polls in the last week of an election would normally be regarded as a winning margin therefore well “big enough”.
A 5% swing at the ballot box is a landslide.
Year Incumbent Last Poll Result
1998 Coalition 49.02 Returned
2001 Coalition 50.95 Returned
2004 Coalition 52.74 Returned
2007 Coalition 47.30 Labor win
2010 Labor 50.12 Labor majority in hung parliament
2013 Labor 53.49 Coalition win
2016 Coalition 51.07 Returned
The difference is 1) more stable voter turnout due to compulsory voting and 2) Australia hasn’t yet had an election where disingenuous respondents in polls were a significant factor.