After the storm
We shall not cease from exploration And the end of all our exploring Will be to arrive where we started And know the place for the first time. -T...
The Polarization Hypothesis Passes The “Access Hollywood” Test
*Video surfaces of Donald Trump killing and eating a guy* Trump surrogate on CNN: UM SILENCE OF THE LAMBS WON BEST PICTURE IN 1992
— Aaron Chewning (@AaronChewning) October 12, 2016
Polarization is so strong that other than Debate #1, which moved opinion by about four percentage points, it is looking like no existing story line can alter the trajectory of the Clinton-versus-Trump race. The primary exhibit is national polls, which have not yet shown any measurable aftermath from the Access Hollywood video or Debate #2:
(Methods: each day, take the median of all national polls that sampled on that day. Then apply a median filter. The gray zone indicates +/- 1 standard deviation.)
Beyond the four-point swing from Debate #1, there has been no further movement. It is as if the race has reached some ceiling that will be hard for Hillary Clinton to get past.
Obviously, the Access Hollywood event is not without consequence. It rocked Republican officialdom, highlighting divisions within the party and pitting Trump and noTrump factions against one another. That internal division, and any decline in GOP voter morale, may affect downticket races. Such effects might show up in Senate/House polls, though it hasn’t become apparent yet.
However, note that we don’t know the effect of get-out-the-vote activities. This year, those effects may be asymmetric, considering the relative absence of such efforts on the Trump side. We won’t really have an idea until after the election.
Also related to polarization, this year’s race has been unusually stable all year. Some commenters are still focused on the possibility that this year’s Presidential race is somehow volatile. As I said here the other day, the answer is no. Also see my American Prospect piece, which points out that the last few Presidential races are more stable than they’ve been since 1952.
To add to the picture a little bit, let us take one more look at the statistics. First, see the national-horserace graphs for 2008, 2012, and 2016 (links to RealClearPolitics). It is possible to extract a few simple measures: standard deviation (as I have reported before), total peak-to-trough variation throughout the election year*, and the largest single change in a one-week period:
| Year | SD | Total swing | Largest change |
|---|---|---|---|
| 2008 | 2.5% | 10% | 7.6% |
| 2012 | 2.0% | 7% | 4.5% |
| 2016 | 2.2% | 12%* | 5.9% |
By these measures, 2016 is a year of middling stability – by recent standards. But again, keep in mind that such high stability is a post-mid-1990s phenomenon.
High stability of opinion fits well with the extreme emotional nature of this year’s campaign. After what everyone has been through this year, not many voters seem likely to change their minds every few days. Other than some Johnson and Stein voters straggling in, minds are made up. This is polarization at work.
*Leaving out the primary season, the total swing in 2016 has been 7%.
Yet, for some reason, all the commentators calling the horserace continue to look at an individual poll to justify that its still up in the air.
Maybe, besides a class in basic civics, all citizens should take a class in probability and statistics. Not holding my breath.
That’s actually a great idea. The appalling lack of education among U.S. citizens is responsible for this lack of reason and knowledge about presidential candidates and the country’s electoral system.
That would be amazing. Frankly I’d be happy with just teaching a basic understanding of the scientific method and critical thinking.
Given that Trump’s base voters (and perhaps the candidate himself) take Alex Jones/Infowars stories as actual journalism, I’m afraid too much damage has already been done.
Nice title. By same metric might also say “2016 = the low information voter’s election”
The pundit/data hybrid agrees: “I’m not sure I can keep up the gag of pretending that Trump has some sort of rational inner monologue.” Not sure if he’s a victim of the gag or the perpetrator, or both.
Regardless, here’s hoping access hollywood plays a smaller role in 2020. Happy to say I did my part to this end by voting friday.
You hit the nail on the head. I’ve been very dismayed this election at how little most people understand about polling methodology. It just provides ample fodder for the wingnuts to claim that this survey or that, or all of them, is “rigged.”
Thanks for addressing the variability in margin issue, as distinct from the variability in outcome issue. The latter has never changed, while the former has shown some significant, but apparently not abnormal, bouncing around. One question of clarification – your graphs shows a max margin after debate of 6% – but I don’t recall the meta-margin ever getting that high – or did I miss something?
The Meta-Margin measures state polling, while the graph above measure national polls. The two numbers, if given enough time, will eventually come more or less into alignment, but they are not exactly the same. (Today, Hillary’s lead in four-way polling is about 5-6%, while the Meta-Margin is just under 5%.)
The graph shows national polls, which have better time resolution.
Sam,
Thanks for the clarification. Makes sense to go with the numbers that have better time resolution.
Last five national polls, Clinton +12, +11, +4, +5 +10
Median: +10
http://elections.huffingtonpost.com/pollster/2016-general-election-trump-vs-clinton
The Senate is genuinely up the air! People looking for drama could concentrate on that.
Could it also be that without the Access Hollywood tape, ongoing drip of accusations and fresh Trump conspiracy pronouncements, his poll numbers would have risen while Hillary’s fell?
In other words, perhaps its not that polarization has prevented these things from affecting voter perceptions, but that voters are responding to a simultaneous bombardment of negative press toward both candidates(with Trump admittedly getting the lion’s share of it).
Wikileaks is no match for sex allegations in terms of click bait, but may be playing some role in stabilizing Trump’s numbers.
What you are missing is that there were 2 concurrent storylines (the Trump tape/accusations and the Clinton campaign emails from WikiLeaks) exerting opposite forces on the electorate. I wonder if looking at the polls by gender would tell a different story.
There’s always some complicated reason for people to disbelieve a basic – and to me, obvious – point. “Imagine multiple underlying variables, which happen to cancel out” seems like yet another installment of the parlor game “Make Work For Data Pundits.”
The further prediction one would extract from this is that the Access Hollywood tape, Wikileaks dumps, etc. will only serve to harden existing allegiances and reduce variability.
That is, the final EV count should be very close to what we see today.
Pity we can’t call the Senate on that basis.
Unless the Trump campaign collapses entirely, which I believe is possible. Trump’s lack of a GOTV organization plus Trump constantly claiming the election is rigged could drive down participation from his supporters.
From my analysis I think 402 EVs are entirely possible. I’ve even been predicting Utah being in play since June at another political site where I participate.
If the Trump campaign collapses entirely, I think that could affect the Senate races.
Sam, have you looked at Google Correlate data to see see how it’s performing relative to the polls?
with trump just running a scorched earth campaign at this point, he won’t gain any new voters.
he is his own worst enemy , since that debate #1.
https://www.facebook.com/VICE/videos/1414301661936421/
I’d posit the polarization has to do with things like facebook algorithms that trap us each in an information bubble tailored for our individual views and preferences.
The walls on both sides are now high enough and sound proof that any changes are highly unlikely:
http://assets.amuniversal.com/c6fa9f6066f60134ca3a005056a9545d
Sorry for being dense, but why exactly did we expect the Access Hollywood tape to have an effect? It does not illuminate his take on any policy issues (like, say, Romney’s 47% exposed his thinking on the public safety net).
If Trump had been overheard saying something policy relevant, say “The Wall is just a metaphor”, or “Social Security should be managed like 401k’s” then perhaps his numbers would have dropped.
By this argument, the first debate should also have had no effect. But it did, according to Sam’s analysis.
Trump reinforced a lot of unorthodox policy positions during the first debate, which had not been fully articulated (in front of such a large audience) before. This may have caused the polls to move.
For instance he reiterated his lack of support for NATO, and advocated seizing another nation’s assets by force (this used to be called “pillaging” in the era of warlords). He also failed to soften his position on immigration and deportation.
Of course he also expressed his opinion on human rights and dignity to be afforded to women, specifically those in his beauty contests. While not a “policy” one would think was still under debate, he made it obvious his stance had not been misunderstood in the following days.
Given this entire package from the first debate, the Access Hollywood tape had precious little to add.
I am concerned about possible post-election havoc. Diehard Trump supporters are not going to fold their tents and go home. I envision lots of rallies about “rigged elections”, possibly some deranged violence.
I can envision publication of the names, addresses and phone numbers of those people who are actual electors, with lots of harassment before the Electoral College makes it all “official”.
“Rigged elections” could turn out to be the greatest vote suppressor of all in this election. Trump gets support from blue collar workers who are going to have to decide to maybe take time off work and lose money or stand in long lines after they’ve worked all day. Why bother if the election is rigged? They’ve also attended ‘massive’ rallies so their one vote won’t make a difference, right?
Trump is right. As of now, there’s nothing on the NY Times website about the new email.
At this point, I don’t see anything moving the election either way,
I once saw Phil Donahue jump a guest at the top of one of his shows and remain combative through out. Highly uncharacteristic of him. The real rub was the guest was taking the side of something that Donahue would typically champion. It turns out the guy had what we would call “attitude” today. Being correct doesn’t mean you get to smash it in peoples faces in a smug, arrogant manner. Donahue felt defensive and responded accordingly. I see a lot of polarization taking place the in the same light. People feel defensive and defend their positions. The more bitter the rhetoric the more entrenched they become until you get to the point you have people defending and rationalizing sexual assault.
Obama tried to turn down the volume. Clinton is doing the same thing, at least for now. Obama started going off message after about 6 years and gave up the hope of bipartisanship.
” Obama started going off message after about 6 years and gave up the hope of bipartisanship”
I might choose to reverse cause-and-effect on that. I think the failure of even a single Republican to support the very “Governor Romney-like” Obamacare and McConnell’s frequently stated top goal of making Obama a one term president created much bad blood between the parties.
Beyond rocking Republican officialdom, the Access Hollywood tapes led to Trump’s denial to A. Cooper of having acting on these brags, which led to women coming forward with their denunciations, which led to the candidate’s approval ratings dropping still lower – surely this has some quantifiable effect, even if data suggests stasis – it may also be that national misogyny is such that a late-stage refusal to elect a woman president is setting in.
So, Silver just posted an article that says:
“As an aside, FiveThirtyEight’s forecast models use a t-distribution rather than the more common normal distribution. The t-distribution has wider and fatter tails and is appropriate in cases like presidential elections where you have smaller sample sizes. Since our model is trained based on only 11 elections (1972 through 2012), we can’t say all that confidently what the chance might be of, say, a 10-percentage-point polling error. The t-distribution makes more conservative assumptions about this than the normal distribution does. ”
Which do you use, Sam? It struck me as an interesting technical issue…
We use t-distributions (3 d.f.).
“However, note that we don’t know the effect of get-out-the-vote activities. This year, those effects may be asymmetric, considering the relative absence of such efforts on the Trump side. We won’t really have an idea until after the election.”
I’ve been fascinated by this question of the impact of GOTV efforts. This year seems to be the ideal year to measure this, but how? I’m struggling to see how we will know the impact of GOTV efforts. Any ideas?
Here’s an admittedly coarse but helpful way to think about it:
Florida is probably about 1-2% to the right of the country as a whole, so if Clinton goes into election day leading Trump nationally by 6%, we’d expect her to win FL by something like 4-5%.
If, instead, she wins FL by more like 7%, you could attribute some (but probably not all) of the difference to GOTV.
FWIW, GOTV operations are definitely effective at the margins in battleground states but probably have little to no effect overall on the rest of the country. You might get an extra 70,000 people to turn out for you across the state of North Carolina, for example, and that would be enough to get you the extra 1% you need to win it. But 70,000 votes is a drop in the bucket compared to the 140 million total votes that will be cast.
there doesnt seem to be any actual statistical analysis of OFA impact on the 2012 election.
Republicans claim there was effectively NO GOTV because of the effective fail of ORCA.
so that would be the same situ, one party with zero GOTV.
Confession: I worked on OFA and i was romantically involved.
Project Ivy doesnt impress me so far.
Well, there are several published analyses of 2012, the most recent being ‘The Ground Game in the 2012 Presidential Election’ by Seth Masket, John Sides & Lynn Vavreck, published in the journal ‘Political Communication’.
Their conclusion is that “in the 2012 presidential election, the ground game appeared to affect the vote. In particular, the Obama campaign’s field operation was associated with increased vote share for him.” Their estimate is that there was an average increase of 0.3% in the Obama vote per field office within the county in which it was located.
Both the Huffpollster and RCP averages show a sudden increase in the margin in the last week or so, but the above graph shows flat. Is there a different method being used?
Sam,
Just finished reading a piece in the NYT by Michael Barbaro titled “Is This Election Over?” in which he cites polling psychologist Kyle Dropp who says Trump’s last hope might be that during such a polarizing election that there might be some shy voters who are reluctant to tell pollsters for whom they really plan to vote.
Do you have any thoughts regarding this element of “shy voters,” and if they exitst, do you think they could have an impact?
Carl Kruse
This reminds me of the persistent myth that the bulk of Trump’s support comes from poor and working-class whites. Trump primary supporters had a median household income of $72,000 a year, which is $11,000 higher than that of non-Hispanic whites in general:
http://www.vox.com/policy-and-politics/2016/10/15/13286498/donald-trump-voters-race-economic-anxiety
In other words, Trump voters aren’t predominantly poor white guys who lost their steelworker jobs due to globalization. They’re people who are second- or third-generation white-flighters whose forebears moved to the suburbs fifty years ago to avoid looking at black people.
“Mr. Trump also has healthy shares of support from the affluent and the well educated — that point should not be lost. But in the places where support for him runs the strongest, the proportion of the white population that didn’t finish high school is relatively high. So is the proportion of working-age adults who neither have a job nor are looking for one. The third-strongest correlation among hundreds of variables tested: the preponderance of mobile homes.”
http://www.nytimes.com/2016/03/13/upshot/the-geography-of-trumpism.html
I have to admit that I am astounded by these results. The Access Hollywood video had such a visceral result that I would have expected to have been reflected in the polls. So I am left with 3 possibilities
1. The feedback loops that establish and maintain partisanship are truly robust even in light of extraordinary events.
2. There were countervailing stories (wikileaks) that obscured the reaction.
3. Poll aggregating methods are susceptible to spam polls if conducted in high enough volume.
I suppose there could be a 4th option. Maybe the people who would be affected were already turned off after the first debate.
I think the bump after debate #1 disproves the polarization hypothesis, whatever that is. If Prof. Wang can do cherry picking, why not me? It’s called confirmation bias.
This comment misses the point, which is quantitative. Polarization implies a low amount of variation – not zero variation. The post-debate-#1 bump is 4 percentage points, which is small by historical standards. Compare it to any pre-1996 campaign. And the amount of variability (i.e. standard deviation) for 2016 has not increased.
To put it another way, here is how the bump actively does fit with the polarization idea: the bump got national polls and the Meta-Margin to a level equal to their pro-Clinton extremes for the entire year. However, opinion did not go beyond the narrow range we have seen so far.
I largely agree with the argument, but even if we accept that polarization limits and inhibits movement, that doesn’t necessarily mean Access Hollywood had no discernible effect. It seems plausible that by now we would have expected the high from after debate 1 to regress towards the mean (which seems a bit lower) and the fact that it has largely not done so could be attributable to the tape.
Or not – it’s hard to say. But I wanted to point out that no change does not necessarily mean no effect, particularly if we might normally have expected change in the other direction.
I’m curious. I can see GOTV making a big difference in states like Nevada, where there is a large Latino vote and unions are strong. My question is: do polls take into consideration the likelihood of people answering pollsters’ calls? I, for one, never answer them, and I’m for Clinton. Maybe there are many like me. On the other hand, Trump supporters, especially the avid ones, probably answer every call they get. I wondered the same thing about Bernie Sanders followers. Do polls adjust for the likelihood of people answering the phone?
Professional pollsters make extensive use of re-weighting to account for a variety of demographic factors.
Likelihood of staying on the phone is one remaining factor that may affect response rates enough to cause day-to-day fluctuations in the result. Andrew Gelman and others have written about this particular point. I do not think it makes much difference…certainly it would be inadvisable to imagine that one’s own side has a large hidden bonus. That can only lead to disappointment on Election Day.
Sam:”There’s always some complicated reason for people to disbelieve a basic – and to me, obvious – point. “Imagine multiple underlying variables, which happen to cancel out””
The polarity hypothesis presupposes that such multiple variables will cancel out. That’s because it treats the electorate under conditions of polarization as a physical system equilibrating itself. That’s why it is predicted to revert to a mean, and why lurches of voter preference n one direction or another can be treated as randomness.
But one could instead treat a complex human system like voters in organic rather than physical terms, as a self-organizing system. Such a system does not reach a static point of equilibrium, but instead evolves. As polarized as voters unarguably are, there is still a development to the ongoing narratives that are the two candidates, and a growth in the familiarity of voters with these candidates. These evolving dual narratives may be reflected in perturbations in voter preference that , rather than random noise in a mechanism, are organized, meaningful transformations which under certain circumstances may transcend the original conditions of polarization, tilting overall preference haphazardly but progressively in the direction of one candidate.
Bottom line; the electorate may amount to a system that is both wilder and more complexly ordered than that presumed by a physicalistic, mean reverting machine model.
The hypothesis seems to me to be inescapably individualistic. “Very few people change their minds.”
If so, the order of the states from strongest Democratic to strongest Republican should remain close to what it was in 2012. I’m working on quantifying that.
Meanwhile, Iowa is much more Republican and NC much less on the aggregators’ charts.