The Hardened Divide: Why Donald Trump Is Mitt Romney Lite

October 7, 2016 by Sam Wang

Update: Thanks to today’s Washington Post story by David Farenthold about Trump’s lewd comments about what sounds like sexual assault, we may see a test of my thesis that voters are close to immovable. However, I should point out that the Meta-Margin could go to Clinton +7% and still be consistent with my argument.
In The American Prospect, I present data that PEC readers know well: this year’s Presidential race is the least volatile in 65 years of modern polling. Despite all of this year’s venom and caterwauling…or maybe because of it, voters are well entrenched.

The same is true geographically. In the piece I give a state-by-state correlation coefficient between the Clinton-Trump margin and Obama-Romney 2012 outcomes. That correlation has strengthened in the last few weeks: it’s now +0.96, again a 65-year record high and very close to the maximum possible value, +1.00.

That is not to say that there have not been shifts. As Ron Brownstein and many others have pointed out, various groups have shifted allegiances, such as less-educated white men (toward Trump), suburban women (toward Clinton), and nonwhites (toward Clinton). Many of these effects cancel one another. Think of this year’s race as being the extreme culmination of what we’ve seen since the mid-1990’s. The polarization that voters under the age of 40 may regard as normal is in fact a weird kind of stasis.


Ari Goldberg says:

Here’s what I don’t get (and I’m just a lowly, amateur poll-watcher): If the volatility is so low, how did one measly bad week for Hillary move the polls so dramatically? So she caught a pneumonia and made one clumsy comment about deplorables and, BOOM!, a solid lead was suddenly erased and Trump pulled ahead in a number of polls. I would assume, given a non-volatile race, that shouldn’t happen. Please englighten, thanks!

Sam Wang says:

Your error is that you think a shift of 2 percentage points in the margin is a massive shift. Watch the data, not the drama. Alternately, look at any of the graphs I have posted here in the last month.

Tony Asdourian says:

Ari, I don’t want to speak for Sam, but I think his comment is referring to the fact that the Meta-Margin was around 4% for a number of months, then went up to 6% after the conventions, then went down to 2% after Hillary’s bad week. But that entire up and down is within 2% of the 4% it had been at for months. That’s why he was predicting, when she was at her lowest, that the Meta-Margin would go back up for her– he expected regression to the mean of 4% that she had maintained for so long. He doesn’t see dips and peaks as particularly significant– thus his “look at the data, not the drama”. If I’ve got this wrong, I’d appreciate it if someone corrected me, but I think I have it right.

Sam Wang says:


A says:

Sam, if what you’re saying is true–then it’s possible we will see another 2 percent shift back towards Trump before this is all over…which would seem like a HUGE shift. Causing folks like me to panic and tear our hair out.
Right now if I click the Trump +2% map on your sidebar, Clinton only gets 263 electoral votes.
So even if the race is polarized and less volatility than ever, isn’t the end result still somewhat unpredictable, given that a 2% shift either way could have enormous ramifications on the outcome?
Or am I missing something?

A says:

Although looking closer, if she hangs on to even one of the swing states up for grabs in that +2 map, she still wins it…so I guess maybe thats what I was missing.

anonymous says:

Yes, close-to-even races are more unpredictable because winning or losing can be within the margin of errors of the polls. Small shifts in such close-to-even races can flip the winner. I think Sam’s point is that the races after 1996 are less volatile when compared to historical races, which his graph shows to be unambiguously true. I don’t think he is saying that the races after 1996 were necessarily more predictable. Given the close final results, it is clear that he 2000 race was probably not predictable using the polling data.

AkS says:

A, I think you are missing a crucial point. With 2% shift toward Trump, you are right that HRC will be at 263, but Trump will still be only at 215. In order for Trump to get to 270, he will have to win FL,NC,WI, and NV, whereas HRC will have to win any one of those.
Mathematically speaking, assuming that 1) each of those four states is a toss-up, i.e. probability of winning each state is 50%, and 2) states are perfectly non-correlated, Trump has a chance of 6.25% (1/16) of winning all four states (same as probability of landing heads four times straight in a coin toss). On the other hand, HRC has to win only one and her probability is 50%.
So, contrary to your statement that the result is unpredictable after a 2% shift toward Trump, I would argue that, in fact, HRC has a fairly strong command on this race. That is to say, even if 2% of the population were to shift toward Trump, HRC would likely still win, which means that her position right now is very strong.

Sam Wang says:

This is the point of the Meta-Margin, which is at Clinton +3.3%. A 2% swing only gets to Clinton +1.3%.
I recommend a close reading of my posts on the Presidential predictor.

A says:

Good point. I think I tend to hear “more predictable” when someone says “less volatile.”
And that’s my mistake.

Sam Wang says:

Well…actually, I think the election is also more predictable. There has been no moment in time in 2016 when Trump led Clinton in national or aggregated state polls.

DonC says:

Anything is possible but I think the big moves are over. The thesis of the piece, which seems inarguable at this point, is that most people are inclined to support one candidate or the other. The true persuadables are maybe 8% – 10% , and they won’t all break the same way. Consequently, the remaining issue is what will electorate look like. Or stated differently, who will vote.
Hillary Clinton had a bad couple of weeks, and those expected to ultimately support Trump decided in their own minds that they were going to support Trump. This was similar to what happened when Romney performed well in the first debate in 2012. And as in that case, the change in the polls resulting from Republican voters coming home was described as a big move. Then Hilary Clinton had a good debate and Donald Trump had a very bad week. This led those expected to support Clinton to decide in their minds they were going to support her. The resulting change in the polls was again described as a shift but was just people expected to come home declaring they were coming home. Since the electorate has returned to home base, there aren’t enough voters left to effect a big shift in the polls.
Time is also short. Election day is still a month out, but by election day 40% of the votes may already be cast. If Clinton is ahead by 6% in these early votes, it’s more or less impossible for Trump to catch up. In this regard, the early voting suggests a much better electorate for Clinton than most suspected.
Finally there aren’t a lot of chances left to change the race. All the “big moments” — conventions, VP nominations, and the like — are all in the past. The first debate was described as being 75% of the remaining campaign, which leaves very little opportunity to change votes from now to election day.

Phillip says:

One possibility worth thinking about is that it’s more about the specific candidates who have been nominated in recent elections. I wonder if there would be more volatility if Democrats nominated a southern governor akin to Carter or Bill Clinton? Those two seemed to create fairly unusual maps. Maybe John Bel Edwards or Jay Nixon? (Of course, it could be that partisan sorting at the local level means those types of candidates are less available)

Sam Wang says:

If you look at election-on-election correlations, only Carter was unusual. Not Clinton. Search this site for “realignment myth”.

AP says:

Of course “dramatic” is an ill-defined concept, but the median EV went from 290 to 323. Now 33 electoral votes to me sound a lot bigger than a 2% shift, but that’s what swing states do. To me the most understandable difference is in terms of probabilities. PEC doesn’t seem to keep an history of victory probs that I could find, but I think it went as far down as 79% at one point. The upshot went from 70% to 82%, similar to most other meta-pollsters. 538 are still deciding which one is their best model. Are these big or small differences? My personal utility function is P(nuclear war). That has roughly halved in the last few weeks, and that’s a big difference, for me.

Sam Wang says:

All of those swings constitute the smallest ever, including the EV swing.
You do bring up an important concept: the cost, which is defined as the probability multiplied by the payoff/penalty. 10% of Armageddon is still a lot of downside risk – much more than, say, 50% of a bummer of a Supreme Court.

liberal says:

While I’m steadfastly for Clinton, P(nuclear war) is decidedly nontrivial if she wins also, given her views on Ukraine and Syria.

Robert says:

As long as nations have nukes, I’d never look at P as nontrivial. Even in the “best” of times like in 1995 when Yeltsin almost attacked.

Jeremiah says:

@liberal Do you mean Russia’s involvement in the Ukraine and Syria? If so, why is the P(nuclear war) non-trivial, as opposed to the probability of any other armed conflict? This “assessment” seems to be just more Clinton bashing rather than thinking about the nuclear posture of the US in the world since the end of WWII.

liberal says:

“Do you mean Russia’s involvement in the Ukraine and Syria?”
Russia is involved in Syria at the invitation of the legal government of Syria. If you think that upending all the premises upon which the international state system has been built, going back to Westphalia, then I can’t say anything except that this is extremely unwise.
As for Ukraine, non-Russian apologist and realist IR scholar John J. Mearsheimer wrote a solid article entitled “Why the Ukraine Crisis Is the West’s Fault.” It’s a must-read.

JayBoy2k says:

Great Article, Sam A really easy read, lots of factual data. As I thought about the Red States – Blue States, this would seems to equally apply to the Senate. For the swing state on the right column, which are the most deviating (Blue state-Red Candidate or Red State Blue Candidate) is the race this year? Thanks for the article

Joe Arnold says:

FiveThirtyEight has an article that says this is the most volatile election in recent history because of the much larger # of Undecideds (16% compared to 5-7% in the past couple elections).
Any thoughts on their analysis and the discrepancy?

Sam Wang says:

Ok, so they think it *might* be volatile someday. If that’s their evidence I don’t have anything to say, except that I am using actual volatility data.

Matt McIrvin says:

Undecideds aren’t even anywhere near that high, unless you explicitly ask about Johnson and Stein and lump them in with the actual undecideds. By that standard, any election that had a real major third-party candidate beats this one.

Ben says:

Something I’ve noticed that hasn’t been much discussed here or elsewhere–the extent to which blue states are becoming redder while red ones bluer. The correlation between the ’12 vote share and the change in vote share between ’12 and ’16 (projected) is highly negative. Like -0.9ish. This is highly unusual for recent cycles.
So while the map hasn’t changed much this year, it seems highly possible this election is laying the groundwork for a future recalibration.

Matt McIrvin says:

It’s hard to say whether that’s a trend or if it’s something to do with the particular candidates this year.
Maybe some of both. I think the upper Midwest is probably trending red over the very long term, and that the Democrats are going to win back bigger and bigger chunks of the coastal South as the strength of their minorities+educated white professionals coalition spreads there.

Lorem says:

Good article and argument, but I’m pretty annoyed with the discontinuity in the graph. It’s kind of assuming the conclusion instead of letting the viewer draw it. Not to mention that even if the conclusion is true, it’s still “continuous”, just with a significant change.
Why not just draw it normally and then put a little dashed vertical line where you want the reader to look or something like that?
(Directed at whoever made the decision, who I realize is quite likely to not be Sam.)

Josh says:

The “discontinuity” stems from the extreme volatility of the 1992 election, which featured a rare, viable, third-party candidate who received almost 20% of the national popular vote.

Paul says:

Indeed, if anything I’d say the chart gets _more_ interesting without the connecting lines! Take a look:

George says:

I think the idea of volatility depends a bit on perspective. Looked at over history and from the standpoint of change against the full (0 to 100) possible outcomes, yes, things look “not volatile.” But when one focuses ONLY on the meta-margin in isolation, we see that from the beginning of the conventions to now it has swung from somewhere in the low to mid “2”s – all the way up to mid to high “6”s, then back down to slightly below 1.1, and now back to 3.3, not far from where it was in June and early July. So yes, regression back to the mean, but compared to the months before the conventions, a whole lot of up and down with what looks to have been a 5+ point drop from peak to valley over the course of only about a month. And that FEELS volatile as you go through it.

Jay Bryant says:

Thanks for this, Sam. I wouldn’t realize this is happening if I didn’t stop by your site. You bring a clarity that no one else provides.

A says:

Sam, reading your 2012 prediction, you went with an SD of 2.2.
However, for this race you went with an SD of 3.
That is my understanding, going back through those old posts…unless I missed something (again).
My question to Sam is, whether knowing what we know now about volatility continuing to match 2012, do you think that going with the previous SD of 2.2 that was used in 2012 would have been appropriate or even preferable?

Sam Wang says:

No regrets. SD=3% is at the upper end of 1952-2012 post-Labor-Day variation, therefore a cautious choice.

Bob Wallace says:

During the Reagan/Bush I years did we experience a consolidation of racists and conservative Christians into one party, turning the parties into more uniform voting blocks?
If so, by the time we start tracking polls the two main candidates probably have signaled their values and voters signed on.

steve rosenberger says:

Thank you for PEC – a beautiful island of tranquil reason floating in an angry sea of sensationalism. It seems there is an inverse relationship between electoral volatility (plummeting) and media melodrama (soaring). Corporate media needs advertising revenue, advertisers need eyeballs. So a rollercoaster ride (She’s up! He’s down! She’s down! He’s up!) – even if it is mostly manufactured – drives up viewership. A presentation of policy differences or societal values or (gods forbid!) a focus on forging consensus on solutions to the nation’s problems… would be a buzzkill. But look at his HAIR! She shook her head, is it Parkinsons?! His third wife was an undocumented immigrant sex worker! Her husband was a philandering cad!
This country is doomed if we don’t stop sacrificing education for entertainment. Ignorance is never an excuse, but it will be our downfall.

maye says:

“Watch the data, not the drama.”
Sam Wang, 2016
Words to live by.

Jay Sheckley says:

And yet today is full of October Surprises: info-rich drama [as opposed to noise] which may swiftly push our data charts back to the median., where Sam sent me to watch the Marriage Equality fight, showcases most of this week’s action: Russian complaining about our election, Washington Post’s coverage of Trump’s self-made outrageous comments about women, the never-ending legit investigations into a certain candidate’s business dealings…. Thanks again, Sam!

Ed Wittens Cat says:

a black swan has appeared on the event horizon

Sam Wang says:

I do not think this is a black swan, which I thought was an event that nobody could have possibly predicted, and that completely upended the existing dynamic.

Scott J. Tepper says:

I disagree. While this latest recording is confirmatory of Trump’s sexism and misogyny, it is also so outrageous and a celebration of rape culture, that it has to make all but the most craven of women think twice about supporting someone who is a monster. Senator Kirk has just called him a clown. I look forward to Paul Ryan declining to share the stage with him as planned.
This is, in my view, a Black Swan event simply because it is so bad. (But now I bet there is a lot more out there.) And it came at the worst possible time, just two days before a debate where Trump must perform well.

Ed Wittens Cat says:

nope, hard to predict (with statistical analysis) but not impossible.
“The black swan theory or theory of black swan events is a metaphor that describes an event that comes as a surprise, has a major effect, and is often inappropriately rationalized after the fact with the benefit of hindsight.”
Now u will see ppl saying that we should have known a tape like was bound to surface– post event rationalization.
ur an abaci. im not.
I think all blackswans are avalanches, while not all avalanches are blackswans.
i just cant prove it yet 😉

Ed Wittens Cat says:

its not just Trump saying something nasty abt women– hes done that plenty.
its a rubicon for the GOP.
Clinton campaign is going to force Ryan and Priebus to say if Trump should be president, not just declaiming against what he said.
It plays into the whole rape culture Stanford swimmer theme.
what happens to the down ballot if the GOP stands by Trump?

Barb says:

It was completely predictable to anyone paying attention to Trump’s behavior over the years. Shoot, it was completely predictable to anyone only paying attention to what he’s said recently. The man tells us who he is – it’s not like we should be surprised when there is confirming evidence.

Jeremiah says:

I completely agree with Barb. We already knew he was a sexist, racist, xenophobic bigot.

Perry says:

I would consider this normal Donald Trump. This is exactly the kind of person I have been seeing for the past year.
He has “issues” way beyond not being polite! Black Swan…. Not so much.

Matt McIrvin says:

Swanny the Exceedingly Obvious White Swan. But a whole lot of US voters evidently didn’t have a clear idea of who Donald Trump was until this point.

Some Body says:

Event horizon? Are you insinuating Trump is a black hole, or is it just that you think there’s some form of singularity about him? Seriously, now, I’m quite sure Trump’s polling will be getting worse, but much less sure that decline will represent much of an underlying shift in opinion.

Mike Zwick says:

A colleague and I were discussing the campaign volatility graph, and he noted that volatility decreased at about the same time that internet arrives on the scene. Could an explanation for the pattern you observe be related to the quality or extent of polling pre-internet as compared to post-internet? Couldn’t such an effect be a confounder in your interpretation of the data?

Damien says:

It also coincides with the rise of the echo chamber that is Rush Limbaugh, Fox News, and conservative media in general.

Kenny says:

I’m with Damien on this one. I think the internet contributed to echo chamber effect. There was a time when we all watched the same news and all believed the same facts. That’s not true anymore. I think both sides are guilty of this to some degree, but I do believe that the conservative side is more prone to seek out conservative media and believe conspiracy theories… where left-leaning folks may watch Rachel Maddow and Democracy Now, but are also like to read, listen, and watch more mainstream news sources and generally trust the news.

Reginald says:

Great article, as always Dr. Wang. I think that this may answer for me why people who I think are otherwise reasonable are voting for Trump. It’s not that they are not appalled by the things he says, it’s that nobody switches parties anymore.

Anthony says:

Well after this latest bomb to drop from the Trump campaign just now (the leaked audio tape), Sam I think you can use the 2016 election results to represent the absolute floor of the vote share you will get by simply being the Republican nominee (which I suspect to be about 42%). I’m not sure how that data point will have any effect on your future projections, but it will at least be an interesting data point.

bks says:

Who knew that Trump could blow away a hurricane?

Amitabh Lath says:

The leaked audiotape reinforces Sam’s point in this post; it’s unlikely to shift any states. Maybe Utah.

538 Refugee says:

The presidential race was over anyhow. If it doesn’t affect down ticket then it did nothing of consequence in terms of the election. Supposedly the GOTV was anemic on the Republican side anyhow so does it hurt the morale of the field offices enough to matter?

Joel W. says:

About the only good thing the tape did for Trump was obliterate attention to his contention that the Central Park Five are guilty. His inability to accept reality—an inability Trump has demonstrated a number of times with respect to a variety of issues—is a weakness that a president cannot afford to have.

Five Tool Player says:

Aside from Dr. Wang’s relevant, always insightful statistical analysis of the election, my main — admittedly subjective, pro-Hillary — worry at this stage of the election, is that Trump drops out of the race. Sure, I realize that my fear is unrelated to statistics and therefore not testable at this point, but due to the statistical stability of the 2016 presidential race, my wild a**ed guess at this point in the election cycle is that this election is essentially over.

Dave Pilon says:

Sam, what kind of a shift in the meta-margin would it take to disconfirm your hypothesis? Half a standard deviation above Clinton’s highest polls to this point?

Joel says:

Hard to focus back on the topic at hand, but how much of the volatility in pre-1992 elections can be explained by the lack of polling? Certainly, small sample sizes can have distorting effects.

Sam Wang says:

The data come from Wlezien and Erikson. There was plenty of polling, enough to get a time series. Organizations like Gallup polled repeatedly, so the methods are even better matched over time than they are now.

Joel says:

@ Sam
But with a single pollster, any sort of sample error related to the method of polling will simply be reiterated in a time series, right?

Sean Patrick Santos says:

Hi Joel,
If the old polling was dominated by organizations that tended to produce a wide variance in results, *and* new pollsters tended to be less volatile (both over time and by agreeing with each other), *and* this change happened within a few years in the mid-90’s, that could explain it. But there doesn’t seem to be much reason to believe that all of these things are true.
On the other hand, we do know that the country became far more polarized at around this time, both because the geographical realignment that started in 1964 was completed, and because the rise of conservative media polarized news media consumption. So it seems more likely that the decrease in volatility is real, and has to do with cultural and policy divergence between the parties.

Dave James says:

Regarding 2005 TrumpTape:
I have yet to see one post on Facebook or Twitter from a Trump supporter that is anything close to a change in who they will vote for.

AAF says:

Certainly core Trump supporters won’t be affected.
One broad effect this may have on non-core Trump supporters is more indirect:
Over the course of the campaign Trump has slowly shifted from being a completely socially unacceptable candidate (other than to his supporters), to being perceived by Republicans as a basically mainstream candidate who sometimes runs off at the mouth. If the Republican establishment has finally found something that forces it to really distance itself from Trump and really treat him as unacceptable again, this general shift may be reversed. More Republicans may just feel uncomfortable with him in a way that they had stopped feeling when various Republican leaders started cozying up to him.
It’s not that the rank and file would say “because of this tape I can’t vote for him anymore”; it’s that the vague sense of “it’s ok to vote for him” semi-consciously shifts to “he’s just not the kind of person people like me vote for”. For the most part these people will deny that they ever intended to vote for Trump. There will retroactively declare that they have been undecided all along (and they’ll believe it). There may be a few people out there who were skeptical of Trump’s misogyny, and thought it may just be anti-Trump spin, until they see it for themselves on video and audio. Probably not a huge number of those.

AAF says:

Perot blew up the volatility measures for just one year – 1992. Other than that year, this decline in volatility is a phenomenon that goes back at least another 16 years before 1996.
In 1992 Ross Perot was an unusually strong third party candidate, then suddenly dropped out for a couple months, then re-entered the race. Sort of a quintessential black swan event if you’re measuring volatility of a race.
If you eliminate 1992, volatility has been declining pretty steadily since at least 1980. I don’t think we need to come up with a special explanation of something that specifically happened between 1993 and 1995 that explains current low volatility.

Paul Quirk says:

Nate Silver today says the campaign has been especially volatile. Can someone clarify–what is the difference in measures of volatility?

Neil says:

I was just looking at those two quotes myself and wondering how exactly to compare them to empirical data.

Alan Cobo-Lewis says:

Care to reconcile your low-volatility conclusion with the hohh-volatility conclusion at
Any chance it’s high volatility with high temporal frequency, which is essentially suppressed by the low-pass filter imposed by using state polls?

Sam Wang says:

No, because national polls show the same phenomenon.

AAF says:

About two paragraphs after Nate’s assertion that the 2016 campaign is characterized by high volatility, he says:
“The volatility in the polls in 2016 is pretty average by historical standards, in fact.”
So maybe this isn’t a big Nate vs. Sam discrepancy. It’s Nate vs. Nate, and one of those Nates agrees with Sam, if by historical standards he means post-1992.

Sam Wang says:

That’s only true if by “history” he means “2004-2012.”

Stanley Chow says:

Great site, I have been following PEC for years, but first time commenting. (Full disclosure, I am Canadian, so I can only be affected by these elections but I cannot vote in them.)
Given the low volatility, an interesting question occurs to me: with all the meta-margin and so on, how many “voter-changes” are being tracked?
It is well known that voters in deep red/blue states are irrelevant and only voters in swing states matter; but the question is, how many? Let’s looks the “Median EV estimator”. The EV estimate had a maximum swing of around 50 EV, roughly 10% of 538. We assume this translates to states with 10% of the population.
Now look at the “History of the Meta-Margin”, by construction, the meta-margin only cares about swing states. This had a maximum peak-to-trough of 5%, which means only 2.5% of voters changed. Putting the two together, we care about 0.25% of the voters changing. At 60 million votes cast, that is 150K voters.
Another way to calculate this is to look at the “core” or die-hard vs independent voters. By all accounts, independent voters are around 10%, but only a few percent are truly independent and changeable. Factor in the fact that only independent voters in swing state matter, the number comes out roughly the same.
To put it another way: the presidency is decided by less than a quarter million people (at least in the post-1996 era). Given the number of sites like PEC, I dare say there are more “avid watchers”, than there are “watchees”. A sobering thought.

Leave a Reply

Your email address will not be published. Required fields are marked *