Saturday, October 29, 2016

Think This Month Has Been A Rollercoaster? Think Again.

If you've been keeping up with the news of the campaign this month, you probably feel like you've riding a roller coaster with all its twists and turns and ups and downs.

Each day there's been some new revelation about one or the other's campaign: Trump made another troubling remark; another Wikileaks dump of embarrassing Clinton campaign emails; another accuser comes forward alleging Trump groped/kissed/said a rude thing to her; another new poll showing the race is running neck-and-neck.

The undulations in this thrill ride we call a presidential election campaign have seemed to be relentless.  And now we have the latest stomach-flipping corkscrew in the track: The FBI has new evidence related to Hillary Clinton's email investigation that it's looking into.

There's a dramatic Breaking News alert every time one of these "bombshells" drop, and it's treated by the news media as if it is the game-changer that will potentially change the race, and fundamentally alter the course of U.S. history.   

Except when it doesn't. 

...which is most, if not all, of the time.

You see, voters don't really make sudden, roller coaster moves during a campaign, even when you would think otherwise, especially this late in a campaign.  The reason for this is pretty simple, and has some pretty deep roots in a psychological phenomenon known as cognitive dissonance.

You've most likely heard of cognitive dissonance before, even if not by name. If your parents, or a significant other, or even you, have ever said "you only hear what you want to hear," you're referencing the idea of cognitive dissonance.

About 60 years ago, a psychologist named Leon Festinger discussed what happens when humans encounter information that conflicts with something they already believe. It creates an uncomfortable state he referred to as cognitive dissonance. Since it is uncomfortable, people naturally try to minimize the importance of that information.  They'll ignore it, or try to rationalize it away.

On the other hand, when information reinforces what people already believe, there's no dissonance. Quite the contrary: they are all too willing to accept it because it makes them feel more justified in holding their views. It's not uncomfortable; it's like curling up with a cozy blanket.

Commentators on TV news programs like to use the term "baked in" a lot to talk about voters' attitudes about certain considerations, like their impressions of a candidate's qualifications, temperament, or honesty.  It's an overused term, but it does get to the heart of what I'm discussing here.

In Political Science there has been a long-standing proposition that campaigns actually have a minimal effect on the outcome of elections.  It's undergone some slight revision in the past couple of decades, but the simple point remains: By this point of the campaign the vast majority of voters have already made up their mind, and when events happen they usually hit the wall of people's pre-existing perceptions.

People will view campaign events through their own psychological filters.  If the events reinforce their predispositions, they will hold them up as highly relevant. If they create dissonance, they will be discounted as irrelevant.

So, what does that tell us about how the events of the past several weeks will affect the election on November 8th?  What it suggests is that "October surprises" often do very little to shift voter intentions.  As a case in point, take a look at the figure below. It presents the movement of the polls so far this month along with the "Breaking News" events we've seen so far.

You'll notice that the individual polls (the red dots) show a great deal of variability.  That's normal. Each individual poll is likely to have some error to it. That's to be expected, because each one represents a small subset of the population, and it is reasonable to expect it to have some error.  That's why we talk about a "margin of error." It's an acknowledgement that it's just a sample and could be wrong.  Even if a poll is conducted perfectly, it's likely to have some unintended bias in its sample.

That's why any savvy consumer of poll data will tell you that instead of focusing on any single poll, you should take a look at the average of polls.  Some polls will have error overstating one candidate's level of support, while others will have error overstating his/her opponents level of support. Sampling bias happens. It's not necessarily a nefarious attempt at generating a phony result.  It is just a simple by-product of not talking to everyone, because no one has time for that. But, generally speaking, that bias is going to be random.

That's why I've added the moving average line to the chart. It shows the ebb and flow of public opinion while smoothing out the random sampling errors of each of the individual polls.  The most important thing to notice is how little movement it shows.  Yes, it has moved and it has moved in somewhat predictable fashion.  It is possible to discern that there may have been an uptick of support for Clinton following the Access Hollywood tape coming out, but what's more telling is that it was pretty small. Only 2 points separate the peak (the week following the Access Hollywood tape) from the valley (the week right before the Access Hollywood tape).  For the most part, however, the polls have hovered within that range.

This is pretty consistent with research that has been done on the subject.  My forecasting partner, Tom Holbrook at the University of Wisconsin-Milwaukee, published a book a number of years ago entitled Do Campaigns Matter? and his conclusion was remarkably similar to what I've shown you here: That events in the course of a Fall campaign can move the polls somewhat in predictable fashion, but generally only do so a couple points, and often they move in opposite directions and cancel each other out.

It's this basic fact that makes it possible for an election forecast model like ours to generate fairly accurate predictions of the outcome a month in advance of the election.  By the end of September, most people already know how they're going to vote so they've already got their cognitive defenses built up.  So whatever comes in October often is blunted by their perceptual screens.

So, yes, the polls do move in response to "Breaking News" events, but not to degree that you would think, especially given the attention they are given on the cable news programs.  That's because, at this point in a campaign, the die is pretty much cast for the vast majority of voters.

Sunday, October 23, 2016

Evan and Maurice's Excellent Adventure

Given the sudden interest in Utah as a potential *gasp* swing state in the 2016 presidential election, I thought it might be interesting to take a look at what made this happen.

At first it might be surprising that Donald Trump would be struggling in what is usually the deepest red of the red states. Even so, the signs were there as early as March that there might be a problem for him here. Trump did very poorly in the Utah Republican caucuses, coming in a very distant third behind Ted Cruz and John Kasich. He garnered a paltry 12% of the vote and was soundly trounced by Cruz, who beat him by a margin of 55% in one of the few speed bumps he encountered as he steamrolled his way to the Republican nomination.  Furthermore, head-to-head polls between Clinton and Trump in the Spring showed them essentially tied in the state, with a lot of undecideds.

Clearly, Utah Republicans would have preferred someone else to be their party's nominee.  Even so, party identification usually prevails and polls conducted in June through September showed Trump opening up a substantial lead over Clinton.

Utah, it seemed, was going to have its usual red color on the electoral map.

Even so, there were signs of weakness. Libertarian Gary Johnson was doing unusually well in the polls, regularly drawing support in the double digits. Given that the Libertarian candidate usually gets around 1% of the vote or less in Presidential elections in Utah, this was a sign that there was a certain degree of wariness among Utah Republicans about their nominee.  This was further demonstrated by the fact that the enigmatic "Other" often polled even better than Johnson did when it was offered as an alternative in surveys.  Clearly, it seemed, many Utah voters didn't know who they wanted, they just knew they didn't want Trump or Clinton.

Enter Evan McMullin.

He wasn't brash. He wasn't politically incorrect.
He wasn't Donald Trump.

He was a Utah native. Brigham Young University graduate. Conservative.
And, unlike the Republican nominee, someone with actual government experience.

He announced his candidacy on August 8th and seemed to be just what many Utah Republicans would have preferred. The first poll conducted in the state just a couple weeks after his announcement had him at 9% support. Not too bad for a candidate with no real campaign, but McMullin, it seemed, would be an interesting footnote to the 2016 election. 

But when the infamous Access Hollywood tape became public, McMullin appeared to become a safe harbor for many disaffected Utah Republicans.

I decided to take a dive into the polls that have been conducted in Utah to see if I could surmise the basis of McMullin's support. He'd publicly stated that his main strategy was not to win an Electoral College majority; that clearly wasn't going to be possible. He couldn't get on the ballot in enough states for that to even be a possibility. Instead, he put forward a plan that he would try to win enough states to prevent either Trump or Clinton from getting to the necessary 270 Electoral Votes, which would trigger the 12th Amendment's provisions to force the election into the House of Representatives in January.  

It was a pretty far-fetched plan. It would only work if he could draw support from both Clinton and Trump enough to be able to deny states to both of them. Just on its face that didn't seem likely.  

And we have a 20th century French sociologist named Maurice Duverger to thank.

Well, OK... it's not really Duverger's doing, but rather the Political Science "law" named after him. Duverger's Law is phenomenon with which you're likely familiar, but never knew it had a name.

Duverger argued that in winner-take-all electoral systems like we have in the United States, third parties don't usually survive because there's a strong incentive to vote for the "lesser of two evils" from the two major parties. As long as the threat of the "worse evil" winning remains strong, a minor party will struggle. 

By September, to most observers, Clinton seemed likely to win but it wasn't necessarily a sure thing. Then The Tape came out and seemed to remove all doubt.

And just like that, the incentive to prevent the "worse evil" from winning appeared to evaporate. 

The  Y² Analytics poll was the first to come out after The Tape was released and it showed a virtual three-way tie between Trump, Clinton, and McMullin in Utah. He became a viral phenomenon and it was the first indication that McMullin might be much more than just an interesting footnote to the 2016 election. 

Many Utahns were jumping on the McMullin bandwagon.  The only real question was where were they jumping from.

I examined all of the polls taken in Utah from the past three months and modeled McMullin's support as a function of support for Trump, Clinton, Johnson, "Other," and undecideds.  The result showed a pretty clear pattern and confirmed my suspicions: McMullin's 12th Amendment Strategy wasn't going to work. He might be able to prevent Trump from winning Utah's six Electoral Votes, but the chances that Trump was going to get to 270 were already pretty slim.  And the chances he'd block Clinton from getting to 270 were even slimmer.  

The figure below shows why.  It shows the estimates of where McMullin's support is coming from based on my analysis.  Simply put, he hurts Trump a lot more than he hurts Clinton. If anything, he's going to make it easier for Clinton to get to 270, not more difficult. 

Keeping with the dynamics of Duverger's Law, Clinton supporters have very little incentive to defect to McMullin. So the likelihood that he'll keep her from getting to 270 Electoral Votes is pretty low.  If he's pulling votes from her, he's pulling a lot more votes from Trump.  

McMullin’s rise in the polls has primarily come from three places: Those who had chosen the “Other” option in previous polls, but mostly and almost equally from both Johnson and Trump.  Both the Republican and Libertarian candidates have seen a substantial decline in support when McMullin was added to the survey. But most importantly, Hillary Clinton’s already low level of support in Utah barely moved. 

So things look pretty good for Evan McMullin in Utah, but there's very little reason to believe that his support will extend too far outside the state. It's an interesting phenomenon here in Utah, but it's not clear that he'll have much of an impact elsewhere In what limited polling that has been done in other states, McMullin is barely registering, failing to get above 1%. He's on the ballot in 11 states, six of them traditionally red: Arkansas, Idaho, Kentucky, Louisiana, South Carolina, and Utah. Of those states, South Carolina is the only one where Trump's average lead in the polls is less than 7%.

But South Carolina is a state that Trump won rather handily during the Republican primaries, so the dynamics there don't quite match those of Utah.  Of the remaining states where McMullin is on the ballot (Colorado, Iowa, Minnesota, New Mexico, and Virginia) Clinton is expected to win, with the possible exception of Iowa.  There, Trump holds a small lead in the polls, and it was a state that Trump lost narrowly to Ted Cruz.  Whether or not that will be enough to tilt the outcome there.

It still remains to be seen whether or not McMullin will win in Utah. Duverger would suggest that he won't and the forecast model that Tom Holbrook and I have developed still says it is highly likely that Trump will take Utah despite McMullin's rise. 

More than anything else, it seems pretty clear that McMullin won't be a spoiler. He's not going to deny Trump the Presidency. He's not the cause of the split in the Republican Party, he's a symptom of it, just as Trump's nomination itself is. So, in the end, as fascinating as this has been, McMullin's wild ride with Maurice Duverger may very well end up just being an interesting footnote after all.