Thursday, 10 November 2016

2016 US Presidential Results: Avoiding Absolutes, Inspecting Change, and Income

For my second blog post of the day on the US Presidential Election results, I'd like to point out some big flaws in the analysis of many of my colleagues and co-commentators.

I am seeing a vast amount of claims that victory was or was not granted to Donald Trump by a particular group of voters, based on the proportion of that social group which voted Republican according to the exit polls.

For example, many have been claiming that this was "not a Brexit 2.0", because "lower income groups actually voted more Democrat than Republican".

Further, I have seen claims that the outcome was perhaps due to racial voting (or racism), with the figures suggesting that less than 1 in 10 Black voters went for Trump, as opposed to a majority for Trump among White voters.

The same can be said of gender, with more women voting Clinton versus men for Trump. There are suggestions there that we can categorise the result as a battle between men versus women.

While each of these observations are entirely correct, using them to explain how Donald Trump won the Presidency is horrendously misleading, and at worst plain wrong. Principally, I just don't think these explanations capture what went on at all.

Such analysis of these headline figures completely rejects change, and misses the part that swing has to play in tipping battlegrounds and changing elections. If we want to understand why states such Wisconsin, Iowa, Florida, and Pennsylvania flipped to Republican, we have to be seeking change based explanations. Why did those states swing? Which voters were pushed away from the Democrats, and toward the Republicans? And why?

Income Groups:

The most oft floated claim I have seen is that Trump's victory was not a case of left behind, lower income groups 'revolting'. This is the frame used to describe the rise of UKIP and Brexit in the United Kingdom, and has been suggested as a potential explanation for Trump's victory on Tuesday.

The argument against however supposedly goes that because lower income groups in fact voted mostly for the Democrats, that a working class revolt must be dismissed as a potential explanation for Clinton's defeat. Indeed, those earning under $30,000 voted 53-41% in Clinton's favour, according to the exit poll data (shown below, courtesy of the New York Times). There is a similar margin for voters earning between $30,000 and $49,999. Brexit 2.0 this was not, some say.

However, looking at the above figure if we ignore the absolutes and inspect the change, we see a completely different picture than to the "this was not a working class revolt" frame. Among those lowest earners, there was a 16 point swing to the Republicans. 16 points. That is in fact the biggest swing of any aggregated group in the exit polling. Among the second lowest income group, the swing is much smaller but still there: 6%.

So how then can this definitely not be a class based explanation?

Further, and crucially, many of the States which flipped from Democrat to Republican are very much below average in terms of income and wealth. Where else but there should we expect to see lower income groups punish the political establishment?

Indeed, in Wisconsin exit polling suggested that Trump had almost closed the gap in terms of lower income voters. In 2012? A 62-37% lead for the Democrats. Wisconsin turning Red of course was a complete surprise, and was the State which pushed Donald Trump over the 270 line.

In Minnesota, which was very much in the balance for most of the night despite it being a traditional Democrat stronghold, the Democrats enjoyed a lead of 55-42% in lower income groups in 2012. However, look ahead to 2016 and this lead has completely disappeared.

Finally, in Ohio we see the exact same trend of rapidly narrowing Democrat leads in lower income groups between 2012 and 2016, according to the exit poll data. All across the Mid-West, exit polling suggests that lower income groups switched in big numbers to support Donald Trump.

Democratic leads among low income voters might not have disappeared, but the big swings in key Rust Belt States is certainly a huge part of the explanation as to why so many of them changed hands from Democrat to Republican. I'd suggest that from these figures, we could very, very much be looking at Brexit 2.0.


Another claim I would like to argue against is that the victory was necessarily about white voters. Again, looking at the exit polling it seems a fairly innocuous suggestion that it was the White vote which put Trump in office. Nearly 60% of Whites voted Republican.

But again, this absolute figure has barely changed since 2012. Trump has done no better among white voters than Mitt Romney did in 2012.

Equally, while there are clear majorities for the Democrats among all Ethnic Minority groups, these margins are actually down on 2012 when Obama stood for a second term. To be expected with the first Black President leaving office, perhaps? Certainly, but swings in ethnic minority support away from Democrats means that fascinating on Trump's large White vote as an explanation for his victory is misguided.

Most interestingly, despite everything the Republican nominee said about them in the run up to polling data, Trump actually won a higher percentage of Latino voters than Romney managed in 2012.

In Florida, the Democrats' failure to dominate the Latino vote in the same way they did the Black vote arguably could have cost them big time. State level exit polling in Florida from this year suggests that 35% of Latinos voted Trump. This was only slightly down from 39% in 2012. The Democrats managed to gain 2% more of this vote, but it was not enough to tip the State over into the Democrat column.

Such a meagre increase, despite all that had been said by Trump about Hispanics/Latinos, will be very disappointing for the Democrats. If they had made more progress with this group, they could well have won here and held on to the 29 electoral college votes Florida offered.

So, was the result racialised? Perhaps: but very much in the opposite direction as to that which you might expect. While whites certainly overwhelmingly voted Trump, they also voted overwhelmingly for Romney. And more importantly, the Democrats appear to have lost ground with Ethnic Minority groups.


Finally figures from the Exit Polling on sex are below. They suggest that while men indeed did tend to vote for Trump over Clinton, and that this swung up by 5% in the Republicans' favour from 2012, the percentage of female voters backing the Democrats has scarcely moved an inch.

So men were more likely to go Republican this time around than last at the national level, but it doesn't appear that women were more likely to go Democrat. Clinton's candidacy did not draw in female voters, according to these figures.

This rather mixed picture comes through again in those battleground states. In Michigan 2012 exit polls reported that 48% of men had voted Republican, and 57% of women had voted Democrat. In 2016, 53% of men voted Trump, an increase of 5%. But also, only 53% of women voted Clinton - a decrease in the Democrat vote share among women of 4%. In fact, in Michigan the same percentage of women voted Republican in 2016 as in 2012 (42%).

In Wisconsin the same trend bears out: 51% of men voted Republican in 2012, according to the exit poll. This increased to 54% in 2016. A 3 increase in favour of Trump. However, the Democrat voting figures among women show an even bigger change: 57% in 2012 down to 53% in 2016. Indeed, there was apparently 1% increase in Republican voting among women.

A majority of men voting Republican and a majority of women voting Democrat is not exceptional, and the latter trend has changed very little from 2012 to 2016. Could that overall 5% swing in men have pushed Trump over the line? Possibly, it's something for sure, but it's nothing on the 16 point swing of lower income groups above.

What this certainly was not is a mass movement of men versus women, or that women had flocked to vote for Clinton and abandoned the Republicans. This contest was not really any more gendered than a usual Democrat versus Republican contest normally is.

Potential Explanations - Lower Income Group Power:

Overall, I would suggest that the working class revolt, Brexit 2.0 style explanations have far more explanatory power than many commentators and pundits are suggesting. Lower income groups have swung dramatically, and analysis of exit polling data suggests that this swing occurred as much in many of the Rust Belt states which flipped to the Republicans as it did anywhere else.

Race has also certainly played a part, but I think in a more nuanced way than some are claiming. Will it have been predominantly Whites on lower incomes in the Rust Belt who flipped the States red? Quite probably. But equally, it looks like the Democrats did not capture the Latino vote in Florida (and of course probably elsewhere too) in anywhere near the fashion that they would have hoped. Overall, the Democrats nationally did not perform as well with Ethnic Minorities as they might have hoped, given who they were up against and given their traditional success with these groups.

I would also suggest that gender will have acted as a compounding factor. Perhaps being lower income and male made you more likely to flip, but it seems certain that a large number of women flipped too. Moreover, I believe income had a much stronger effect than gender.

In short, at this early stage it's my belief that swinging lower income voters (perhaps more likely to be white and male) coupled with the Democrats coming up shorter than they would have liked with the Latino vote is the best explanation for the States which flipped, and thus this result.

As a caveat: it's important to remember that these exit polling figures are only polls, not results. They're not likely to be completely accurate in terms of the marginals at least - the general movements and swings you can most likely count on.

2016 US Presidential Results: Primary Warnings

As we all conduct our polling and forecasting post-mortems and try to figure out just how Donald Trump won the US Presidential Election on Tuesday, I propose turning our attention back to primaries for a potential explanation.

On the night, news that Clinton was really struggling in Michigan and Wisconsin were almost the final nail in her coffin - the Rust Belt was abandoning her when her earlier failure to pick up Florida and North Carolina meant that winning these states was crucial. Eventually it turned out that even Pennsylvania would turn its back on the Democratic nominee. The Blue Wall had crumbled.

In theory, that shouldn't have happened. These states have been solidly Democrat since the 1980s, and polls conducted there suggested Clinton had a sizeable lead. So much was her confidence, that Clinton did not once visit Wisconsin, and appeared in Michigan just a handful of times, and sent only surrogates in the final weeks in an attempt to see off Trump's advances there.

But of course it did. Wisconsin voted Trump, and Pennsylvania followed. Iowa flipped back to the Republicans, along with Ohio. Clinton seems to have just about scraped a victory in Minnesota, but Michigan is still on a knife edge as we move into Thursday.

While some might have found this surprising, there was a clear pattern: Clinton was struggling in Northern States where Bernie Sanders either won or ran her to the wire in during Primary season. The map below demonstrates this perfectly.

County-level Primary Results: New York Times

Looking at those North East and Mid-Western states, we are instantly reminded that Sanders in fact won Wisconsin, Minnesota, and Michigan. He ran Clinton down to the wire in Iowa, and was much closer to Clinton in Pennsylvania than the polls were predicting. Sanders also won in New Hampshire - another State where Clinton is currently still unable to claim as the count goes down to the final few precincts.

It's not that Clinton is struggling in all States that Sanders collected in the Primary contests: she's done very well in Vermont, Colorado, Nevada, Oregon and Washington for example. But where Clinton has struggled, particularly in the Mid-West and North-East, it seems to correlate very highly with Sanders support (with the exception of Ohio).

What does this mean? Have Sanders supporters in those States abandoned her and stayed at home - or worse turned out and voted Trump?

Possibly, maybe in the case of a few. But far more likely I think hat the Primary results in these states were a warning shot across Clinton's bow, which she completely ignored. They captured a feeling, demonstrated conflicting emotions and values which Clinton and her campaign should have taken very seriously.

I travelled around Wisconsin while the Primary battle-buses were in town, and the anti-Clinton feeling there was huge. Democrats were very much pro-Sanders, particularly in the ever-blue city of Milwaukee. They of course went out and overwhelmingly backed him ahead of Clinton.

Republicans, though mostly supporting Ted Cruz, were vehemently anti-Clinton, citing her as an embodiment of entitlement and privilege, a shining light of success while their State struggled and failed economically to keep up with its neighbours.

But among independents too, taxi drivers, students, and bar goers who had no affiliation, there was a tangible anti-Clinton feeling. Everyone I spoke to after the result was pleased that Sanders had won on the Democratic side. Everyone.

It's my view that the results of those primaries should have been a shot in the arm to Clinton - here were the kind of areas where perhaps life wasn't going so well, where the liberal, free market values of the modern Democratic Party establishment were not at all shared.

I don't believe that voters who backed Sanders in those Primary contests went out en masse and voted for Trump on Tuesday, but I do believe that they captured, or that they personified, a feeling. An attitude and a set of values that were totally at odds with everything they believed Clinton was about.

Voters in places like Wisconsin had already taken one opportunity to take aim at Clinton, and she totally ignored them. Was it all that surprising that they ended up doing it again?

Tuesday, 8 November 2016

US Presidential Prediction

For my first Blog post in a while, I'm offering a prediction for today's US Presidential Election.

The vast majority of predictions have Hilary Clinton favoured heavily as the winner. Even the usually highly cautious 538 forecast (ran by Nate Silver) now has a Clinton win predicted with over 70% probability. Other forecasters are ranging in the high nineties, with some as close to 100% as their mathematics can allow. For an excellent overview of what each of the prediction and betting markets are saying, this piece by Dr. Steve Fisher at Elections Etc is an excellent resource.

My prediction isn't the result of a mathematical model, just my reading of state level polls and other prediction houses and betting markets.

Principally, I am also calling a Clinton victory (no surprise there), but with a slightly higher Electoral College count than 538, closer to some of the less-cautious alternatives. My map of projected outcomes (put together using 538's prediction tool) is below.

In terms of the popular vote, I'm expecting Clinton to fetch just shy of 50% to Trump's 45%. My projection actually hasn't changed in about a week, since the impact on the polls of the latest round of negative Clinton news bottomed out and favourable state-level polls came out for Clinton in many key states.

Donald Trump seems to have recovered from scares in Arizona and Georgia, and looks solid in Ohio, so I'm predicting him to win there. Since Trump seems to be doing well in the Buckeye State, I also expect him to cling on to North Carolina in a Romney-like fashion. This is the state I am least certain about, but early turnout information from this key battleground seems to suggest a decrease in black voters heading to the polls, which won't be good news for Clinton.

Contrary to some expectations, as well as holding on to New Hampshire, Michigan, Virginia and Pennsylvania, I believe Clinton will sweep up in all three of Colorado, Nevada, and Florida. The latter is by far the least certain, with most polls on a knife edge. However, an increasingly ethnically diverse electorate which appears to turning out in high numbers is encouraging news for Clinton. This is the same trend which looks to be delivering Nevada for her.

Such is the nature in which the States are likely to be called over the course of the night, by my prediction the contest will be all but over by 1am GMT when Florida comes in, with Trump's path to victory pretty much closed at that point.

And here's one to look out for: although I'm not predicting it to come in, I have a little bet on Evan McMullin to win in Utah.