Monday 31 October 2016

British lessons for the American election?


Many British observers looking on at the US Presidential election have started to feel hairs standing up on their necks. A front-runner most people think is now the inevitable winner? An establishment choice that it seems inconceivable won't make it over the line somehow? Polling that mostly points in one direction, but which seems more divided and confused than ever before? No-one who's had to suffer both through the 2015 UK General Election, which so many people thought would bring Labour leader Ed Miliband to power, and the 2016 European referendum, which most voters thought 'Remain' would win, can fail to see the parallels with Britain's recent travails. Liberals, progressives, radicals, internationalists: they all thought that they were at or over the line. They discovered that the country they lived in wasn't quite what they thought it was. Might the same thing happen now, with Hillary Clinton (above) the victim of a similar polling failure?

The parallels actually don't fit very well. There are four reasons why, and each of them is important in giving greater statistical clarity to the present US election, as well as thinking about the British contests just gone. Let's have a look at them, in no particular order:

1. The British polls weren't actually all that wrong. 

Remember that the 2015 General Election looked very tight, giving off a nip-and-tuck, neck-and-neck quality all the way through... If you trusted the polls. That meant that a far-from-massive miss turned a projected Hung Parliament in which Labour might have sought to form a government into a small Conservative majority - a polling miss of about 6.5% if we add up the inaccuracy on the Labour and Conservative scores. Polls got slightly closer in the EU referendum, with statistical averages on the eve of the results registering an exactly-equal deadlock to a Remain lead of four points (Leave, of course, won by 3.8%). In fact, if we just average the last poll from each company in the field, we get a Remain lead of 2.5% - and a total polling error, from polling lead to final numbers, of 6.3%. Note here, though, that UK election polls don't actually have that fantastic a record when compared to their US cousins. British polls usually overstated the government by a long way in the 1980s, and then again under New Labour: there was a five-point overall miss in 2001, for instance. The Americans, as we shall see in a moment, have always done rather better. And they haven't been disgraced in any of the last few contests - even very tight ones such as 2000 and 2004. So we can't even be that confident that our actually rather narrow range of error can be read across to the new contest on the other side of the Atlantic.

Then there are differences in the detail. The Brexit vote continuously threw up a so-called 'mode effect': online polls seemed to show Leave narrowly ahead, while some phone pollsters showed a large Remain lead. There's not such a large difference in the current Presidential election: YouGov's online panel, indeed, is returning a fairly constant four-point lead for the Democrats' candidate, which is right in the average range that we'd expect from all polling. The two cases are very dissimilar, and even a May 2015-style error in the US might not allow Mr Trump to prevail. US pollsters are savvy, experienced and multifarious. At the time of writing, Mrs Clinton leads on average by between three and seven per cent. It will be a large and unusual polling miss - even by British standards - to overwhelm that lead. But beware: that's exactly what we got in 2015 and 2016. And the British polls had been getting more accurate, in 2005 and 2010 not ending up all that far off if we look at the overall average. Then look what happened. The miss can come out of the blue. It could be there, but it will have to be pretty big.

2. The USA just has a lot more, and more accurate, polling. 

It's true that British General Elections bring in a lot of numbers. But American elections throw up a lot more information - many thousands of polls at state as well as national level which just about confirm the national picture we've got of a not-particularly-large, but for now stable, Clinton lead. And it's usually right. Take figures from the venerable Gallup Organisation, which has now stopped horse-race polling after a chastening polling miss in 2012 (the two-party result was a total of five points different from their final poll). If we look back to 1980, their final numbers were never more than 6.8% away from the actual outcome - and were as near as 0.2% in 1984. Polling averages since 1968 have usually predicted the final scores within a range from exact accuracy to a four point miss. In 2012 the polls were on average 3.2% away from the actual election numbers.

We also cannot emphasise enough the sheer amount of lower-level polling conducted among the individual states that make up the US. Yes, polling of individual Westminster seats got a terrible reputation because of their apparent 'miss' in 2015 - but bear in mind that this was an unusual experiment, funded by the Conservative peer and philanthropist Lord Ashcroft, and which usually made its most egregious misses highlighting the wrong question in Liberal Democrat seats - namely, 'would you vote for your local MP', rather than 'would you vote Liberal Democrat'? More general regional polls, for instance of the Liberal Democrats' apparent stronghold in the South-West of England, got things spot on. State polls in the US, on the other hand, are telling us a story that is intuitively understandable: Donald Trump is doing better than Mitt Romney among blue-collar white voters in the Upper Midwest, while Secretary Clinton is doing better than President Obama in the South with Latino voters and college-educated whites who detest Mr Trump's plunge into Know Nothing politics. That doesn't mean the picture's right. But it is an important confirmation of the national numbers from experienced pollsters who've mostly done all this before. We didn't see much of that in Britain.

3. Drawing in new voters probably won't help Donald Trump. 

It's actually a mistake - one which we shared at the time - to think that the British polling misses are to do with 'shy' anyone. In 2015, there was a large framing error which brought too many politically-aware young people into the numbers; in 2016, there was a large turnout of unlikely and unusual voters who thought that at last someone might listen to them about immigration, border control and national sovereignty. So Republicans hoping for a 'shy Trump' effect, under which Americans too embarrassed to tell anyone they are voting Trump nonetheless pull the lever for him in the secrecy of the polling booth, might be waiting a long time for their surprise. Mr Trump didn't manage this in the primary season: indeed, his voters were actually a bit better off than the other Republican candidates' were. He actually underperformed his polls (on average). And there's no evidence at all that he's pulling in new voters or new registrations. Indeed, British readers should in this respect remember that, as we draw more people into the American electorate, it's likely to become more Democratic as poorer voters get pulled into the sample. There's not all that much data to say that such people are more likely to favour Donald Trump.

4. The early vote doesn't look the same. 

It's illegal to snoop around the early postal voting in Britain, and it's still more frowned on to report its content. But there's not much that the election authorities can do - beyond threatening prosecution - to stop albeit-vague news of how it's going leaking out. In 2015, Labour officials knew pretty early on that the swing that they needed wasn't happening. In 2016, Labour MPs in particular - and especially those with seats in the North of England - thought that the contest was over pretty much before it began, as they picked up talk of a huge Leave vote from their constituents that even a last-minute switch back to Remain couldn't make up for. Things in the US right now don't sound anything like that, and in fact those who know think that on balance things look pretty good for Mrs Clinton. They look particularly good in Nevada, not-too-bad in North Carolina (a state without which it will be very, very difficult for Mr Trump to win overall) and close in Florida - though not so hot for the Democrats in Iowa, a state in which Secretary Clinton has struggled all year.

It's not all good news for Mr Trumps's opponents. African-American enthusiasm appears to be down. There's a large and unpredictable rise in unaffiliated, non-partisan early voting in some places, though as these Americans in some key places lean Democratic and are often pretty young, that might just confirm the overall picture. In general, if you'd offered Team Clinton those impressions a few months ago, they would gladly have accepted them. Now it's possible that the unlikely and marginal voters Mr Trump is depending on will turn up in person on the day, but it's not as likely as the 100% certitude that these early Democratic votes have already been banked.

The British and American cases don't look alike, really. The political context is different. The closest parallel we can offer is if the Brexit campaign had been led by Nigel Farage, UKIP's not-very-popular leader, and if Conservatives Boris Johnson and Michael Gove had stayed loyal to Prime Minister David Cameron and not lent their backing to Leave. Almost certainly, Leave would have lost, tarnished by the United Kingdom Independence Party's perceived extremism and beyond-the-pale dog whistling on immigration. Without Mr Johnson, inexplicably one of the most popular politicians anywhere, and without Mr Gove, the Leavers would have backed both personal appeal and apparent intellectual weight. They would have ended up pretty much where Mr Trump is now - without, perhaps, being weighed down by the fact Mr Trump is for most of the electorate a deeply feared and loathed narcissist, fantasist, sociopath and misogynist.

Even so, the dissimilarity between the British and American cases should certainly not give anyone confidence that Secretary Clinton is sure to win, even if she goes into polling day four or five or so points ahead (given her ongoing email drama, she probably won't anyway). Polling is getting harder. Let's look at a couple of recent examples. Last year's Israeli election was a notable polling miss. The ruling Likud Party was supposed to win a number of Knesset seats in the low 20s. They actually got 30. The rival Zionist Union was thought likely to win a number of seats in the mid-20s, which they just about got to in the end (they won 24): but Benjamin Netanyahu, Likud's leader, had managed to draw in harder-line nationalist voters from other right-wing parties.

Here's another example: the Icelandic election we've just had gave us an 8.6% gross polling miss for the two main parties going into the election. The centre-right Independence Party's last all-pollster average stood at 25.4% going into polling day: they actually got 29% of the vote. The rival Pirate Party had a polling average of 19.5%: they actually got 14.5%, and came third. Now, none of those countries look much like the United States. Israel bans polls for the last five days before an election, which makes it very difficult indeed to pick up any last-minute shift. Iceland is a tiny country going through enormous political changes emanating from its recent crisis. But the lesson is the same: these days, if things look close, you can't definitely trust the polls.

This American election doesn't look all that much like the British General Election of 2015 or the Brexit vote. US polling is more reliable, more resilient, more voluminous. The politics of Clinton-versus-Trump looks very different from the two most recent UK contests. The likely electorate, and the dynamics of change, don't look all that similar. But is there room for a great polling miss? Yes. What we've discussed here is the effect of known unknowns from the recent British past: a framing error (like 2015) and a big expansion of the electorate that helps the more conservative side (like 2016). Neither look vastly likely. There's not that much evidence to say that the two cases are alike. The British experience says that we'll need to be hit by something we haven't yet thought of. But that doesn't mean that Democrats should rest easy, as the Israeli and Icelandic examples demonstrate. The US hasn't seen a polling disaster like that presently needed by Mr Trump since the 1980 Presidential contest between President Carter and Ronald Reagan. But perhaps there's an unknown unknown lurking out there, that we haven't thought of. Or perhaps Mr Trump will just overtake Secretary Clinton in the polls as well as in the actual voting booths. Both possible. American and especially European liberals have long taken a Trump victory to be impossible just because it is rather unlikely - and because they find it impossible to understand many of the moral impulses, and much of the language, of people who like Mr Trump.

So: a Trump Presidency is not impossible. It is, in fact, an entirely credible next step from here. We're within the spectrum of recent international polling error, and within the range of British misses - though in the latter case, only just about. Polling is more difficult than it used to be. Right now the betting markets and Nate Silver's 538 give you roughly a 23-24% chance of a Trump victory. That's about the same likelihood, as you stand with a coin in your hand, of turning up a heads twice in a row. It's not exactly a comforting thought, is it?