This election has caught everybody off guard, including the pollsters. The polling houses hadn’t yet decided on a common approach in response to the wreckage of the 2015 General Election. A side benefit of this for social scientists is the natural experiment playing out in the heat of the campaign.
Before we get into that, it’s useful to remember why 2015 failed so miserably (and it was miserable for those of us who worked GOTV all day on polling day from 0530 until 2200, changed our clothes and dressed up smart, affixed the red rosette, and arrived at the Guildhall for the count only to see the exit polls coming in suggesting a Conservative majority, who then went on to witness our losing of one Labour seat here in Plymouth combined with losing what we were fairly confident of a Labour pickup by only 523 votes). It wasn’t because of the “shy tory” effect that plagued British pollsters in the 1992 election (which far too many people still believe was the fault in 2015). Rather, it was a sampling / likely voter model failure:
In looking at what’s changed it’s probably best to start with what actually went wrong and what problem the pollsters are trying to solve. As all readers will know, the polls in 2015 wrongly overstated Labour support and understated the Conservatives. The BPC/MRS inquiry under Pat Sturgis concluded this was down to unrepresentative samples.
Specially, it looked as if polls had too many younger people who were too engaged and too interested in politics. The effect of this was that while in reality there was a big difference between the high turnout among old people and the low turnout among young people, among the sort of people who took part in polls this gap was too small. In short, the sort of young people who took part in polls went out and voted Labour; the sort of young people who weren’t interested and stayed at home didn’t take part in polls either.
UK Polling Report has a useful explanation of who is doing what methodologically for 2017 (the above quote is excepted from this post). This is especially critical in being able to interpret any given poll from a reasonably educated perspective, as within the last week alone we have seen the range of estimate Conservative leads from a high of 14% to a low of 3%. As I wrote on Wednesday, while the individual estimates are all over the shop floor at the moment, one thing that is solid is the shrinking Conservative lead.
At least two sources are now publishing predictive models at the constituency level. We briefly discussed the methodology of the YouGov model in the comments of Wednesday’s post (shout out to commenter Gregor Sansa for correctly predicting their use of multilevel regression and poststratification, or which I call in an old school sense hierarchical linear modelling). YouGov’s seat-level probabilities can be found here, and it’s updated daily. Using a different methodology (the several paragraph explanation thereof isn’t geeky / detailed enough to pass judgment on the relative merits) the Britain Elects model can be found here. The two models suggest quite different results. At time of writing (0630 BST 2 June) YouGov is estimating 317 seats for the Tories (with an error band of 285 to 353 seats), while Britain Elects is going with 362 (330 to 394). Both models also have significant, indeed wildly different predictions for the specific constituency where I campaign. Of course, the quality of the predictive model is limited by the quality of the data fed into it; those of us who blithely predicted in the media last November that Clinton would win the Electoral College were undermined by the data coming out of WI, MI, and PA. Hence, while the YouGov model makes it appear that all our hard work in Plymouth Sutton & Devonport is finally paying off, in reality the YouGov model is a captive of the underlying YouGov data. Labourites need to understand that since the 2015 GE, Labour has only been even or ahead (by one point) in three out of well over 250 polls now, and all three were YouGov. Britain Elects takes a broader view by using data from Ipsos Mori, YouGov, TNS, Panelbase, GfK, ICM, Survation, ComRes, Opinium, ORB, BMG and Lord Ashcroft, combined with data from the British Election Study. Ashcroft also has a probabilistic model here, but it’s already a week old.
Is one better than the other? We won’t know until the 9th of June.
One other, more personal / rant note about polling and the Labour Party. There is a wing of the party that harbours deep suspicion of all polling, especially (ironically) YouGov because they have / had connections with the Conservative Party. Before this election was called, in this community polls were relegated to the status of Tory Conspiracy to demonstrate how unelectable Jeremy Corbyn is thus brainwashing the mindless sheep of the electorate from the one true path. It’s entertaining to witness all the new converts to polling data when YouGov has shown our deficit narrow to 5%, then 3%. Kids (writing to the Labour Party, not the sophisticated and discerning readers of LGM of course), this instinctive rejection of, and newfound love for YouGov is called confirmation bias, and don’t do that.
Yet people still do. I was ranted at, defriended, and blocked (I nailed the trifecta!) by a local Labour Party member (and a parliamentary candidate in the 2015GE no less) on fb yesterday because I had the temerity to point out in response to the suggestion that “we all know” polling drives public opinion that the evidence does not support that assertion (and yes, I might have pulled rank in a sense by pointing out that I used to teach survey research here, but I do have an understanding of the subject better than the average bear given that I’ve published on it). In a Labour forum on fb yesterday, it was asserted that the polling houses are under strict orders by Conservative High Command to make it appear close to scare Tories into voting given the latest YouGov poll has us only 3% behind. Apparently, when the polls show a huge Conservative lead, the fix is in to demonstrate that the LP is unelectable, when it’s close the fix is in to mobilise Conservative voters, and when the sun rises in the East the fix is in to show that the Tories are, indeed, the masters of the universe.
Conspiracy theories: for the love of God and all that is good and pure, don’t do that.
In response to some of the comments from Wednesday’s post (usually I have the time to read the comments to my posts, but lately I haven’t had the time to read the front page of LGM, let alone the comments to my own, rare, posts):
When British pollsters have failed spectacularly in a general election (e.g. 1992, 2015), it’s generally been in overestimating Labour support and underestimating Conservative support, but for different reasons (response bias / social desirability in 1992, sampling / likely voter models in 2015).
I’m not going to get into a discussion on the merits of the current Labour Party leadership at this point. For five more days of campaigning, and a day of GOTV on polling day, we have an election to fight. I’m under no illusions that the Labour Party will be full of peace, love, and understanding form the 9th of June, but until then, at least in Plymouth, we’re doing a commendable job at party unity. While both data and anecdotal evidence suggests that around the country the post-2015 intake of members (which is a large majority of the LP’s membership, it should be stated) are proportionately less likely to do the grunt work of politics and more likely to be clicktivists, here in Plymouth we have seen a large number of new faces all working towards electing our decidedly un-Corbynista candidates to Parliament. We’ll see where we stand on the 9th of June.
jben writes: “I mean, I agree that even Corbyn and co. are the lesser evil compared to the malicious jerks who are running the country now. (I assume David also agrees, or he wouldn’t be campaigning for Labour.)”
I voted for Corbyn in 2015, and wrote about it here at LGM. The TL;DR version of that post is “not due to any sense of ideological purity, but because I think he has the best chance to lead Labour to victory in 2020”. Needless to say, that choice has been tested since, and I did not vote for him last summer. That said, I do not consider Corbyn the lesser of two evils. I have reservations about the man, in terms of his political nous, the choices he makes in advisors, and on some bits of foreign policy, but given the binomial choice environment, I do not have one ounce of reservation that a Labour Government, a Corbyn-led Labour Government, is infinitely preferable to a Conservative government. But, given I’m active in the Plymouth Labour Party, my loyalties are also, if not predominantly, local. Local councillors, activists, and our parliamentary candidate for my constituency are also friends of mine. Over the past several elections, local and general, I’ve canvassed, plotted, attended counts, and drank with these people. I’d be doing what I’m doing regardless of the name of the party leader. I’m loyal to the party, not the leader du jour.
Ronan writes “As a matter of seats available, considering they’ve lost Scotland, what do Labour need to do to get a majority”. During the 2014 independence referendum, I wrote a series of posts, including one that explored the impact on the Labour Party if Scotland were to go its separate way. At the time, Labour held 40 or 41 seats in Scotland (this is from memory). It turns out that the Labour Party only required its Scottish bulwark once to form a government. However, obviously, we’d take 30 or 40 seats in Scotland, as well as the political talent offered by a strong Scottish Labour Party (John Smith, Gordon Brown, to name just two). Given we’re likely to only win one seat in Scotland (again), for Labour to get to 326 we’d have to run the table in the urban constituencies, hold if not expand in Wales, do better than predicted in the Midlands, as well as pick up seats in the South East of England. Even the optimistic (for Labour) YouGov model predicts a top estimate of 285 seats, an extra net 40 Scottish seats would get us to the precipice of an outright majority. However, that 285 represents the absolute top end of the most Labour-friendly of the various predictive models.
Finally, let’s return to the first line of this post, about how this election has caught everybody off guard. This is the first snap election in the UK since October 1974, so it’s a rare treat for anybody, let alone a politically active American political scientist, to be right in the middle of this campaign. Aside from the very occasional recall election in the US at the state or local level, there is no comparison in American politics (for better or for worse).
UPDATE: I meant to add in a completely unrelated note that two years ago today, I was at what I believe was the last ever Replacements gig at the Roundhouse in London.