Polls
The Modelers Have No Clothes
Predicting elections is an occult science—with a shoddy track record.
How many ‘centrists’ are there? (On Twitter and off it)
‘Centrists,’ Jeremy Corbyn’s former official spokesman tells us, ‘love being talked down to by people in power as it provides false reassurance that they must be better than them.’ The centrist, another Corbynite B-lister explains, ‘has the weary, lecturing tone of a frustrated parent absolutely nailed on.’ But why be so mealy mouthed? As the man who came closer than anyone to embodying the spirit of the ongoing neo-Tankie renaissance memorably put it, ‘better a thousand honest fascists than some glistening sleaze who’s “neither left nor right.”‘
Is it wise to alienate voters who don’t identify with the left or the right? Some have taken last year’s general election result to demonstrate that it is. The Hammer of the Moderates himself, Owen Jones, has argued that an appeal to moderation can make no electoral sense ‘at a time when more than 80% of the electorate voted for a left-led Labour party or the Brexiteer Tories.’ But given that Labour lost that election by a wider margin of seats than it did under Gordon Brown, that argument seems a little odd. Surely losing an election to the Conservatives after publicly abandoning the centre doesn’t indicate that publicly abandoning the centre was the right thing for Labour to do?
But I think I’ve got it figured out now. These people live their lives on Twitter, where the best way of getting your voice heard is to start a fight that others will want to join in. Moderation never gets much traction there. It’s not like in the outside world:
Data courtesy of the British Election Study.
Voters in the 2017 general election – and how they voted previously
This is the third and final part of my preliminary analysis of groups of voters defined by the choices they made in the 2015 general election, the 2016 European Union membership referendum, and the 2017 general election (c.f. Stephen Bush’s nine voter groups), using an English subset of responses to the British Election Study’s post-election face-to-face survey. In the first part, I looked at the ten largest groups, from Conservative-Leave-Conservative to Conservative-Remain-Labour, both in terms of their size and in terms of their self-declared likelihood to vote for various parties in future, and found that Labour Remainers were not only more numerous but (on their own assessment) more likely to be poached than Labour Leavers, while the smaller group of Conservative Remainers who had switched to voting Labour were quite likely to switch again. In the second part, I looked at six groups of voters who had in common that they could have voted but did not in the 2015 general election, finding that most of them did not vote either in the 2016 referendum or the 2017 general election, and that only the minority who voted Remain in the 2016 referendum were more likely than not to have voted in the 2017 general election.
To finish up for now, here’s a single chart showing all voter groups which participated in the 2017 general election (weighted by demographic group and by 2017 vote). Each quarter of the chart below shows the members of the sample who voted for one of the four main parties. These voters are further subdivided into columns to show how they voted in the referendum and into coloured blocks to show who they voted for in 2015 (note that black covers both non-voting and voting outside the four main parties, which most often meant voting Green as the data are from England only):
What does this chart tell us? It tells us that, at least within this random sample of 1874 voters (1777 after weighting)…
- Most of those who voted for the two right-wing parties in 2017 had voted Leave in 2016, while most of those who voted for the two liberal-left wing parties in 2017 had voted Remain – and this is true whether we focus on voters retained or on new voters gained
- However, the Remainer proportion of the 2017 Conservative vote is substantially greater than the Leaver proportion of the 2017 Labour vote
- The Conservative Party picked up most of the (Leave-voting) UKIP vote from 2015
- But it picked up more votes from Labour Remainers than it did from Labour Leavers
- And so (less surprisingly) did the Liberal Democrats
- In fact, the Labour Party and the Liberal Democrats both picked up far more votes from Remainers than from Leavers – both from the Conservatives and from each other
- Those 2017 voters who did not vote in the 2016 referendum mostly ended up voting Labour
- Every party except UKIP both gained and lost a substantial proportion of voters between 2015 and 2017
- UKIP definitely lost voters, but you have to zoom into the chart to see its minuscule gains
To summarise: on the evidence of these data, the only really notable movement among Leavers seems to have been the metamorphosis of most 2015 UKIP voters into 2017 Conservative voters. Those voters are unlikely to go back to UKIP because, on the brink of financial ruin and with no credible leadership, UKIP is in no position to win them back. Otherwise, the movement that has taken place appears to have been largely among Remain voters: more Remainers than Leavers would seem to have gone in both directions between Labour and the Conservatives and between Labour and the Liberal Democrats, while the Liberal Democrats picked up a not inconsiderable number of formerly Conservative Remainers and while a larger block of Remain voters who had not voted for any of the four main parties in 2015 simultaneously fell into orbit around Labour.
The Labour and Conservative Parties did nothing to court the Remain vote last year, and have done nothing to court it since. But it looks like it’s the Remain vote that is volatile now.
Non-voters from the 2015 general election and what they did next
On Friday, I posted some analysis of groups of English voters defined by the combinations of choices they made in a succession of votes. That was the first installment of a multi-part response to Stephen Bush’s recent article on why we should stop focusing so obsessively on people who voted Labour in 2015 and then voted to leave the European Union in 2016. I’d now like to take a look at those who didn’t vote at all in the 2015 general election.
Excluding those who did not vote because they were ineligible, there were 290 GE2015 non-voters in the dataset that I’m using: an English subset of the post-election 2017 face-to-face survey carried out as part of the long-running and hugely respected British Election Study. The 290 become 311 or more if we weight for demographic group, as I did for Friday’s analysis – which indicates that the non-voters were from demographic groups that were under-represented in the sample as a whole. (It’s only slightly less difficult to get non-voters to answer a survey than it is to get them into a polling booth, as we see from the fact that just 15% of the sample did not vote in an election with 66% turnout.) But because 290 is a small sample and weighting tends to magnify the effect of sampling error, I’ve used unweighted counts throughout this post (not that weighting made an appreciable difference to any of the patterns I will talk about below). The following alluvial diagram (created using the R package, ggalluvial) tracks the voting behaviour of sampled 2015 non-voters post-2015:
That diagram is a pretty clear illustration of the truism that non-voters don’t vote. Most members of the sample who didn’t vote in the 2015 general election also did not vote in the 2017 general election. And while slightly more of them voted in the 2016 EU membership referendum, most of them didn’t vote in that either. This is despite the fact that some of those who didn’t vote in 2015 will likely have been habitual voters who – for one reason or another – did not happen to vote in that particular election, yet subsequently returned to past form.
Here’s an interesting point, though. The majority of those GE2015 non-voters in the sample who did vote in the 2016 referendum voted Leave, and the majority of those who voted Leave did not vote in GE2017. But the Remain-voting minority bucked the overall trend. Of the mere 48 GE2015 non-voters in the sample who voted Remain in the 2016 referendum, 26 went on to vote in the next general election, while of the 179 who did not vote in the referendum, just 25 went on to do the same.
What are we to make of this?
A few thoughts:
- The sample of GE2015 non-voters appears to consist, to a fairly great extent, of people who are generally disinclined to vote (though not all non-voters in any particular election or referendum will belong to that group, and the same is probably true of this specific sample of non-voters from that specific election)
- On the evidence of their subsequent voting history, these people seem to have been slightly less disinclined to vote in the 2016 EU membership referendum than in the 2017 general election. Perhaps that’s because referenda generally seem more meaningful to them than elections do, or perhaps it’s because of something specific about that particular referendum
- It seems possible that voting Remain in the 2016 EU membership referendum somehow led previously habitual non-voters into higher levels of political involvement from that point onward
- However, it may simply be that those habitual voters who just happened not to vote in 2015 (and therefore became part of the analysis presented here despite not being habitual non-voters) were over-represented within the minority of GE2015 non-voters who voted Remain the following year (whether in the wider population or only in this sample of it). If this is the case, then we may not be looking at a ‘Remain effect’ among previously habitual non-voters, but at a tendency towards Europhilia on the part of some habitual voters who temporarily got mixed up with them
- Because we don’t know how any of these people voted in 2010, their voting histories as recorded in these data would seem to offer little prospect of helping us to decide between the interpretations in points 3 and 4, even if we could be confident that the different proportions were not the result of sampling error (which we can’t, because the absolute numbers are so small)
Now for the statistic that many of you will have been waiting for. Of the 24% of GE2015 non-voters in the sample who did vote in the 2017 general election, a remarkable 61% voted Labour. But here’s the rub: that’s 61% of 24% of 15%, which means that we’re only talking about 41 actual people out of a sample of 1874. And this has to be seen in context of an overall picture that reveals losses to non-voting as well as gains from it. In 2017, Labour picked up 41 members of the sample who hadn’t voted in 2015, but 55 of its GE2015 voters from the same sample went the other way and didn’t vote. The difference between those two figures is less than 1% of the sample. In the bigger scheme of things, it just doesn’t matter.
From the point of view of political sociology, it’s vitally important to understand non-voters. And if I were going to pick any single ‘voter’ group for my research moving forward, I think that I could do much worse than to focus on the voluntarily self-disenfranchising.
But habitual non-voters are not going to have a decisive impact on the next general election.
Too few of them vote.
Ten voter groups: combinations of EU referendum and general election votes in the BES 2017 face-to-face survey
Like many, I read with interest Stephen Bush’s recent article on ‘The nine voter groups who are more important than Labour Leavers’. If Bush were a grant awarding institution, there would be money available for researching those groups. Well, he isn’t, so there isn’t, but I like a challenge so I’m going to make a start anyway – using open data from the British Election Study (henceforth, BES). To be more specific, I’ll be using the BES 2017 face-to-face survey, which was conducted after the election and uses what should probably be considered a more genuinely random sample than the online waves.
If we focus only on England (because the other parts of the UK have really quite different political systems), this gives us 1874 respondents, or 1839 after weighting for demographic group. Even after weighting, Labour voters are over-represented, but we’re not trying to predict last year’s election – we’re trying to understand why people made the voting choices that they did, and to use that information to derive hints about what voting choices they might make in future.
Bush’s general approach was to identify groups by how they have voted in recent years. Groups identified in this way are more or less important, if I understand him correctly, according to how much they might sway the result of future elections. And a group’s ability to do that would presumably depend upon (a) its size and (b) its likelihood of flipping from one party to another, given some sort of predictable event such as a political party coming out for some particular policy.
What sort of a policy might that be? Respondents to the survey gave a pretty clear hint of that in their responses to the open question, ‘As far as you’re concerned, what is the single most important issue facing the country at the present time?’ Although Jeremy Corbyn and Theresa May largely succeeded in avoiding discussing Brexit policy during the 2017 election campaign, Brexit was clearly the most popular answer. And well it might have been, I might add, because all the other issues in the top five clearly depend on what kind of a Brexit the UK actually gets. (Note for the nerds: what I’m actually counting here is the number of respondents who gave each of these words as a one-word answer, plus the number of respondents who began their answers with one of these words, or a morphologically related word, e.g. ‘immigrants’ was counted under ‘immigration’. Also, throughout this article I’m using the BES team’s wt_demog weighting.)
The overwhelming importance given to Brexit is analytically convenient, because one of the last three important UK-wide votes was the 2016 referendum on continued membership of the EU, so if we use that referendum to define our groups (as Bush did in some cases), then the definition of the groups themselves will give us an indication of how they might feel about the policy area that their members seem to regard as most important.
So now to those groups.
There are some that I’d like to come to at a later date (in particular, ‘Conservative voters in 2017’ and ‘Labour 2017, Gives An Answer Other Than “Jeremy Corbyn” When Asked Who Would Make The Best Prime Minister’). But right now I’m intrigued by the possibility Bush raises of dividing up the electorate by combinations of 2015 General Election, 2016 EU referendum, and 2017 General Election votes
Bush lists four of these: ‘Conservative 2015, Remain 2016, Labour 2017’, ‘Conservative 2015, Remain 2016, Conservatives 2017’, ‘Non-voting until 2016, Remain 2016, Labour 2017’, and ‘Non-voting until 2016, Leave 2016, Non-voting 2017’ (the latter of which he lumped together with an implied fifth group, ‘UKIP 2015, Leave 2016, Non-voting 2017’). But even if we limit ourselves to the four main parties (Conservative, Labour, Liberal Democrat, and UKIP) plus ‘other’ as a catch-all plus non-voting in the two general elections, plus the three possible responses to the 2016 referendum (Leave, Remain, and non-voting), there are an awful lot of potential groups: 108 of them, to be precise. Or 126 if we distinguish those who didn’t vote in 2015 from those who couldn’t vote, e.g. because they were too young. (I made this distinction in my analysis but it didn’t seem to make any difference to the overall picture.) Most of these groups are going to be too small as to be important. If voters are evenly distributed between all 108 or 126 groups, then less than 1% will fall into each, and if they are not, then less than 1% will fall into each of most of them.
With R’s dplyr package, it’s quite easy to divide up survey respondents according to combinations of votes (using group_by), and then calculate each group’s average responses to the question ‘How likely is it that you would ever vote for each of the following parties?’ (on a scale from 0 to 10) to give a sense of their probability of flipping in the next election (using summarise). Here are the ten largest:
As the table shows, the largest groups are of people whose voting behaviour was the same in 2017 and 2015, regardless of what they did or didn’t do in 2016. That’s not particularly surprising, because the British electorate isn’t known for shopping around: by and large, Conservative voters vote Conservative, Labour voters vote Labour, and non-voters don’t vote.
But the table contains a very clear hint that Bush might be right in his main thesis that ‘Labour Leavers – that is, voters who backed Labour in 2015 and Leave in the referendum of 2016 – have received an outsized share of attention and analysis’. As we see from the above, the ‘Labour 2015, Remain 2016, Labour 2017’ group is much, much bigger than the ‘Labour 2015, Leave 2016, Labour 2017’ group, and – given the probability scores – the former looks much more amenable to being poached by the (anti-Brexit) Liberal Democrats than the latter does by the (pro-Brexit) UKIP and Conservative Party. Meanwhile, there was no other combination beginning ‘Labour 2015, Leave 2016’ that was large enough to make the top 10, indicating that there has been no major exodus of Leave voters from Labour to elsewhere.
When we remember the recent survey finding that ‘fewer than one third (32%) [of Labour Leavers] think [that leaving the EU is] very important’ while ‘over half (51%) [of Labour Remainers] say [that staying in the EU] is very important’, there is therefore clearly now a substantial body of evidence to indicate the greater potential for Labour Remainers than Labour Leavers to influence the result of a future general election – provided that somebody makes a bid for their support. Whoever it is, it probably won’t be the Labour leader, a lifelong Eurosceptic who called for the invocation of Article 50 before even Nigel Farage. But as I said last June, Paris might be worth a mass.
Looking further down the list, ‘Conservative 2015, Remain 2016, Labour 2017’ looks very much like a floating vote, giving very similar probability scores for all parties except UKIP, which it would seem to regard as beyond the pale. In fact, the only top 10 group giving UKIP a score higher than 3.1 was ‘UKIP 2015, Leave 2016, Conservative 2017’: the Leave-voting UKIP-Conservative switchers who decimated Paul Nuttall’s already shaky credibility last June. Those switchers might switch again. But, based on the probability scores, they don’t seem to like Labour or the Liberal Democrats very much, and right now the chances of UKIP’s winning back anybody’s vote with its current leadership and financial difficulties are looking pretty remote. And even the 3.1 came from ‘None 2015, Leave 2016, None 2017’: a small group that is, like the much larger ‘None 2015, None 2016, None 2017’, unlikely to vote for anyone in the near future (not only on the evidence of its avowedly low likelihood of voting for any of the four main parties but on the evidence of, ahem, past behaviour). Those two groups can, I think, fairly safely be written off as unimportant, at least in the sense defined above. (I know that sounds elitist, but if people don’t want to vote for anything that’s available, then they can’t influence elections.) The Conservatives might be able to attract the ‘UKIP 2015, Leave 2016, UKIP 2017’ group next time around, but it was too small to make the top ten above and therefore its departure from the UKIP fold probably won’t have much of an impact on anything now that the bulk of 2015 UKIP voters has already left for pastures new.
So it looks as if UKIP voters have had their (admittedly gigantic) effect and are now a spent force, and – all in all – the groups to watch from now on are the very large group of Labour Remainers – only a small proportion of which would need to peel off and vote for another party to make an impact at the polls – and the much smaller group of Remain-voting Conservative-Labour switchers, whose votes have migrated once and (even if we did not have the evidence of the probability scores above) might therefore be assumed to be more than averagely predisposed to migrate again. Labour Leavers are much smaller in number than one of the aforementioned, and – it seems – more likely than both to stay put.
Now to the inevitable caveat. While the approximate relative sizes of these groups are probably a robust finding (because the poll itself was so large), the average probability scores for a group become less and less reliable as the size of the group falls: for example, the averages in the final row in the table above are the findings of what was in effect a poll of 29 people, so the margin of error will be huge, and I mention the scores above only because they are more-or-less what we might expect given that particular group’s voting history.
We could find greater numbers of people falling within each group from the BES online panel, but that has representativeness problems of its own, and what we’re coming to here is an essential problem of the approach of splitting survey respondents up into discrete groups of successively smaller size. I’ve done some more analysis that takes a slightly different approach in order to mitigate that problem, but I’ll save it for a future post.