Like all good things in life, this piece is too long for email. If you’re viewing this in Gmail, click on the title to open it in your browser.
“Democracy is the theory that the common people know what they want and deserve to get it good and hard.”
-H. L. Mencken
For decade, with reams of scientific studies and hours upon hours of polls, political scientists have disagreed on what partisanship really means. Do people say they’re Democrats because they’re liberal and they like the Democrats’ liberal policies? Or are they Democrats first and support whatever policies their party puts forward?
The canonical understanding, the one you’d probably find in a civics textbook, suggests that voters make choices between political parties based on their positions on different issues, or based on how far they are from the party on a left-right scale. Ultimately, this idea - like other products of the Enlightenment - holds that voters are rational evaluators, weighing all the options and choosing whom to support.
The other school of thought challenges these ideas of a rational, omniscient voter. Partisan identities, they say, are more like emotional attachments than rational choices. “[A] partisan behaves more like a sports fan”, as one paper put it, “than a banker choosing an investment”1. Partisanship, in their telling, is an enduring identity - strengthened by social identities like gender, race and sexuality - which creates emotional attachments to political parties.
Whichever of these ideas appeals to you, it’s easy to see the thing itself: partisanship is extremely sticky, and people vote consistently with a particular party, even in multi-party democracies. It’s pretty difficult to get them to vote another way from their party, let alone form strong issue preferences that aren’t their party’s mainstream position.
And yet, that happened right in front of our eyes: about seven years ago this month, the Brexit referendum had opened up a huge chasm in British politics. This was a position on a policy issue - whether or not to stay in the EU - that came to not just dominate British politics for years, but to alter the balance of how people identified with political parties.
How did this happen? How do people come to care so much about an issue that it overrides their partisan identity?
The answer, as we’ll see, is in the question itself.
The first thing we need to understand this is a quick primer on Brexit. So, let’s talk about Brexit. (SIR KEIR faints, is carried off stage right.)
When the campaigning for the EU referendum started in 2016, both parties were roughly split on this issue. David Cameron and the Conservative government (“Tories”) officially backed Remain. The MPs of the party were initially not allowed to participate in the campaign, but lobbied hard to be allowed to campaign for any side they wanted. The morning after the referendum, Cameron resigned, triggering an almost overnight race in the party towards euroscepticism. The elections to replace Cameron had all the candidates promising to follow the results of the referendum; when she won, Theresa May created a department for Brexit and started laying out her lines for negotiation.
Almost overnight - based on a vote that had gone 52 to 48 - the biggest party in the UK had taken a hard turn from being broadly split but supporting Remain to a total commitment to leaving the EU.
Meanwhile, the Labour party, torn by its rapidly diverging coalition of educated professional Remainers and working-class Leavers2, tried to maintain strategic neutrality to avoid alienating their base, and started bleeding both Leave and Remain voters. As late as 2019, 71% of voters described Labour's policy on Brexit as "unclear and confusing"3.
The second thing we need to understand this is a deeper understanding of the debate about partisanship. So: what do the people care about? Do they care about policy positions? Or do they just care about their party like fans of the Chennai Super Kings?
In the 2000 US presidential election, George W. Bush supported privatization of Social Security, and made it an extremely important issue in the campaign. In the last few weeks of the campaign, both candidates ran lots of ads in battleground states about this issue, and by the end votes for Bush aligned neatly with support for privatisation. But as the researcher Gabriel Lenz later showed, this relationship wasn’t about people deciding whom to vote for based on their position on privatising Social Security. It was the other way around: people had already decided to vote for Bush, and they were just learning about his position and adopting it as their own4.
Over time, a ton of evidence5 has emerged to support the understanding that partisanship is a lot more about identity than it is about policy. People twist their understanding of reality to support their partisan ideas; they take up their party's positions as their own; they view their party as ideologically closer to themselves than it really is6; they inherit their partisanship from their parents, and it shapes a part of who they are and how they see themselves. We seem to be strongly wired to form these identities and senses of “us” vs “them” - in a landmark 1970 study, even boys who were randomly assigned to groups started favouring group members over others, and thinking they were better than those not in their group7.
Almost all politics is identity politics, whether we like it or not.
But if this is all true, if people have almost tribal attachments to their political parties and decide their positions based on partisan cues, how do positions on whether to leave the EU come to tear British political parties at the seams?
In their book Democracy for Realists, political scientists Christopher Achen and Larry Bartels talk about the Southern realignment, when the Civil Rights Movement triggered an exodus of White Democrats in the Southern US towards the Republican party in the 1960s. These White Democrats opposed policies like racial integration in schools, and they believed enough in this that they switched parties.
Or did they? Achen and Bartels find that the way they voted was way out of date with the way they identified - Republicans didn’t outnumber Democrats in the South until the 1990s, when the Democrats had stopped being competitive there for decades. Partisanship is a funny thing: voting for Republicans at every level can still be very different from identifying as a Republican. Identities die hard.
What’s more: in this entire shift, actual policy opinions mattered for very little. Even among those who opposed racial integration of schools - the main group who moved to the Republicans - a majority identified as Democrats well into the 1980s. And even White Southerners who favoured integration were leaving the Democrats.
As it turned out, they were White Southerners more than they were Democrats. But that took them thirty years to accept.
In 2021, Bryan Schonfeld and Sam Winter-Levy at Princeton examined partisan identities and Brexit8. They analysed polling data from the British Election Studies (BES)9, immediately pre- and post-referendum, to ask whether voters changed parties based on their Brexit policy opinions.
After a careful analysis, they argued that a relationship exists: “We find substantial evidence that voters are policy motivated:”, they wrote, “Euroskeptic voters flocked to the newly Euroskeptic Conservatives, while formerly Conservative Europhiles defected from their original party.”
“[…] On at least some issues—European integration, in this case—most voters do not simply follow their psychological attachments to their political parties. Instead, on both sides of the debate over Britain’s role in the EU, voters’ policy preferences shaped their partisan loyalties, in ways more or less consistent with conventional normative accounts of democratic accountability.”
On big enough issues, they argued, policy positions do come into the picture. On issues voters feel strongly enough about, they are willing to override their partisan identities and the cues they provide.
Are they?
The BES involves a module of questions designed to measure a strength of social identification of voters. For instance, see this one about party identification:
Earlier you said that you tend to identify as [<party>]. Thinking about this party, how much do you agree with these statements?
When I speak about this party, I usually say “we” instead of “they”.
I am interested in what other people think about this party.
When people criticize this party, it feels like a personal insult.
I have a lot in common with other supporters of this party.
Strongly Disagree / Disagree / Agree / Strongly Agree / Don’t know
It’s designed, implicitly, to tease out of voters just how much they identify with members of their party as a social group: how much it forms not just a label but a coherent social identity. In 2016 for example, right after the referendum, this was the split of answers of how many people said “we” when talking about their side - a simple marker of group identification - about their political parties:
In the same survey, the BES also asked voters the same question about their Brexit identities - Leave or Remain. Here’s the split of those answers10:
See a difference? Look at the number of people who “Agree” or “Strongly Agree” with the statement combined: for their party, this number is just 21 per cent. For their side on Brexit it is 51 per cent!
And so, the missing piece of this puzzle might just be the one word we keep coming back to: Identity.
Over the course of the referendum, and in the following years, Leave and Remain - two choices on a ballot to start with - morphed into something else altogether. They became social identities of how people saw themselves in a changing UK, identities consistently stronger than the partisan identities they started out with11. Schonfeld and Winter-Levy remark on this idea in a footnote - citing a paper from the LSE which argues12 that Leave and Remain are social identities - but they don't actually address it. Do these identities affect the way people identify with parties?
So, let’s start with our basic group: People who believed the UK should leave the EU, immediately after the referendum in 2016. There’s about 13.6k of them in our data. They’re fairly split between parties, but a plurality identify as Tories to begin with.
They’re also, as you’d expect, pretty strongly Eurosceptic. The BES asks respondents to rate their own euroscepticism on a scale of 1-10: “Some people feel that Britain should do all it can to unite fully with the European Union. Other people feel that Britain should do all it can to protect its independence from the European Union. Where would you place yourself on this scale?” - where 1 is “Unite fully with the EU” and 10 is “Protect our Independence” (9999 is Don’t know). These are the ratings for our group. Remember, the members get more and more Eurosceptic as you go rightward in the graph.
Now let’s track this group through the year after the referendum, the election year of 2017. We can see a shift on two levels.
As a group, they started liking the Tories more.
The BES asks respondents to rate their “liking” of different entities - all the major parties, the Prime Minister, etc - on a scale of zero to ten, zero being “strongly dislike” and ten being “strongly like”. On average, people in our group started liking the Tories about one point more over the year, on a scale of 10.
Look at the distribution of these ratings for the Conservative party in our group for those who answered both surveys (and didn’t respond “Don’t Know”): post-referendum on top vs post-2017 on the bottom. From left to right, you’re going from loathing Tories to loving them. Observe both graphs carefully, and a small but definite shift is apparent, all across the scale from those who hated the Tories to those who loved them.More of them identified as Tories in 2017 than in 2016.
In fact, this is a trend that continues on to 2019: More and more people in our group of Leavers identified as Tories with time13.
(Both of these shifts mostly carry forward even if you take the 3-year time period of 2016-2019 instead of the one year - so they’re likely to be a trend, and not one spurious increase.)
In response to the Tories’ turn towards euroscepticism, evidently, Leavers as a group were also shifting towards the Tories. This much should perhaps be a bit obvious, since the Tories won the elections in both 2017 and 2019 (“Get Brexit Done”).
The main challenge is identifying what drove these shifts. And doing that means untangling the knot between views and identity. As we saw, most Leavers were strong Eurosceptics to begin with - of course they were! So how do we decide which of these two things were actually driving them towards the Tories? Was their Euroscepticism attracting them towards the Tories who were taking increasingly Eurosceptic positions, or was their Leave identity attracting them to a party that increasingly aligned with Leavers and was increasingly led by Leavers?
For this we turn to our trusty, simple old friend, multiple regression. Think of it this way. A simple regression tries to account for a variation in Y using variation in X - say variation in increased liking for Tories using variation in Euroscepticism - and assumes the rest to be noise. When we use a multiple regression and bring in another variable - say the Leave group identity - we’re bringing it out of the random noise and accounting for it explicitly. So, when we analyse the liking for Tories against Euroscepticism and Brexit identity, our model accounts for the influence of Euroscepticism without varying Brexit identity. If we see that the coefficient for that - the unit change in Tory liking per unit change in Euroscepticism - is still significantly different from zero, we can be pretty sure it has some predictive effect apart from its intersection with Brexit identity, and vice versa. It’s not a causal analysis, but it’ll do for what we’re trying to do here.
Imagine that our data looks like this (as a made-up example):
Pooling this over lots of people from our group, we can look at the association between the increased liking of Tories as a target and the other two variables as predictors, and see where there’s a significant effect.
What does that analysis tell us?
For the increased liking for Tories, both of these variables are significant predictors (p<< 0.001). But for the increased Tory identification, the strength of their Brexit identity is a strong predictor (p=0.01), but their Euroscepticism in 2016 isn’t (p=0.26)14.
This is powerful to visualise. If you got a hundred Leavers in a room in July 2016, and you asked them how much they agreed with the statement “When I talk about other Leavers, I say ‘us’ instead of ‘them’”, their answer, and the strength of social identity it showed you, would have a small but definite part in predicting whether they would come to identify as Tories a year later. More than their answer to how much they thought the UK should stay away from Europe - their actual position on the referendum question - would.
Over the year after the referendum, Leavers started to identify as Tory partisans. But they weren’t doing this based on policy opinions. They were doing it based on their social identity as Leavers, because they were different from all those Remainers over there.
Come on, I hear you asking. Does that really matter? Does it matter that people were making this choice based on identities they had formed that sort of aligned with their position on this issue, and not on the position itself? Isn’t the end result the same?
The truth is, it matters - because identities make people do funny things. Because when people identify as Leavers first, they start perceiving immigration as decreasing after the referendum, even if the actual numbers haven’t changed15. When people don’t just believe the UK is better staying in the EU but identify as Remainers, they perceive the economy doing badly immediately after the referendum, even if the indicators haven’t changed - and Brexit hasn’t happened yet.
Our views and identities are often murkily entwined in each other, but very often - much more than you’d think - the push goes the other way. Group identities are complex, weird, funny things, and we’re wired to form them. Far too often, the push and pull in politics is not around principles and policy, but around tribes and vibes.
Look at the politics around you. You’ll sometimes see people attacking or defending strange things, and believing strange things about their out-groups - sometimes even factually wrong things16. More often that not, they’re being guided not by well-thought policy principles, but by their group pulling them a certain way.
Partisanship is a very dominant identity. And that also means that when people move away from their party, it’s very unlikely they’re setting off into the wilderness guided by their opinions; it’s far more likely that they’re following a closer-knit group as it moves away.
Something funny shows up if you look at our group of Leave voters in 2019, for those who still identified as Labour supporters. They had been defecting. Only 60% of them actually voted Labour in 2017. In 2019, a majority had voted for parties other than Labour - hell, more of them had voted for the Tories than for Labour - 36 to 35%.
And yet in 2019, three years and two general elections after the referendum itself, after just having mostly voted for non-Labour parties in the recent election, 61% of these Leavers - a solid majority - thought of themselves as Very Strong or Fairly Strong Labour supporters.
Identities die hard.
https://www.democratic-erosion.com/wp-content/uploads/2018/03/Mason-2015.pdf
https://www.nytimes.com/2019/03/11/world/europe/jeremy-corbyn-labour-brexit.html
This reference, and a part of the explanation of Brexit, comes from the Princeton paper linked below.
https://ukandeu.ac.uk/wp-content/uploads/2019/01/Public-Opinion-2019-report.pdf
UK in a Changing EU has done so much interesting work about Brexit, and some of their data-based work prompted the direction of this analysis Anand Menon once wrote an absolute banger of a Tweet that made me chuckle for hours.
https://press.uchicago.edu/ucp/books/book/chicago/F/bo11644533.html
~Page 60 for the Social Security research [Lockbox: Social Security in the 2000 US presidential election].
https://bracket.substack.com/p/partisan-identities-are-more-complex
Yes, I did just add my own article as a footnote. I’m now going to get full of myself and stop picking up all your calls. Anyway, that article has a long, detailed explanation of this debate and why I think there is overwhelming support for one side.
https://press.princeton.edu/books/hardcover/9780691169446/democracy-for-realists
I’m going to avoid writing a paean to this book again, but I think it is one of the definitive books to read that will change your view of how the world works.
https://www.jstor.org/stable/24927662
https://static1.squarespace.com/static/57b3ced6b8a79b78f934ff9f/t/610c4b9acab3a41d243c62d6/1628195739659/JOP+Brexit.pdf
This paper is very well done. I’ve often talked about studies that fuck up, where proper researchers try to extrapolate too much or not consider why they could be wrong. This is what doing a lot of those things well looks like: the researchers try to examine the issue, try to come up with potential pitfalls and objections for themselves and verify whether they’re true. Regardless of whether you agree with what they’re saying - and personally as I argue here I think the truth is more about identity than policy - this is to their great credit.
https://www.britishelectionstudy.com/data/
The BES is a huge survey in British politics, the way something like the ANES is in the US.
This might seem like a big thing to premise on people’s answers to one question about we vs they, but that’s only for simplicity. The findings presented here hold up even if you create a “Brexit identity score” adding up the answers to all the questions in this module. I just didn’t want to complicate the narrative further by adding four more variables.
https://whatukthinks.org/eu/wp-content/uploads/2018/10/WUKT-EU-Briefing-Paper-15-Oct-18-Emotional-legacy-paper-final.pdf
https://eprints.lse.ac.uk/103485/1/Divided_by_the_vote.pdf
This is where the sample needs to be weighted to match population parameters, and obviously we start out with a large group and it narrows down in successive years as only a few of them respond in those waves. I’ve read conflicting ideas about which weights should be used to treat this as nationally representative - should we use the original weights from 2016 throughout because that’s the set of people we’re originally starting with and everyone else is a subset? Or should we use the weight from each wave? I think the first is the right way to do this because we’re certainly starting out with one group and following it through, but in either case this particular finding shows up any way you weight the respondents.
This is a result that depends on whether you weight the data or not. In an unweighted sample this result is actually the other way around. but I think this is why applying panel weights so the sample is representative is important.
https://www.cambridge.org/core/journals/political-science-research-and-methods/article/abs/tale-of-two-peoples-motivated-reasoning-in-the-aftermath-of-the-brexit-vote/0029F77B845C75BC45DFAB98620E5691
The authors have also written an article about their findings that might be easier to understand.
https://grist.org/wp-content/uploads/2010/04/thinking.pdf