Blog/Analysis

What’s the effect of boundary changes?

The new boundaries on which the next election will be fought were released in stages over the last couple of months. These changes have been made to constituencies in England, Wales and Scotland (NI remains unchanged) to equalize the number of voters in each seat, accounting for population movements and change over the last decade. Since constituencies convert votes to seats, the composition of new seats can have a significant effect on the outcome of a general election.

The question is, if the 2019 election had been held on the new boundaries, what would have been the result? Working out the effect of the new boundaries is not straightforward. The new seat boundaries overlap with the old in a sort of huge, messy national Venn diagram. And while we know how people voted in each old constituency, what we don’t know is how people voted in the precise areas of overlap between old and new seats. In other words, when looking at the parts of old constituencies that make up a new one, we don’t know exactly how people voted in each of them.

The solution? This first step was to use demographic data. We started with demographic metrics for each seat: health, wealth, education, ethnicity and job occupation, among others. We used these metrics to model the link between demographics and vote share for each party, in 2019. We then applied these models to each part of the UK where an old and new constituency overlap - there were 1,412 of these areas - in order to estimate how people voted in each of them. At this point, we also adjusted each estimate to take account of the error in the original model predictions for 2019. Finally, we combined the estimates for the overlap areas to obtain figures for the result in each new constituency. Political scientist Chris Hanretty conducted a similar exercise for his EU Referendum Leave vote share estimates across all UK constituencies.
‍
For those of you that are interested in the results in full, they’re available here (if you subscribe!). We’re likely to update these estimates as they are a work in progress - and would love feedback on them.

The new boundaries have a marginal effect on current seats - mainly in Wales and South East England

So what’s the answer? Broadly, if the 2019 election had been run on the new boundaries, the effect is net positive for the Conservatives (up 6) at 371 seats. Other parties barely change their seat totals; Labour down 3 at 199, Liberals down 1 at 10 and Plaid interestingly down 2, halving the number of their seats.

The where and the why of the changes is revealed when we look at the gains and losses by region. Essentially, the main change across the UK is the addition of 7 seats to South East England, while 8 seats have been removed from Wales - proportionately 20% of its seat count. Other regions have gained or lost far fewer seats. These changes are based on population movements. So what’s happened is that proportionally more people are now living in the South East, in broadly Conservative - but Remain leaning - areas. The new boundaries are going to give these voices more political weight and representation. This has had the effect of removing Labour and Plaid seats, and adding Conservative ones.

This all might seem much ado about nothing. A six seat swing would make scarcely any difference to the makeup of the current parliament - especially with the polls suggesting a double digit swing to Labour, and various MRP models suggesting a catastrophic fall in seat count for the Conservatives. What’s the relevance, you might think, of these handful of gains and losses? It’s worth remembering, though, just how much Theresa May would have liked six more seats in the close-to-hung parliament after the 2017 election.

And that said - how would the new boundaries have affected past elections? We re-fitted our model to the various party vote shares in 2010, 2015, and 2017, and re-predicted vote share on the new boundaries for each. Of course this historical exercise is more for fun, given that the new seats reflect population change across the decade, but it underlines the point that the boundaries consistently give the Conservatives a small but noticeable bump.

A few other Easter eggs can be found in the historic modeling - in 2015 the boundaries might have given the Greens an extra seat, picking up the new constituency Bristol Central. Meanwhile, “Cameroon” Conservative type constituencies - think Witney - will have much stronger representation in the next parliament. Hence why the 2015 election would have been so drastically different on new boundaries (345 Conservative seats!), and the collapse of the Liberal Democrats could have been even more acute than it was.

Beneath the surface things look much better for the Conservative Party…

The real story, though, is not in the number of seats, but in distribution of votes within those seats. When we compare the spread of each party’s vote share across the seats where it stands, we can see that the new boundaries have a pretty significant effect, one that may have notable consequences for the result of the next general election if the polls become narrower than they are currently.

The spreads shown below give an idea of how the new seats affect vote shares differentially by party. The smaller parties – the Lib Dems, Greens, Brexit, Plaid – get very few votes in most seats, and are only competitive in a few, thanks to tactical voting. Labour and the Conservatives, by contrast, are competitive (vote share >30%) in many more seats, while the SNP waste very few votes and are competitive in almost every seat they stand.

Comparing the coloured bars (new boundaries) against the black outline (old boundaries), you can see some changes. Plaid are losing out in a number of seats in which they are competitive, reflecting the fact that the new Welsh boundaries cross historic Welsh language borders in a way that’s deeply unhelpful to Plaid. Meanwhile the number of seats where Greens and Brexit are not at all competitive (vote share <10%) will increase. The Lib Dems can be happier - they will become competitive in slightly more seats than before.

Where this will really tell, though, is when we consider the potential effect of a swing in the vote. What we’ve looked at so far is running the 2019 election again, on the new boundaries. But very few people expect voting patterns to remain the same. In the event (highly likely) of a swing in the national vote to Labour, how many seats Labour can expect to actually gain depends on the size of other parties’ majorities in seats they hold. Given Labour’s electoral map, what this means is the size of Conservative majorities in Labour-Conservative marginals. And the majorities in those very marginals stand to change on the new boundaries.

When we’re considering who will form the next government, this is the graph that really counts. The number of Conservative marginal seats, where they have a majority less than 5% of total votes, will materially decrease from 43 to 29. When the Conservatives are playing a defensive game, the new boundaries give them a significant advantage, compared to the old ones.

What’s the net effect? The new boundaries means that the swing (average change in both Labour vote share up, and Con down) that Labour need to be the largest party (~275 seats) moves from 6% to 8%. The swing Labour would need to have an overall majority (326 seats) moves from 10% to 12.5%. That doesn't matter today where the swing is closer to 15% - but if the polls tighten it could make a material difference. This calculation changes significantly if Labour can gain seats in Scotland from the SNP based on mass-switching in the central Scottish belt, unpeeling some of the initial Labour collapse of 2015. Current polls in Scotland, though, reveal no consensus on whether this is likely to happen.

The new seats are very different

One final note on the new boundaries is that the composition of the seats is split: around half of seats are almost identical to legacy seats while the other half are significantly reconstituted. We estimate that around 135 seats are left completely unchanged. Around 307 seats (out of 632) will have at least 9 out of 10 of their electors coming from a single pre-boundary change constituency. What’s fascinating though are the 70 or so new seats which are made up of 60% or less of a single old legacy seat. Some of these new seats have been given quite grandiose and cumbersome names by the various boundary commissions; St Neots and Mid Cambridgeshire comes to mind. A big unknown factor in all of this will be the loss of an incumbency effect in some of these heavily restructured seats - potentially exacerbated by MP retirements.

Boundary changes or not, two years out there’s absolutely no way of being certain where we will end up in 2024 – a Conservative majority, a hung parliament, or a Labour landslide. But, returning to one of the UK’s Golden Rules of politics: leader ratings matter (particularly when measured by Ipsos Mori’s historic dataset). The analyst Matt Singh first made this pattern very clear in the 2015 election. Given leader ratings, you can predict the governing party’s election vote share pretty accurately for all UK elections since 1983. Today - like in 2015 - there is a massive difference between where the Conservatives are polling, and the polling that would be implied by the Prime Minister’s leader ratings. PM Sunak is much more popular than Conservative vote share would suggest. This could go either way between now and the election, of course: Sunak could get less popular, or the Conservative vote share could increase. What’s unprecedented is the gap between the national government lead implied by PM Sunak’s ratings (c.1-2% above Labour) and where they currently are (between 15-25 points behind depending on the pollster!). It’s going to be interesting to see whether these two metrics converge, or whether a differential will persist until election day - even if it’s smaller in size.

Mixed-Ethnicity Partnerships vary wildly by area

I’ve been taking a look at one variable in particular in the new 2021 census data: the percentage of households with mixed-ethnicity partnerships. That’s partly a personal interest, but it’s also not one I’ve seen analysed too much before. I’m interested in it because it seems to me as good a proxy as any for cultural integration and assimilation. Looking at the number of mixed-ethnicity partnerships I was struck by how loose the correlation was to overall levels of diversity in an area. In other words, among areas of Britain that are relatively diverse and multicultural, some are more integrated, others less so.

Looking at the top 20 and bottom 20 seats for mixed ethnicity partnerships a couple of things struck me immediately. The bottom 20 are, without exception, deprived, heavily Leave, whiter than average, concentrated in South Wales and North East England, and until 2019 all Labour, but trending Conservative. The top 20 seats for mixed ethnicity partnerships are by contrast all in London, all heavily Remain, and a mixture of Conservative, Liberal and Labour marginals with a fair amount of political volatility. In some ways the top and bottom tables provided contrasting snapshots of the realignment in UK politics in the past 10 years.

What’s driving this discrepancy? What factors are linked to mixed-ethnicity partnerships? To find out, I modeled the proportion of households with mixed-ethnicity partnerships, by constituency, using a range of constituency-level demographic information (not yet all available at seat level from the latest census). The proportion of households with only one person was controlled for.

The proportion of non-white individuals is positively associated with mixed-ethnicity partnerships - unsurprisingly, given that in order to have mixed-ethnicity partners, you need a supply of individuals of different ethnicities. The proportion of non-white individuals, however, doesn’t explain all the variance. Richer constituencies (in terms of median income and/or house prices) have far more mixed partnerships, while constituencies with more home ownership have less. Professional and technical occupations are also linked to more mixed partnerships, but this effect is caused by degree-level qualification, which is the strongest predictor of all (the collinearity between these two predictors is responsible for the wider uncertainty intervals for these two coefficients).

Taken together, these factors are very predictive, giving an R-squared of 0.86. Initial analysis suggests that whether you’re in a mixed partnership is strongly determined by levels of education, income, home ownership as well as the number of non-white (and therefore addressable pool of) people in your area. Population density (which we covered in a previous newsletter) was also included as a predictor, but - a little surprisingly - once these other factors are controlled for it has almost no further effect.

Given that the most diverse constituencies are not necessarily the most well integrated or assimilated, it’s worth looking at how these two dimensions interact. We can plot constituencies on these two axes, to make a diversity/assimilation ‘map’ of England and Wales.

This is the map: most constituencies are almost entirely white, and only a few are either >20% non white, or have >5% of households in a mixed partnership. The clump in the bottom left are the majority of seats which are mostly white, while the most diverse constituencies are towards the right, and those with high levels of mixed partnerships towards the top.

The pattern of cultural assimilation looks fairly politically aligned. Very diverse areas are almost all Labour. However, for any given rate of ethnic diversity it appears that areas with more assimilation are overwhelmingly less likely to be Labour (except in the most diverse constituencies). This isn’t really reflected in much commentary. Modeling shows that this is mostly because of income: for a given rate of ethnic diversity, more income leads to both more assimilation (as in the model above) as well as more Conservative votes.

This version of the ‘map’ uses log scales to spread out the clump of heavily white constituencies in the lower left above, to make the relationships clearer. The pattern becomes more obvious if you divide the proportion of mixed partnerships by the proportion of non-white to create a single metric of relative assimilation. The higher this score, the more people are in mixed partnerships in a constituency given the background level of diversity. You can see in the chart below that this measure is strongly linked to right-wing voting.

Meanwhile, the same map gives a different perspective on the link between diversity and Remain vote. While it is true that more diverse constituencies voted Remain, the more powerful link is between assimilation and Remain. In fact, once this is controlled for, the link between more diverse constituencies and Remain voting completely disappears. It is, in fact, the constituencies with the most assimilation that voted Remain. Leave voting is strongest in heavily white constituencies, but not only there. Leave was also strongly supported in some of the most diverse constituencies, that have relatively low mixed partnerships given their high levels of non-white people. The effect still remains even after you take account of income, qualification, and population density.