Regression Analysis 2017

This page first posted 24 June 2017

The shock result of the 2017 election came as a surprise to most pollsters and pundits, including Electoral Calculus. It was originally thought that the Conservatives would gain from the Brexit effect as Leave voters switched to them from other parties, whilst Remain voters were less inclined to switch party.

But a careful look at the election results reveals what was going on with the public to create the surprise result.

In this analysis, we have regressed the Conservative to Labour swing against various census demographics and political attitudes. This can be used to calculate the correlation between demography and voting patterns, which can give an indication of what sort of people switched to the Conservatives and what sort switched to Labour. It can also be used to determine the statistically most important factors.

The analysis was performed at a constituency level, focusing primarily on English seats only. The situation in Scotland was noticeable different and was driven by different factors.

The results of the analysis show that the top three factors influencing voting behaviour are:

IndicatorDescriptionCorrelation
with Swing
UK BornPercentage of voters in the seat
who were born in the UK
57%
Remain/LeaveResult of 2016 EU Referendum
in the seat
49%
NationalismAverage score of the seat on the
internationalist/nationalist axis
of our Two-D political attitudes
39%

Here "Swing" is defined as half of the change in the difference between Conservative and Labour vote share in the seat from 2015 to 2017. That is

Swing = ( CON(2017) − LAB(2017) − [ CON(2015) − LAB(2015) ] ) / 2.

In other words, the swing is the fraction of voters who change from one of the big parties to the other. In this election the average swing was 2.5pc to Labour, in that 2.5pc of voters nationally switched from the Conservatives to Labour. (The reality is a bit more complicated since both Conservative and Labour vote shares increased due to the decline in the smaller parties, but the measure is still a good indicator of the national and local picture.) But the situation in individual seats can be very different with swings ranging from 16pc to Labour in Bristol West all the way to a 9pc swing to the Conservatives in Ashfield.

Our aim is to explain some of the drivers of these differences in local swing. Taken together these three factors alone explain 56pc of the variance of seat-by-seat swing. This means that over half of the variation is due to just these three factors – Country of birth, EU Referendum vote, and Nationalist attitude.

A graphical demonstration of this correlation can be seen in the following image. This shows the swing in each seat plotted against the seat's Leave vote share in the EU referendum. The picture shows clearly that the Remain seats generally swung more to Labour than the national average, and that Leave seats were more favourable to the Conservatives. The straight light blue line indicates the best fit through the data, and is clearly upwards sloping. It passes through the centre of gravity of the cloud of seats which is located at a Leave share of 52pc and a swing of -2.5pc.

Swing 2015-2017 against Leave share

The next most important three factors are:

IndicatorDescriptionCorrelation
with Swing
ABC1 ClassPercentage of voters in the seat
who are in the "middle class"
social grades of A, B or C1
−50%
Average AgeAverage age in years of adults
in the seat
38%
Good HealthPercentage of voters in the seat who
have very good health
−35%

We can see an example of these graphically. The next figure shows a plot of swing in each seat against the percentage of people in the seat classed as "middle class" by the 2011 census. In this case the best fit line has a negative slope, running from the top-left to bottom-right. This shows that seats with more Working than Middle class voters were more favourable to the Conservatives. Conversely, Middle class seats were likely to swing more to Labour.

Swing 2015-2017 against Middle Class fraction

These data tells us about the sort of people who voted Conservative in 2017 as opposed to those who voted Labour, particularly about those who changed preference since the election in 2015. The next table summarises the types of people who switched to either of the big parties:

IndicatorConservativeLabour
UK BornMore UK bornLess UK born
Remain/LeaveVoted LeaveVoted Remain
NationalismNationalist
attitudes
Internationalist
attitudes
ABC1 ClassWorking ClassMiddle Class
Average AgeOlderYounger
Good HealthUnhealthyHealthy

This table does not mean that all the supporters of the two big parties fall into these neat categories. There were still many Middle Class people who voted Conservative and many Working Class people who voted Labour. But the swing voters who changed party were more likely than not to belong to the categories shown in the table.

Additionally there were some less important, but still significant, other indicators. These were:

IndicatorDescriptionConservativeLabourCorrelation
with Swing
EducationPercentage of voters with
A-levels or better
Poor EducationGood Education−60%
British IdentityPercentage of voters who
feel British rather than English
EnglishBritish−59%
Good JobPercentage of voters with a job
in the top 4 ONS categories
Less Good JobGood Job−48%

Possibly as important as the indicators with high correlation was one indicator which was much less correlated with voting than expected. The "Economic" political attitude of our Two-D political mapping measures classic economic attitudes on a left/right axis with high tax and spending on the left, and lower tax and spending on the right. This indicator had a very weak correlation with swing voters in either direction. This suggests that either voters were not focused on economic issues, or else they did not expect the election to make a difference to the economic outlook.

Summary

From this regression analysis, we see that the Conservatives have gained votes from the archetypical Leave voters: older, less well-educated, working class, English-identifying nationalists. And Labour has gained votes from the typical Remain voters: younger, better educated, middle class internationalists.

Strangely economic issues did not seem to play a major role in voters' decisions. The Conservative campaign downplayed economic issues, though Labour stressed themes of increased public spending and renationalisation.

In summary, the Conservative campaign seems to have successfully targeted Leave voters, including ex-UKIP supporters, with a relatively nationalist message. But they have lost support from more traditional Conservative groups such as the well-off and well-educated middle classes.

This was always billed as the "Brexit" Election and it lived up to that name as voters changed sides according to their EU referendum vote and nationalist attitudes, whilst ignoring the usual economic and social factors.


Return to Psephology home page.