Where parties win votes at the next general election will matter as well as how many they win. Recent opinion polls suggest that since 2010 the Conservatives have suffered a small set back and Labour have made a correspondingly modest recovery. Much more dramatic has been the collapse of the Liberal Democrats and the rise of the United Kingdom Independence Party (UKIP). But more startling still are the Scottish polls which show 20+ point swings from Labour to the Scottish Nationalist Party (SNP) which suggest around a half of all seats in Scotland could change hands.
Much could change in public opinion between now and the election, but supposing further changes in overall party popularity before the next election are not large (between + or – 5 points) then it is appropriate to start asking whether the overall gains and loses of parties will be replicated equally across constituencies. In particular, how appropriate are traditional uniform change assumptions when thinking about how votes might translate into seats?
This paper does not provide a forecast but instead seeks to identify the main patterns of variation across constituencies in party performance and discuss their potential implications for possible seat outcomes in broad terms. The methodological approach is based on the latest (provisional) data from the British Election Study (BES) online panel survey, which were collected by YouGov in September and October. These data have been subject to minor reweighting to the share of the vote in recent GB, Scottish and Welsh opinion polls, and then linked to the 2010 constituency results in order to estimate constituency-level change in vote shares. The BES was not designed for this purpose and for various reasons it is important to focus on only the very broadest and clearest patterns. It is certainly inappropriate to ask of these data what might happen in particular constituencies or very small groups of constituencies.
Nations apart
The table below shows average changes in party support across constituencies for each of the three British nations and for GB as a whole. The dramatic 25 point average increase in the SNP vote in Scotland is accompanied by a big average drop of 18 points for Labour. Meanwhile it is striking that the rise of UKIP in Scotland, at 5 points on average, has been much more muted than its 14 point average rise in England and Wales.
The pattern of change in Scotland is so very different from that in England and Wales that it is much more appropriate to separate them for the purposes of analysis. However, because there are only 59 constituencies in Scotland it is difficult to be confident about the pattern of change within Scotland from these data. So the remainder of the analysis in this paper will focus on variation within England and Wales.
Although the table above does not show it, there is very little difference between England and Wales in the pattern of change, with the exception of Labour who appear to have made little or no advance in Wales but were up on average 5 points in English constituencies. Importantly for the party this means that GB vote intention polls are understating the increase in Labour support in England and so their potential to take seats from the Conservatives. Relative to uniform swing assumptions, this will offset partly but not fully their losses to the SNP.
Another strange death of Liberal England (and Wales)?
The most significant factor affecting party performance at the constituency level is prior Liberal Democrat strength. The following graph shows how party performance varies with the 2010 Liberal Democrat constituency vote share across England and Wales. If classic uniform swing assumptions held all the lines in the graph would be flat. They very roughly are for the Conservatives and UKIP, but the Liberal Democrats are clearly loosing most in the seats where they started strongest and losing least where they started weakest.
Partly this is inevitable. There are over 100 seats where the Lib Dems got less than 16% of the vote in 2010 and so their vote share cannot fall by this much. Moreover it is unlikely that the party will fall exactly to zero even where it does very badly. So if the GB polls are right overall, the Liberal Democrats must be falling more where they started stronger, and the BES data suggest the drop is broadly proportional to their prior strength. This mirrors the pattern of change at the local authority level at the European Parliament elections this year, adding confidence that the effect is real.
The implications for Liberal Democrat seats are straightforward. If they are indeed losing most heavily in the seats they are defending they are set to lose several more seats than national polls with uniform swing would predict.
Maybe closer to the election the Liberal Democrats will benefit from voters focusing more on the specific situation in their constituency, with tactical voting and incumbency effects kicking in. Constituency polls by Lord Ashcroft suggest that prompting people to think about the candidates in their constituency when asking people whom they will vote for results in much more Liberal Democrat voting in Liberal Democrat seats. But there is a danger that such prompting over-states incumbency advantage. For many Liberal Democrat MPs to hold on to their seats they will need to become even more personally popular than they were in 2010: a tough task under the circumstances.
It is also worth noting that even with a much smaller sample size, a similar analysis of the 2009 wave of the BES internet panel rightly suggested little difference in Liberal Democrat 2010 performance in the seats they previously won compared with those where they came second in 2005. So the above finding for 2014 is not just a usual pre-election campaign poll finding. Moreover the constituency variation in the 2010 BES data corresponds well to the actual result. So there are various good reasons to believe the Liberal Democrats are facing a bigger uphill battle in their own seats. The 2014 BES data, suggest they will do 10 points worse in their own seats compared with where they were previously second. Even if this difference is somewhat attenuated by the time of the election, it is quite a gap to close.
Labour are the clear beneficiaries of this pattern of Liberal Democrat decline. This fits previous research on the pattern of vote switching at the individual level. While more votes are clearly welcome for Labour, this particular constituency pattern is not particularly advantageous for the party. Most Liberal Democrat seats have the Conservatives in second place and it will be tough for Labour to come from third to win. There is a danger for Labour of not winning so many seats as uniform swing would suggest if their vote gains are disproportionately concentrated in areas where the Liberal Democrats did relatively well in 2010.
The rise of little Englandism
The third key issue is where UKIP are doing relatively well and relatively badly and which other parties that hurts.
In line with previous research, especially by Robert Ford and Matthew Goodwin, the BES data aggregated to the constituency level suggests that UKIP are doing rather better in places with more pensioners, and more people without any educational qualifications. Conversely UKIP are doing worse where there are more students, more graduates and more people between the ages of 18 and 24. But the single best predictor of constituency performance for the BES data seems to be population density; UKIP is doing much less well in urban constituencies.
Cities have fewer of the kinds of people who are attracted to UKIP and more of those who are not. It is also worth noting that according to the 2011 census, cities have larger populations of EU immigrants than more mixed and rural areas. Even though UKIP mobilize support with an anti-immigrant platform, they seem to get it in areas where there are fewer such migrants. Perhaps this is a sign that those who have more contact with EU immigrants are least hostile towards them.
So what does this geographical pattern of UKIP support imply for the three main Westminster parties? The graph below shows how their net performance since 2010 varies according the change in the UKIP share of the vote since 2010. It shows almost no association with Liberal Democrat change, but Labour, and much more so the Conservatives, do worse the more UKIP rises. This is as expected given that the individual-level survey data suggest that UKIP have taken more votes from the Tories than from Labour, but it was not clear that this would play itself out at the constituency level.
One consequence of this pattern is that the Conservatives are doing worse where they started strongest and particularly in seats they are defending. Labour are also doing worse the better they did in 2010, but mainly because they are getting more Liberal Democrat votes where they started weaker and not so much because they are shedding votes to UKIP more where they started stronger.
The impact of the UKIP rise on seat outcomes may be more muted than some suppose. Neither regressions models capturing the patterns above nor classic uniform change models predict UKIP to win any seats. But there are a number of constituencies where UKIP are the largest party among BES respondents, which certainly signals the potential for the party to win a handful of seats. At the same time, however, nowhere does UKIP achieve the kind of large constituency vote shares in the BES that the two main parties manage where they do well. Local campaigns can change things but at the moment, the BES data suggest UKIP would struggle to get more than a dozen seats.
The other sense in which the impact of UKIP will be more muted than some suppose regards the contest between Labour and the Conservatives. Focusing on the seats where the Conservatives and Labour finished first and second (in either order) in England and Wales, there is little association between UKIP performance and the difference between the Conservative and Labour share in 2010. Despite taking more votes from the Tories than Labour overall, there is little sign that UKIP are damaging Tory chances more in the key Con-Lab or Lab-Con marginals than elsewhere. There is little reason to expect this, but it is nonetheless important to check whether UKIP are managing to do particularly well in (certain kinds of) key marginals to test uniform change assumptions. The two main parties are in the low 30s in the polls now in large part because of UKIP, but uniform change still seems to provide a reasonable guide as to how those overall vote shares translate into seats among key Con-Lab and Lab-Con marginals in England and Wales.
Conclusion
If this analysis of the BES data is indicative then dramatic changes are in store for the geography of electoral competition in May 2015 with important implications for seats. The doubling of SNP support in Scotland presages big Labour losses north of the border and potentially a pivotal role for the SNP in a hung parliament. The geography of Liberal Democrat support in the BES suggests even greater losses for the party than the already dire overall vote share in the polls. Finally, despite the important effects of the substantial rise of UKIP on the shares of the vote for the two main parties, UKIP are not set to win many seats nor have a disproportionate effect on the competition for seats between the two main parties in their key marginals.
The findings here illustrate the potential wonders of the UK’s first-past-the-post electoral system. It seems likely that UKIP will become the third largest party in votes, followed by the Liberal Democrats and then perhaps the Greens before the SNP. But of these four parties the SNP might be both smallest on votes and the largest on seats. Moreover, both the SNP and the Liberal Democrats are set to win more seats despite fewer votes than UKIP. Although Labour losses in Scotland are likely to undo some of the pro-Labour bias in the electoral system and so reduce the chances that the party will emerge largest on seats but not votes, that scenario is still possible. Finally, despite a hung parliament being the most likely outcome and a fairer reflection of the fragmented distribution of votes, the two main parties are still on course to win 90% of British seats from a combined share of just two-thirds of the vote.
Acknowledgements
Thanks to Michael Addelman, John Curtice, Robert Ford, Jane Green and the British Election Study team for comments on earlier drafts.