Adrian Kavanagh, 24th May 2017

There are just over two weeks to go until the general election that is scheduled to be held in the United Kingdom on Thursday 8th June 2017. What do the latest opinion poll figures suggest in relation to the likely number of seats that each party will win in that contest? As the last general election in the United Kingdom showed, this is a hazardous task to engage in, as the votes to seats ratio can be skewed by geography and spatial differences in terms of support gains/losses for different parties. For instance, Labour’s overall share of the vote increased in 2015, relative to the 2010 General Election, by 1.5%, but the extent of the party’s support/seat losses to the Scottish National Party in Scotland (returning with just 1 of the 41 seats that the party had won there in 2010) meant that most of the 701,147 votes won by Labour in Scotland were effectively wasted votes.

For the purposes of this analysis, I have averaged out the estimated support levels for the five political parties in the United Kingdom (Conservatives, Labour, United Kingdom Independence Party, Liberal Democrats and the Green Party) that are covered in all of these polls. As no consistently accurate poll figures are presented here for the regional parties, or for the very small parties, I have assumed – for the purposes of this model – that support levels for these parties/groupings will remain “as was” relative to the 2015 contest. Averaging out the party support estimates in the most recent ICM/Guardian, Kantar, Survation/Mail on Sunday, YouGov/Sunday Times, ComRes/Sunday Mirror and Opinium/Observer opinion polls (as published on the UK Polling Report website) estimates support levels for these five parties to stand as follows: Conservatives 47.3% (up by 10.4% relative to 2015), Labour 30.3% (down by 0.2% relative to 2015), United Kingdom Independence Party 5.4% (down by 7.3% relative to 2015), Liberal Democrats 9.0% (up by 0.9% relative to 2015) and Green Party 2.7% (down by 1.0% relative to 2015).

Using the same constituency-level analysis model that I employ to study the likely impact of Irish opinion poll figures on Dail-seat estimates, I estimate the following number of seats would be won by these parties – and other parties in England, Scotland and Wales – in June 2017, should the latest opinion poll figures be replicated exactly at the general election contest (As noted below, I also apply the figures in the most recent Scottish regional poll to generate separate support/seat estimates for the 59 Scottish constituencies): Conservatives 388 (up by 58 seats relative to 2015), Labour Party 187 (down by 45 relative to 2015), Scottish National Party 44 (down by 12 relative to 2015), Liberal Democrats 7 (down by 1 relative to 2015), Plaid Cymru 4 (up by 1 relative to 2015), Green Party 1 (no change relative to 2015) and United Kingdom Independence Party 0 (down by 1 relative to 2015). (As the national-level opinion poll figures seem to largely ignore Northern Ireland, I cannot factor in support/seat estimates for the political parties in Northern Ireland into this analysis, unfortunately.)

Improving poll fortunes for Labour over the past few weeks have seen the party almost return to the level of support that they had at the 2015 election and the Liberal Democrats are expected to improve on their 2015 result (although this party has lost some ground in recent weeks). However, the extent of the swing to the Conservative Party, as measured in the most recent opinion polls, would see both of these parties losing seats to the Conservatives if these poll figures were to be replicated at the real election on 8th June.  The Green Party, by contrast, would see an overall loss in their national support levels if the most recent poll figures were replicated at the June 8th contest, but would be expected to hold their one seat in Brighton Pavilion – although that seat would become increasingly marginal and vulnerable to a gain by the Conservatives or Labour.

The model predicts that the Conservative Party would gain seats in the following 46 constituencies:

Most of these gains would be made at the expense of the Labour Party, but gains would also come at the expense of the United Kingdom Independence Party (Clacton), the Scottish National Party (Berwickshire, Roxburgh & Selkirk) and the Liberal Democrats (Carshalton & Wallington and Southport) should the support/seat estimates in this analysis pan out in the actual general election on 8 June 2017.

The model predicts that Labour would lose seats in the following constituencies – with most of these losses being made to the Conservatives, but with losses also predicted to be made to the Liberal Democrats (Cambridge), as well as to Plaid Cymru in Wales (Ynys Môn):

While the Green Party are expected to hold their one seat in Brighton Pavilion, based on these figures, this constituency is emerging as a significant battleground constituency, with Labour and the Conservatives, based on the model support estimates, both in contention to win this constituency.

Update (Scotland): I have now factored in the results of the latest Scottish regional opinion poll (the YouGov/Sunday Times poll of 19 May) into the support/seat estimate figures for the 59 Scottish figures. This poll estimates support levels for the five aforementioned parties and the Scottish National Party as follows: Scottish National Party 42% (down by 8.0% relative to 2015), Conservatives 29% (up by 14.1% relative to 2015), Labour 19% (down by 5.3% relative to 2015), United Kingdom Independence Party 1% (down by 0.6% relative to 2015), Liberal Democrats 6% (down by 1.5% relative to 2015) and Green Party 2% (up by 0.7% relative to 2015). These support/seat estimate figures suggest that the Conservatives would make gains in Scotland at the expense of the Scottish National Party in the following constituencies if these support figures were to be replicated at the June 8th election across the 59 Scottish constituencies: Angus, Banff & Buchan, Dumfries & Galloway, East Lothian, East Renfrewshire, Edinburgh South-West, as well as Berwickshire, Roxburgh & Selkirk, Moray, Ochill and South Penrithshire, Perth and North Perthshire, Stirling and West Aberdeenshire & Kincardine, as well as Berwickshire, Roxburgh & Selkirk, as already noted above. The Conservative Party in Scotland would also be predicted to gain Edinburgh South at the expense of Labour on the basis of the party support estimates relating to the YouGov/Sunday Times poll.

How this model works: As with my analyses of opinion poll figures in Ireland, constituency support estimates for different parties and groupings form the basis of the general approach taken with this analysis. This approach seeks to ask the following question in relation to different opinion poll results – what do these poll figures mean in terms of the likely number of House of Commons seats that could be won by the different parties and groupings on those national support levels? As the electoral system employed in the United Kingdom is most definitely not a proportional electoral system, the proportion of seats won by parties will not measure up exactly to their actual share of the first preference votes. Geography will have a very significant impact here, while first past the post electoral systems, such as the one used in the United Kingdom, will always act to the advantage of the largest parties and will significantly disadvantage the smaller parties, unless their votes are strongly clustered in a specific geographical region – e.g. Scotland, Wales or Northern Ireland.

First preference votes need to be filtered through the system of the United Kingdom’s general election constituencies . In order to address the question, I estimate what the party first preference votes would be in the different constituencies, assuming similar (proportional) changes in party vote shares in all constituencies to those that are being suggested by a particular opinion poll, or – in this case – an average of the most recent polls. This of course is a very rough model and it cannot take appropriate account of the fact that changing support levels between elections tend to vary geographically 9as occurred in the 2015 contest). Thus constituency support estimates for different parties/groupings will be over-estimated in some constituencies and under-estimated in others, but the expectation would be that the overall national seat figures figures estimated will be relatively close to the true level, given that over-estimates in certain constituencies will be offset by under-estimates in others. Based on these estimated constituency support figures, I proceed to estimate the destination of seats in the different constituencies. Due to unusually high/low support levels for some parties or political groupings in certain constituencies in the previous election, the model may throw up occasional constituency predictions that are unlikely to pan out in a “real election”, but of course the estimates here cannot be seen as highly accurate estimates of support levels at the constituency level as in a “real election” party support changes will vary significantly across constituency given uneven geographical shifts in support levels.

The point to remember here is that the ultimate aim of this model is to get an overall, national-level, estimate of seat numbers and these are based, as noted earlier, on the proviso that an over-prediction in one constituency may be offset by an under-prediction in another constituency. This model does not aim, or expect, to produce 100% accurate seat level predictions for each of the 650 constituencies (and indeed cannot do so for the 18 constituencies in Northern Ireland). These analyses simply estimate what party seat levels would be, should such national support trends be replicated in an actual general election. For a variety of reasons (including the impact of high levels of undecided voters in a specific poll), the actual result of an election contest may vary from the figures suggested by an opinion poll, even if the poll is carried out relatively close to election day, or on election day itself as in the case of exit polls, but the likelihood of such variation is not something that can be factored into this model.