PA has a long and proud history of covering UK elections and has been a trusted source for elections data since the 19th Century. Over time it has established conventions for handling this data that differ in some respects from the approaches adopted by some other media organisations and elections pundits.
Customers should be aware of the differences as these can help explain apparent data discrepancies.
PA normally uses the most recent state of representation at the time of an election as its baseline for seat changes rather than the state at the outcome of the previous main election. But other statistical calculations – e.g. of percentage changes in vote share and majorities, or swing calculations – do use the previous main election as the baseline.
In the case of a Westminster general election or by-election, or a devolved assembly election or by-election, a first-past-the-post constituency will be regarded as being held by the party of the sitting Member rather than whichever party won it at the previous general election or main election.
In other words, if a constituency has changed hands since the previous general election or main election because of a by-election, defection or switch to independent, PA will reflect that position in its baseline data.
For example, if a seat was won by Labour at the previous general election but then lost to the Conservatives in a subsequent by-election and the Tories still held it at the time of the new general election, PA would record the seat as being held by Conservatives in its baseline data.
If Labour then regained the seat at the new general election, PA would show this as a Labour gain from Conservative.
However, others using the previous general election as their baseline for representation, disregarding any changes since then, would show such a result as a Labour hold.
By way of illustration, here is the 2017 UK general election result for Richmond Park, as reported by PA:
Note that the result is shown as a Conservative gain from Lib Dem as the Lib Dems took the seat from the Conservatives at a 2016 by-election. It is not shown as a Conservative hold, which would be the case if the 2015 general election was used as the baseline and the intervening by-election was ignored.
In PA’s overall 2017 general election State of the Parties statistic, the seat was counted as a gain for the Conservatives and a loss for the Lib Dems.
But note that PA’s statistical calculations for the Richmond Park result do use the preceding general election in 2015 as the baseline and not the intervening by-election. The Conservatives’ vote share is shown as 13.07% down on the 2015 result while the Lib Dems’ vote share is up 25.8%. The swing calculation is from 2015 and the increase in turnout of 2.68% is from that in 2015 – and not from the 2016 by-election.
PA believes this approach is justified because the public will want to know first of all where and how many seats have changed hands as a result of the new election. But where people wish to know the statistical changes they primarily will want to see them in relation to the previous main election rather than any intervening by-election. These statistics then will be in line with those done for all the other seats where no by-elections have taken place and will contribute to an overall picture without the distortion caused by the large variations that sometimes can be seen at by-elections.
Similarly, in the case of council elections, PA will use the most recent political composition of the council as its baseline rather than the composition at the outcome of the previous comparable election.
For example, if the current line-up for a particular council at the time of a new all-up election is Conservative 20, Labour 20, Lib Dem 5, Green 5, UKIP 5, Independent 5, and the outcome of the new election is C 18, Lab 18, LD 6, Green 5, UKIP 6, Ind 7, then PA would record the changes as: C lose 2, Lab lose 2, LD gain 1, UKIP gain 1, Ind gain 2.
However, those using the previous comparable main election as their baseline may well record the changes differently. To take a more extreme example to illustrate the point, if the outcome for this council at the previous election was C 17, Lab 17, LD 5, Green 6, UKIP 7, Ind 8, and the numerous seat changes since then because of by-elections and defections were disregarded, then the gains and losses would be recorded as: C gain 1, Lab gain 1, LD gain 1, Green lose 1, UKIP lose 1, Ind lose 1.
Note that this difference in approach carries over into overall calculations of seats – and councils – gained and lost by the various parties in the election.
PA’s final State of Parties statistic for English council elections in May 2017 appeared thus:
However, a similar statistic using the previous comparable election in 2013 as the baseline instead of (PA’s preference) the council line-ups immediately before polling day would appear with many differences in the seats change figures, e.g. Conservatives up 319 instead of 313, Labour down 142 instead of 146, Lib Dems down 28 instead of 25 and UKIP down 143 instead of 114.
A few discrepancies may be the result of differences in the recording of candidates’ party descriptions but most would be because of changes in council line-ups since the previous comparable election following by-elections, defections and switches to or from independent.
Note that the tally of seats gained will not necessarily match that of seats lost because of changes to the overall number of seats on some councils following boundary changes.
PA believes its convention allows its reporting on elections to provide an immediate, more meaningful reflection of the impact of changes. For example, a governing party that narrowly won a general election, then lost its majority because of a split but regained it at the next general election would be shown by PA as having gained the seats it needed to put it back into power – rather than simply holding seats it did not actually have at the time of the new poll.
The situation is similar in the case of council elections. For example, a council may have been under no overall control (NOC) after the previous election but later came to be controlled by one of the main parties which obtained a majority because of by-election wins and defections. If in the new election the council went back to NOC, PA would record this as a loss for the main party concerned, which would be regarded as significant. A different approach, using the previous comparable election as the baseline, would show the council control situation as unchanged.
PA’s council elections data coverage is significantly different from that it provides for other polls such as general or assembly elections, or mayoral or police commissioner elections. At council elections, PA provides the overall picture for nominations and results at each authority where polls are taking place, summarised by party. PA does not provide detailed ward/division-level nominations and results data for council elections.
Coverage is limited to higher-tier authorities such as unitary, metropolitan and non-metropolitan authorities and county councils. PA does not provide elections data coverage of lower-tier authorities such as town and parish councils. Elections for sui generis authorities – the City of London and the Isles of Scilly – are excluded from data coverage as they have special arrangements and do not conform to type.
Note that where a party did not field a candidate in a particular seat at the previous election, no up or down calculation will be made for their candidate in the result for the current election. Some other media outlets will provide a percentage increase figure from a baseline of zero for such a candidate but PA believes this could give a misleading impression of an increase in support from voters.
An example from the 2017 UK general election is the Basingstoke result, where the Green and Libertarian parties did not field candidates at the preceding general election in 2015. The new result reported by PA gives no up or down percentage for these parties – see below. But some others may report the result with the Greens given +1.98% and Libertarians +0.38%.
Note that the XML version of PA’s results also provides a majority percentage change calculation which is not given in the wire message version.
A swing calculation measures the shift in voter support from one party to another in an individual result or combined across all results in so far. Overall swing calculations can provide tools for establishing which seats should fall to one party or another on an even nationwide swing and can help with forecasting an election outcome.
In an individual result, a swing calculation can be done between any two parties that fielded candidates in that seat at the previous related election. The increase in the vote share of one is added to the decrease for the other and divided by two. If both parties have seen their vote shares increase or decrease, the lower figure is subtracted from the higher and divided by two.
In PA’s individual results (apart from council elections where results comprise summaries by party), the swing calculation is performed between the winning party and the losing party where the seat has changed hands or between the winning party and the second party where the seat has not changed hands.
If the losing party did not field a candidate this time the calculation will be between the winning party and the second party – or between the winning party and the party that won the seat at the previous main election.
Note the result is given as a Labour gain from Ind but the swing calculation is between Labour and the second party, the Conservatives.
Note that the result is given as a Conservative gain from Independent but the swing calculation is between UKIP and Conservative as UKIP won the seat at the 2015 GE.
Note that if the winning party or the second party did not field a candidate at the previous main election no such swing calculation can be done and it will not be provided in the result.
Example – Bexhill & Battle at the 2015 UK general election where the second party, UKIP, did not field a candidate at the preceding 2010 GE:
PA includes the seat of the Speaker of the House of Commons in the tally for the party for which he or she previously was elected as an MP – even though they will be shown with the party description “The Speaker” in the individual result for their constituency.
So, for example, at the 2015 UK general election, Speaker John Bercow was shown in PA’s nominations and results data with the party description “The Speaker”. His Buckingham seat was shown as being held by The Speaker at the time of the poll and the result was shown as a hold for The Speaker.
However, as Mr Bercow was elected for the Conservatives as MP for Buckingham prior to becoming Speaker, PA counted him among the Conservatives in its key State of the Parties statistic.
Buckingham was included as a Conservative seat in the party’s final tally of 331 seats, providing an overall Conservative majority of 12 in the Commons.
However, those who excluded the Speaker from the Conservative tally gave the party a final figure of 330, providing a Tory majority of 10.
The PA convention takes account of the parliamentary arrangements between the two sides of the Commons, where the offices of the Speaker and three deputies are divided equally between them. This means the votes of the four MPs concerned are effectively cancelled out and do not affect the Government’s working majority. The Speaker therefore cannot be regarded in isolation when it comes to the arithmetic of the House.
PA has specific measures in place to handle independent candidates and sitting or elected Members. These are necessary to deal with the following exceptional points:
Independents are not members of a party called “Independent” – they are different from party representatives
Whereas parties will field only one candidate per available place in first-past-the-post elections it is possible to have more than one independent candidate standing.
When it comes to a statistical overview such as the key State of Parties statistic all independents should be grouped together.
PA uses the party description and party abbreviation fields in its elections data to cover the situation.
In order to differentiate one independent candidate from another, the party description will be given as “Independent” followed by the first name and surname of the candidate, e.g. “Independent Sean Smith”. However, the abbreviation used will be simply “Ind” – the same as for all other independent candidates.
This means that customers can differentiate independent candidates using the disambiguated party description but provide the party abbreviation, “Ind”, when publishing nominations and results.
At the same time, grouping candidates with the party abbreviation, “Ind”, for statistical purposes will provide meaningful figures.
Note that disambiguation is applied only in cases of multiple independents standing in the same constituency. Where there is just one independent standing in a particular seat disambiguation is not necessary and the candidate will appear with “Independent” as the party description with the abbreviation “Ind”.
The same considerations apply when it comes to “no description” candidates and sitting or elected Members. PA has adopted the same measures as for independents. In other words, the party description will be given as “No description” followed by the first name and surname of the candidate, e.g. “No description Mary Jones”. However, the abbreviation used will be simply “ND” – the same as for all other “no description” candidates.
May 2019 note regarding UK Euro elections coverage: an exception is made in this case with regard to the party descriptions of independent candidates standing in this type of election, which uses a party-list system. To make clear such candidates are individual and not on an “Independent” party list, the party description is given as Independent followed by the candidate’s name or surname and the abbreviation is given as Ind followed by the surname (or an abbreviated version of the surname). However, these candidates are still grouped under “Ind” for statistical purposes.
PA makes use of “parent-child” structures in some cases to cover some specific party relationships.
All independents, including those with disambiguated party descriptions, nevertheless are grouped as subsets (or “children”) of what is regarded for statistical purposes as the Independent “party” (or “parent”) with the abbreviation “Ind”. The same applies for “No description” candidates.
Another notable case where this applies is in the coverage of Labour Co-op candidates and sitting Members. Where a candidate sponsored by the Co-operative movement is standing as “Labour Co-op” this will be reflected in PA’s party description and the abbreviation “Lab Co-op” will be used.
But this party will be regarded as a subset (or “child”) of the Labour party (the “parent” party) and all Labour Co-op candidates and sitting or elected Members will be included under Labour for statistical purposes.
Parliamentary and assembly constituency boundaries and local government ward/division boundaries are reviewed periodically by the various boundary commissions with the aim of equalising the number of voters in each voting area as much as possible, according to set criteria.
The process is deemed necessary to take account of changes in the size of electorates and to try to make sure votes are weighed equally – except where there are special considerations such as in the case of some island constituencies.
When boundary changes are implemented it can complicate the reporting of results. For example, a Commons constituency currently held by the Conservatives may have its boundaries redrawn and lose some areas known to favour the party while taking in others known to lean more towards Labour. The effect of the changes may be so dramatic that it would be unreasonable to regard the seat as Conservative held. If in the ensuing election the seat falls to Labour, it could be misleading to report it as a Labour gain from Conservative.
In the interests of fair and accurate reporting, PA makes use of analysis of ward-level voting from the previous election which provides a notional result showing how the new constituency would have voted at the previous election if the new boundaries had been in force at that time.
In the above mentioned example, the analysis could show that if the previous election had been fought on the new boundaries the seat would have gone to Labour and not Conservative. PA’s baseline data therefore would show the seat in the new election as notionally held by Labour from the previous election. If Labour then took the seat in the new election, PA’s results would show it as a notional Labour hold rather than as a Labour gain from Conservative.
Example of PA result for a boundary change constituency at the 2010 UK general election:
Wide-ranging changes to Commons constituencies are due to be implemented for the next UK general election under the 2018 Review which proposes to reduce the number of seats from 650 to 600.
Prior to the calling of the snap UK general election for June 2017, PA joined a media consortium with BBC News, ITV News and Sky News in commissioning Professors Michael Thrasher and Colin Rallings to calculate a set of “official” notional results from the 2015 general election.
The academics, emeritus professors of politics at the University of Plymouth and Associate Members of Nuffield College, Oxford, are providing new calculations to cover the 2017 general election.
These will form the basis of election reporting and coverage by all the members of the media consortium and their clients if the 2018 review is implemented in time for the next general election now due in 2022.
See the original press release about these plans at: https://www.pressassociation.com/company-news/uk-media-commission-common-standard-election-reporting/
The situation is simpler when it comes to local government boundary changes as PA does not provide detailed ward or division level nominations and results for council elections but provides summaries by party for each council.
Where boundary changes have taken place at a particular council, this will be indicated with the text, “Boundary change”, in PA’s council result. The result will give the new council line-up and provide up and down figures by party calculated from the previous council line-up even where the total number of councillors has changed.
Examples – 2016 council election results for Knowsley (council reduced from 63 to 45 after boundary changes) and Peterborough (increased from 57 to 60) – all seats up at this poll:
For comparison, two examples of “normal” results for councils where no boundary changes have taken place (both with one third of their seats up at this election):
Note that in these “normal” cases the seat changes are given as gains or losses rather than the up or down applied when boundary changes have taken place.
It is normal procedure during election counts for the Acting Returning Officer to announce the “turnout” or “turn-up” figure and/or percentage after the verification process is complete. PA does not provide this in its elections data although it may be covered in editorial coverage if it is regarded as significant (e.g. if it is an early indication of a generally high or low turnout).
What PA does provide in its results data is a turnout figure and percentage calculated from the result itself. In this case the turnout figure will be the total number of valid votes cast. In other words, it excludes spoilt ballot papers (which are included in the official turnout/turn-up figure).
PA’s turnout percentage will be related to the final or “nearly final” official electorate figure released before polling day.
The exclusion of spoilt ballot papers from PA’s turnout data helps speed the delivery of results and avoids the complications that would arise from trying to include and account for figures for rejected votes.
Note that PA provides a dynamic wire-only statistic for turnout (valid votes cast) for the overall vote in Westminster and devolved assembly elections. Users of PA’s elections data in XML format for digital products can calculate this statistic themselves from the results.
Example of POLL Turnout message:
PA does not use real candidate data in the elections tests it provides for customers in the run-up to polling day. This is necessary to avoid complaints from candidates and particularly sitting Members who may be shown as doing badly or losing their seats in the test data, which uses computer-generated randomised voting figures.
This means there are some significant differences between the test set-up and the real one, particularly when it comes to the PA IDs for elements such as parties, candidates and elections in XML format for digital products. See the advisory on XML format elections messages at: electionssite.wpengine.com/xml-format-3