Methodology - "Weighted Swing"
Constituency % = Base +/- (National Swing * Final Weight)
We currently hold the results of each constituency from the 2017 General Election. We use these results as a baseline to which we apply our methodology.
New Parties – Brx and TIG baselines are currently calculated using average % swing data from other parties over the last month of polls. These are revealed within pollsters polling tables. For example, Brx could be comprised of 33% of 2017 Con voters, 80% of UKIP 2017 voters, 20% of 2017 Lab voters. Using this information, we will project an estimated 2017 value as the new party’s baseline and feed this into the overall projection methodology.
The final weight number is comprised of various minor weights. These are added together and applied to the national swing per constituency. Each weight is explained below.
Relative Party Strength Weight
Each party has it’s mean support calculated by adding up all of its % support in the previous election and dividing it by the number of constituencies that is stood in. This weight is the relative strength of constituency compared to that mean with a cap on those over performing outliers. The cap has been calculated by comparing the highest constituency swings with the general election swings.
There is no lower cap as we cannot have a party which gained 3% in a constituency losing 15% of the vote, therefore the lowest performing seats compared to the average would see a far smaller decrease in vote than that of the larger seats. This is as we rarely see the major parties achieve a 0% voting record in any constituency they stand in.
The reason for the upper cap is that we see parties such as the Greens holding large outliers to their mean support, without the cap this weight would see such outliers boost their support in such areas enormously within the methodology where it has no basis of increasing by such a margin.
The leftover weight from each overperforming seat is not thr/own away, it is redistributed over all constituencies the party is standing in. For example, if the party has 4 outliers totalling an extra weighting of 8 after the cap is applied, this would be divided by the total number of seats standing and that number added to each constituency weight for said party including those seats that were initially capped.
We currently hold the estimated Brexit vote % per constituency and the current polling of voting intention in the event of a new referendum. Before using the constituency Brexit vote, we apply a uniform swing to each constituency. For example, if the current average is 49-51 in favour of remain, we would move each Constituency Brexit vote 3% towards remain.
We can now calculate the ratio of leave / remain constituencies in the area we are projecting (Usually GB). This is used in the calculation of Brexit weight as the numbers are barely ever even thus if we -1% due to Brexit stance in 200 constituencies, but +1% in 400 constituencies, the national % would skew. Something we have to avoid as much as possible when projecting.
Now that we have a general estimation on how leave / remain a constituency is and the ratio, we can calculate the Brexit weighting of the constituency itself. This is simply the amount by which the Constituency voted brexit or remain divided by 100 and then applying our ratio. This leaves us with a positive value for remain constituencies and a negative value for leave constituencies.
We must now apply a party Brexit weighting to this value which is also calculated by average polling, however this uses % of Leavers and % of Remainers voting for each party. This calculation leaves us with a negative value for parties being voted for majority by leave and a positive value for parties voted for by a majority remain.
We then multiply these numbers together which would give us the final brexit weight for each party in each constituency. However before applying this to our final overall weight we have to reverse it if a party has polled a negative swing.
This would result in higher positive swings in like minded constituencies and lower positive swings in constituencies with opposing views when the national swing is positive. The reverse is true with a negative national swing as a like minded constituency would take a lower hit than that of a constituency with opposing views to the party.
Age Density Weighting
We currently store the estimated amount of people of a given age within each constituency. This data is updated based on releases thr/ough the ONS, NRScotland and NISRA. Each pollster releases breakdowns of voting intention by age groupings. We use an average of these groupings between various pollsters to give a weight to each party on how likely a given age group will vote for said party.
*As a note there are four commonly used age groupings, 18-24, 25-49, 50-64, 65+. Each grouping has a proportionate amount polled for the years in the range. E.g. there are roughly 4 times more 25-49 polled to 18-24
From these numbers we can calculate the proportions of a party’s voting intention an age group fills. For example, although say 53% of 65+ will vote Conservative, this will constitute 39% of the overall Conservative vote. Giving the conservatives a +0.14 initial weight with the 65+ grouping, in contrast say 18% of 18-24 grouping vote conservative this will constitute only 5% of the Conservative overall vote. Equalling a -0.20 weight for the 18-24 grouping. We can then apply the percentage each grouping constitutes in a given seat based on the aforementioned population statistics to each party age weight, we end up with the full age density weighting for a given party in a given seat.
As previously mentioned we hold the constituency results from the 2017 general election, we also hold the 2015 results. With these sets of results we can see the volatility within each seat visualised as the % either over or under the average swing per party.
This is the easier of the weightings to calculate.
If a party achieves an average national swing of 5%, a constituency change of 15% would equate to a 0.10 weight whilst a -5% would equate to -0.10 weight. We calculate this for both 2015 and 2017 elections and average the result. This helps pick up on some constituencies that are diverging from the trend.