Bailey and English - Forecasting GE2019

Estimating voting intention and House of Commons seat wins for the 2019 UK General Election

Jack Bailey and Patrick English

Voting intention

Below is the latest estimated voting intention figures from our polling model. This regression-based approach statistically models, rather than simply aggregating, polling data from all companies since the start of the year in order to produce daily estimates. The figures represent where we believe the parties stand today, according to the figures and trends in the latest polls.

Note: Figures are rounded median and credible interval estimates.
Full methodology is available below.

House of Commons seats

Select a forecast model from the options below. These models are projections of seat wins for each party in Great Britain, worked out using the latest estimated vote intention according to the polling model (above). All models are swing based, but rather than using a Uniform National Swing (UNS), we calculate four regional swings: GB-wide, London, Wales, and Scotland.

Applying these swings to 2017 results data produces the Regional Swing Model. The Vote Transfer models adjust the estimated regional swing according to assumptions about tactical voting and party entry/exit, assuming that the large majority of voters follow tactical voting advice of exiting parties. Users can select forecasts which take into account only one of the Remain or Leave alliances to see their individual impacts on the overall forecast.

Note: Figures are rounded median and credible interval estimates.
Full methodology is available below.

According to this seat forecast, assuming the aggregation of polls reflects an accurate picture of current voting intention, the probabilities of each headline result for the December 12th election are as follows:

The full 95% credible interval range of seats for each party is as follows:

Implied constituency-level vote intention

Select a constituency from the 632 British Westminster seats to view our implied vote intention for that specific location, based on the results of the full vote transfer model swing figures.

Keep in mind that unlike multiple regression and post-stratification (MRP) models, swing-based models such as ours are not particularly designed to produce completely accurate constituency level forecasts. Though we do make adjustments at the constituency level for parties standing down and votes transferring, we do not include constituency level polls nor make use of MRP as the logic of swing-based models is that they work best at the aggregate level (predicting the overall result).

Constituency names must be typed exactly as they appear in our database, where we use the following naming conventions: 'and' rather than '&' (e.g. Heywood and Middleton), no commas (e.g. Ross Skye and Lochaber), and place names proceeding compass points (e.g. Devon East, Chester City of).


Polling model - modelling voting intention

The polling model collects the latest available polling data published by all British Polling Council guideline compliant pollsters and models voting intention. The process is more advanced than a simple aggregation, in that we use a Bayesian dirichlet regression which controls for the House effects, fits a spline for time (reducing overall sensitivity to random variation in headline voting intention), and pools data from regional polling into the estimates. The result of the model is a smoothed fit of estimated voting intention over time, including credibility intervals.

The Regional Swing Model - calculating seat wins from regional swings

Our Regional Swing Model (RSM) picks up the full distribution of voting intention results - including regional level data - and subtracts 2017 aggregated results from each level (GB, London, Wales, Scotland). This leaves us with 4,000 estimated ‘swings’ for each, which we then apply to the 2017 results at the constituency level (with London projections only applied to London constituencies, and the same for Wales and Scotland). This in turn creates 4,000 projected results for each constituency across Great Britain. Each ‘row’ of the constituency level results represents one possible election outcome, which the model then calculates the winner for and sums across parties. Repeated 4,000 times, this gives us a matrix of 4,000 potential election results based on the polling model data. Extracted from these distributions are the median, 2.5%, and 97.5% quantiles for number of seats one. This creates our central estimate and lower and upper credible interval bounds for each party respectively, which are represented in the graphic. Note: Brexit Party swing is calculated using the UKIP 2017 baseline.

The Vote Transfer (Regional Swing) Model - calculating the impact of local alliances

We attempt to model the impact that local alliances on both the 'Remain' and 'Leave' front will have on the overall results by including a mechanism for the transfer of vote shares when a particular party drops out. There are two separate models available looking individually at the Unite to Remain and Brexit Party withdrawal impact on our forecast, and a further model which combines the two and adds 1) Green exit vote transfer to SNP in Scotland, and 2) some special cases (such as Lib Dem vote transferring to Dominic Grieve in Beaconsfield). The simple transfers of shares figures are informed by vote transfer and second preference polling results. The model assumes that 100% of voters involved in Unite to Remain constituencies will back the parties they are instructed to, while when the Brexit Party withdraw we expect 75% of voters to switch to the Conservatives while 25% will move to Labour. This approach, it must be reiterated, is a fairly simplistic way of estimating these effects but will nonetheless give a more accurate picture of what might happen at the constituency level than the RSM, which is unable to account for specific patterns of party entry and exit.

Questions, Queries, Contact

For any questions or queries about the model, or if you'd like to use any of the predictions or forecasts in your own work, publications, or presentations, please just reach out to either one of Jack (@PolSciJack) or Patrick (@PME_Politics) on Twitter, or consult our institutional webpages for email addresses.

This application is hosted on a server machine at the Q-Step Centre, University of Exeter. We would like to thank Robert O'Neale, Travis Coan, and the Politics Departments at the Universities of Manchester and Exeter for their support.