LucidTalk Westminster General Election seat predictor: The safe and the precarious
Hopefully you may have seen our first article covering the results from the first running of our automated computer systems based ‘Seat Predictor Model’ for the upcoming NI Westminster Election.
This was last used very successfully for the UK General Election in 2015. This computer based predictor model has been constantly improved and enhanced, and we’re again confident that it will be able to closely predict this upcoming Westminster election in NI.
The model uses as input the three most recent major NI election results - currently: UK (NI) General Election 2015, and the NI Assembly Elections 2016 and 2017, and also the most recent LT polling i.e. our first pre-Westminster election NI-Wide poll: 27th – 29th April. The model also considers and weighs-in factors such as incumbency (who holds the seat at the moment), and what poll participants believe will be the results in each seat (i.e. apart from what way they will be voting themselves in their own seat).
The model then then predicts likelihood in percentage terms, of seats being won by the various parties, broken down by the 18 Westminster constituencies. We’ve run the model, using the three most recent NI elections and our April 2017 NI-Wide representative ‘Tracker’ poll.
The model generated the results shown in the table (see above). Comparisons are shown against our ‘Base Table’ which was generated before the results of this most recent poll were used.
This will be the ‘Version 1 – Poll 1’ table, and we will re-run the seat predictor model again after each of the next two pre-election polls that are scheduled (i.e. using the latest polls data as the main input to the model) – this will enable tracking of trends up or down for each of the political parties in each of the 18 NI Westminster constituencies as the campaign progresses. As such, the seat predictor model should get ‘more accurate’, and up-to-date, as the campaign progresses, as it will be using/analysing the latest poll data.
SEAT PREDICTOR – WHAT DOES THE TABLE SHOW
So what does the model currently show?
Interestingly, the model previously predicted that Sinn Fein were the narrow favourites to re-capture the Fermanagh and South Tyrone seat from the Ulster Unionists – they were rated at 55% for this seat, i.e. before the data/results from this poll were inputted into the model. In Fermanagh and South Tyrone the unionists are running one agreed candidate (Tom Elliott for the UUP) with the support of the DUP. We see after the first LT pre-election poll results were inputted into the model that the Sinn Fein chances in this seat has dropped from 55% to 50% - i.e. basically the UUP and Sinn Fein are now equal chance on 50% each.
The earlier 55% Sinn Fein chance wasn’t surprising, that taking into account that our recent 2017 pre-NI Assembly election polls, and the election itself, all showed and produced a strong Sinn Fein performance. Remember the data from our polls and the recent elections are the key inputs into our computer-based seat predictor model. However, this current poll showed a better performance for the Unionists, and now the seat predictor model says this seat is on a knife-edge/toss-up between Sinn Fein and the current incumbent Tom Elliott of the UUP.
However, this rating may change following our second and third polls during the campaign. Perhaps what’s keeping the Sinn Fein rating for this seat at ‘only’ 50% is that we also have an incumbency weighting built-in to our predictor model – incumbents always have an advantage in Westminster first-past-the-post elections.
East Belfast is showing as a DUP hold, but at a probability of 65%, and at no change from the first run of our pre-election seat predictor model. This is again lower than expected for an incumbent MP (Gavin Robinson) and for someone from NI’s largest party, the DUP. However, the model has probably been impacted by the Alliance Party’s strong showing in the recent 2017 NI Assembly election and in most recent LT poll.
Alliance are the main challengers for this seat, and will have a chance now that their party leader Naomi Long has been selected as the candidate – a 65% DUP win probability is OK, but it’s by no means certain.
Again of course, this win-probability score may change following our upcoming polling. Alasdair McDonnell, who is defending the marginal South Belfast constituency for the SDLP, comes in at 65% probability, which is up 5% from his 60% score in our ‘base table’ reflecting perhaps that it looks as if McDonnell may not be up against one combined agreed unionist opponent in this seat. The further seat predictor model forecasts will obviously be dependent on whether there are election pacts between certain parties (this also applies to other seats).
Not surprisingly, and as in our ‘base table’, nine of the 18 seats are showing win-probability scores of 100% for the named political party (all of them the incumbent party). North Antrim for the DUP, West Belfast for Sinn Fein etc. Unless something dramatic happens we don’t see these scores changing for any of these nine seats. So as we’ve always said it’s not the current percentage scores that are the major factor, but whether they move up or down following our next two pre-election polls (mid May, and start of June), so watch out for the next runs of our Seat Predictor Model.
Notes on the Methodology: - The seat predictor table shows the results generated from LucidTalk’s NI 18 Westminster constituency computer based seat predictor model (as used for the 2015 General Election).
This automated predictor model takes into account recent NI election results, and the most recent LT NI polls. - This is the Version 1 – Poll 1 Table generated using the first pre-NI Westminster Election LT Opinion Panel NI-Wide representative ‘Tracker’ Poll: 27th-29th April 2017. The predictor model also uses as input results from: May 2015 General Election, and the NI Assembly Elections 2016 and 2017.
Minus percentage probability represents the probability of the named party in the ‘2017 Forecast-Party’ prediction column of winning the named seat, e.g. 90%+ means the named party is almost certain of winning the named seat, 70%-85% means the named party is ‘reasonably probable’ of winning the named seat, 55%-70% means the named party is ‘favourite but not certain’ of winning the named seat etc.. -
As this is the ‘Version 1 Table’ for the General Election 2017 incorporating the first LT pre-Westminster election polling – the ‘change since last forecast’ i.e. before this first pre-election LT poll, is detailed where applicable. This table will be updated after each of the next two planned LT polls to take place as part of the UK General Election campaign (April-June 2017).
As indicated above, changes will be shown compared to the immediate previous Seat Predictor table by % probability changes e.g. +5%, -10% etc..
This table, does not, indeed cannot, take into account unknown factors at this stage of the campaign e.g. political party pacts, specific candidates running or not running. These factors will be built into the automated predictor model, as and when they become clarified during the course of the campaign – as such, later LT seat predictor model tables will be ‘more accurate’, and reflect the current status e.g. candidates, pacts etc.
Bill White, is Managing Director of Belfast based LucidTalk Polling and Market Research. You can follow LucidTalk on Twitter at @LucidTalk.