I wrote a week or so ago about using some statistical methodology to try and predict the forthcoming General Election in the UK. I’ve now undertaken my first run of the model I developed and the results are below.
Remember, the model works on the following basis:
- A statistical model that looks at historical trends for each party of the movement of the polls from x months before the election to the actual point of the election
- Multi-variate analysis looking at the spread of the party’s support in the current polls
- Based on 1 and 2, a projection simulation of the actual election next year. I run 2,000 random simulations and analyse the results to produce the results below.
Remember, the model is based on the party’s opinion poll support moving as expected before the next General Election. If the support doesn’t move as expected over the next few months then the model will self-correct and respond to the new situation.
The following table is the mid-point prediction – i.e. of the 2,000 simulations this is the most likely outcome.
|Party||Vote %||Seats||Prediction Range (95% CI)|
|Conservatives||37.3||332||311 – 343|
|Labour||31.3||262||233 – 290|
|Liberal Democrats||16.1||13||7 – 21|
|SNP/PC||2.9||20/4||9 – 40 / 2 – 7|
Other specific predictions
|Conservative Overall Majority||73.4%|
|Exact Seat Tie Con / Lab||0.3%|
|Most Seats in Scotland – Labour||62.2%|
|Most Seats in Scotland – SNP||36.8%|
|Most Seats in Scotland – Tie||1.0%|
I need to do some more work on the Scottish component of the model as I suspect it is over-predicting the SNP share. I will therefore try and get some more data on Scottish polling and see if I can forecast the move from polls to outcome in that area.