Yesterday I predicted a 92% chance of a Conservative majority, which we now know has not happened. I over-predicted the number of Conservative seats, and under-predicted Labour seats (by about 25 seats in both cases). The question then is why – is there a problem with the model, or the polling data that it was based on?
The answer, of course, is a bit of both.
Running my model with the actual national voting shares (with 645 seats declared) gives the following result:
- CON 331 +/- 19
- LAB 253 +/- 20
- SNP 39 +/- 6
- LD 7 +/- 2
Which is the prediction I would have made if the polls had been perfect, along with a 63% chance of a Conservative majority. This is closer, but of course still not quite right. So what’s wrong with the model?
The model applies a uniform swing in votes to every constituency – assuming any national swing is reflected exactly in every seat. This is obviously not the case, with large variations in swing across the country, such as 10% to Labour in Canterbury.
There will be a lot of analysis over the next few days discussing why people voted the way they did – such as Labour doing well in Remain areas, and the Conservatives in Leave areas.
But of course the first question is who will form a government, and that’s a matter of politics, not data, so I can’t help you there!