Posted by Simon Hix and Nick Vivyan.
We have updated our election prediction model, based on national voting intention results from all polls with fieldwork up to and including 26 April. Here is a figure showing the time trends in our “pooling the polls” analysis. The dots in the figure show the results from the various polls and the shaded areas around the lines are the 95 per cent confidence intervals around the mean standings of the parties.
As of 26 April, the national standing of the parties was 33.3 per cent for the Conservatives (up 1.0 per cent from our 19 April analysis), 26.5 per cent for Labour (down 0.3 per cent), and 29.3 per cent for the Lib Dems (down 0.8 per cent). So, in relative terms the Tories may have gained very slightly since our 19 April analysis, but all of these changes are within the margin of error.
Here’s how these national vote shares might translate into seats under three different assumptions: (1) a uniform change in party support across constituencies; (2) differential changes in party support in each region of the country, based on the latest regional polling data from YouGov; and (3) differential changes in party support in key marginal seats, using the latest data from Ipsos-Mori polling data in 57 Labour-held constituencies where the Conservatives need a swing of 5-9 per cent to win.
Hix-Vivyan Seat Projections for 28 April
Hix-Vivyan pooling-the-polls model | Comparison with other predictions | ||||
---|---|---|---|---|---|
Assuming a uniform national swing | Assuming different regional swings | Assuming different marginal seat swings | UK Polling Report Seat Projection | Electoral Calculus Seat Projection | |
Last Checked | 28 April | 28 April | 28 April | 27 April | 27 April |
Conservatives | 265 | 278 | 266 | 283 | 265 |
Labour | 255 | 223 | 248 | 238 | 260 |
Liberal Democrats | 98 | 117 | 103 | 97 | 94 |
Other Parties | 32 | 32 | 33 | 32 | 32 |
No. of seats Cons short of a majority | 61 | 48 | 60 | 43 | 61 |
No. of seats Labour short of a majority | 71 | 103 | 78 | 88 | 66 |
With a week to go until polling day, the Conservatives look set to be the largest party in the Commons but well short of a majority of seats.
I’d be very interested to see what your opinion is of the work that Nate Silver and Renard Sexton are doing at fivethirtyeight.com. They very accurately projected the results of the US election in 2008, and now they’re trying their hand at the one here. Currently their model is projecting Conservatives with 299 seats, Labour with 199 and libdems with 120, they are also trying to get away from a uniform national swing calculation, though I think they do it a bit differently from you… The URL is: http://www.fivethirtyeight.com/2010/04/uk-seats-projection-tories-299-labour.html
Can you provide a detailed description of how your Hix-Vivyan model works? From a previous post it seems that you sample from various election scenarios compatible with the overall vote shares indicated by the polls, and then extract seat predictions from aggregating constituency results. I encourage you to share the method.
Also, it would seem an obvious thing to include margins of error, or confidence intervals on your seat number predictions. If your predictions come from averaging (mean, median?) simulations then you must know your error estimate already.
It would also be interesting to see typical scenarios sampled from your model, and not just the average of the results.
Thanks. That’s extremely helpful to those of us who make a living from betting on politics.
I am struck by how close Labour remains to snatching Most Seats and also by the fact that the betting markets appear to be underestimating the likely number of LD Seats in te new Parliament.