So, in the most unpredictable election, we predictably got a result that nobody predicted, but how bad were the predictions?

I look here at the averaged polls, and seat predictions, from May2015.com.

Firstly, the polls. The polls weren’t hugely massively wrong, but they did overestimate support for Labour and underestimate it for the Tories. They were closer with the other parties, but overestimated it for all of them, except the SNP (however, the SNP vote share, though underestimated on a national level, was actually overestimated on a Scottish level — 51.2% predicted, vs 50% actual — I believe this is because turnout was so much higher in Scotland than elsewhere, which pushed the SNP vote share up nationally, even though it was slightly down on predictions in Scotland).

So, how about the predictions? Well, these are worse. They were just about bang on for the small parties, but really underestimated the number of seats the Tories would win, and really overestimated the number of seats Labour would win. They also got it really quite badly wrong on the number of seats the Lib Dems would keep. May2015.com had predictions including or excluding Lord Ashcroft’s polls of marginal constituencies, so I’ve compared these. They also had two different swing prediction models, but these made little difference, so I have only used the “Strong Transition” one.

So, it would seem from this that the predictions were pretty terrible, and whilst the Ashcroft polls led to slightly more realistic predictions for Labour and the Tories, they increased the overestimation of Lib Dem seat counts.

I thought it would then be interesting to see how the prediction models did with the actual results, putting in the actual vote shares, the prediction was closer, but still overestimated the number of seats Labour would win, and underestimated the number of Tory seats. The Lib Dem seats were also still overestimated. This time, the Ashcroft polls made the predictions worse for the Tories (underestimating) and the Lib Dems (overestimating), but better for Labour (overestimated more without Ashcroft polls).

So, it’s hard to tell if including the Ashcroft polls was a good or bad idea. It would be interesting to see how accurate they ended up being for each seat he polled, but that’s too long a task for now. (I should also note that May2015.com says “Ashcroft polls in English and Welsh seats change over time as national polls change”, I’ve never fully understood what this means, I think it means they apply swing in national polls to the results for Ashcroft’s constituency polls, but it’s not clear how this is done.)

The data for all the above charts are shown in this table:

Finally, I thought I’d try and show how overestimated or underestimated the vote and seat share predictions for each party were. This is shown in the graph below, which show the difference, for each party, between their predicted vote and seat share, and their actual vote and seat share.

So, in essence, the polls were wrong, but the predictions were worse.

The data for the above chart are shown in the table below: