Four things have changed in the polls – and in election predictions – since this site launched in early September.
Labour lost their 3-4 point poll lead and are now resolutely tied with the Tories; Ukip have gradually dipped since November, but are still set to win 4 million votes; and the Greens nearly caught the Lib Dems in the polls before fading.
But by far the most significant change has, of course, been in Scotland. If the SNP surge had never happened, Labour would be set to win more than 300 seats and take power in May. Instead, we are predicting the SNP will win 55 seats, and an average of different forecasts hands them 46. In 2010 they won 6. There are only 59 seats in Scotland.
Few pundits believe these predictions.
Few pundits believe these predictions. Most people hesitate to give the SNP more than 30 seats (as a recent survey of hundreds of political academics proved). Very few can conceive of more than 40. And almost no one predicts the SNP will win at least 50.
And yet no poll has implied the SNP will win as few as 35 seats, and the vast majority suggest they will win more than 40. 18 Scotland-wide polls have been published over the past five months, and they have almost all suggested the same thing: the SNP will win 45-50 seats, and Labour will lose around 30 of its 41.
Labour are collapsing in their heartlands
At their most favourable, the polls suggest Labour will only lose half their Scottish seats. But no poll has been so favourable to Labour since February began, and constituency polls published since then have suggested that national polls are underrating the scale of Labour’s collapse.
Lord Ashcroft has polled 40 per cent of Scotland’s seats over the past two months. His polls have been devastating for Labour. He has polled 19 of Labour’s 41 seats. 16 of those polls have been in the harder half of seats for the SNP to win – the ones where Labour are protecting majorities of at least 29 points (for comparison, Ed Miliband won his seat, Doncaster North, by 26 points in 2010).
Ashcroft’s polls suggest Labour will lose also but 3-4 of their seats.
14 of those 16 polls have put the SNP ahead, all by at least 3 points (in other words, almost all are outside the margin of error). These polls imply Labour will lose all but 3-4 of their seats. That is significantly fewer than the 9-11 seats that national polls imply Labour will hold.
In other words, the swing to the SNP in Labour’s heartlands is even greater than the swing in national polls. The swing in national polls is 20.5 points, which means any Labour seat with a majority of less than 41 per cent would turn SNP on an uniform swing. 
But Ashcroft has shown an even greater swing than 20.5 points in the 19 Labour-held seats he’s polled. The swing he’s found has been just under 25 points, which would wipe out all Labour seats where they hold majorities of less than 50 points (David Cameron won his seat, Witney, by 39 points in 2010). The graphic below shows the percentage point difference between what national polls imply for the SNP in every seat and what Lord Ashcroft’s polls have actually shown.
The SNP are, on average, doing eight points better in Labour-held seats than national polls imply. In other words, the swing is four points greater, hence it is nearly 25 points. This is laughable. It’s the kind of swing one might type into our seat calculator for fun, along with predictions of a Green majority or extra seats for the Lib Dems. Yet this is what almost every poll has implied, and Ashcroft’s polls have more than confirmed.
National polls imply Labour will hold 9-11 seats. This underrates the swing in Labour heartlands.
This is why May2015’s election-forecasting model suggests the SNP will win 55 seats in one month, as it has since Ashcroft first polled Scotland in February (we actually predicted 56 seats at first – a figure that was later widely reported). That prediction is based on both Ashcroft and national polls. We use Ashcroft’s numbers in the seats he’s polled, and otherwise use a variation on uniform swing for the other 35 seats in Scotland.  If we ignored Ashcroft’s polls, we would be forecasting 48-49 seats for the SNP.
Why we’re confident in SNP 50+
We are forecasting an extra 6-7 wins for the SNP because of this extra swing in the heartlands, and we are increasingly confident in doing so. Ashcroft’s polls are not the only evidence of this.
John Curtice, who runs the exit poll room on election night and is the BBC’s chief psephologist, has used recent national polls to suggest the same thing. He has been given access to the disaggregated results of these national polls – in other words, he can break down the national poll into regions. By doing this, Curtice has also argued the swing in Labour’s heartlands is even greater than swing shown by national polls.
John Curtice, the UK’s pre-eminent psephologist, has used recent national polls to suggest this too.
He recently went so far as to suggest Labour could lose all but two of their seats, and end up with as few as the Lib Dems and Tories. (As we speculated a few weeks ago.) Academic forecasts have also been inching up towards 50 SNP seats since January. Elections Etc – the model created by Oxford psephologist Steve Fisher, another of the hallowed few that run election night’s exit poll room – now predicts 47 seats for the SNP.
Election Forecast ‘only’ predict 42 for the SNP, but that’s because they are discounting current polls based on how UK-wide polls have moved in the past. But there is little basis for doing this. The SNP surge is a post-referendum, once-in-a-generation change. It is an out of sample event. It is a stretch to suspect the SNP will fade because smaller parties have faded in UK-wide elections before.
If Election Forecast offered a ‘nowcast’, and just projected current polls into an election result rather than predicting how the polls will change, they would be predicting more than 50 seats for the SNP. The Guardian, whose model works on very similar principles to ours, already do. As for the bookies, in January they were predicting just 25-26 SNP seats. Now they are forecasting 42-43 (so much for following the money – the money moved as polls and predictions moved). We’re still taking the ‘over’ on that bet.
 The SNP are polling 26 points higher than in 2010 and Labour 15 points lower = 20.5 point swing.
 Because Scotland polls have scarcely moved in the past two months and there are comparatively few of them, Ashcroft’s polls are ‘static’. We do not change them in-line with changes in national polls, as we do for Ashcroft seats in England and Wales.