How to analyse these is an issue. There is the simple view of allowing each party's share of the vote to rise/fall uniformly. So, if party X's share of the national vote was 20% at the last election, and opinion polls show it to be 15% now, we should expect the share of the vote to fall by 5% across the board. So, if it got 80% in one seat, it should get 75% now. And if it got 4% in another seat, it should get, er, minus 1% now.
Clearly, negative shares of the vote don't happen - not even opposition parties in communist regimes do that badly in elections. So, should we say that party X loses a quarter of its support to bring its national share of the vote down from 20% to 15%? That sounds more logical. In that seat where it got 80%, it should be looking at 60%, and where it got 4%, it shoudl expect 3%.
Most opinion polls have detailed tables under the headline figures, which show where the vote is going. And one common feature is that more people who voted Liberal Democrat at the May 2010 general election intend to vote Labour than remain with the Liberal Democrats, leading to one obvious consequence - the Liberal Democrats would face complete wipeout.
Now, there are several Liberal Democrat seats (Birmingham Yardley; Bradford East; Brent Central; Bristol West; Burnley; Cambridge; Cardiff Central; Dunbartonshire East; Hornsey & Wood Green; Inverness, Nairn, Badenoch & Strathspey; Leeds North West; Manchester Withington; Norwich South and Redcar) which were won from Labour at the May 2005 or 2010 election.
One thing that has been pointed out to me is that it might be sensible to assume that this shift from the Liberal Democrats to Labour in the opinion polls begins with the voters that the Liberal Democrats have picked up from Labour in 2005 and 2010 returning, and hence these are the seats we should be expecting Labour to recapture. While on the other hand, is it reasonable to think that seats like Argyll & Bute, Gordon, or even Sheffield Hallam (whose MP is the Liberal Democrat leader and Lord President of the Council, Nick Clegg) would elect their first Labour MP in 2015?
Although saying that, looking at Edinburgh West, which the Liberal Democrats won from the Conservatives at the May 1997 general election, and where Labour have moved into second place, it is entirely possible that this is a seat that will go Labour for the first time in 2015.
If we take the results of the May 2011 election to the Scottish Parliament and project them onto the Westminster constituencies we actually get the Liberal Democrats losing all their seats to the Scottish National Party, except for Berwickshire, Roxburgh & Selkirk (whose MP is the Scottish Secretary, Michael Moore) falling to the Conservatives, and Orkney & Shetland being held by the Liberal Democrats.
So, how would we model this? The simple answer is with great difficulty.
Suppose at an election, party B increased its vote from 9 million to 10 million, and party C from 5 million to 6 million, and we assume that these 2 million voters were ones that had voted for party A at the previous election (or the election before the previous election - for a reason which I will make clear later, although you might have guessed from the list of seats that were once Labour and now Liberal Democrat).
We are then told that 4% of people who voted for party B are planning to vote for party A. We can make the reasonable assumption that these 400,000 people are from the group of people who previously switched from party A to party B.
Suppose in two seats, party B has 20,000 votes. We could simply transfer 800 of them in both seats to party A. But, let's be a bit more sophisticated, and work on the principle that 40% of those who switched from party A to party B at the last election are planning to switch back to party A.
In the first seat, we assume that 1,000 of party B's voters switched from party A. So, we transfer 40% of these (i.e. 400 voters) back to party A.
In the second seat, we assume that 10,000 of party B's voters had switched from party A, and in this seat we transfer 4,000 of them back to party A.
Now throw into the mix party D, which was a fringe party at the previous election, but is suddenly growing in support. And opinion polls show that 20% of people who voted for party B (that is 2 million people) plan to vote for party D.
So, in these two seats, simply transfer 20% of party B's votes to party D? That would be 4,000 in both constituencies. Or assume that these 2 million voters come from people who did not switch to party B from elsewhere at the previous election?
In this case, we assume that 2/9ths (about 22.2%) of the 9 million people who stuck with party B are now planning to switch to party D.
In our first seat, with 1,000 of party B's voters being switchers from party A, we have 19,000 voters who had stuck with party B. We transfer 4,222 of them to party D. While in our second seat, with 10,000 of party B's being switchers from party A, we have 10,000 voters who had stuck with party B, and we transfer 2,222 of them to party D.
Now assume that we are also told that a third of people who voted for party C now plan to switch to party A. That makes 2 million people. Following the method we used for party B, we can make the assumption that the 1 million people who switched from party A to party C plan to switch back, and 1 million of the 5 million people who stuck with party C (i.e. 20%) now plan to switch to party A.
So, consider two seats where party C got 20,000 votes.
In the first one we assume that 1,000 of these are people who switched from party A, leaving 19,000 who stuck with party C. We transfer the 1,000 switchers back to party A, as well as 20% of the 19,000 who stuck with party C - that is 3,800. So in this seat, 4,800 people who voted for party C plan to vote for party A.
In the second one we assume that 5,000 of these are people who switched from party A, leaving 15,000 who stuck with party C. We transfer the 5,000 switchers back to party A, as well as 20% of the 15,000 who stuck with party C - that is 3,000. So, in this seat, 8,000 people who voted for party C plan to vote for party A.
To put this into practice, we divide the Conservative and Liberal Democrat vote in each constituency, by separating out those who switched from Labour since the June 2001 general election. The reason for this is when we look at the 2005 and 2010 election results, we have Labour losing office in a two stages.
The first stage, in 2005, sees the significant rise in Liberal Democrat support and seats. While 2010 sees the Conservative vote rising significantly (and the Liberal Democrat vote only rising a small amount).
Of course, this method cannot deal with voters who transferred from parties other than Labour - so people who switched from the Liberal Democrats to the Conservatives in 2010 are simply thrown in the group of people who stuck with the Conservatives. To deal with these combinations would involve an extra set of assumptions and spreadsheets.
In addition, to do this we would need the 2001 election results projected onto the current constituencies. For Scotland this is easy, as the current constituencies are the same as the 2005 ones, and such data are available. For England and Wales, this will be a lot trickier, but would be an interesting challenge.
There will also be seats where we cannot divide the Liberal Democrat vote this way. For example, Chesterfield, which Labour won from the Liberal Democrats at the 2010 election, is a seat where the Liberal Democrat vote has fallen since the 2001 election; or Wyre Forest, which the Liberal Democrats did not contest at the 2001 or 2005 elections.