For softfruit

[ciphergoth]

Another diagram showing the fortunes of the Lib Dems - I had another go to see if I could show more of what's going on.

Again, the position in the diagram (and the percentages) indicate the share of major party vote that each party gets, and the pie charts indicate the proportion of seats that each gets.

Looking at those polls, the Lib Dems had better hope that Clegg is going to completely turn around the fortunes of the party, or they are facing total irrelevance, which is a long way from the bright hopes entertained after the 2005 elections.

Comments

[softfruit.livejournal.com]

That's all true but actually the opposite of what the diagram shows - it's based on "straight swing" while those are the factors (incumbency etc) that mean Paul's chart prediction of 9 Lib Dems on that opinion poll almost certainly underestimates things: people elected in 2001 and 2005 as Lib Dems would gain an incumbency factor and so be less likely to lose their seats.

[zotz]

The Lib Dems were a random example. I don't offhand know what prediction method Paul's using - it must take these factors into account, given the observed asymmetries.

Whether 9 is a plausible figure given the current poll performance is a good question - electoral wipeouts on that scale have certainly happened, although I'm not aware of a British example.

[softfruit.livejournal.com]

The asymmetries are based on the 2005 result; the predictions all are straight 3-way swings, so don't factor in the incumbency factors built up from 2005 to (next GE). Those are hell to calculate anyhow, especially given the boundary changes and unpredictability of which PPCs will be incumbents in areas of the new seats. Arguably they're not the role of a swingometer anyhow - to me after the votes are cast a swingometer should highlight that seat X would have changed hands but for local factors like incumbency.

I'm not sure if he's taking the 2005 results "as they were" or a projection of them onto the new constituencies that the next GE will be fought on, which IIRC shift about ten seats net from Lab to Con.

[zotz]

to me after the votes are cast a swingometer should highlight that seat X would have changed hands but for local factors like incumbency.

Well, the aim should surely to predict an overall outcome, which would mean assuming an overall incumbency effect - in any given seat the question would be whether there's more or less incumbency than average, or indeed a negative effect if the voters have taken against their member.

So what method is actually being used, then?

[ciphergoth.livejournal.com]

The model assumes that the total number of votes cast does not change, that the votes given to all other parties do not change, and that the proportion that a given party achieves in one constituency compared to another doesn’t change.

This isn't quite "straight swing" - uniform swing is mathematically messy with lots of irritating discontinuities. However, I don't suppose it gives radically different results.

[ciphergoth.livejournal.com]

BTW, no, I'm not using new boundaries. Can I download a voting numbers spreadsheet from somewhere which has done this projection, do you know?