Thursday, 5 August 2010

Introducing the Alternative Vote swingometer

I've now written an online version of the calculator from this post, so you can try it out too.

Alternative Vote Swingometer for England (requires Javascript)

Enter the first preference percentages for the three major parties and "others", and then enter the transfer rates. Press the calculate button to give the results. Alternatively, use one of the "quick calculations" links to automatically complete the form and run the calculations based on an opinion poll.

For each party you'll be told:

  1. The number of English seats won
  2. The change from the previous election
  3. The proportion of this change that was caused by changes in first preferences (UPS Component)
  4. The proportion of this change that was caused by the candidate ahead on first preferences losing on transfers (UT Component)

So, for example, in the Populous/Times poll of 23 June, Labour win 198 seats, an increase of 7. On first preferences alone, they gain 19 seats, but lose most of these again after transfers. The UT Component can be viewed as a rough measure of "who benefits from Alternative Vote in this scenario".

Below the summary table is a list of all the constituencies that are changing hands.

It uses approximately the same algorithm as before: various notes are below.

Currently it is of course analysing an election that will almost certainly never happen - if the general election takes place under Alternative Vote, it will also be taking place under new boundaries and for 50 fewer MPs. Once the boundary review is done, I'll switch to the notional figures for the new constituencies.

  • I've added quick links to calculate based on various polls that ask about second preferences.
  • Transfer rates remain stateless - votes will transfer away from an eliminated candidate in the given proportions regardless of their original source. There is no useful third preference polling out there (there's only one useless one, for that matter), so I'm assuming this based on my previous experience of AV counts.

    In practice this will almost always make no difference. "Others" collectively will be eliminated first in all but the weird constituencies. Their votes will transfer almost evenly anyway. The real determinant is the transfers from the third-placed major party in the constituency, almost all of which will be second preferences rather than third or later preferences.

    You will be able to find very marginal constituencies where this assumption being significantly false causes a change in result, but I'd be surprised if you can find one where adjusting the first preference votes of one of the major parties by 0.5% (well inside the margin of error of polling) didn't stop this happening.

    If we had four or five equally-matched parties in most constituencies, rather than two or rarely three, then third preferences would be worth a closer look. For now, they're not worth the effort.

  • Polling (and experience) suggests that between 20% and 40% of votes will be untransferable ("transfer to nowhere") at any stage. You can adjust this rate since it seems to vary between parties.
  • There's some polling now on whether people's AV first preferences would be the same as their FPTP first preferences. In most cases, the answer is 'Yes'. The number of people who do differently is relatively small and the net effect is almost certainly smaller than the margin of error in the first preference polling.
  • Looking at the recent polling, transfers from Others do seem to be split relatively evenly. There might be a slight bonus to the Lib Dems, but it's well within the margin of error.
  • Results for Wyre Forest and Brighton Pavilion (respectively lost and won by 'Others' at the last election) are going to be extremely inaccurate. Buckingham (the Speaker's constituency) is also ignored. Some other constituencies can be won by 'Others' in the right circumstances; take that with several pinches of salt.
  • I'm using proportional swing rather than uniform swing, because that won't give negative vote totals for a party. [Edit 8 April: You can now choose which you prefer. Uniform swing is probably better]
  • There's still no usable polling for Scotland or Wales (and there's never usable polling for Northern Ireland), so for now the swingometer is for English seats only.

I've tried to make it as accessible as I can, but because of the limitations of Blogger it has to be in Javascript, which makes this quite a bit more difficult. If you have problems using it, let me know in comments and I'll try to fix it. Other bug reports and suggestions for improvements are also welcome.