Wednesday 12 May 2010

Alternative Vote

So, we have a Conservative/Lib Dem majority coalition in government now. Part of the deal was that there'd be a referendum on changing to the Alternative Vote (AV) before the next election.

Alternative Vote - called Instant Runoff Voting in some countries - is a good system for electing a single representative. It could be applied to the UK with minimal legislation - no need to change boundaries at all, for instance - and would not need significant changes to the design of ballot papers.

It is not a proportional representation system, and shares most of the same problems of the current "First Past the Post" system for parties without geographically concentrated support.

How bad a deal is this for the Conservatives? It's the minimum offer they had to make - the Lib Dems needed some voting system reform to avoid looking weak to their own supporters, and Labour were offering (possibly with not much credibility) AV and a referendum on proportional representation. It's also, in practice, the maximum offer they could make and still be able to avoid too many defections from their own side.

Looking at the 2010 election, I think they would have been okay - about 20 seats down, which still leaves them the largest party - and most of that change going to the Lib Dems. Enough of a change to make a Lab-Lib coalition viable, but this was an exceptionally close election. In most cases, it wouldn't make much difference, and in some it could benefit them.

An analysis of some scenarios follows below (warning: may cause extreme boredom in people who don't like discussing voting systems), and then a quick guess about what this means for UK politics.

(skip past the boring and speculative bit)

Andrew at Generalising, as part of discussions on "how many seats did UKIP take off the Conservatives?", mentioned this ComRes poll (PDF). The interesting data is on page 21 of the PDF - second preferences of voters for the major parties.

The data for UKIP is based on 19 answers - far too small a sample to be statistically useful - but the transfers for the three main parties are probably usable if somewhat error-prone.

So, I put together a simulator that uses that data to re-run the 2010 election under AV. It only covers English seats (data on transfers to and from the Scottish National Party and Plaid Cymru was too small to be usable). I got the constituency data from the Guardian DataBlog. Here's how it works.

  • Assume that the transfer percentages between the three main parties are accurate. What else can I do...
  • Assume that people who "refused" or "didn't know" wouldn't transfer at all. Specifying of partial preferences is very common in AV, and this doesn't seem unreasonable.
  • Group all "Other" parties together. There are very few English seats where a fourth party or an independent is in contention, and the local situation is sufficently unusual there to make mass predictions pointless anyway. Assume their voters will transfer equally to all three main parties, and so have no effect on the result.
  • Assume that first preference votes don't change. Tactical voting tends to hurt the third and fourth parties anyway, so this shouldn't generally change the results, and AV will unwind the tactical voting anyway.
  • Assume that votes with a particular party will transfer out in the same proportions regardless of their original source. Transfers to parties that have already been eliminated are just discarded. There's no data on third preferences, so this will have to do.
  • Then run an AV count in each constituency.

Here's what it comes up with for the English seats, for a re-run of 2010 under AV:

PartySeatsChange as a result of switching voting system
Conservative276-20
Labour193+2
Lib Dem61+18
Others (Green)10

The Conservatives lose 20 constituencies, slightly more to the Lib Dems than to Labour. Labour gain some from the Conservatives, but lose almost as many to the Lib Dems as well. Here's the full list of changes.

ConstituencyReal resultPredicted under AV
Amber ValleyConLab
AshfieldLabLib
BosworthConLib
Bristol North WestConLib
BroxtoweConLab
Camborne and RedruthConLib
ChesterfieldLabLib
Colne ValleyConLib
Harrogate and KnaresboroughConLib
HendonConLab
Hull NorthLabLib
Lancaster and FleetwoodConLab
Newton AbbotConLib
Oldham East and SaddleworthLabLib
Oxford West and AbingdonConLib
RochdaleLabLib
Sheffield CentralLabLib
SherwoodConLab
St AlbansConLib
Stockton SouthConLab
ThurrockConLab
Truro and FalmouthConLib
Warwickshire NorthConLab
WatfordConLib
Weston-Super-MareConLib
York OuterConLib

That particular change makes a Lab-Lib coalition also numerically viable, though this was an unusually close election to start with.

Having got the simulator working, it's not hard to plug in alternative assumptions.

What if the coalition government collapses, and Labour's vote recovers - in effect, we move back to the 2005 vote shares for England. I'll use Uniform National Swing (UNS) because it's easy to calculate and isn't significantly worse than other models, and because I've been assuming Uniform Transfers anyway. This would probably affect transfer preferences as well, but it's a relatively small change in first preference vote, and those preferences are only approximate anyway.

PartySeatsChange as a result of switching voting systemChange as a result of UNS
Conservative191-27-78
Labour276+7+78
Lib Dem62+20-1
Others20+1

The Conservatives lose most of the gains they made this time round in England, and then lose another 7 (compared with AV with no swing), though one of those is Wyre Forest lost to "Other" which would be unlikely to happen now. Again, the net effect is a transfer of seats from the Conservatives to the Lib Dems.

What if it works, though? Five years of Con-Lib coalition moves the Lib Dems a bit more press coverage - and favourable coverage from the right-wing papers, too. Labour works well in opposition, but the arguments currently ongoing between Labour and Lib Dem activists about whose fault this all is break the two parties further apart, and the "anyone but Tory vote" deserts the Lib Dems. Conversely, the "anyone but Labour vote" strengthens.

I assumed that transfers from Labour to Lib Dems halved, the difference going to 'Others' instead. Similarly, I gave half of the Conservatives' current transfers to 'Others' to the Lib Dems. For Lib Dem transfers, I swapped the Labour and Conservative values (not that big a change).

PartyPredicted voteChange from using AV
Conservative295-1
Labour175-16
Lib Dem59+16
Others2+1

The effect is roughly reversed. The Lib Dems take a few seats from the Conservatives, but only as many as the Conservatives take from Labour. The Lib Dems also pick up a few from Labour, and you can probably ignore Wyre Forest (Other +1) again.

Modifying the above, let's assume that the Lib Dems pick up a few more points of the "anyone but Labour" vote as first preferences that went to the Conservatives in 2010, but that Labour manages to regroup a bit better in opposition too.

PartyPredicted voteChange from using AVChange from UNS
Conservative259+6-43
Labour198-24+31
Lib Dem72+18+11
Others20+1

Labour win a lot of seats from the swing on first preferences, but then lose most of them back after transfers. The Conservatives inevitably lose quite a few seats as their vote share reduces, but gain some back from transfers from the Lib Dems. The Lib Dems benefit from both factors.

It's not all good for the Lib Dems. What if the confusion over who they'd go with strikes again, and the supporters of other parties decide not to transfer to them. In this simulation, I dropped the transfer rate from both Labour and Conservative to the Lib Dems to be not much more than the transfer rate to each other (10% each). It's still enough to change a few close marginals, but that's all.

PartyPredicted voteChange from using AV
Conservative287-9
Labour197+6
Lib Dem46+3
Others10

Combine this with the tactical vote they may have received already backing off, and it looks even worse. I took 5% off their vote, and split it 3:2 to Labour and Conservatives (to roughly match how the second preferences split).

PartyPredicted voteChange from using AVChange from UNS
Conservative293-8+5
Labour216+8+17
Lib Dem210-22
Others100

Labour pick up a few seats from the Conservatives on transfers, but the UNS component means that the Lib Dems really suffer.

What might this mean?

AV is unequivocally an improvement on the current system. It's a largely non-partisan improvement, and so I'd urge you to vote for it if we get a referendum regardless of your affiliation (unless you're a UKIP voter trying to drag the Conservatives rightwards by costing them seats, in which case we're still trying to figure out if you did or not...)

It's also not a very big improvement. Don't expect any more non-default MPs as a result of it, for instance, or results that are more proportional to the vote share (it won't usually make things worse either, at least)

The biggest implication of the change is likely to be that much more intensive polling will be needed, with second preference (and perhaps third) as a routine question. There will be virtually no point to polls covering all of mainland Britain together - I'd expect to see those shrink to England-only, with some having parallel Welsh and Scottish polls. It will certainly make the swingometers more fun.

Alternative Vote raises the stakes. The Lib Dems are probably in the best place to gain from it, taking seats from both Conservatives and Labour. However, there's also usually a noticeable effect where one of the Conservatives and Labour gains seats from the other as a result of Lib Dem second preferences. Which one wins from this depends on how the Lib Dem voters are feeling. Previously they split about 3:2 for Labour.

It's certainly not true that it necessarily harms the Conservatives - indeed, if the Conservative-Labour balance (both nationally and more crucially within Lib Dem voters) swung more towards the Conservatives (as it has been in the past), it could easily help them.

What this means for what people wanting the Conservatives out at the next election should do I don't know. Really, a change from FPTP to AV, though an improvement, is not going to be a major factor.