Saturday, 19 February 2011

Arguments against Alternative Vote

Since people keep finding my earlier post on AV and proportionality when searching for "arguments against alternative vote", I thought it was probably time to actually set out the list of those arguments. (Arguments in favour of AV get a separate page, though in some cases whether an argument is positive or negative depends on what you're looking for in a voting system)

You should probably skip the rest of this post if the merits of various electoral systems don't interest you.

For each point, I'll describe a problem with Alternative Vote (AV) (which might not necessarily be a problem for you personally), and say which other voting systems can be used to avoid this problem.

Since a lot of the recent interest is probably to do with the upcoming referendum, I'll also explicitly say how the other referendum option, First Past The Post (FPTP), does on the same problem.

All voting systems are inevitably flawed in some way - there is no one perfect correct way to count votes. So picking a voting system is about considering the problems of the possible voting systems, and ignoring those that you don't think are problems.

I also list a few things that are not arguments against AV, but are sometimes presented as such.

AV is not proportional

Because Alternative Vote is only suitable for electing a single candidate, to elect multiple candidates to a body, to use AV you need to divide the electorate up into multiple constituencies, each electing only a single candidate.

This operation means that generally the strongest one or two parties get by far the highest number of seats, with other parties getting far fewer or even none. The distribution of the voters within the constituencies can also make a large difference, with parties with geographically-concentrated support doing better than parties with geographically-separated support.

Two parties could get identical numbers of first preference votes (and indeed second and further preference votes) across the electorate as a whole, but could get vastly different numbers of seats.

In the UK this effect significantly benefits Labour and the Conservatives (by a similar amount), and harms every other party.

To avoid this, use...

The only sensible voting systems that can avoid this problem are ones using constituencies returning multiple candidates. Either Single Transferable Vote or one of the List Voting systems will give a result where the number of candidates a party gets is roughly proportional to the number of voters for that party.

What about FPTP?

I cover this in far more detail in the earlier post. In summary, neither AV or FPTP are proportional systems. Which gives the result closer to proportionality will depend on fine details of party affiliation, geographical distributions, constituency boundaries, and so on - and will probably vary from election to election anyway.

There is no reason to pick AV over FPTP, or FPTP over AV, on the grounds of proportionality - if you don't want proportionality, pick either. If you do want proportionality, pick a different system if you can, or find another way to decide if you can't.

Let's call this one a draw.

AV is not monotonic

Monotonicity is an obscure property of voting systems, but it's generally considered quite important. It states that, if candidate A is currently winning the election, and a number of voters change their votes to be more favourable to candidate A, this should not cause candidate A to lose the election.

Or, more simply: getting more votes shouldn't make one's situation worse. "Monotonicity" is also called "Non-perversity", for obvious reasons.

AV, unfortunately, is not monotonic. Consider the following small election with three candidates - Alice, Bob and Cath - and a few voters. To make the example simple, all the voters follow one of three patterns for their ballot papers.

Here are the votes
Number of ballot papersFirst preferenceSecond preferenceThird preference
10 votersAliceCathBob
9 votersBobCathAlice
8 votersCathAliceBob

There are 27 votes, so 14 votes are needed to be elected. No-one has 14 votes yet, so Cath, who has the lowest number of votes, is eliminated from the election. The second preference of Cath's 8 votes are looked at, and they're all for Alice. The votes are added to Alice's total, and on the second round of counting she is ahead 18 votes to 9, which is enough to win the election.

But what would have happen if Alice's campaign had been a bit more effective, and two of the Bob voters, rather than putting Alice last, had put her first, Bob second, and Cath last?

The new vote totals are
Number of ballot papersFirst preferenceSecond preferenceThird preference
10 votersAliceCathBob
2 votersAliceBobCath
7 votersBobCathAlice
8 votersCathAliceBob

Still no-one wins on first preferences alone, but rather than Cath being the last-placed candidate, Bob is. So Bob is eliminated instead, and those 7 votes get transferred to their second preference candidate - Cath.

Cath now wins by 15 votes to 12 on the next round of voting. By getting some voters to give her a first preference rather than a third, Alice has ended up losing the election!

How big a problem is this in practice?

It sounds very serious, but fortunately, in practice, this doesn't happen that much. (If it did, no-one would ever use Alternative Vote).

[Edit 7 April: For a more detailed analysis of how common non-monotonicity is, see the later post. You can also use the swingometer to find out which seats might be affected.]

In almost all cases, getting more votes will improve a candidate's chances of winning the election.

The major reason that it isn't a problem is that it's both rare and very unpredictable. Unless you have almost perfect polling, seeing when an election might give a non-monotonic result is very difficult. In the situations where your candidate ends up in a non-monotonic 'trap' and loses, it's possible to get out of the 'trap' by picking up extra votes or by losing some votes - and it's always safer to try for even more votes.

The consequence is that it's a rare election that's close enough to have this behaviour at all, and an exceptionally rare election where another candidate can actually take advantage of it. (The exception is when the vote is not a secret ballot, and later voters can incorporate earlier votes into their plans - in this case, a sufficiently quick-thinking candidate might just be able to use this)

To avoid this, use...

...just about anything else. Single Transferable Vote has the same problem - though even less predictably. Every other voting system that I know of is monotonic.

What about FPTP?

FPTP is monotonic. AV is non-monotonic in rare situations. If monotonicity is important to you, FPTP has a narrow advantage here.

AV allows tactical voting

'Tactical' voting is where a voter completes their ballot paper in a way that is not their real preference order, with the aim of getting a better result than they would have done if they were honest.

For instance, in the election above where Alice wins, Bob's voters generally do not like Alice. If some of them put Cath rather than Bob as their first preference, then Bob still won't win, but neither will Alice, and Cath is at least preferable.

It's easy in theory, but it's rarely actually done in practice, because it means deliberately throwing away a good position in the election to benefit a different candidate.

So, for instance, in a Labour/Conservative marginal in the UK, where Labour are going to be ahead on first preferences, and the transfers of votes from Lib Dems and minor parties are not predicted to close the gap completely, people who would "honestly" put the Conservative candidate first might instead put the Lib Dem candidate first, and push the Conservative candidate down into third place. Conservative transfers of second preferences are likely to be overwhemingly in favour of the Lib Dem rather than the Labour candidate, so the Lib Dem wins rather than the Labour candidate.

(This example works equally well with Labour and the Conservatives reversed, by the way).

That's the theory. In practice, though, it only works in seats which are extremely marginal to begin with. In those seats, the choice is between:

  • throwing away a good second place (or even first place if your polling is a bit off), and so letting the third-placed candidate win.
  • campaigning harder for second preferences from the third-placed candidate and winning on transfers yourself.

It's generally considered better to lose narrowly in second place - and so be a definite contender for the seat next time, with all the central party resources that implies - than get knocked back into a distant third place and have the party focus its resources elsewhere next time.

It's also really difficult to campaign for - Labour candidates going around saying "Vote Lib Dem to keep the Conservatives out" are going to get very odd looks (and probably thrown out of the party, too).

To avoid this, use...

It's very difficult to find a voting system which isn't at all susceptible to tactical voting. For single-seat constituencies, Condorcet is the best option. For multiple-seat elections, any PR system will be almost immune, but not completely.

AV isn't bad for single-seat elections, though - I think only Condorcet is more resistant.

What about FPTP?

Both FPTP and AV are susceptible to tactical voting, but FPTP has it far worse.

In AV, the tactical vote is to throw away a possible first or second place to improve the chances of a somewhat acceptable winner coming up from third. It's very difficult to do correctly, and no-one is likely to try.

In FPTP, the tactical vote is to throw away a possible fourth or fifth place (or weaken a third place), to improve the chances of the better of the top two. It's relatively common and somewhat effective.

If you want to avoid the need for tactical voting, AV is by far the better of the two.

AV is biased against extreme candidates

Is this a flaw? It is if you are extreme.

Extremist candidates are not going to get 50% of the first preference votes (if they were doing that well, it would be somewhat misleading to call their political views extreme in the context of that election). Because they're so extreme, though, it's also very rare that anyone will give them a decent preference for transfers, which means that they're very likely to go out of the election in an early round.

To avoid this, use...

Any proportional representation system, except those which require a high minimum percentage of the vote to get any seats.

What about FPTP?

FPTP and AV both have a tendency to benefit parties around the centre, but these tendencies behave in different ways.

FPTP benefits the two largest parties near the centre (in the UK, Labour and the Conservatives), and makes things very difficult for anyone else.

AV also benefits those parties, but gives a smaller benefit to parties between the two biggest (such as the Lib Dems).

Its effect on parties that are not extremist but are outside the space between the big two is hard to tell, but the evidence suggests that AV would make things a little easier (though not much) for most left and right parties such as the Greens and UKIP, but would make things much harder for extremists such as the BNP.

Which of the two systems you prefer on this basis probably depends on who you consider extremist, whether you are one yourself, and whether you think having a few extremists elected is good or bad for the government as a whole.

Selecting a preference order is complex

It's certainly true that numbering candidates in order of preference is more complicated than just picking a single favourite candidate.

Depending on the design of the ballot paper, this may also cause accessibility problems.

To avoid this, use...

Either a non-preferential system such as FPTP or a List system, or a system such as Approval Voting where voters place identical marks by all the candidates they approve of.

What about FPTP?

FPTP ballot papers are a bit easier to complete. Whether this matters to you depends on how difficult you think the electorate are likely to find "number the candidates in order of preference, starting with '1' for your most preferred candidate.", and how much impact you think voter education is likely to have on this.

In an optional-preference AV system, which is the one proposed for the UK, the problem is less severe as there's no need for voters to specify more preferences than they want to (and they can only specify one, if they want).

False arguments against AV

Edit: 2 April 2011 - since this post was written, another false argument has come up: Would elections be decided on BNP second preferences? (My conclusion: no more than they are now)

AV requires voting machines or other electronic or mechanical voting assistance"

No. AV was invented and used long before any of that technology existed. Most of the AV elections I've participated in have used normal paper ballot papers, counted by hand, without any problems at all.

"AV is complex to count"

No. An AV count will usually take longer than a FPTP count unless the result is very obvious, but in practice not by much. It will be considerably quicker than the count for most other election systems except List systems and FPTP. Also, while the count will take slightly longer, the counting process is not significantly more complex than that for FPTP.

I've participated in several AV counts, and been in charge of a few. While the voting process is a little more complex, the counting process is not.

"AV gives some people multiple votes"

No. Alternative Vote is intended as an alternative to holding multiple rounds of run-off voting, by collecting all the information needed on a single ballot paper. Rather than holding the votes on separate days, requiring multiple rounds of elections, and delaying the final result by weeks, the voters preferences are collected, and used to find out what would have happened - the ballot paper acts as a proxy for the voter.

(In US English, Alternative Voting is called Instant Runoff Voting, which makes this somewhat clearer)

A typical ballot paper might read as follows (the voter has chosen not to use all their preferences):


This means:

  1. In the first round of voting, vote for Cath.
  2. If there's a second round of voting, and Cath hasn't been eliminated, vote for Cath again. Keep voting for Cath if possible.
  3. If Cath gets eliminated, vote for Dan instead in the next round.
  4. If Cath and Dan both get eliminated, vote for Alice.
  5. If Cath, Dan and Alice all get eliminated, don't bother showing up for the next round of voting.

Rather than the voters having to go back and forth to the polls every week until someone gets 50% of the votes, the voters fill in a ballot paper that acts as their proxy in the voting. That ballot paper will then vote in every round it can until either someone gets 50% or it runs out of instructions.

Every voter gets one vote each round, if they want. (People who don't specify all their preferences won't get a vote in future rounds once their specified preferences run out - which might never happen, of course - just as people who don't show up to vote in the later rounds of a run-off election don't get a vote)

AV doesn't give some people multiple votes - it gives everyone one vote, but means that people who are happy with the way their vote is going don't need to go back and keep saying that.

AV will lead to permanent coalition government

No. It's difficult to predict the effects of a change in voting system, but the only recent UK general election in which AV would have been likely to give a coalition government and FPTP didn't anyway is the 1992 general election (and given that Major's Conservatives lost their majority by the end of that Parliament anyway, it might be more accurate to say that AV would only have given a coalition sooner in that election).

Australia, which uses AV, has had considerably fewer coalitions1 than the FPTP-using UK. Conversely Canada, which uses FPTP, has had considerably more coalition governments than either the UK or Australia.

Whether AV or FPTP leads to coalition governments depends far more on the strength of the third and fourth parties than on which of the two systems is used.

AV would make coalition governments marginally more likely in the UK, but only very marginally. Majority Conservative or Labour governments would still be the norm.


1 Edit 17 March: As Tim Roll-Pickering points out in comments, this depends on what you count as a coalition. To be more specific, short-term alliances between parties for the duration of a single Parliament, needing to be agreed only after the votes are cast and the majorities known, are not significantly more likely under AV in the UK context. Long-term alliances made possible by a change in voting system are a different matter, and beyond prediction for now.