Annoyed by politics

[andrewducker]

I keep seeing articles talking about alliances between the Lib Dems and Conservatives, either for the next election, or for the Oldham East and Saddleworth by-election that's happening a week today. In the latter case, many of the Conservatives basically seem to be saying "We don't have a chance of winning, so you Tory voters should vote LibDem instead, so that Labour don't win."

Not only do I disagree over there being alliances over elections (because it denies people a free choice), but I object to the fact that the current system encourages them. If we had AV then Conservative voters could vote the way they want to (Conservative) and then vote Lib-Dem as a second choice _if that's what they want_. Similarly, Lib-Dem voters could vote Lib-Dem first, and then either Conservative or Labour depending on which they preferred as a second-choice, etc., etc.

That way the parties could concentrate on standing for themselves, and not what other parties are doing, and electoral bargaining could at least wait until _after_ the votes were in.

As it is, the election results won't actually tell us what the honest choices of the electorate are. People will be voting tactically, to keep out the people they oppose, based on guesswork over who has the most chance of winning. It's a horribly broken system.

(Not that I think that AV is the bees knees, but it's decidedly better than FPTP. I think my ideal system would probably be AV with an AMS top-up, but that's a completely different debate.)

Comments

[andrewducker]

Sadly, it fails the "an ordinary person can understand how it works" test, as I've seen repeatedly when people try to explain Condorcet.

[ciphergoth.livejournal.com]

How would you go about testing this assertion? If you were doing a survey to test a particular explanation of STV vs a particular explanation of Condorcet, what questions would you ask the subjects after they'd seen the explanation?

[andrewducker]

I'd ask them to explain it to me.

People seem to get how the comparisons work in AV/STV relatively easily, whereas every discussion I've seen of Condorcet has been full of comments from people getting it wrong, or just being baffled.

(I may be somewhat biased here, as I felt similarly baffled for a long time by it.)

[pete stevens (from livejournal.com)]

Get twelve people to whom you've explained it, give them the set of results for a large election, let them each individually calculate the winner without reference to the explanation. If they all agree then you're in with a chance, otherwise it's hanging chads and lawsuit time.

This is the test of if it came up in court, would the jury be able to get the right candidate elected.

[ciphergoth.livejournal.com]

How would you even *present* the results of a large STV election to your subjects? With Condorcet you just present the win matrix, but for STV?

[andrewhickey.livejournal.com]

List of the candidates, number of first-preference votes received, how the second, third, etc broke down, as a chart.
(Don't have graphviz installed on this computer, but imagine
Bill Bloggs (Birthday) 3000 -> Dave Bloggs (Birthday) 1000
Bill Bloggs (Birthday) 3000 -> Fred Bloggs (Birthday) 1000
Bill Bloggs (Birthday) 3000 -> Alan Smithee (Madeup) 1000

and so on, as a simple DAG. Easier to view than to describe verbally).

[andrewducker]

I'd settle for 50 actual votes. If they can do it with 50 votes then they understand it.

[andrewhickey.livejournal.com]

Oh, I absolutely agree there. And I've *never* found anyone who didn't understand AV/STV style preferentiality when I've explained it. But was answering the precise question of how you present all the results to someone.

[andrewducker]

Which reminds me - with STV, once a candidate is elected, do you just take _new_ votes for them and distribute them to others?

I ask because I can see a problem with skewing the data. If the first half of the votes arrive in from precinct one, where people voted LibDem then Labour, and the second half of the votes arrive in from precinct two, where people votes LibDem then Conservative, then if the first lot are counted, electing the candidate, then the second lot will skew the voting towards the Conservatives. (I know voting is for individual candidates, but I hope you see my point).

How does it work in practice? Get all of the votes in and shuffle them before you start?

[andrewhickey.livejournal.com]

Good question. I would imagine the shuffling or something like it would have to take place, but someone like matgb would know better than I how these things happen in reality.

[andrewducker]

I've dropped him an email...

[andrewducker]

drdoug has handily enlightened us below...

[matgb]

Heh. How complicated do you want it?

Normally, you get someone to do the maths. I've actually done this for committee elections while half drunk, it's a lot easier to do than it is to explain.

Say there are 5 seats, and 599,000 votes. To win, you need 100,000 votes, as if 5 candidates get 100,000 then no one else can get more than 99,000.

Say it's Sheffield at the last GE, when everyone agreed with Nick.

So Nick Clegg gets 250,000 first preferences. Paul Scriven, 2nd LD candidate, gets maybe 10 first preferences and 150,000 of Nick's voters give him their 2nd preference.

You eliminate anyone that's past quota first. So Nick's excess quota votes are shared out.

Now, to ensure it's fair, rather than randomly picking, or using the top of the pile, all votes are redistributed, but are now worth less each.

In this case, Nick got 250K but needed 100K. So all of his 250K votes are redistributed, but each is now worth 150/250 (3/5ths, obviously) of a vote.

So Paul gets 3/5ths of 150,000 added to his score, and the rest go off to wherever they were going to, likely spread around all the other parties.

Essentially, most modern STV systems never remove a ballot from the contention, but each individual ballot is worth less and less as counting goes on.

There's software to do it for you (Colin Rosenstiel wrote one, he's the guy that does all the Lib Dem internal counts), or you can actually do it manually by writing the current value of each ballot on the paper sheet, which is a bit of a PITA but doable.

Very very easy to actually do in practice as long as you can pass remedial maths. Very difficult to explain in words.

[drdoug.livejournal.com]

In the STV elections I've counted, the order of counting votes is entirely irrelevant, and you count up *all* the votes for each round each time. You can't complete the first round until you have all the votes in. (Obviously, you can be pretty confident of how it's going to go once you have 95% of the votes!)

First you count up all the votes (for security checking).

Then you count first preferences, discarding any invalid. Total of valid first preferences gives you N, you calculate the quota, and declare elected anyone over the quota (or eliminate the lowest if nobody made the quota).

When someone's declared elected, you distribute their surplus. You take the whole pile of their first preference votes, and redistribute them to their next-available preference candidates. You don't just add those totals to the other candidates though - you multiply those totals by the surplus divided by the number of votes.

So, say the quota is 1000 votes, and Candidate A gets 1500 first preferences. Their surplus is 500. All 1500 papers are redistributed to next available preferences - say 500 go to B, and 1000 to C. The transfer value (the value of a redistributed vote) is 500/1500, or 1/3. So B's score goes up by 500/3, and C's by 1000/3.

So if you voted for candidate A, it's as if 2/3 of your vote gets counted for them, and 1/3 is distributed to your next preference.

[andrewducker]

Thank you. That makes perfect sense. Takes a little longer, but gives you a better result.

[drdoug.livejournal.com]

You're welcome. If you're really interested, the Electoral Reform Society procedures for the count itself under STV are here:
http://www.electoral-reform.org.uk/votingsystems/stvrules.htm

That's pretty much how I've done it in practice. For big elections, there's a load of counters sitting in a big hall at tables facing the front, and the Returning Officer sits at the front like teacher. A big groan goes up late on in the process when they announce you have to transfer a surplus from a candidate with a huge pile of votes but only a small surplus - you know that transferring their votes isn't going to get anyone else elected, and you'll have to transfer someone else before the count is done.

Happily the number of votes to re-count tends to tail off through the process - lots of people only rank two or three candidates. But towards the end of a big count, you're stuck with the hard core who have listed every single bloody candidate. (Which is what I do if I vote in such elections.)

[pete stevens (from livejournal.com)]

Isn't this a strong argument that STV would fail my jury test if you who understands it aren't entirely certain how to count the ballots?

[andrewducker]

I'd never taken an interest past "and then excess votes are redistributed" before. So I wouldn't say I understood it. Having had the 30 second explanation from two people, it seems trivial.

[andrewhickey.livejournal.com]

No.
There are several good ways to count the ballots, any one of which can be explained in relatively few words (e.g. 'shuffle them before you start'). The only question was which one of these good ways is actually chosen in practice.

[matgb]

Colin does the full tables with all the elimination rounds, here's a fairly good one:
Interim Peers List Top-up Election.

Only 1500 votes as it was voting reps only, small electorate and some of us (*cough*me*cough*) couldn't be arsed voting, but the principle works fine for any number of votes, it's how the LDs do candidate selection for lists elections and similar.

I'm sure the Irish have very reader friendly ways of doing it by now. I probably ought to go look at some point.

[ciphergoth.livejournal.com]

Right, but that presentation basically gives away the answer. Obviously a presentation that gives away the answer for Condorcet could also be devised, but at that point it's not much of a test.

[matgb]

Ah, I was answering an entirely different question.

Not sure how to do it for Pete's hypothetical, but I am sure that it could be done, you might need to give them all the ballot papers, but...