[EM] IRV vs Plurality

[Abd ul-Rahman Lomax]

At 09:23 AM 1/8/2010, Stéphane Rouillon wrote:
> > Therefore IRV/STV is no better than plurality, but has extra very
> > serious flaws, inequities, and vagaries that plurality does not have.
>
>I definitively disagree. Plurality is worst than IRV.
>The flaws that IRV does have are real.
>But these problems appear very less often than 
>the splitting-vote issue of FPTP.

Stephane, as to abstract theoretical voting 
systems, naively analyzed, and also as to certain 
real-world situations -- but not others -- you'd 
be correct. But notice that Kathy Dopp claimed 
that IRV is "no better than plurality." That's 
because, in nonpartisan elections, it appears 
that IRV closely reproduces the results of 
plurality. We have tended to think in terms of 
neat factions, arranged in a spectrum, so that 
you can predict vote transfer patterns with IRV, 
but nonpartisan elections don't work that way.

Generally, in nonpartisan elections in the U.S., 
vote transfers with IRV do not alter the 
preference order among the remaining candidates. 
Exceptions may occur when races are very close.

On the other hand, in one-third of nonpartisan 
top-two runoff elections, which IRV supposedly 
simulates, the runner-up in the primary goes on 
to win the runoff, a "comeback election," 
according to a FairVote study. It simply does not happen with IRV.

If you have top-two runoff as a system in use, 
and you replace it with IRV, for nonpartisan 
elections, you might as well replace it with 
plurality, you will get the same results. That's what is being said.

The recent election in Burlington, Vermont, 
though, was a partisan election. There, Kiss was 
trailing Wright in first-preference votes, but 
Kiss obtained enough vote transfers from Montrose 
supporters to pass up Wright in the second round 
of counting. Kiss is Progressive, Wright 
Republican, and Montrose is a Democrat.

But looking at the actual voting data, which is 
available, we can see that Montrose was, in fact, 
the Condorcet winner, and, as it's been pointed 
out, had a few of the Write supporters voted for 
Montrose in first place instead of in second, 
Montrose would have won. In other words, IRV will 
punish you (as does plurality) for voting your 
conscience; but with Plurality, it's obvious and 
everyone would know that voting for a Republican 
in Burlington would be a wasted vote (where the 
leading party is Progressive), so they'd have 
compromised and voted accordingly and Montrose would quite likely have won.

Also, there is good reason to believe that most 
voters would vote according to the same patterns 
if the method were Bucklin. The ballot would have 
been the same, three-rank. With Bucklin, first 
round results would have been same as IRV, 
presumably (and assuming that nobody did, with 
IRV, vote strategically already, we can assume 
that with the limited experience with IRV, few 
would have known to do so). Data is from a quite 
good video Kathy Dopp pointed to, http://www.youtube.com/watch?v=bPCS-zWuel8

candidate       1st     2nd
Montrose        2554    3556
Kiss            2982    1827
Wright          3297    1138
------------------------------
                 8843    6521

The ballots show third rank data, but my view is 
that this isn't meaningful, many voters may 
actually be thinking that they are voting 
*against* a candidate by ranking in third place. 
(There were other minor candidates on the ballot 
and the data in the video is obviously 
oversimplified, but it serves as an example.)

As you can see, no candidate gained a majority in 
first preference. There is serious vote-splitting 
between Montrose and Kiss, quite likely. With 
IRV, Montrose is eliminated before the second 
rank votes for him are counted. That's 3556 votes that weren't counted!

With Bucklin, all the votes are counted up to the 
ranks necessary to find a majority. The majority 
is 4422. Adding in the second rank votes, we get

Montrose        6110
Kiss            4809
Wright          4435

It's not even close! Montrose is the first or 
second choice of roughly three-fourths of the 
voters. This is Bucklin voting, supremely easy to 
count, just add up the expressed preferences at 
each rank. It's Instant Runoff Approval.

It's true that there might not be such heavy 
usage of second rank with Bucklin (though already 
2312 voters "truncated," not expressing a second 
preference). However, there are two possible ways to use Bucklin.

We can generally assume that the votes in the 
Burlington election were sincere. They might not 
stay that way if Burlington Republican voters 
realize they've been had. Because there are no 
candidate eliminations in Bucklin, though, 
supporters of minor candidates can safely vote 
their conscience in first rank, because their 
vote will either help their candidate win 
(unlikely by the conditions) or will cause 
majority failure or will be moot in any case. 
There is no need for Favorite Betrayal, as it's called.

What we have in Burlinton is a three party 
system, with the Republicans being, slightly, the 
largest. Naturally, they might prefer Plurality, 
except that they know they won't win, because 
they'd need more than a third of the voters. I'd 
expect Burlington to see a lot of runoffs if top-two runoff is used, straight.

But consider top-two runoff with Bucklin used in 
the primary (and I believe that it would be wise 
to allow write-ins in the runoff and use Bucklin 
there too to prevent the spoiler effect).

The voters would have -- would learn that they 
have -- a choice: add second rank (or third rank) 
votes if you approve of additional candidates, 
even though you have some stronger preference, or 
see a runoff election. The circumstances actually 
encourage a form of range voting, whether or not 
you'd add a second or third rank vote depends on 
*how much* you prefer your favorite over the 
others. This would amalgamate to show average 
preference strength against an actual 
inconvenience. In Bucklin, it's true, if your 
favorite doesn't win in the first round, your 
second rank vote can cause your favorite to lose. 
IRV allows you to think you are avoiding this 
possibly undesirable outcome, but only because it 
takes your candidate and eliminates him.

Were the Wright voters in Burlington happy 
because their vote for Wright was "protected" from "hurting" Wright?

The 2009 Burlington outcome was truly outrageous, 
and the votes show it. It was a classic center 
squeeze situation, and the possibility of this is 
precisely why Robert's Rules of Order criticizes 
IRV and considers true repeated balloting 
(without eliminations!) superior. RRO doesn't 
consider other forms of preferential voting 
though it notes that they exist. I understand 
that this is because RRO is a manual of actual 
practice, not of theoretical recommendations, but 
there are much, much better voting systems.

Bucklin, to me, has these advantages:

1. It's been widely used in the U.S., about 
eighty years ago. It was very popular, and much 
more widely used than the current IRV fad. Why 
was it dumped? Good question. I wish I knew. Most 
likely answer: it worked, and some people didn't 
like that, such as the Minnesota Supreme Court.

2. It's cheap to canvass. Just add up votes, no 
complicated handling, totals can be summed by precinct easily and transmitted.

3. It preserves the ability to vote for more than 
one candidate but simultaneously indicate 
preference, unlike Approval. (Bucklin is really 
Approval voting with a "virtual runoff" feature, 
so that approvals are added in as needed.)

4. It satisfies the Majority Criterion, which is 
politically desirable. It does not satisfy, 
technically, the Condorcet Criterion, though my 
sense is that Condorcet failure would be rare and with low preference strength.

Bucklin would have allowed the Republican voters 
in Burlington to vote for Wright without 
suffering the consequential loss of their second 
choice to their lowest preference. Someone should tell them!

I think it's worth looking at how voting strategy 
might work. Some candidates might encourage their 
supporters not to add lower ranked votes for 
their major opponent. But we already see that 
many of the voters in Burlington did not vote the 
standard politically predictable patterns. We had 
some Wright supporters voting second rank for 
Kiss. Did that mean that they really preferred 
Kiss to Montrose. Maybe. Or they believed that 
this would somehow help Wright. Likewise, we had 
Kiss voters voting second rank for Wright. But in 
both cases these numbers were fairly small.

I would indeed expect truncation to increase a 
bit with Bucklin, maybe even a lot. However, not 
enough, I'm practically certain, to alter the 
result. Second rank voting would have had to 
decline by 1689 votes for Montrose not to gain a 
majority, from his 3556 second-rank votes as 
shown. He'd still have a plurality. If a majority 
were required, he'd be in the runoff, certainly 
(whereas with a vote-for-one primary, he might be eliminated).

If runoffs are held when there is majority 
failure, voters should know that they should not 
vote for a candidate, at any rank, unless they 
prefer the election of that candidate to a runoff 
being held (with its costs, inconvenience, and 
risks). Voters should also be able to leave lower 
ranks blank, deferring the counting of a lower 
ranked vote until later in the process. (It's a 
little more protection against "harming your 
favorite.") They should also be able to vote for 
more than one candidate at any rank, for reasons 
I won't explain here, but it is a good strategy 
if you really don't have a strong preference 
between two candidates. But, of course, they 
should never be able to vote more than once for 
any given candidate, should they mark the same 
candidate in lower ranks, those additional marks 
would simply be disregarded, they should not 
invalidate the ballot. A vote for a candidate 
will be counted at the highest rank expressed....