Thoughts on Maine IRV

[Clay Shentrup]

So, this ballot initiative is apparently going to be in 2016. I'm curious to hear people's thoughts.

On the plus side, Maine currently has *plain Plurality Voting* for governor, and has thus elected one of the country's craziest elected officials twice in a row—meaning that IRV here would obviously be a far more substantial and obvious improvement than in the other areas where it has been adopted, as a replacement for delayed runoffs.

On the minus side, its proponents are telling all the standard lies about it:

> Ranked choice voting, also known as instant run-off voting, has all the advantages of an actual run-off election

Obviously false. E.g. the immunity to tactical behavior in the second round for instance. Or precinct summability (within each round at least).

> if their 1st choice does not win, their vote will go to their next choice.

Not if your next choice was eliminated *before* your 1st choice.

But then they go into incredibly frustrating statements like this:

It minimizes the “spoiler effect.” A candidate who receives fewer votes than others does not “spoil” the outcome of the election, because there is an instant run-off among the candidates who receive more votes.

They started with "minimizes" the spoiler effect instead of "eliminates"—which makes you think they're going to maybe be the first IRV proponents to actually make SOME factually accurate statements. However, they then effectively negate that in the next sentence. And that pattern continues in the page.

But, ignoring the extreme dishonesty of its proponents, should we generally view this as an improvement? What chance is there that this actually HURTS voting reform by leading to a big non-summability mess and then getting repealed a few years later?

I tend to think there is some value in having statewide use, because it gets the whole issue of alternative voting methods in the public consciousness.

[Clay Shentrup]

Here's there web site.

http://www.rcvmaine.com/

[Warren D Smith]

In Maine, our own polling study and other polls indicated that in the
last gov election 2014,
won by LePage, the winner probably would have been Cutler using IRV,
approval, borda,
condorcet, or score voting.

In the election before that, also won by LePage, Cutler probably would
have won with approval or IRV.

In both cases Cutler ran as an independent and LePage as a republican.
LePage has earned a reputation as being pretty nuts.



--
Warren D. Smith
http://RangeVoting.org <-- add your endorsement (by clicking
"endorse" as 1st step)

[Warren D Smith]

I think IRV is an improvement versus plain plurality (at least if
complexity ignored).

If FairVote is pushing IRV for Maine gov, that may be the first (?)
time they actually tried to replace plain plurality with IRV,
which hence may actually be the first time they actually
tried to improve a voting system.
Yay. Unfortunately if Maine adopts IRV then quite likely based on
history and polls
they later will repeal it and revert to plain plurality.

[Rob Wilson]

It is really frustrating that IRV is the only option being voted on, but I if I were in Maine, I'd support it. We can't continuously keep trying to undermine every voting reform because it isn't our voting reform (even though ours is objectively better). If we did that, we'd never get nothing done. If this passes, we should still try to advocate for approval voting in Maine. Hopefully, if it ever gets repealed we could try to get a head start and replace it with approval voting instead of plurality. Any chance we can try to get a truce with fairvote or misguided greens to support all initiatives calling for voting reform? It would have been nice if the Green Party supported the OR initiative.

I don't like that they say things that are misleading, but it is hard to sell voting reform with all the nuance – especially on twitter or on bullet points that you want people to read. I'm for top-two approval voting, but when I try to sell it, I say that you can always vote for your favorite without penalty. That isn't 100% true, but it is close enough. Your favorite could make it to the top two, but lose to someone who wasn't your second favorite while your second favorite would have won, but odds are that the winning candidate wouldn't be too objectionable. Also, if your favorite makes it into the top two, he'll at least be very likely to be viable.

I also try to sell approval voting as avoiding vote splitting. This can be true or false depending on what you mean by vote splitting. I never say that it avoids spoilers.

Btw, I hope this doesn't go out to everyone. It has come to my attention that my posts have been unintentionally emailed out to everyone. If this happens again, I apologize.

[Clay Shentrup]

On Monday, August 10, 2015 at 9:01:36 AM UTC-7, Warren D. Smith (CRV cofounder, http://RangeVoting.org) wrote:

Unfortunately if Maine adopts IRV then quite likely based on history and polls they later will repeal it and revert to plain plurality.

I think that would be an interesting and informative experiment. If that were to happen, maybe some of the more rational IRV proponents would finally be a little more receptive to simpler systems like Approval Voting, which would be less likely to backslide.

Drew? 

[Clay Shentrup]

On Monday, August 10, 2015 at 9:45:48 AM UTC-7, Rob Wilson wrote:

It is really frustrating that IRV is the only option being voted on, but I if I were in Maine, I'd support it.

I guess I'd have to say the same.

 

We can't continuously keep trying to undermine every voting reform because it isn't our voting reform (even though ours is objectively better).

I don't really try to "undermine" IRV. I just try to fact check the deluge of lies pumped out by its proponents. I'm even open to the idea that it could be, on the whole, a little better than TTR.

> Any chance we can try to get a truce with fairvote or misguided greens to support all initiatives calling for voting reform?

I don't even know what that means. You will NEVER get them to behave rationally. At least, not Rob Richie. He's the most dogmatic person I've ever encountered. No amount of evidence will ever ever ever change his mind. One very mild mannered person I know called him "ineducable".

I don't like that they say things that are misleading, but it is hard to sell voting reform with all the nuance

Yeah. But I think they could have made the same basic arguments along the lines of, "reduces the chance for spoilers" without being any less effective. That certainly would have reduced the number of op-eds I'd have written regarding their Duluth proposal for example. (Though I assume it will pass anyway.)

 

I'm for top-two approval voting, but when I try to sell it, I say that you can always vote for your favorite without penalty. That isn't 100% true, but it is close enough.

IMO, "close enough" means "telling the truth". If you made a claim like that, I'd write an op-ed to fact check you.

Btw, I hope this doesn't go out to everyone. It has come to my attention that my posts have been unintentionally emailed out to everyone. If this happens again, I apologize.

That only happens if you CC other email addresses.

[Warren D Smith]

--nah.
"Experiment" would take 10-30 years. By that time they'll all have forgotten,
and I may be dead.

Any argument made about Maine's repeal would be dismissed as irrelevant
or ancient or special or something by those who do not want to learn from it.

Score presently seems the least likely to backslide based on
http://rangevoting.org/WhatVotersWant.html

[Rob Wilson]

IMO, "close enough" means "telling the truth". If you made a claim like that, I'd write an op-ed to fact check you.

Would you say that the claim that approval voting stops vote splitting is true or false?  Technically it is true in that it allows voters to not split the vote, but voters may tend to avoid supporting lesser liked candidates to help their favorite.

[Drew Spencer]

I'm not going to speculate as to what might happen  some day if Maine first adopts and then years later decides to repeal RCV. Of course, I don't see it as very likely that they will repeal it. We've seen an escalation in places adopting it, and we've seen people in places with it becoming more comfortable with it, especially among younger voters.

More pertinently, once it is adopted, people will realize that there are different ways of conducting elections. In my opinion, the biggest hurdle to reform is the attitude among far too many people that the system used now is "just the way it is." A statewide implementation will get attention to the fact that things can change, and so I'd bet that it will be that much easier to get another place to seriously consider something like approval or score after.

Drew.

--
You received this message because you are subscribed to the Google Groups "The Center for Election Science" group.
To unsubscribe from this group and stop receiving emails from it, send an email to electionscien...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Andrew Spencer

[Jameson Quinn]

Personally I agree with the sentiment that it would be a step up for Maine and a step forward for the cause of voting reform, although obviously approval would be better on both counts. 

In terms of their language: 

It minimizes the “spoiler effect.” A candidate who receives fewer votes than others does not “spoil” the outcome of the election, because there is an instant run-off among the candidates who receive more votes.

I think that this is arguably correct, under certain (in my opinion slightly strained) definitions of the terms. I think we should acknowledge that FairVote is taking a step forward here by saying "minimizing", and not try to insist that they use our (correct) definition of "spoil".

It might be good for Electology (that is, the CES) to officially endorse this effort and/or write it up in our newsletter. I'd be willing to write such an article, mentioning IRV's flaws but focusing on the positive aspects here, if others agree with that.

Jameson

[Warren D Smith]

On 8/11/15, Drew Spencer <drew.s...@gmail.com> wrote:
> I'm not going to speculate as to what might happen some day if Maine first
> adopts and then years later decides to repeal RCV. Of course, I don't see
> it as very likely that they will repeal it. We've seen an escalation in
> places adopting it, and we've seen people in places with it becoming more
> comfortable with it, especially among younger voters.

--actually, I believe more places have repealed+rejected IRV than adopted it;
and there is a large history of places adopting then later repealing
ranked systems.

[Andy Jennings]

In terms of their language: 

It minimizes the “spoiler effect.” A candidate who receives fewer votes than others does not “spoil” the outcome of the election, because there is an instant run-off among the candidates who receive more votes.

I think that this is arguably correct, under certain (in my opinion slightly strained) definitions of the terms. I think we should acknowledge that FairVote is taking a step forward here by saying "minimizing", and not try to insist that they use our (correct) definition of "spoil".

Also, note the wording "a candidate who receives fewer votes than others does not spoil...".  It's true.  The candidates with the fewest first place votes don't spoil the election.  It's only when extremists get lots of first place votes that they can "spoil" the election by getting all the compromise candidates eliminated.  And as Jameson says, that's a controversial definition of "spoil" (though I also believe it correct).

I think they're really making an effort to be technically accurate here, and I do appreciate that.

[Clay Shentrup]

On Tuesday, August 11, 2015 at 4:52:33 AM UTC-7, Drew Spencer wrote:

In my opinion, the biggest hurdle to reform is the attitude among far too many people that the system used now is "just the way it is." A statewide implementation will get attention to the fact that things can change, and so I'd bet that it will be that much easier to get another place to seriously consider something like approval or score after.

I suspect you're probably right actually.

[Clay Shentrup]

On Tuesday, August 11, 2015 at 6:16:25 AM UTC-7, Jameson Quinn wrote:

I think we should acknowledge that FairVote is taking a step forward here by saying "minimizing", and not try to insist that they use our (correct) definition of "spoil".

Yes, a step forward. It at least seems to acknowledge that they're cognizant of the complex reality, as opposed to deliberately misleading people.

It might be good for Electology (that is, the CES) to officially endorse this effort and/or write it up in our newsletter. I'd be willing to write such an article, mentioning IRV's flaws but focusing on the positive aspects here, if others agree with that.

That actually seems reasonable.

[Clay Shentrup]

Someone on Twitter just responded with this link after I mentioned Score Voting.

http://archive.fairvote.org/rangevoting.pdf

Sigh. It's a reminder of what a terribly dishonest and hubristic person Rob Richie can be. I'll just try to remind myself that Rob Richie's personality traits have nothing to do with the merit of this initiative. But still, I think FairVote would really benefit from a more likable/humble/honest leader.

[Jan Kok]

Would RCV have changed the 2010 or 2014 governors election results? I think that is a key question that will be of interest to Mainers considering whether to adopt RCV.

My thoughts:

1. According to the Bayesian Regret metrics, IRV (and hopefully this version of RCV) is better than plain Plurality voting with less-than-100% strategic voters, thus, RCV seems like an improvement over Plurality.

2. But, it's better in a probabilistic sense. There may be some elections where the results are clearly non-optimal (such as Burlington, VT 2009), or even worse than what Plurality would deliver. Mainers need to be informed about those possible bad results, in order to mitigate the fallout if/when they occur.

3. The rcvmaine.com website looks like more of the same-old same-old from Fairvote. They oversell RCV and fail to warn about the possible bad results, which could lead to a loss of credibility for the entire voting reform movement if those bad results occur.

I'd suggest that CES prepare some web pages that parallel the rcvmaine.com site, give OUR reasons for supporting RCV, but also addresses the rcvmaine site point by point and gives more accurate and complete info about each point.

Sort of like the patient information sheet you get when you buy prescription drugs. "May cause vomiting, cancer, strokes, or heart attacks. But at least you won't be depressed." :-)

[Warren D Smith]

How's LePage been doing as governor for Maine?

http://rangevoting.org/LePageDumb.html

[Jan Kok]

On Thu, Aug 13, 2015 at 12:45 PM, Jan Kok <jan.k...@gmail.com> wrote:

2. But, it's better in a probabilistic sense. There may be some elections where the results are clearly non-optimal (such as Burlington, VT 2009), or even worse than what Plurality would deliver.

Maybe I was not thinking clearly. Is there any plausible situation where RCV could produce worse results than Plurality?

[Jameson Quinn]

Using the same ballot set: no.

But if plurality led to reasonably good strategy*, while IRV led to naive honesty, then plurality could be better.

*By "reasonably good", I mean that the definition of "frontrunners" has not been hijacked by minorities such as the rich, the media, or self-defined intellectual elites.

I doubt it would happen very often; and I still think we should support IRV in this case. But it is not totally unrealistic.

[Rob Wilson]

Could we please not cede the term Rank Choice Voting to IRV advocates? It is too ambiguous and it isn't fair to other rank choice voting advocates. If Maine were advocating for a Condorcet method, I would be less conflicted in supporting the initiative.

 

[Brian Goldman]

Is there any plausible situation where RCV could produce worse results than Plurality?

With 4 candidates its possible for plurality to elect the Condorcet winner while IRV does not using the same ballots. However, I can't come up with a very plausible election without using 5 candidates. Consider this election:

7 Center > Left > Right > Far Left > Far Right (The true centrist)

6 Left > Center > Right > Far Left > Far Right (Left-of-center)

6 Right > Center > Left > Far Right > Far Left (Right-of-center)

2 Far Left > Left > Center > Right > Far Right (Far Left)

3 Far Right > Right > Center > Left > Far Left (Far Right)

The plurality results are Center=7, Left=6, Right=6, Far Left=2, Far Right=3, meaning Center wins. Center is also the Condorcet winner as Center>Left on 7+6+2 vs 6+3, Center > Right on 7+6+2 vs 6+3, and the "Far" groups are only better than Center on 2 and 3 ballots.

However, in this election Center is NOT the IRV winner. Far Left has the fewest first round top votes and is eliminated, giving us:

7 Center > Left > Right > Far Right

8 Left > Center > Right > Far Right

6 Right > Center > Left > Far Right

3 Far Right > Right > Center > Left

Far Right has the least top votes in round 2 and is eliminated, giving us:

7 Center > Left > Right

8 Left > Center > Right

9 Right > Center > Left

At this point Center is squeezed out in round 3, giving us:

15 Left > Right

9 Right > Left

So IRV elects Left, even though 2/3rds of voters preferred Center>Left AND the plurality winner was Center.

--

[Clay Shentrup]

On Friday, August 14, 2015 at 2:10:24 PM UTC-7, Rob Wilson wrote:

Could we please not cede the term Rank Choice Voting to IRV advocates?

Amen.

[Jan Kok]

Thanks, Brian.

So now can anyone estimate the frequency that such elections occur, where the IRV results are worse than would be obtained with Plurality using the same ballot set?

[Clay Shentrup]

Brian,

I don't know if you saw these.

http://scorevoting.net/ClayIrv.html

http://scorevoting.net/ClayIrv2.html

http://scorevoting.net/ClayIrv3.html

[Brian Goldman]

Clay, I hadn't seen those. Neat. However, I'm not sure if any look particularly plausible.

In thinking about it, I determined any election matching the following rules always result in Plurality and Condorcet agreeing, but IRV disagreeing, about who should win. Elections which do not meet these rules may still have this outcome, but these are sufficient to make it happen.

1. All voters and the 5 candidates exist on a one-dimensional spectrum.
2. "Center" receives the most top votes.
3. "Far Left" and "Far Right" receive less top votes than any other candidate.
4. When looking at top votes, "Far Left" + "Left" > "Center" and "Far Right" + "Right" > "Center".

Point 2 means "Center" is the plurality winner. Points 1+3 mean the "Far" candidates must be eliminated in rounds 1 and 2. Points 1+4 means "Center" has to lose in round 3, and is therefore not the IRV winner.

To show "Center" is always Condorcet, assume instead that some candidate on the left is the winner. Due to the one-dimensional spectrum, only ballots from "Far Left" and "Left" supporters can have any left candidate higher than "Center", meaning "Far Left" + "Left" > "Center" + "Right" + "Far Right" if a left candidate is the Condorcet winner.

From Point 2 we know that "Far Left" < "Center" and "Left" < "Center" which means "Far Left" + "Left" < 2 * "Center". From Point 4 we know that "Right" + "Far Right" > "Center" which means "Center" + "Right" + "Far Right" > 2 * "Center". Combining those two facts we get "Far Left" + "Left" < 2 * "Center" < "Center" + "Right" + "Far Right", contradicting the assumption required for any left candidate to be the Condorcet winner. This reasoning is symmetric for any candidate on the right, meaning no candidate other than "Center" can be the Condorcet winner.

These conditions always induce favorite betrayal situations. Consider that if "Left" wins, the "Right" voters could improve their happiness by switching to "Center", with vice versa also true.

To me these are relatively plausible conditions for real world elections with 5+ candidates.

--

[Warren D Smith]

> Brian Goldman:

> In thinking about it, I determined any election matching the following
> rules always result in Plurality and Condorcet agreeing, but IRV
> disagreeing, about who should win. Elections which do not meet these rules
> may still have this outcome, but these are sufficient to make it happen.
>

> 1. All voters and the 5 candidates exist on a one-dimensional spectrum.
> 2. "Center" receives the most top votes.
> 3. "Far Left" and "Far Right" receive less top votes than any other
> candidate.
> 4. When looking at top votes, "Far Left" + "Left" > "Center" and "Far

> Right" + "Right" > "Center".

--WDS:
These conditions (which are linear inequalities) define a convex polytope
in 4-dimensional space. (There are 5 dimensions giving the top-rank
counts for each of the 5 candidates, but since they sum to 100% that
restricts to a 4-dimensiona hyperplane.) We can write them as
C>F, C>G, C>L, C>R, L>F, R>F, L>G, R>G, F+L>C, G+R>C
among 5 variables C>0,L>0,R>0,F>0,G>0 with C+L+R+F+G=100.

Here I'm taking the "1D spectrum" aspect for granted and not deriving any
inequalities from it.

Now one thing we can do is to solve the linear program
MAXIMIZE s SUBJECT TO
{ C>F+s, C>G+s, C>L+s, C>R+s, L>F+s, R>F+s, L>G+s, R>G+s,
F+L>C+s, G+R>C+s, C>s, L>s, R>s, F>s, G>s, C+L+R+F+G=TotalVoters }
to find the "most extreme" or "clearest" election of Goldman's type;
I tried to do that by hand (there actually are linear programming
hacks online now, though, e.g.
http://www.phpsimplex.com/simplex/simplex.htm?l=en
if you want to be surer), which if my hands succeeded gave the solution
C=4, L=3, R=3, F=2, G=2, TotalVoters=14


Another thing we can do is compute the 4-volume of our polytope as a
fraction of the
entire hyperplane. That volume-fraction will arguably equal the
"probability" that
this scenario will arise in 1D-spectrum-politics 5-candidate elections.

[Warren D Smith]

> --WDS:
> These conditions (which are linear inequalities) define a convex polytope
> in 4-dimensional space. (There are 5 dimensions giving the top-rank
> counts for each of the 5 candidates, but since they sum to 100% that
> restricts to a 4-dimensiona hyperplane.) We can write them as
> C>F, C>G, C>L, C>R, L>F, R>F, L>G, R>G, F+L>C, G+R>C
> among 5 variables C>0,L>0,R>0,F>0,G>0 with C+L+R+F+G=100.
>
> Here I'm taking the "1D spectrum" aspect for granted and not deriving any

> inequalities from it...

>
> Another thing we can do is compute the 4-volume of our polytope as a
> fraction of the
> entire hyperplane. That volume-fraction will arguably equal the
> "probability" that
> this scenario will arise in 1D-spectrum-politics 5-candidate elections.

Of the 120=5! possible orderings of the counts C,L,R,F,G,
exactly 4 orderings (from most to least) CLRFG, CRLFG, CLRGF, CRLGF
satisfy the inequalities

C>F, C>G, C>L, C>R, L>F, R>F, L>G, R>G

among 5 variables C>0,L>0,R>0,F>0,G>0.
So the chance of that is 1/30 if all 120 equally likely.
Now among Goldman's 2 further inequalities
F+L>C, G+R>C
obviously if x,y are random uniform in [0,z]
the chance their sum exceeds z is exactly 1/2 by
reflection symmetry. Which, putting it all together, suggests
the chance all Goldman's necessary conditions will be satisfied simultaneously
is about 1/120.

In other words, in a 1D political spectrum 5-candidate election, the
chance plurality
will yield a better winner than IRV, will be at least about 1/120.

One could assess the chance more precisely and reliably by computer simulation.
I have not tried.

Meanwhile consulting
http://www.rangevoting.org/IrvParadoxProbabilities.html
we find the chance IRV and plurality winners differ in 3-candidate elections

T: Plurality and IRV winners differ 24.4661% 12.3263% 25.0000%

in its three probability models, the last of which was a 1D politics model.

[Brian Goldman]

It occurs to me there are at least 2 more permutations that will always result in this problem: CLFRG and CRGLF. This because my Point 3 can be amended to: "Left" > "Far Left" and "Right" > "Far Right". This point still ensures that either "Far Left" or "Far Right" must be eliminated in round 1. For simplicity, say "Far Left" is the lowest. After "Far Left" is removed, "Left" > "Center" due to Point 1+4. From Point 2, "Center" > "Right", and from Point 3 "Right" > "Far Right". Therefore "Far Right" loses round 2. As always, the reverse is true.

[Jan Kok]

Does Brian's last email combined with Warren's analysis mean that the probability of having this problem is 4x greater, i.e. 1/30?

The reason why I asked about probability that IRV gives worse results than Plurality, is that that is the kind of election result that is most likely to result in repealing IRV and returning to Plurality.

When Plurality would choose the Condorcet winner but IRV chooses someone else, that makes it easy to attack IRV.

But apparently ANY difference between Plurality and IRV results can lead to repealing IRV. Witness Burlington VT 2009. IRV failed to choose the Condorcet winner, but chose a better winner than Plurality would have, based on the IRV ballots and assuming honest voting. Yet Burlington voters promptly repealed IRV and returned to their previous voting method (which was Plurality with a runoff if no one gets at least 40% of the votes).

On Sun, Aug 16, 2015 at 8:11 PM, Brian Goldman <gold...@msu.edu> wrote:

It occurs to me there are at least 2 more permutations that will always result in this problem: CLFRG and CRGLF. This because my Point 3 can be amended to: "Left" > "Far Left" and "Right" > "Far Right". This point still ensures that either "Far Left" or "Far Right" must be eliminated in round 1. For simplicity, say "Far Left" is the lowest. After "Far Left" is removed, "Left" > "Center" due to Point 1+4. From Point 2, "Center" > "Right", and from Point 3 "Right" > "Far Right". Therefore "Far Right" loses round 2. As always, the reverse is true.

--

[Warren D Smith]

On 8/18/15, Jan Kok <jan.k...@gmail.com> wrote:
> Does Brian's last email combined with Warren's analysis mean that the
> probability of having this problem is 4x greater, i.e. 1/30?

--I would think 1.5x greater, i.e. 1/80, since he added 2 more perms beyond
the 4 I already mentioned.

I think Brian's latest is making it be a nonconvex polytope, though,
(union of 2 convex ones)
which if so is kind of annoying for some purposes.
If you really want to know the chance in some probability model, then
run a computer sim of the IRV and plurality processes in that model,
counting how many elections satisfy whatever criterion, to easily obtain
accuracy of +-0.01%.