How will Instant Runoff Voting in Maine effect voting reform?

[Blahface]

OK, so the instant runoff voting initiative passed in Maine and it will be implemented statewide. I think IRV is much better than FPTP, but I do not think it is a very good system. I am worried that it may leave a bad taste in the mouths of the voters and it may lead to a repeal and subsequently poison the well for further voting reform.

The first problem I have is that the results aren't intuitive and it can give weird results. It has been argued that you can vote for your favorite without worry, but this isn't really true because order of elimination matters. I'm afraid that people won't understand the results and blame the voting system. This happened in the Burlington, Vermont mayoral race in which the Democrat would have beaten the Progressive head to head, but ending up winning because Republican voters put the Republican as their first choice instead of strategically voting for their second choice - the Democrat . The voters voted to repeal it after the election.

Another problem I have with it is that I think voters may feel that they have to do more homework on candidates and just choose not to vote. There is also a problem that a lot of people just don't understand how it works.

I am crossing my fingers that this works though. I hope other states will go forward with better and simpler voting systems like approval voting. I am really scared that there will be voting reform fatigue and opponents will argue "you said IRV was going to fix the problem and it didn't. Why waste money on another failed failed system?"

 

[Jayjay]

OK, so the instant runoff voting initiative passed in Maine and it will be implemented statewide. I think IRV is much better than FPTP, but I do not think it is a very good system. I am worried that it may leave a bad taste in the mouths of the voters and it may lead to a repeal and subsequently poison the well for further voting reform.

The first problem I have is that the results aren't intuitive and it can give weird results. It has been argued that you can vote for your favorite without worry, but this isn't really true because order of elimination matters. I'm afraid that people won't understand the results and blame the voting system. This happened in the Burlington, Vermont mayoral race in which the Democrat would have beaten the Progressive head to head, but ending up winning because Republican voters put the Republican as their first choice instead of strategically voting for their second choice - the Democrat . The voters voted to repeal it after the election.

Isn't that kind of strategic voting exactly what IRV is supposed to get rid of? Or do you mean that the Republicans omitted the second choice altogether because they didn't understand the ballot?

EDIT: Nevermind, checked the Wikipedia article. But still, to me it sounds like the system worked as it should.

 

[zorq]

I think that Burlington election played out just fine. There were clearly more "liberal" voters and among the liberal voters, more of them were "progressive." The fact that the Republicans would have preferred the more moderate Democrat to win than the more extreme Progressive is inconsequential. Why should BROAD appeal in the voting base be more important than DEEP appeal or vice versa? To win in IRV you need both broad AND deep appeal. This is a feature, not a flaw.

 

[Blahface]

I think that Burlington election played out just fine. There were clearly more "liberal" voters and among the liberal voters, more of them were "progressive." The fact that the Republicans would have preferred the more moderate Democrat to win than the more extreme Progressive is inconsequential. Why should BROAD appeal in the voting base be more important than DEEP appeal or vice versa? To win in IRV you need both broad AND deep appeal. This is a feature, not a flaw.

Would you be happy if it were the other way around and instead of a Progressive getting elected, it was someone from the Constitution Party?

 

[zorq]

I think that Burlington election played out just fine. There were clearly more "liberal" voters and among the liberal voters, more of them were "progressive." The fact that the Republicans would have preferred the more moderate Democrat to win than the more extreme Progressive is inconsequential. Why should BROAD appeal in the voting base be more important than DEEP appeal or vice versa? To win in IRV you need both broad AND deep appeal. This is a feature, not a flaw.

Would you be happy if it were the other way around and instead of a Progressive getting elected, it was someone from the Constitution Party?

We're never happy when our preferred candidate doesn't win. If the Constituition Party wins the Election, they probably deserved to win it in that district. They needed lots of support to get there. Any liberal voters in the Constituition District weren't going to get anything they wanted anyway.

Why should the most moderate candidate necessarily get a bonus? If your electoral system only allows for moderate candidates to win, then voting for 3rd party candidates will STILL be just as superfluous as it is in FPTP voting.

 

[Valjean]

IRV would be a huge improvement, IMHO. Vote for the candidate you actually support without worrying about wasting your vote or supporting some miscreant blackguard. Without strategic voting the vote would actually reveal public sentiment.
Hopefully this will spread to other states.

Now if we could only institute some system of proportional representation instead of winner-take-all, this might start looking like a democracy.

 

[Blahface]

Would you be happy if it were the other way around and instead of a Progressive getting elected, it was someone from the Constitution Party?

We're never happy when our preferred candidate doesn't win. If the Constituition Party wins the Election, they probably deserved to win it in that district. They needed lots of support to get there. Any liberal voters in the Constituition District weren't going to get anything they wanted anyway.

Why should the most moderate candidate necessarily get a bonus? If your electoral system only allows for moderate candidates to win, then voting for 3rd party candidates will STILL be just as superfluous as it is in FPTP voting.

If the Constitution Party candidate couldn't beat the Republican head to head, then he shouldn't win the election. I don't want to just elect moderates - I want to elect the best person who accurately represents the district. The Condorcet winner should win as much as possible if there is one.

Also, IRV was sold as being safe to vote for your favorite. I'm worried about the backlash when voters find out that this isn't true. A lot of people voted for Hillary in the primary because they thought she was more electable. That is the same strategic voting that we could see with IRV.

 

[zorq]

We're never happy when our preferred candidate doesn't win. If the Constituition Party wins the Election, they probably deserved to win it in that district. They needed lots of support to get there. Any liberal voters in the Constituition District weren't going to get anything they wanted anyway.

Why should the most moderate candidate necessarily get a bonus? If your electoral system only allows for moderate candidates to win, then voting for 3rd party candidates will STILL be just as superfluous as it is in FPTP voting.

If the Constitution Party candidate couldn't beat the Republican head to head, then he shouldn't win the election.

That's FPTP thinking. Have you considered it the other way? That if the moderate candidate is the least popular one in the district among three candidates then they shouldn't win the election?

I don't want to just elect moderates - I want to elect the best person who accurately represents the district. The Condorcet winner should win as much as possible if there is one.

Condorcet values broad support much more than deep support and therefore gives a bonus to the moderate candidates.

Also, IRV was sold as being safe to vote for your favorite. I'm worried about the backlash when voters find out that this isn't true. A lot of people voted for Hillary in the primary because they thought she was more electable. That is the same strategic voting that we could see with IRV.

IRV was safe for republicans to vote for their favorite. In Burlington, they weren't going to win in IRV, OR Condorcet. I'd argue that they weren't going to win in FPTP Either, as a big chunk of Progressive voters would have followed FPTP wisdom and voted strategically for the Democrat. In no situation was the Republican candidate winning that seat. The Republicans are just pissed that the Progressive got in instead of the Democrat that they would have preferred. But I say Boo Hoo to that. There was more deep support for the Progressive than the Democrat even if the Democrat had more broad support. So we are heare again, why should we value broad support more than deep support?

 

[Blahface]

That's FPTP thinking. Have you considered it the other way? That if the moderate candidate is the least popular one in the district among three candidates then they shouldn't win the election?

I think you are the one who is thinking in terms of FPTP. If you want to measure popularity, you need to use a cardinal system like approval voting or score voting. You don't have enough information from just first place picks to say who is more popular.

IRV was safe for republicans to vote for their favorite. In Burlington, they weren't going to win in IRV, OR Condorcet. I'd argue that they weren't going to win in FPTP Either, as a big chunk of Progressive voters would have followed FPTP wisdom and voted strategically for the Democrat. In no situation was the Republican candidate winning that seat. The Republicans are just pissed that the Progressive got in instead of the Democrat that they would have preferred. But I say Boo Hoo to that. There was more deep support for the Progressive than the Democrat even if the Democrat had more broad support. So we are heare again, why should we value broad support more than deep support?

It was not safe for Republicans to vote for their favorite. They got a worse outcome by voting for their favorite. The argument that you are making can be used against IRV as well. Imagine there is a FPTP race with a Democrat, a Republican, and a Green. You hate the Republican, but you vote for the Green because you like him best. The Republican ends up winning because the Green siphoned off votes from the Democrat. The Green would have never won in IRV so under your argument it was safe to vote for the Green. In this scenario, the Republican has the most “deep support”.

Even under IRV, you can always add hypothetical candidates that can act as a spoiler. In the Burlington example, you can add in a candidate slightly more left wing than the Progressive. If you do that, you can eliminate the Progressive and then the hypothetical candidate and have the Democrat advancing to the next round. The only thing that changed was adding a new candidate to change the order of elimination.

 

[zorq]

I think you are the one who is thinking in terms of FPTP. If you want to measure popularity, you need to use a cardinal system like approval voting or score voting. You don't have enough information from just first place picks to say who is more popular.

I think the ordinal voting in IRV gives a good enough indication. And again, it depends on whether you value deep support.

IRV was safe for republicans to vote for their favorite. In Burlington, they weren't going to win in IRV, OR Condorcet. I'd argue that they weren't going to win in FPTP Either, as a big chunk of Progressive voters would have followed FPTP wisdom and voted strategically for the Democrat. In no situation was the Republican candidate winning that seat. The Republicans are just pissed that the Progressive got in instead of the Democrat that they would have preferred. But I say Boo Hoo to that. There was more deep support for the Progressive than the Democrat even if the Democrat had more broad support. So we are heare again, why should we value broad support more than deep support?

It was not safe for Republicans to vote for their favorite. They got a worse outcome by voting for their favorite.

(Nore: I am changing my answer a little here and acknowledging that your point here is right)
The race was close enough that Republicans couldn't have known they should vote strategically. In hind sight, Republicans had no chance of their favorite candidate winning regardless of the voting system but in the voting booth they didn't have enough knowledge to make a strategic vote that would affect their first choice pick in IRV. Besides, as far as they could tell, their 2nd place pick, the Dems, were just as likely to advance to the final round as their hated rivals, the Progressives. Strategically voting for the Dems to block the Progressives might have been completely unnecessary and they would have completely sacrificed their vote for their favorite candidate to benefit their second favorite candidare but only in an unknown future scenario that might only have a 50% chance of happening.

My point is that strategic voting in IRV is actually a risky endeavor. Risky enough, that it most people will never bother with it.

The argument that you are making can be used against IRV as well. Imagine there is a FPTP race with a Democrat, a Republican, and a Green. You hate the Republican, but you vote for the Green because you like him best. The Republican ends up winning because the Green siphoned off votes from the Democrat. The Green would have never won in IRV so under your argument it was safe to vote for the Green. In this scenario, the Republican has the most “deep support”.

Oh yes, I understand that fptp is a measure of ONLY deep support and I understand why people vote strategically.

Even under IRV, you can always add hypothetical candidates that can act as a spoiler. In the Burlington example, you can add in a candidate slightly more left wing than the Progressive. If you do that, you can eliminate the Progressive and then the hypothetical candidate and have the Democrat advancing to the next round. The only thing that changed was adding a new candidate to change the order of elimination.

I don't think I understand what you are saying here. I keep referring to Burlington because it is a perfect example of IRV working at its best AND because it is a concrete example we can use to examine the shift in votes after each round. The hypothetical you just suggested Is too vague for me. You can't change the elimination order without drastically changing the vote distribuition in every round.

If you are supposing that all of the Progressive votes had been split evenly between two parties in the first round, "Progressive extreme" and "Progressive mild," then yes, they might have both been eliminated (especially if "Progressive Mild" was eliminated first.) The Democrat might win that election because neither Progressive party would have enough deep or broad support on their own. I don't think this result confounds IRV at all.

 

[Blahface]

Let me a series of elections in an oversimplified scenario to explain what I'm talking about. Imagine there is a single political spectrum from 0 to 100. There is a perfectly uniform voter distribution among this spectrum and everyone votes to candidates closest to them. In any election in this scenario there are potential candidates C0,C1, C2,C3..., C100 and their number indicates where they are on the political spectrum.

Now, let me attempt to use the forum tables to show a bunch of different election scenarios and how the spoilers affect the election under IRV.

Imagine a race between C26 and C40. The results are as follows:

Candidate	RD 1
C26	33%
C40	67%

C40 is the winner by a landslide.

In the next scenario candidate C77 is in the race:

Candidate	RD 1	RD 2
C26	33%	51%
C40	25.5%	0
C77	41.5%	49%

C26 is now the winner just because a spoiler candidate entered the race. Let's see what happens when C16 enters the race.

Candidate	RD 1	RD 2	RD 3
C16	21%	28%	0%
C26	12%	0%	0%
C40	25.5%	30.5%	58.5%
C77	41.5%	41.5%	41.5%

After C16 entered the race, C40 suddenly becomes the winner again. Now, lets add C46 to the race

Candidate	RD 1	RD 2	RD 3	RD 4
C16	21%	21%	31%	46.5%
C26	12%	15%	0%	0%
C40	10%	0%	0%	0%
C46	18.5%	25.5%	30.5%	0%
C77	38.5%	38.5%	38.5%	53.5%

Now, C77 wins. This goes to show how order of elimination matters and how chaotic IRV can be. You can't necessarily measure popularity by where a candidate is in a single round.

That being said, I am still glad Maine passed IRV.

 

[lpetrich]

Although Burlington, VT backed away from IRV, some other US cities continue to use it. The state with the most users is California, with San Francisco, Oakland, Berkeley, and San Leandro using it. Other cities using it are Minneapolis and St. Paul, MN, and Portland, ME.

However, California Governor recently Jerry Brown vetoed Senate Bill 1288, a bill that would have enabled IRV for many cities, counties, and school districts in that state. "Ranked choice voting is overly complicated and confusing," he stated in his veto message (Gov. Jerry Brown makes it crystal clear: He really doesn't like ranked-choice voting - LA Times).

But is IRV much worse than the Electoral College? Or what happens if the EC is tied?

 

[lpetrich]

Let me a series of elections in an oversimplified scenario to explain what I'm talking about. Imagine there is a single political spectrum from 0 to 100. There is a perfectly uniform voter distribution among this spectrum and everyone votes to candidates closest to them. In any election in this scenario there are potential candidates C0,C1, C2,C3..., C100 and their number indicates where they are on the political spectrum.

Blahface, I'd implemented some voting-method software some time ago. It works with preference voting, counting up the ballots in a variety of ways. Here are my results. If a line is a simple list, then the candidates in order from winner to loser. If a list of pairs, then the pair is of (candidate, number of votes or other score). Sequential Runoff is IRV. Schulze is a Condorcet beatpath method, a rather complicated algorithm for disambiguating circular preferences.

Candidates: (26, 40)
Top One
((40, 67), (26, 34))
Top-Two Runoff
((40, 67), (26, 34))
((40, 67), (26, 34))
Sequential Runoff
((40, 67), (26, 34))
((40, 101),)
Schulze
(40, 26)
Borda
((40, 168), (26, 135))

Candidates: (26, 40, 77)
Top One
((77, 42), (26, 34), (40, 25))
Top-Two Runoff
((77, 42), (26, 34), (40, 25))
((26, 52), (77, 49))
Sequential Runoff
((77, 42), (26, 34), (40, 25))
((26, 52), (77, 49))
((26, 101),)
Schulze
(40, 26, 77)
Borda
((40, 227), (77, 192), (26, 187))

Candidates: (16, 26, 40, 77)
Top One
((77, 42), (40, 25), (16, 22), (26, 12))
Top-Two Runoff
((77, 42), (40, 25), (16, 22), (26, 12))
((40, 59), (77, 42))
Sequential Runoff
((77, 42), (40, 25), (16, 22), (26, 12))
((77, 42), (40, 30), (16, 29))
((40, 59), (77, 42))
((40, 101),)
Schulze
(40, 26, 77, 16)
Borda
((40, 299), (26, 266), (77, 246), (16, 199))

Candidates: (16, 26, 40, 46, 77)
Top One
((77, 39), (16, 22), (46, 18), (26, 12), (40, 10))
Top-Two Runoff
((77, 39), (16, 22), (46, 18), (26, 12), (40, 10))
((77, 54), (16, 47))
Sequential Runoff
((77, 39), (16, 22), (46, 18), (26, 12), (40, 10))
((77, 39), (46, 25), (16, 22), (26, 15))
((77, 39), (16, 32), (46, 30))
((77, 54), (16, 47))
((77, 101),)
Schulze
(46, 40, 26, 77, 16)
Borda
((46, 353), (40, 343), (26, 303), (77, 285), (16, 231))

For two candidates, all the methods agree. But they disagree for all the others.

I can send my code to anyone who is interested -- or even try to post it as an attachment.

 

[lpetrich]

California and Washington State now use a "top two" system. This is effectively a delayed top-two runoff system, where the primaries are the first round and the main election is the second round. Let's look at a recent election:
 United States Senate election in California, 2016, for the seat the Barbara Boxer is retiring from.

Numerous candidates from several parties entered the race, and the Wikipedia article lists 11 candidates who filed financial reports with the Federal Election Commission: 4 Democrats, 5 Republicans, and 2 others. The biggest spenders were Kamala Harris and Loretta Sanchez, both outspending all the Republicans combined. KH and LS are both Democrats, KH is California's Attorney General, and LS as US Representative.

In the primary, KH got 3 million votes, LS got 1.4 million votes, and all the Republicans combined got 2 million votes. The top Republican was Duf Sundheim at 0.6 million votes. A Libertarian got 0.1 million votes.

So KH and LS went on to the general election, and KH won that one also.

 

[Blahface]

Although Burlington, VT backed away from IRV, some other US cities continue to use it. The state with the most users is California, with San Francisco, Oakland, Berkeley, and San Leandro using it. Other cities using it are Minneapolis and St. Paul, MN, and Portland, ME.

However, California Governor recently Jerry Brown vetoed Senate Bill 1288, a bill that would have enabled IRV for many cities, counties, and school districts in that state. "Ranked choice voting is overly complicated and confusing," he stated in his veto message (Gov. Jerry Brown makes it crystal clear: He really doesn't like ranked-choice voting - LA Times).

But is IRV much worse than the Electoral College? Or what happens if the EC is tied?

If he thinks IRV is too complicated, then he should be advocating for a better system. Top two primaries with plurality is pretty bad. If it were top two with approval on the other hand, I think that is probably the best solution. It would be much easier to organize voting blocs and pressure politicians to pay attention to important issues.

I plan on making some voting emulation software in the future as well. What I'd like to do is be able to customize the demographics of a district and give different voting properties and behaviors to different groups and see how voting systems compare. Unfortunately I am trying to do something else first and I am massively procrastinating on that.

 

[lpetrich]

Here is an interesting application of IRV: Ranked Choice Poll of GOP Voters Yields Insights - FairVote with graphs of how many people gave what preference rank to which candidate, and also how the IRV count progressed. It might have been interesting to find the poll's Condorcet matrix and maybe also to do a Borda count of it.

The fractions at each preference rank were interesting. Donald Trump had a rather exceptional pattern: from a high of 31% first preferences to 11% second preferences and 6% third preferences to a decline to 2% to 3% to 4% second to last to 22% last. It is evident that Republicans either love him or hate him. The others had much smoother variation. Ted Cruz, Marco Rubio, and Ben Carson were highest at high rankings, while John Kasich was highest at low rankings, and Jeb Bush was rather flat.

In the IRV count, the dropping out was in this order: Rick Santorum, Mike Huckabee, Carly Fiorina, Chris Christie, John Kasich, Rand Paul, Jeb Bush, Ben Carson, Marco Rubio, Donald Trump. Ted Cruz barely won, with 50.68% in the final round.

 

[lpetrich]

I plan on making some voting emulation software in the future as well. What I'd like to do is be able to customize the demographics of a district and give different voting properties and behaviors to different groups and see how voting systems compare.

Seems interesting. You could try to do some stripped-down version of it and gradually add features to it.

I don't know if my software would be helpful. I've implemented preference-voting and proportional-representation algorithms, and I've covered the full spectrum of them. There are a few that I haven't implemented, like  Dodgson's method, because I couldn't think of any simple way of implementing them. I also just found out about the  Maximal lotteries method.

Unfortunately I am trying to do something else first and I am massively procrastinating on that.

You might try setting it aside for a while and then returning to it. Or else thinking of some subset of it that you can implement more quickly.

 

[lpetrich]

Some terminology for various sorts of runoff elections.

The voting can either be in separate rounds or a single overall round with a preference ballot: which candidate first, which one second, etc. If the latter, then instead of rounds of voting, it is rounds of ballot counting.

The rounds of voting or counting come in two main types.

• Top-two: The top two candidates from the first round go to the second round.
• Sequential: All but the bottommost candidate of each round go to the next round.

So one gets a 2*2 table of possibilities.

• Separate Rounds:
  • Top-Two: Two-Round System
  • Sequential: Exhaustive Vote
• Preference Ballot:
  • Top-Two: Contingent Vote
  • Sequential: Instant Runoff Voting, Alternative Vote

Preference voting is sometimes called ranked-choice voting.

 

[lpetrich]

I checked on the  Maximal lotteries method, and I couldn't find any simple way to do it.

I also checked on  Dodgson's method, and I think I have a way to do it.

For N candidates, generate all permutations P of N symbols. For each P, apply P to each ballot and then look for a Condorcet winner W. Then for each P with a winner W, look for the P with the smallest number of interchanges to generate it. Its winner is then the overall winner. One can emit each candidate's minimum number of interchanges to get a ranking of them.

For a m-cycle, the minimum number is (m-1), making a permutation's minimum number (number of symbols) - (number of cycles). This means that if one cannot find a Condorcet winner for some candidate, we can give it N interchanges.

Its inventor, mathematician and logician Charles Dodgson, is better known as Lewis Carroll, author of Alice's Adventures in Wonderland and Through the Looking-Glass, among others.

 

[lpetrich]

(CA Gov Jerry Brown dismissing IRV as "overly complicated and confusing"...)

If he thinks IRV is too complicated, then he should be advocating for a better system. Top two primaries with plurality is pretty bad. If it were top two with approval on the other hand, I think that is probably the best solution. It would be much easier to organize voting blocs and pressure politicians to pay attention to important issues.

 Voting system and  Single-member district have some big tables of comparisons of single-winner methods. One of the more broadly relevant criteria is the  Independence of clones criterion: what happens when there is more than one candidate with similar voter appeal. There are three kinds of effects:

• Spoiling: they hurt each other, typically by splitting their votes.
• Teaming: they help each other.
• Crowding: they affect other candidates.

What's vulnerable to what:

• (none): Approval, Sequential Runoff, Schulze, Ranked Pairs
• Spoiling: FPTP, Top-Two Runoff, Kemeny-Young, Minimax
• Teaming: Borda, Copeland
• Crowding: Copeland

Now for the computational complexity of aggregating and counting the votes. Aggregating votes is a convenience for elections over wide areas.

For aggregating votes, we must turn them into some form that can be summed. The size of the summable data structure varies between methods, and for N candidates, one finds these size factors for it:

• O(N): FPTP, Approval, Separate-Round Runoffs
• O(N^2): Condorcet methods: Schulze, Ranked Pairs, Copeland, Minimax, Kemeny-Young
• O(N!): Preference-Ballot Runoffs (Contingent, IRV)

For the latter one, if one has only P preferences in the ballot, then the number becomes N!/(N-P)!.

Now for counting votes, with the same notation.

• O(N): FPTP, Approval, Top-Two Runoffs
• O(N^2): Sequential Runoffs, Copeland, Minimax
• O(N^3): Schulze
• O(N^4): Ranked Pairs
• O(N!): Kemeny-Young